diff --git a/.buildkite/format_tap.py b/.buildkite/format_tap.py deleted file mode 100644 index b557a9c38e..0000000000 --- a/.buildkite/format_tap.py +++ /dev/null @@ -1,48 +0,0 @@ -# -*- coding: utf-8 -*- -# Copyright 2019 The Matrix.org Foundation C.I.C. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import sys -from tap.parser import Parser -from tap.line import Result, Unknown, Diagnostic - -out = ["### TAP Output for " + sys.argv[2]] - -p = Parser() - -in_error = False - -for line in p.parse_file(sys.argv[1]): - if isinstance(line, Result): - if in_error: - out.append("") - out.append("") - out.append("") - out.append("----") - out.append("") - in_error = False - - if not line.ok and not line.todo: - in_error = True - - out.append("FAILURE Test #%d: ``%s``" % (line.number, line.description)) - out.append("") - out.append("
Show log
")
-
-    elif isinstance(line, Diagnostic) and in_error:
-        out.append(line.text)
-
-if out:
-    for line in out[:-3]:
-        print(line)
diff --git a/.buildkite/postgres-config.yaml b/.buildkite/postgres-config.yaml
new file mode 100644
index 0000000000..a35fec394d
--- /dev/null
+++ b/.buildkite/postgres-config.yaml
@@ -0,0 +1,21 @@
+# Configuration file used for testing the 'synapse_port_db' script.
+# Tells the script to connect to the postgresql database that will be available in the
+# CI's Docker setup at the point where this file is considered.
+server_name: "test"
+
+signing_key_path: "/src/.buildkite/test.signing.key"
+
+report_stats: false
+
+database:
+  name: "psycopg2"
+  args:
+    user: postgres
+    host: postgres
+    password: postgres
+    database: synapse
+
+# Suppress the key server warning.
+trusted_key_servers:
+  - server_name: "matrix.org"
+suppress_key_server_warning: true
diff --git a/.buildkite/scripts/create_postgres_db.py b/.buildkite/scripts/create_postgres_db.py
new file mode 100755
index 0000000000..df6082b0ac
--- /dev/null
+++ b/.buildkite/scripts/create_postgres_db.py
@@ -0,0 +1,36 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from synapse.storage.engines import create_engine
+
+logger = logging.getLogger("create_postgres_db")
+
+if __name__ == "__main__":
+    # Create a PostgresEngine.
+    db_engine = create_engine({"name": "psycopg2", "args": {}})
+
+    # Connect to postgres to create the base database.
+    # We use "postgres" as a database because it's bound to exist and the "synapse" one
+    # doesn't exist yet.
+    db_conn = db_engine.module.connect(
+        user="postgres", host="postgres", password="postgres", dbname="postgres"
+    )
+    db_conn.autocommit = True
+    cur = db_conn.cursor()
+    cur.execute("CREATE DATABASE synapse;")
+    cur.close()
+    db_conn.close()
diff --git a/.buildkite/scripts/test_synapse_port_db.sh b/.buildkite/scripts/test_synapse_port_db.sh
new file mode 100755
index 0000000000..9ed2177635
--- /dev/null
+++ b/.buildkite/scripts/test_synapse_port_db.sh
@@ -0,0 +1,36 @@
+#!/bin/bash
+#
+# Test script for 'synapse_port_db', which creates a virtualenv, installs Synapse along
+# with additional dependencies needed for the test (such as coverage or the PostgreSQL
+# driver), update the schema of the test SQLite database and run background updates on it,
+# create an empty test database in PostgreSQL, then run the 'synapse_port_db' script to
+# test porting the SQLite database to the PostgreSQL database (with coverage).
+
+set -xe
+cd `dirname $0`/../..
+
+echo "--- Install dependencies"
+
+# Install dependencies for this test.
+pip install psycopg2 coverage coverage-enable-subprocess
+
+# Install Synapse itself. This won't update any libraries.
+pip install -e .
+
+echo "--- Generate the signing key"
+
+# Generate the server's signing key.
+python -m synapse.app.homeserver --generate-keys -c .buildkite/sqlite-config.yaml
+
+echo "--- Prepare the databases"
+
+# Make sure the SQLite3 database is using the latest schema and has no pending background update.
+scripts-dev/update_database --database-config .buildkite/sqlite-config.yaml
+
+# Create the PostgreSQL database.
+./.buildkite/scripts/create_postgres_db.py
+
+echo "+++ Run synapse_port_db"
+
+# Run the script
+coverage run scripts/synapse_port_db --sqlite-database .buildkite/test_db.db --postgres-config .buildkite/postgres-config.yaml
diff --git a/.buildkite/sqlite-config.yaml b/.buildkite/sqlite-config.yaml
new file mode 100644
index 0000000000..635b921764
--- /dev/null
+++ b/.buildkite/sqlite-config.yaml
@@ -0,0 +1,18 @@
+# Configuration file used for testing the 'synapse_port_db' script.
+# Tells the 'update_database' script to connect to the test SQLite database to upgrade its
+# schema and run background updates on it.
+server_name: "test"
+
+signing_key_path: "/src/.buildkite/test.signing.key"
+
+report_stats: false
+
+database:
+  name: "sqlite3"
+  args:
+    database: ".buildkite/test_db.db"
+
+# Suppress the key server warning.
+trusted_key_servers:
+  - server_name: "matrix.org"
+suppress_key_server_warning: true
diff --git a/.buildkite/test_db.db b/.buildkite/test_db.db
new file mode 100644
index 0000000000..f20567ba73
Binary files /dev/null and b/.buildkite/test_db.db differ
diff --git a/.github/ISSUE_TEMPLATE/BUG_REPORT.md b/.github/ISSUE_TEMPLATE/BUG_REPORT.md
index 5cf844bfb1..9dd05bcba3 100644
--- a/.github/ISSUE_TEMPLATE/BUG_REPORT.md
+++ b/.github/ISSUE_TEMPLATE/BUG_REPORT.md
@@ -7,7 +7,7 @@ about: Create a report to help us improve
 
 
 
-- **Homeserver**: 
+- **Homeserver**:
 
 If not matrix.org:
 
 
-- **Version**: 
+ What version of Synapse is running?
 
-- **Install method**: 
+You can find the Synapse version with this command:
+
+$ curl http://localhost:8008/_synapse/admin/v1/server_version
+
+(You may need to replace `localhost:8008` if Synapse is not configured to
+listen on that port.)
+-->
+- **Version**:
+
+- **Install method**:
 
 
-- **Platform**: 
+- **Platform**:
  rest/* <=> Client
+        |                  | Rooms*Handler     |              |
+    HS <=> federation/* <==> FederationHandler |              |
+        |      |           | PresenceHandler   |              |
+        |      |           | TypingHandler     |              |
+        |      |           '-------------------'              |
+        |      |                 |     |                      |
+        |      |              state/*  |                      |
+        |      |                 |     |                      |
+        |      |                 v     v                      |
+        |      `--------------> storage/*                     |
+        |                          |                          |
+        '--------------------------|--------------------------'
+                                   v
+                                .----.
+                                | DB |
+                                '----'
+
+-   Handlers: business logic of synapse itself. Follows a set contract of BaseHandler:
+    -   BaseHandler gives us onNewRoomEvent which: (TODO: flesh this out and make it less cryptic):
+        -   handle_state(event)
+        -   auth(event)
+        -   persist_event(event)
+        -   notify notifier or federation(event)
+    -   PresenceHandler: use distributor to get EDUs out of Federation.
+        Very lightweight logic built on the distributor
+    -   TypingHandler: use distributor to get EDUs out of Federation.
+        Very lightweight logic built on the distributor
+    -   EventsHandler: handles the events stream...
+    -   FederationHandler: - gets PDU from Federation Layer; turns into
+        an event; follows basehandler functionality.
+    -   RoomsHandler: does all the room logic, including members - lots
+        of classes in RoomsHandler.
+    -   ProfileHandler: talks to the storage to store/retrieve profile
+        info.
+-   EventFactory: generates events of particular event types.
+-   Notifier: Backs the events handler
+-   REST: Interfaces handlers and events to the outside world via
+    HTTP/JSON. Converts events back and forth from JSON.
+-   Federation: holds the HTTP client & server to talk to other servers.
+    Does replication to make sure there's nothing missing in the graph.
+    Handles reliability. Handles txns.
+-   Distributor: generic event bus. used for presence & typing only
+    currently. Notifier could be implemented using Distributor - so far
+    we are only using for things which actually /require/ dynamic
+    pluggability however as it can obfuscate the actual flow of control.
+-   Auth: helper singleton to say whether a given event is allowed to do
+    a given thing (TODO: put this on the diagram)
+-   State: helper singleton: does state conflict resolution. You give it
+    an event and it tells you if it actually updates the state or not,
+    and annotates the event up properly and handles merge conflict
+    resolution.
+-   Storage: abstracts the storage engine.
diff --git a/docs/architecture.rst b/docs/architecture.rst
deleted file mode 100644
index 98050428b9..0000000000
--- a/docs/architecture.rst
+++ /dev/null
@@ -1,68 +0,0 @@
-Synapse Architecture
-====================
-
-As of the end of Oct 2014, Synapse's overall architecture looks like::
-
-        synapse
-        .-----------------------------------------------------.
-        |                          Notifier                   |
-        |                            ^  |                     |
-        |                            |  |                     |
-        |                  .------------|------.              |
-        |                  | handlers/  |      |              |
-        |                  |            v      |              |
-        |                  | Event*Handler <--------> rest/* <=> Client
-        |                  | Rooms*Handler     |              |
-  HSes <=> federation/* <==> FederationHandler |              |
-        |      |           | PresenceHandler   |              |
-        |      |           | TypingHandler     |              |
-        |      |           '-------------------'              |
-        |      |                 |     |                      |
-        |      |              state/*  |                      |
-        |      |                 |     |                      |
-        |      |                 v     v                      |
-        |      `--------------> storage/*                     |
-        |                          |                          |
-        '--------------------------|--------------------------'
-                                   v
-                                .----.
-                                | DB |
-                                '----'
-
-* Handlers: business logic of synapse itself.  Follows a set contract of BaseHandler:
-
-  - BaseHandler gives us onNewRoomEvent which: (TODO: flesh this out and make it less cryptic):
- 
-    + handle_state(event)
-    + auth(event)
-    + persist_event(event)
-    + notify notifier or federation(event)
-   
-  - PresenceHandler: use distributor to get EDUs out of Federation.  Very
-    lightweight logic built on the distributor
-  - TypingHandler: use distributor to get EDUs out of Federation.  Very
-    lightweight logic built on the distributor
-  - EventsHandler: handles the events stream...
-  - FederationHandler: - gets PDU from Federation Layer; turns into an event;
-    follows basehandler functionality.
-  - RoomsHandler: does all the room logic, including members - lots of classes in
-    RoomsHandler.
-  - ProfileHandler: talks to the storage to store/retrieve profile info.
-
-* EventFactory: generates events of particular event types.
-* Notifier: Backs the events handler
-* REST: Interfaces handlers and events to the outside world via HTTP/JSON.
-  Converts events back and forth from JSON.
-* Federation: holds the HTTP client & server to talk to other servers.  Does
-  replication to make sure there's nothing missing in the graph.  Handles
-  reliability.  Handles txns.
-* Distributor: generic event bus. used for presence & typing only currently. 
-  Notifier could be implemented using Distributor - so far we are only using for
-  things which actually /require/ dynamic pluggability however as it can
-  obfuscate the actual flow of control.
-* Auth: helper singleton to say whether a given event is allowed to do a given
-  thing  (TODO: put this on the diagram)
-* State: helper singleton: does state conflict resolution. You give it an event
-  and it tells you if it actually updates the state or not, and annotates the
-  event up properly and handles merge conflict resolution.
-* Storage: abstracts the storage engine.
diff --git a/docs/code_style.md b/docs/code_style.md
new file mode 100644
index 0000000000..f983f72d6c
--- /dev/null
+++ b/docs/code_style.md
@@ -0,0 +1,169 @@
+# Code Style
+
+## Formatting tools
+
+The Synapse codebase uses a number of code formatting tools in order to
+quickly and automatically check for formatting (and sometimes logical)
+errors in code.
+
+The necessary tools are detailed below.
+
+-   **black**
+
+    The Synapse codebase uses [black](https://pypi.org/project/black/)
+    as an opinionated code formatter, ensuring all comitted code is
+    properly formatted.
+
+    First install `black` with:
+
+        pip install --upgrade black
+
+    Have `black` auto-format your code (it shouldn't change any
+    functionality) with:
+
+        black . --exclude="\.tox|build|env"
+
+-   **flake8**
+
+    `flake8` is a code checking tool. We require code to pass `flake8`
+    before being merged into the codebase.
+
+    Install `flake8` with:
+
+        pip install --upgrade flake8
+
+    Check all application and test code with:
+
+        flake8 synapse tests
+
+-   **isort**
+
+    `isort` ensures imports are nicely formatted, and can suggest and
+    auto-fix issues such as double-importing.
+
+    Install `isort` with:
+
+        pip install --upgrade isort
+
+    Auto-fix imports with:
+
+        isort -rc synapse tests
+
+    `-rc` means to recursively search the given directories.
+
+It's worth noting that modern IDEs and text editors can run these tools
+automatically on save. It may be worth looking into whether this
+functionality is supported in your editor for a more convenient
+development workflow. It is not, however, recommended to run `flake8` on
+save as it takes a while and is very resource intensive.
+
+## General rules
+
+-   **Naming**:
+    -   Use camel case for class and type names
+    -   Use underscores for functions and variables.
+-   **Docstrings**: should follow the [google code
+    style](https://google.github.io/styleguide/pyguide.html#38-comments-and-docstrings).
+    This is so that we can generate documentation with
+    [sphinx](http://sphinxcontrib-napoleon.readthedocs.org/en/latest/).
+    See the
+    [examples](http://sphinxcontrib-napoleon.readthedocs.io/en/latest/example_google.html)
+    in the sphinx documentation.
+-   **Imports**:
+    -   Imports should be sorted by `isort` as described above.
+    -   Prefer to import classes and functions rather than packages or
+        modules.
+
+        Example:
+
+            from synapse.types import UserID
+            ...
+            user_id = UserID(local, server)
+
+        is preferred over:
+
+            from synapse import types
+            ...
+            user_id = types.UserID(local, server)
+
+        (or any other variant).
+
+        This goes against the advice in the Google style guide, but it
+        means that errors in the name are caught early (at import time).
+
+    -   Avoid wildcard imports (`from synapse.types import *`) and
+        relative imports (`from .types import UserID`).
+
+## Configuration file format
+
+The [sample configuration file](./sample_config.yaml) acts as a
+reference to Synapse's configuration options for server administrators.
+Remember that many readers will be unfamiliar with YAML and server
+administration in general, so that it is important that the file be as
+easy to understand as possible, which includes following a consistent
+format.
+
+Some guidelines follow:
+
+-   Sections should be separated with a heading consisting of a single
+    line prefixed and suffixed with `##`. There should be **two** blank
+    lines before the section header, and **one** after.
+-   Each option should be listed in the file with the following format:
+    -   A comment describing the setting. Each line of this comment
+        should be prefixed with a hash (`#`) and a space.
+
+        The comment should describe the default behaviour (ie, what
+        happens if the setting is omitted), as well as what the effect
+        will be if the setting is changed.
+
+        Often, the comment end with something like "uncomment the
+        following to ".
+
+    -   A line consisting of only `#`.
+    -   A commented-out example setting, prefixed with only `#`.
+
+        For boolean (on/off) options, convention is that this example
+        should be the *opposite* to the default (so the comment will end
+        with "Uncomment the following to enable [or disable]
+        ." For other options, the example should give some
+        non-default value which is likely to be useful to the reader.
+
+-   There should be a blank line between each option.
+-   Where several settings are grouped into a single dict, *avoid* the
+    convention where the whole block is commented out, resulting in
+    comment lines starting `# #`, as this is hard to read and confusing
+    to edit. Instead, leave the top-level config option uncommented, and
+    follow the conventions above for sub-options. Ensure that your code
+    correctly handles the top-level option being set to `None` (as it
+    will be if no sub-options are enabled).
+-   Lines should be wrapped at 80 characters.
+
+Example:
+
+    ## Frobnication ##
+
+    # The frobnicator will ensure that all requests are fully frobnicated.
+    # To enable it, uncomment the following.
+    #
+    #frobnicator_enabled: true
+
+    # By default, the frobnicator will frobnicate with the default frobber.
+    # The following will make it use an alternative frobber.
+    #
+    #frobincator_frobber: special_frobber
+
+    # Settings for the frobber
+    #
+    frobber:
+       # frobbing speed. Defaults to 1.
+       #
+       #speed: 10
+
+       # frobbing distance. Defaults to 1000.
+       #
+       #distance: 100
+
+Note that the sample configuration is generated from the synapse code
+and is maintained by a script, `scripts-dev/generate_sample_config`.
+Making sure that the output from this script matches the desired format
+is left as an exercise for the reader!
diff --git a/docs/code_style.rst b/docs/code_style.rst
deleted file mode 100644
index 39ac4ebedc..0000000000
--- a/docs/code_style.rst
+++ /dev/null
@@ -1,180 +0,0 @@
-Code Style
-==========
-
-Formatting tools
-----------------
-
-The Synapse codebase uses a number of code formatting tools in order to
-quickly and automatically check for formatting (and sometimes logical) errors
-in code.
-
-The necessary tools are detailed below.
-
-- **black**
-
-  The Synapse codebase uses `black `_ as an
-  opinionated code formatter, ensuring all comitted code is properly
-  formatted.
-
-  First install ``black`` with::
-
-    pip install --upgrade black
-
-  Have ``black`` auto-format your code (it shouldn't change any functionality)
-  with::
-
-    black . --exclude="\.tox|build|env"
-
-- **flake8**
-
-  ``flake8`` is a code checking tool. We require code to pass ``flake8`` before being merged into the codebase.
-
-  Install ``flake8`` with::
-
-    pip install --upgrade flake8
-
-  Check all application and test code with::
-
-    flake8 synapse tests
-
-- **isort**
-
-  ``isort`` ensures imports are nicely formatted, and can suggest and
-  auto-fix issues such as double-importing.
-
-  Install ``isort`` with::
-
-    pip install --upgrade isort
-
-  Auto-fix imports with::
-
-    isort -rc synapse tests
-
-  ``-rc`` means to recursively search the given directories.
-
-It's worth noting that modern IDEs and text editors can run these tools
-automatically on save. It may be worth looking into whether this
-functionality is supported in your editor for a more convenient development
-workflow. It is not, however, recommended to run ``flake8`` on save as it
-takes a while and is very resource intensive.
-
-General rules
--------------
-
-- **Naming**:
-
-  - Use camel case for class and type names
-  - Use underscores for functions and variables.
-
-- **Docstrings**: should follow the `google code style
-  `_.
-  This is so that we can generate documentation with `sphinx
-  `_. See the
-  `examples
-  `_
-  in the sphinx documentation.
-
-- **Imports**:
-
-  - Imports should be sorted by ``isort`` as described above.
-
-  - Prefer to import classes and functions rather than packages or modules.
-
-    Example::
-
-      from synapse.types import UserID
-      ...
-      user_id = UserID(local, server)
-
-    is preferred over::
-
-      from synapse import types
-      ...
-      user_id = types.UserID(local, server)
-
-    (or any other variant).
-
-    This goes against the advice in the Google style guide, but it means that
-    errors in the name are caught early (at import time).
-
-  - Avoid wildcard imports (``from synapse.types import *``) and relative
-    imports (``from .types import UserID``).
-
-Configuration file format
--------------------------
-
-The `sample configuration file <./sample_config.yaml>`_ acts as a reference to
-Synapse's configuration options for server administrators. Remember that many
-readers will be unfamiliar with YAML and server administration in general, so
-that it is important that the file be as easy to understand as possible, which
-includes following a consistent format.
-
-Some guidelines follow:
-
-* Sections should be separated with a heading consisting of a single line
-  prefixed and suffixed with ``##``. There should be **two** blank lines
-  before the section header, and **one** after.
-
-* Each option should be listed in the file with the following format:
-
-  * A comment describing the setting. Each line of this comment should be
-    prefixed with a hash (``#``) and a space.
-
-    The comment should describe the default behaviour (ie, what happens if
-    the setting is omitted), as well as what the effect will be if the
-    setting is changed.
-
-    Often, the comment end with something like "uncomment the
-    following to \".
-
-  * A line consisting of only ``#``.
-
-  * A commented-out example setting, prefixed with only ``#``.
-
-    For boolean (on/off) options, convention is that this example should be
-    the *opposite* to the default (so the comment will end with "Uncomment
-    the following to enable [or disable] \." For other options,
-    the example should give some non-default value which is likely to be
-    useful to the reader.
-
-* There should be a blank line between each option.
-
-* Where several settings are grouped into a single dict, *avoid* the
-  convention where the whole block is commented out, resulting in comment
-  lines starting ``# #``, as this is hard to read and confusing to
-  edit. Instead, leave the top-level config option uncommented, and follow
-  the conventions above for sub-options. Ensure that your code correctly
-  handles the top-level option being set to ``None`` (as it will be if no
-  sub-options are enabled).
-
-* Lines should be wrapped at 80 characters.
-
-Example::
-
-    ## Frobnication ##
-
-    # The frobnicator will ensure that all requests are fully frobnicated.
-    # To enable it, uncomment the following.
-    #
-    #frobnicator_enabled: true
-
-    # By default, the frobnicator will frobnicate with the default frobber.
-    # The following will make it use an alternative frobber.
-    #
-    #frobincator_frobber: special_frobber
-
-    # Settings for the frobber
-    #
-    frobber:
-       # frobbing speed. Defaults to 1.
-       #
-       #speed: 10
-
-       # frobbing distance. Defaults to 1000.
-       #
-       #distance: 100
-
-Note that the sample configuration is generated from the synapse code and is
-maintained by a script, ``scripts-dev/generate_sample_config``. Making sure
-that the output from this script matches the desired format is left as an
-exercise for the reader!
diff --git a/docs/dev/saml.md b/docs/dev/saml.md
new file mode 100644
index 0000000000..f41aadce47
--- /dev/null
+++ b/docs/dev/saml.md
@@ -0,0 +1,37 @@
+# How to test SAML as a developer without a server
+
+https://capriza.github.io/samling/samling.html (https://github.com/capriza/samling) is a great
+resource for being able to tinker with the SAML options within Synapse without needing to
+deploy and configure a complicated software stack.
+
+To make Synapse (and therefore Riot) use it:
+
+1. Use the samling.html URL above or deploy your own and visit the IdP Metadata tab.
+2. Copy the XML to your clipboard.
+3. On your Synapse server, create a new file `samling.xml` next to your `homeserver.yaml` with
+   the XML from step 2 as the contents.
+4. Edit your `homeserver.yaml` to include:
+   ```yaml
+   saml2_config:
+     sp_config:
+       allow_unknown_attributes: true  # Works around a bug with AVA Hashes: https://github.com/IdentityPython/pysaml2/issues/388
+       metadata:
+         local: ["samling.xml"]   
+   ```
+5. Run `apt-get install xmlsec1` and `pip install --upgrade --force 'pysaml2>=4.5.0'` to ensure
+   the dependencies are installed and ready to go.
+6. Restart Synapse.
+
+Then in Riot:
+
+1. Visit the login page with a Riot pointing at your homeserver.
+2. Click the Single Sign-On button.
+3. On the samling page, enter a Name Identifier and add a SAML Attribute for `uid=your_localpart`.
+   The response must also be signed.
+4. Click "Next".
+5. Click "Post Response" (change nothing).
+6. You should be logged in.
+
+If you try and repeat this process, you may be automatically logged in using the information you
+gave previously. To fix this, open your developer console (`F12` or `Ctrl+Shift+I`) while on the
+samling page and clear the site data. In Chrome, this will be a button on the Application tab.
diff --git a/docs/federate.md b/docs/federate.md
index 6d6bb85e15..193e2d2dfe 100644
--- a/docs/federate.md
+++ b/docs/federate.md
@@ -148,7 +148,7 @@ We no longer actively recommend against using a reverse proxy. Many admins will
 find it easier to direct federation traffic to a reverse proxy and manage their
 own TLS certificates, and this is a supported configuration.
 
-See [reverse_proxy.rst](reverse_proxy.rst) for information on setting up a
+See [reverse_proxy.md](reverse_proxy.md) for information on setting up a
 reverse proxy.
 
 #### Do I still need to give my TLS certificates to Synapse if I am using a reverse proxy?
@@ -184,7 +184,7 @@ a complicated dance which requires connections in both directions).
 
 Another common problem is that people on other servers can't join rooms that
 you invite them to. This can be caused by an incorrectly-configured reverse
-proxy: see [reverse_proxy.rst]() for instructions on how to correctly
+proxy: see [reverse_proxy.md]() for instructions on how to correctly
 configure a reverse proxy.
 
 ## Running a Demo Federation of Synapses
diff --git a/docs/log_contexts.md b/docs/log_contexts.md
new file mode 100644
index 0000000000..5331e8c88b
--- /dev/null
+++ b/docs/log_contexts.md
@@ -0,0 +1,494 @@
+# Log Contexts
+
+To help track the processing of individual requests, synapse uses a
+'`log context`' to track which request it is handling at any given
+moment. This is done via a thread-local variable; a `logging.Filter` is
+then used to fish the information back out of the thread-local variable
+and add it to each log record.
+
+Logcontexts are also used for CPU and database accounting, so that we
+can track which requests were responsible for high CPU use or database
+activity.
+
+The `synapse.logging.context` module provides a facilities for managing
+the current log context (as well as providing the `LoggingContextFilter`
+class).
+
+Deferreds make the whole thing complicated, so this document describes
+how it all works, and how to write code which follows the rules.
+
+##Logcontexts without Deferreds
+
+In the absence of any Deferred voodoo, things are simple enough. As with
+any code of this nature, the rule is that our function should leave
+things as it found them:
+
+```python
+from synapse.logging import context         # omitted from future snippets
+
+def handle_request(request_id):
+    request_context = context.LoggingContext()
+
+    calling_context = context.LoggingContext.current_context()
+    context.LoggingContext.set_current_context(request_context)
+    try:
+        request_context.request = request_id
+        do_request_handling()
+        logger.debug("finished")
+    finally:
+        context.LoggingContext.set_current_context(calling_context)
+
+def do_request_handling():
+    logger.debug("phew")  # this will be logged against request_id
+```
+
+LoggingContext implements the context management methods, so the above
+can be written much more succinctly as:
+
+```python
+def handle_request(request_id):
+    with context.LoggingContext() as request_context:
+        request_context.request = request_id
+        do_request_handling()
+        logger.debug("finished")
+
+def do_request_handling():
+    logger.debug("phew")
+```
+
+## Using logcontexts with Deferreds
+
+Deferreds --- and in particular, `defer.inlineCallbacks` --- break the
+linear flow of code so that there is no longer a single entry point
+where we should set the logcontext and a single exit point where we
+should remove it.
+
+Consider the example above, where `do_request_handling` needs to do some
+blocking operation, and returns a deferred:
+
+```python
+@defer.inlineCallbacks
+def handle_request(request_id):
+    with context.LoggingContext() as request_context:
+        request_context.request = request_id
+        yield do_request_handling()
+        logger.debug("finished")
+```
+
+In the above flow:
+
+-   The logcontext is set
+-   `do_request_handling` is called, and returns a deferred
+-   `handle_request` yields the deferred
+-   The `inlineCallbacks` wrapper of `handle_request` returns a deferred
+
+So we have stopped processing the request (and will probably go on to
+start processing the next), without clearing the logcontext.
+
+To circumvent this problem, synapse code assumes that, wherever you have
+a deferred, you will want to yield on it. To that end, whereever
+functions return a deferred, we adopt the following conventions:
+
+**Rules for functions returning deferreds:**
+
+> -   If the deferred is already complete, the function returns with the
+>     same logcontext it started with.
+> -   If the deferred is incomplete, the function clears the logcontext
+>     before returning; when the deferred completes, it restores the
+>     logcontext before running any callbacks.
+
+That sounds complicated, but actually it means a lot of code (including
+the example above) "just works". There are two cases:
+
+-   If `do_request_handling` returns a completed deferred, then the
+    logcontext will still be in place. In this case, execution will
+    continue immediately after the `yield`; the "finished" line will
+    be logged against the right context, and the `with` block restores
+    the original context before we return to the caller.
+-   If the returned deferred is incomplete, `do_request_handling` clears
+    the logcontext before returning. The logcontext is therefore clear
+    when `handle_request` yields the deferred. At that point, the
+    `inlineCallbacks` wrapper adds a callback to the deferred, and
+    returns another (incomplete) deferred to the caller, and it is safe
+    to begin processing the next request.
+
+    Once `do_request_handling`'s deferred completes, it will reinstate
+    the logcontext, before running the callback added by the
+    `inlineCallbacks` wrapper. That callback runs the second half of
+    `handle_request`, so again the "finished" line will be logged
+    against the right context, and the `with` block restores the
+    original context.
+
+As an aside, it's worth noting that `handle_request` follows our rules
+-though that only matters if the caller has its own logcontext which it
+cares about.
+
+The following sections describe pitfalls and helpful patterns when
+implementing these rules.
+
+Always yield your deferreds
+---------------------------
+
+Whenever you get a deferred back from a function, you should `yield` on
+it as soon as possible. (Returning it directly to your caller is ok too,
+if you're not doing `inlineCallbacks`.) Do not pass go; do not do any
+logging; do not call any other functions.
+
+```python
+@defer.inlineCallbacks
+def fun():
+    logger.debug("starting")
+    yield do_some_stuff()       # just like this
+
+    d = more_stuff()
+    result = yield d            # also fine, of course
+
+    return result
+
+def nonInlineCallbacksFun():
+    logger.debug("just a wrapper really")
+    return do_some_stuff()      # this is ok too - the caller will yield on
+                                # it anyway.
+```
+
+Provided this pattern is followed all the way back up to the callchain
+to where the logcontext was set, this will make things work out ok:
+provided `do_some_stuff` and `more_stuff` follow the rules above, then
+so will `fun` (as wrapped by `inlineCallbacks`) and
+`nonInlineCallbacksFun`.
+
+It's all too easy to forget to `yield`: for instance if we forgot that
+`do_some_stuff` returned a deferred, we might plough on regardless. This
+leads to a mess; it will probably work itself out eventually, but not
+before a load of stuff has been logged against the wrong context.
+(Normally, other things will break, more obviously, if you forget to
+`yield`, so this tends not to be a major problem in practice.)
+
+Of course sometimes you need to do something a bit fancier with your
+Deferreds - not all code follows the linear A-then-B-then-C pattern.
+Notes on implementing more complex patterns are in later sections.
+
+## Where you create a new Deferred, make it follow the rules
+
+Most of the time, a Deferred comes from another synapse function.
+Sometimes, though, we need to make up a new Deferred, or we get a
+Deferred back from external code. We need to make it follow our rules.
+
+The easy way to do it is with a combination of `defer.inlineCallbacks`,
+and `context.PreserveLoggingContext`. Suppose we want to implement
+`sleep`, which returns a deferred which will run its callbacks after a
+given number of seconds. That might look like:
+
+```python
+# not a logcontext-rules-compliant function
+def get_sleep_deferred(seconds):
+    d = defer.Deferred()
+    reactor.callLater(seconds, d.callback, None)
+    return d
+```
+
+That doesn't follow the rules, but we can fix it by wrapping it with
+`PreserveLoggingContext` and `yield` ing on it:
+
+```python
+@defer.inlineCallbacks
+def sleep(seconds):
+    with PreserveLoggingContext():
+        yield get_sleep_deferred(seconds)
+```
+
+This technique works equally for external functions which return
+deferreds, or deferreds we have made ourselves.
+
+You can also use `context.make_deferred_yieldable`, which just does the
+boilerplate for you, so the above could be written:
+
+```python
+def sleep(seconds):
+    return context.make_deferred_yieldable(get_sleep_deferred(seconds))
+```
+
+## Fire-and-forget
+
+Sometimes you want to fire off a chain of execution, but not wait for
+its result. That might look a bit like this:
+
+```python
+@defer.inlineCallbacks
+def do_request_handling():
+    yield foreground_operation()
+
+    # *don't* do this
+    background_operation()
+
+    logger.debug("Request handling complete")
+
+@defer.inlineCallbacks
+def background_operation():
+    yield first_background_step()
+    logger.debug("Completed first step")
+    yield second_background_step()
+    logger.debug("Completed second step")
+```
+
+The above code does a couple of steps in the background after
+`do_request_handling` has finished. The log lines are still logged
+against the `request_context` logcontext, which may or may not be
+desirable. There are two big problems with the above, however. The first
+problem is that, if `background_operation` returns an incomplete
+Deferred, it will expect its caller to `yield` immediately, so will have
+cleared the logcontext. In this example, that means that 'Request
+handling complete' will be logged without any context.
+
+The second problem, which is potentially even worse, is that when the
+Deferred returned by `background_operation` completes, it will restore
+the original logcontext. There is nothing waiting on that Deferred, so
+the logcontext will leak into the reactor and possibly get attached to
+some arbitrary future operation.
+
+There are two potential solutions to this.
+
+One option is to surround the call to `background_operation` with a
+`PreserveLoggingContext` call. That will reset the logcontext before
+starting `background_operation` (so the context restored when the
+deferred completes will be the empty logcontext), and will restore the
+current logcontext before continuing the foreground process:
+
+```python
+@defer.inlineCallbacks
+def do_request_handling():
+    yield foreground_operation()
+
+    # start background_operation off in the empty logcontext, to
+    # avoid leaking the current context into the reactor.
+    with PreserveLoggingContext():
+        background_operation()
+
+    # this will now be logged against the request context
+    logger.debug("Request handling complete")
+```
+
+Obviously that option means that the operations done in
+`background_operation` would be not be logged against a logcontext
+(though that might be fixed by setting a different logcontext via a
+`with LoggingContext(...)` in `background_operation`).
+
+The second option is to use `context.run_in_background`, which wraps a
+function so that it doesn't reset the logcontext even when it returns
+an incomplete deferred, and adds a callback to the returned deferred to
+reset the logcontext. In other words, it turns a function that follows
+the Synapse rules about logcontexts and Deferreds into one which behaves
+more like an external function --- the opposite operation to that
+described in the previous section. It can be used like this:
+
+```python
+@defer.inlineCallbacks
+def do_request_handling():
+    yield foreground_operation()
+
+    context.run_in_background(background_operation)
+
+    # this will now be logged against the request context
+    logger.debug("Request handling complete")
+```
+
+## Passing synapse deferreds into third-party functions
+
+A typical example of this is where we want to collect together two or
+more deferred via `defer.gatherResults`:
+
+```python
+d1 = operation1()
+d2 = operation2()
+d3 = defer.gatherResults([d1, d2])
+```
+
+This is really a variation of the fire-and-forget problem above, in that
+we are firing off `d1` and `d2` without yielding on them. The difference
+is that we now have third-party code attached to their callbacks. Anyway
+either technique given in the [Fire-and-forget](#fire-and-forget)
+section will work.
+
+Of course, the new Deferred returned by `gatherResults` needs to be
+wrapped in order to make it follow the logcontext rules before we can
+yield it, as described in [Where you create a new Deferred, make it
+follow the
+rules](#where-you-create-a-new-deferred-make-it-follow-the-rules).
+
+So, option one: reset the logcontext before starting the operations to
+be gathered:
+
+```python
+@defer.inlineCallbacks
+def do_request_handling():
+    with PreserveLoggingContext():
+        d1 = operation1()
+        d2 = operation2()
+        result = yield defer.gatherResults([d1, d2])
+```
+
+In this case particularly, though, option two, of using
+`context.preserve_fn` almost certainly makes more sense, so that
+`operation1` and `operation2` are both logged against the original
+logcontext. This looks like:
+
+```python
+@defer.inlineCallbacks
+def do_request_handling():
+    d1 = context.preserve_fn(operation1)()
+    d2 = context.preserve_fn(operation2)()
+
+    with PreserveLoggingContext():
+        result = yield defer.gatherResults([d1, d2])
+```
+
+## Was all this really necessary?
+
+The conventions used work fine for a linear flow where everything
+happens in series via `defer.inlineCallbacks` and `yield`, but are
+certainly tricky to follow for any more exotic flows. It's hard not to
+wonder if we could have done something else.
+
+We're not going to rewrite Synapse now, so the following is entirely of
+academic interest, but I'd like to record some thoughts on an
+alternative approach.
+
+I briefly prototyped some code following an alternative set of rules. I
+think it would work, but I certainly didn't get as far as thinking how
+it would interact with concepts as complicated as the cache descriptors.
+
+My alternative rules were:
+
+-   functions always preserve the logcontext of their caller, whether or
+    not they are returning a Deferred.
+-   Deferreds returned by synapse functions run their callbacks in the
+    same context as the function was orignally called in.
+
+The main point of this scheme is that everywhere that sets the
+logcontext is responsible for clearing it before returning control to
+the reactor.
+
+So, for example, if you were the function which started a
+`with LoggingContext` block, you wouldn't `yield` within it --- instead
+you'd start off the background process, and then leave the `with` block
+to wait for it:
+
+```python
+def handle_request(request_id):
+    with context.LoggingContext() as request_context:
+        request_context.request = request_id
+        d = do_request_handling()
+
+    def cb(r):
+        logger.debug("finished")
+
+    d.addCallback(cb)
+    return d
+```
+
+(in general, mixing `with LoggingContext` blocks and
+`defer.inlineCallbacks` in the same function leads to slighly
+counter-intuitive code, under this scheme).
+
+Because we leave the original `with` block as soon as the Deferred is
+returned (as opposed to waiting for it to be resolved, as we do today),
+the logcontext is cleared before control passes back to the reactor; so
+if there is some code within `do_request_handling` which needs to wait
+for a Deferred to complete, there is no need for it to worry about
+clearing the logcontext before doing so:
+
+```python
+def handle_request():
+    r = do_some_stuff()
+    r.addCallback(do_some_more_stuff)
+    return r
+```
+
+--- and provided `do_some_stuff` follows the rules of returning a
+Deferred which runs its callbacks in the original logcontext, all is
+happy.
+
+The business of a Deferred which runs its callbacks in the original
+logcontext isn't hard to achieve --- we have it today, in the shape of
+`context._PreservingContextDeferred`:
+
+```python
+def do_some_stuff():
+    deferred = do_some_io()
+    pcd = _PreservingContextDeferred(LoggingContext.current_context())
+    deferred.chainDeferred(pcd)
+    return pcd
+```
+
+It turns out that, thanks to the way that Deferreds chain together, we
+automatically get the property of a context-preserving deferred with
+`defer.inlineCallbacks`, provided the final Defered the function
+`yields` on has that property. So we can just write:
+
+```python
+@defer.inlineCallbacks
+def handle_request():
+    yield do_some_stuff()
+    yield do_some_more_stuff()
+```
+
+To conclude: I think this scheme would have worked equally well, with
+less danger of messing it up, and probably made some more esoteric code
+easier to write. But again --- changing the conventions of the entire
+Synapse codebase is not a sensible option for the marginal improvement
+offered.
+
+## A note on garbage-collection of Deferred chains
+
+It turns out that our logcontext rules do not play nicely with Deferred
+chains which get orphaned and garbage-collected.
+
+Imagine we have some code that looks like this:
+
+```python
+listener_queue = []
+
+def on_something_interesting():
+    for d in listener_queue:
+        d.callback("foo")
+
+@defer.inlineCallbacks
+def await_something_interesting():
+    new_deferred = defer.Deferred()
+    listener_queue.append(new_deferred)
+
+    with PreserveLoggingContext():
+        yield new_deferred
+```
+
+Obviously, the idea here is that we have a bunch of things which are
+waiting for an event. (It's just an example of the problem here, but a
+relatively common one.)
+
+Now let's imagine two further things happen. First of all, whatever was
+waiting for the interesting thing goes away. (Perhaps the request times
+out, or something *even more* interesting happens.)
+
+Secondly, let's suppose that we decide that the interesting thing is
+never going to happen, and we reset the listener queue:
+
+```python
+def reset_listener_queue():
+    listener_queue.clear()
+```
+
+So, both ends of the deferred chain have now dropped their references,
+and the deferred chain is now orphaned, and will be garbage-collected at
+some point. Note that `await_something_interesting` is a generator
+function, and when Python garbage-collects generator functions, it gives
+them a chance to clean up by making the `yield` raise a `GeneratorExit`
+exception. In our case, that means that the `__exit__` handler of
+`PreserveLoggingContext` will carefully restore the request context, but
+there is now nothing waiting for its return, so the request context is
+never cleared.
+
+To reiterate, this problem only arises when *both* ends of a deferred
+chain are dropped. Dropping the the reference to a deferred you're
+supposed to be calling is probably bad practice, so this doesn't
+actually happen too much. Unfortunately, when it does happen, it will
+lead to leaked logcontexts which are incredibly hard to track down.
diff --git a/docs/log_contexts.rst b/docs/log_contexts.rst
deleted file mode 100644
index 4502cd9454..0000000000
--- a/docs/log_contexts.rst
+++ /dev/null
@@ -1,498 +0,0 @@
-Log Contexts
-============
-
-.. contents::
-
-To help track the processing of individual requests, synapse uses a
-'log context' to track which request it is handling at any given moment. This
-is done via a thread-local variable; a ``logging.Filter`` is then used to fish
-the information back out of the thread-local variable and add it to each log
-record.
-
-Logcontexts are also used for CPU and database accounting, so that we can track
-which requests were responsible for high CPU use or database activity.
-
-The ``synapse.logging.context`` module provides a facilities for managing the
-current log context (as well as providing the ``LoggingContextFilter`` class).
-
-Deferreds make the whole thing complicated, so this document describes how it
-all works, and how to write code which follows the rules.
-
-Logcontexts without Deferreds
------------------------------
-
-In the absence of any Deferred voodoo, things are simple enough. As with any
-code of this nature, the rule is that our function should leave things as it
-found them:
-
-.. code:: python
-
-    from synapse.logging import context         # omitted from future snippets
-
-    def handle_request(request_id):
-        request_context = context.LoggingContext()
-
-        calling_context = context.LoggingContext.current_context()
-        context.LoggingContext.set_current_context(request_context)
-        try:
-            request_context.request = request_id
-            do_request_handling()
-            logger.debug("finished")
-        finally:
-            context.LoggingContext.set_current_context(calling_context)
-
-    def do_request_handling():
-        logger.debug("phew")  # this will be logged against request_id
-
-
-LoggingContext implements the context management methods, so the above can be
-written much more succinctly as:
-
-.. code:: python
-
-    def handle_request(request_id):
-        with context.LoggingContext() as request_context:
-            request_context.request = request_id
-            do_request_handling()
-            logger.debug("finished")
-
-    def do_request_handling():
-        logger.debug("phew")
-
-
-Using logcontexts with Deferreds
---------------------------------
-
-Deferreds — and in particular, ``defer.inlineCallbacks`` — break
-the linear flow of code so that there is no longer a single entry point where
-we should set the logcontext and a single exit point where we should remove it.
-
-Consider the example above, where ``do_request_handling`` needs to do some
-blocking operation, and returns a deferred:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def handle_request(request_id):
-        with context.LoggingContext() as request_context:
-            request_context.request = request_id
-            yield do_request_handling()
-            logger.debug("finished")
-
-
-In the above flow:
-
-* The logcontext is set
-* ``do_request_handling`` is called, and returns a deferred
-* ``handle_request`` yields the deferred
-* The ``inlineCallbacks`` wrapper of ``handle_request`` returns a deferred
-
-So we have stopped processing the request (and will probably go on to start
-processing the next), without clearing the logcontext.
-
-To circumvent this problem, synapse code assumes that, wherever you have a
-deferred, you will want to yield on it. To that end, whereever functions return
-a deferred, we adopt the following conventions:
-
-**Rules for functions returning deferreds:**
-
-  * If the deferred is already complete, the function returns with the same
-    logcontext it started with.
-  * If the deferred is incomplete, the function clears the logcontext before
-    returning; when the deferred completes, it restores the logcontext before
-    running any callbacks.
-
-That sounds complicated, but actually it means a lot of code (including the
-example above) "just works". There are two cases:
-
-* If ``do_request_handling`` returns a completed deferred, then the logcontext
-  will still be in place. In this case, execution will continue immediately
-  after the ``yield``; the "finished" line will be logged against the right
-  context, and the ``with`` block restores the original context before we
-  return to the caller.
-
-* If the returned deferred is incomplete, ``do_request_handling`` clears the
-  logcontext before returning. The logcontext is therefore clear when
-  ``handle_request`` yields the deferred. At that point, the ``inlineCallbacks``
-  wrapper adds a callback to the deferred, and returns another (incomplete)
-  deferred to the caller, and it is safe to begin processing the next request.
-
-  Once ``do_request_handling``'s deferred completes, it will reinstate the
-  logcontext, before running the callback added by the ``inlineCallbacks``
-  wrapper. That callback runs the second half of ``handle_request``, so again
-  the "finished" line will be logged against the right
-  context, and the ``with`` block restores the original context.
-
-As an aside, it's worth noting that ``handle_request`` follows our rules -
-though that only matters if the caller has its own logcontext which it cares
-about.
-
-The following sections describe pitfalls and helpful patterns when implementing
-these rules.
-
-Always yield your deferreds
----------------------------
-
-Whenever you get a deferred back from a function, you should ``yield`` on it
-as soon as possible. (Returning it directly to your caller is ok too, if you're
-not doing ``inlineCallbacks``.) Do not pass go; do not do any logging; do not
-call any other functions.
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def fun():
-        logger.debug("starting")
-        yield do_some_stuff()       # just like this
-
-        d = more_stuff()
-        result = yield d            # also fine, of course
-
-        return result
-
-    def nonInlineCallbacksFun():
-        logger.debug("just a wrapper really")
-        return do_some_stuff()      # this is ok too - the caller will yield on
-                                    # it anyway.
-
-Provided this pattern is followed all the way back up to the callchain to where
-the logcontext was set, this will make things work out ok: provided
-``do_some_stuff`` and ``more_stuff`` follow the rules above, then so will
-``fun`` (as wrapped by ``inlineCallbacks``) and ``nonInlineCallbacksFun``.
-
-It's all too easy to forget to ``yield``: for instance if we forgot that
-``do_some_stuff`` returned a deferred, we might plough on regardless. This
-leads to a mess; it will probably work itself out eventually, but not before
-a load of stuff has been logged against the wrong context. (Normally, other
-things will break, more obviously, if you forget to ``yield``, so this tends
-not to be a major problem in practice.)
-
-Of course sometimes you need to do something a bit fancier with your Deferreds
-- not all code follows the linear A-then-B-then-C pattern. Notes on
-implementing more complex patterns are in later sections.
-
-Where you create a new Deferred, make it follow the rules
----------------------------------------------------------
-
-Most of the time, a Deferred comes from another synapse function. Sometimes,
-though, we need to make up a new Deferred, or we get a Deferred back from
-external code. We need to make it follow our rules.
-
-The easy way to do it is with a combination of ``defer.inlineCallbacks``, and
-``context.PreserveLoggingContext``. Suppose we want to implement ``sleep``,
-which returns a deferred which will run its callbacks after a given number of
-seconds. That might look like:
-
-.. code:: python
-
-    # not a logcontext-rules-compliant function
-    def get_sleep_deferred(seconds):
-        d = defer.Deferred()
-        reactor.callLater(seconds, d.callback, None)
-        return d
-
-That doesn't follow the rules, but we can fix it by wrapping it with
-``PreserveLoggingContext`` and ``yield`` ing on it:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def sleep(seconds):
-        with PreserveLoggingContext():
-            yield get_sleep_deferred(seconds)
-
-This technique works equally for external functions which return deferreds,
-or deferreds we have made ourselves.
-
-You can also use ``context.make_deferred_yieldable``, which just does the
-boilerplate for you, so the above could be written:
-
-.. code:: python
-
-    def sleep(seconds):
-        return context.make_deferred_yieldable(get_sleep_deferred(seconds))
-
-
-Fire-and-forget
----------------
-
-Sometimes you want to fire off a chain of execution, but not wait for its
-result. That might look a bit like this:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def do_request_handling():
-        yield foreground_operation()
-
-        # *don't* do this
-        background_operation()
-
-        logger.debug("Request handling complete")
-
-    @defer.inlineCallbacks
-    def background_operation():
-        yield first_background_step()
-        logger.debug("Completed first step")
-        yield second_background_step()
-        logger.debug("Completed second step")
-
-The above code does a couple of steps in the background after
-``do_request_handling`` has finished. The log lines are still logged against
-the ``request_context`` logcontext, which may or may not be desirable. There
-are two big problems with the above, however. The first problem is that, if
-``background_operation`` returns an incomplete Deferred, it will expect its
-caller to ``yield`` immediately, so will have cleared the logcontext. In this
-example, that means that 'Request handling complete' will be logged without any
-context.
-
-The second problem, which is potentially even worse, is that when the Deferred
-returned by ``background_operation`` completes, it will restore the original
-logcontext. There is nothing waiting on that Deferred, so the logcontext will
-leak into the reactor and possibly get attached to some arbitrary future
-operation.
-
-There are two potential solutions to this.
-
-One option is to surround the call to ``background_operation`` with a
-``PreserveLoggingContext`` call. That will reset the logcontext before
-starting ``background_operation`` (so the context restored when the deferred
-completes will be the empty logcontext), and will restore the current
-logcontext before continuing the foreground process:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def do_request_handling():
-        yield foreground_operation()
-
-        # start background_operation off in the empty logcontext, to
-        # avoid leaking the current context into the reactor.
-        with PreserveLoggingContext():
-            background_operation()
-
-        # this will now be logged against the request context
-        logger.debug("Request handling complete")
-
-Obviously that option means that the operations done in
-``background_operation`` would be not be logged against a logcontext (though
-that might be fixed by setting a different logcontext via a ``with
-LoggingContext(...)`` in ``background_operation``).
-
-The second option is to use ``context.run_in_background``, which wraps a
-function so that it doesn't reset the logcontext even when it returns an
-incomplete deferred, and adds a callback to the returned deferred to reset the
-logcontext. In other words, it turns a function that follows the Synapse rules
-about logcontexts and Deferreds into one which behaves more like an external
-function — the opposite operation to that described in the previous section.
-It can be used like this:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def do_request_handling():
-        yield foreground_operation()
-
-        context.run_in_background(background_operation)
-
-        # this will now be logged against the request context
-        logger.debug("Request handling complete")
-
-Passing synapse deferreds into third-party functions
-----------------------------------------------------
-
-A typical example of this is where we want to collect together two or more
-deferred via ``defer.gatherResults``:
-
-.. code:: python
-
-    d1 = operation1()
-    d2 = operation2()
-    d3 = defer.gatherResults([d1, d2])
-
-This is really a variation of the fire-and-forget problem above, in that we are
-firing off ``d1`` and ``d2`` without yielding on them. The difference
-is that we now have third-party code attached to their callbacks. Anyway either
-technique given in the `Fire-and-forget`_ section will work.
-
-Of course, the new Deferred returned by ``gatherResults`` needs to be wrapped
-in order to make it follow the logcontext rules before we can yield it, as
-described in `Where you create a new Deferred, make it follow the rules`_.
-
-So, option one: reset the logcontext before starting the operations to be
-gathered:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def do_request_handling():
-        with PreserveLoggingContext():
-            d1 = operation1()
-            d2 = operation2()
-            result = yield defer.gatherResults([d1, d2])
-
-In this case particularly, though, option two, of using
-``context.preserve_fn`` almost certainly makes more sense, so that
-``operation1`` and ``operation2`` are both logged against the original
-logcontext. This looks like:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def do_request_handling():
-        d1 = context.preserve_fn(operation1)()
-        d2 = context.preserve_fn(operation2)()
-
-        with PreserveLoggingContext():
-            result = yield defer.gatherResults([d1, d2])
-
-
-Was all this really necessary?
-------------------------------
-
-The conventions used work fine for a linear flow where everything happens in
-series via ``defer.inlineCallbacks`` and ``yield``, but are certainly tricky to
-follow for any more exotic flows. It's hard not to wonder if we could have done
-something else.
-
-We're not going to rewrite Synapse now, so the following is entirely of
-academic interest, but I'd like to record some thoughts on an alternative
-approach.
-
-I briefly prototyped some code following an alternative set of rules. I think
-it would work, but I certainly didn't get as far as thinking how it would
-interact with concepts as complicated as the cache descriptors.
-
-My alternative rules were:
-
-* functions always preserve the logcontext of their caller, whether or not they
-  are returning a Deferred.
-
-* Deferreds returned by synapse functions run their callbacks in the same
-  context as the function was orignally called in.
-
-The main point of this scheme is that everywhere that sets the logcontext is
-responsible for clearing it before returning control to the reactor.
-
-So, for example, if you were the function which started a ``with
-LoggingContext`` block, you wouldn't ``yield`` within it — instead you'd start
-off the background process, and then leave the ``with`` block to wait for it:
-
-.. code:: python
-
-    def handle_request(request_id):
-        with context.LoggingContext() as request_context:
-            request_context.request = request_id
-            d = do_request_handling()
-
-        def cb(r):
-            logger.debug("finished")
-
-        d.addCallback(cb)
-        return d
-
-(in general, mixing ``with LoggingContext`` blocks and
-``defer.inlineCallbacks`` in the same function leads to slighly
-counter-intuitive code, under this scheme).
-
-Because we leave the original ``with`` block as soon as the Deferred is
-returned (as opposed to waiting for it to be resolved, as we do today), the
-logcontext is cleared before control passes back to the reactor; so if there is
-some code within ``do_request_handling`` which needs to wait for a Deferred to
-complete, there is no need for it to worry about clearing the logcontext before
-doing so:
-
-.. code:: python
-
-    def handle_request():
-        r = do_some_stuff()
-        r.addCallback(do_some_more_stuff)
-        return r
-
-— and provided ``do_some_stuff`` follows the rules of returning a Deferred which
-runs its callbacks in the original logcontext, all is happy.
-
-The business of a Deferred which runs its callbacks in the original logcontext
-isn't hard to achieve — we have it today, in the shape of
-``context._PreservingContextDeferred``:
-
-.. code:: python
-
-    def do_some_stuff():
-        deferred = do_some_io()
-        pcd = _PreservingContextDeferred(LoggingContext.current_context())
-        deferred.chainDeferred(pcd)
-        return pcd
-
-It turns out that, thanks to the way that Deferreds chain together, we
-automatically get the property of a context-preserving deferred with
-``defer.inlineCallbacks``, provided the final Defered the function ``yields``
-on has that property. So we can just write:
-
-.. code:: python
-
-    @defer.inlineCallbacks
-    def handle_request():
-        yield do_some_stuff()
-        yield do_some_more_stuff()
-
-To conclude: I think this scheme would have worked equally well, with less
-danger of messing it up, and probably made some more esoteric code easier to
-write. But again — changing the conventions of the entire Synapse codebase is
-not a sensible option for the marginal improvement offered.
-
-
-A note on garbage-collection of Deferred chains
------------------------------------------------
-
-It turns out that our logcontext rules do not play nicely with Deferred
-chains which get orphaned and garbage-collected.
-
-Imagine we have some code that looks like this:
-
-.. code:: python
-
-    listener_queue = []
-
-    def on_something_interesting():
-        for d in listener_queue:
-            d.callback("foo")
-
-    @defer.inlineCallbacks
-    def await_something_interesting():
-        new_deferred = defer.Deferred()
-        listener_queue.append(new_deferred)
-
-        with PreserveLoggingContext():
-            yield new_deferred
-
-Obviously, the idea here is that we have a bunch of things which are waiting
-for an event. (It's just an example of the problem here, but a relatively
-common one.)
-
-Now let's imagine two further things happen. First of all, whatever was
-waiting for the interesting thing goes away. (Perhaps the request times out,
-or something *even more* interesting happens.)
-
-Secondly, let's suppose that we decide that the interesting thing is never
-going to happen, and we reset the listener queue:
-
-.. code:: python
-
-    def reset_listener_queue():
-        listener_queue.clear()
-
-So, both ends of the deferred chain have now dropped their references, and the
-deferred chain is now orphaned, and will be garbage-collected at some point.
-Note that ``await_something_interesting`` is a generator function, and when
-Python garbage-collects generator functions, it gives them a chance to clean
-up by making the ``yield`` raise a ``GeneratorExit`` exception. In our case,
-that means that the ``__exit__`` handler of ``PreserveLoggingContext`` will
-carefully restore the request context, but there is now nothing waiting for
-its return, so the request context is never cleared.
-
-To reiterate, this problem only arises when *both* ends of a deferred chain
-are dropped. Dropping the the reference to a deferred you're supposed to be
-calling is probably bad practice, so this doesn't actually happen too much.
-Unfortunately, when it does happen, it will lead to leaked logcontexts which
-are incredibly hard to track down.
diff --git a/docs/media_repository.md b/docs/media_repository.md
new file mode 100644
index 0000000000..1bf8f16f55
--- /dev/null
+++ b/docs/media_repository.md
@@ -0,0 +1,30 @@
+# Media Repository 
+
+*Synapse implementation-specific details for the media repository*
+
+The media repository is where attachments and avatar photos are stored.
+It stores attachment content and thumbnails for media uploaded by local users.
+It caches attachment content and thumbnails for media uploaded by remote users.
+
+## Storage
+
+Each item of media is assigned a `media_id` when it is uploaded.
+The `media_id` is a randomly chosen, URL safe 24 character string.
+
+Metadata such as the MIME type, upload time and length are stored in the
+sqlite3 database indexed by `media_id`.
+
+Content is stored on the filesystem under a `"local_content"` directory.
+
+Thumbnails are stored under a `"local_thumbnails"` directory.
+
+The item with `media_id` `"aabbccccccccdddddddddddd"` is stored under
+`"local_content/aa/bb/ccccccccdddddddddddd"`. Its thumbnail with width
+`128` and height `96` and type `"image/jpeg"` is stored under
+`"local_thumbnails/aa/bb/ccccccccdddddddddddd/128-96-image-jpeg"`
+
+Remote content is cached under `"remote_content"` directory. Each item of
+remote content is assigned a local `"filesystem_id"` to ensure that the
+directory structure `"remote_content/server_name/aa/bb/ccccccccdddddddddddd"`
+is appropriate. Thumbnails for remote content are stored under
+`"remote_thumbnails/server_name/..."`
diff --git a/docs/media_repository.rst b/docs/media_repository.rst
deleted file mode 100644
index 1037b5be63..0000000000
--- a/docs/media_repository.rst
+++ /dev/null
@@ -1,27 +0,0 @@
-Media Repository 
-================
-
-*Synapse implementation-specific details for the media repository*
-
-The media repository is where attachments and avatar photos are stored.
-It stores attachment content and thumbnails for media uploaded by local users.
-It caches attachment content and thumbnails for media uploaded by remote users.
-
-Storage
--------
-
-Each item of media is assigned a ``media_id`` when it is uploaded.
-The ``media_id`` is a randomly chosen, URL safe 24 character string.
-Metadata such as the MIME type, upload time and length are stored in the
-sqlite3 database indexed by ``media_id``.
-Content is stored on the filesystem under a ``"local_content"`` directory.
-Thumbnails are stored under a ``"local_thumbnails"`` directory.
-The item with ``media_id`` ``"aabbccccccccdddddddddddd"`` is stored under
-``"local_content/aa/bb/ccccccccdddddddddddd"``. Its thumbnail with width
-``128`` and height ``96`` and type ``"image/jpeg"`` is stored under
-``"local_thumbnails/aa/bb/ccccccccdddddddddddd/128-96-image-jpeg"``
-Remote content is cached under ``"remote_content"`` directory. Each item of
-remote content is assigned a local "``filesystem_id``" to ensure that the
-directory structure ``"remote_content/server_name/aa/bb/ccccccccdddddddddddd"``
-is appropriate. Thumbnails for remote content are stored under
-``"remote_thumbnails/server_name/..."``
diff --git a/docs/metrics-howto.md b/docs/metrics-howto.md
new file mode 100644
index 0000000000..32abb9f44e
--- /dev/null
+++ b/docs/metrics-howto.md
@@ -0,0 +1,217 @@
+# How to monitor Synapse metrics using Prometheus
+
+1.  Install Prometheus:
+
+    Follow instructions at
+    
+
+1.  Enable Synapse metrics:
+
+    There are two methods of enabling metrics in Synapse.
+
+    The first serves the metrics as a part of the usual web server and
+    can be enabled by adding the \"metrics\" resource to the existing
+    listener as such:
+
+        resources:
+          - names:
+            - client
+            - metrics
+
+    This provides a simple way of adding metrics to your Synapse
+    installation, and serves under `/_synapse/metrics`. If you do not
+    wish your metrics be publicly exposed, you will need to either
+    filter it out at your load balancer, or use the second method.
+
+    The second method runs the metrics server on a different port, in a
+    different thread to Synapse. This can make it more resilient to
+    heavy load meaning metrics cannot be retrieved, and can be exposed
+    to just internal networks easier. The served metrics are available
+    over HTTP only, and will be available at `/`.
+
+    Add a new listener to homeserver.yaml:
+
+        listeners:
+          - type: metrics
+            port: 9000
+            bind_addresses:
+              - '0.0.0.0'
+
+    For both options, you will need to ensure that `enable_metrics` is
+    set to `True`.
+
+1.  Restart Synapse.
+
+1.  Add a Prometheus target for Synapse.
+
+    It needs to set the `metrics_path` to a non-default value (under
+    `scrape_configs`):
+
+        - job_name: "synapse"
+          metrics_path: "/_synapse/metrics"
+          static_configs:
+            - targets: ["my.server.here:port"]
+
+    where `my.server.here` is the IP address of Synapse, and `port` is
+    the listener port configured with the `metrics` resource.
+
+    If your prometheus is older than 1.5.2, you will need to replace
+    `static_configs` in the above with `target_groups`.
+
+1.  Restart Prometheus.
+
+## Renaming of metrics & deprecation of old names in 1.2
+
+Synapse 1.2 updates the Prometheus metrics to match the naming
+convention of the upstream `prometheus_client`. The old names are
+considered deprecated and will be removed in a future version of
+Synapse.
+
+| New Name                                                                     | Old Name                                                               |
+| ---------------------------------------------------------------------------- | ---------------------------------------------------------------------- |
+| python_gc_objects_collected_total                                            | python_gc_objects_collected                                            |
+| python_gc_objects_uncollectable_total                                        | python_gc_objects_uncollectable                                        |
+| python_gc_collections_total                                                  | python_gc_collections                                                  |
+| process_cpu_seconds_total                                                    | process_cpu_seconds                                                    |
+| synapse_federation_client_sent_transactions_total                            | synapse_federation_client_sent_transactions                            |
+| synapse_federation_client_events_processed_total                             | synapse_federation_client_events_processed                             |
+| synapse_event_processing_loop_count_total                                    | synapse_event_processing_loop_count                                    |
+| synapse_event_processing_loop_room_count_total                               | synapse_event_processing_loop_room_count                               |
+| synapse_util_metrics_block_count_total                                       | synapse_util_metrics_block_count                                       |
+| synapse_util_metrics_block_time_seconds_total                                | synapse_util_metrics_block_time_seconds                                |
+| synapse_util_metrics_block_ru_utime_seconds_total                            | synapse_util_metrics_block_ru_utime_seconds                            |
+| synapse_util_metrics_block_ru_stime_seconds_total                            | synapse_util_metrics_block_ru_stime_seconds                            |
+| synapse_util_metrics_block_db_txn_count_total                                | synapse_util_metrics_block_db_txn_count                                |
+| synapse_util_metrics_block_db_txn_duration_seconds_total                     | synapse_util_metrics_block_db_txn_duration_seconds                     |
+| synapse_util_metrics_block_db_sched_duration_seconds_total                   | synapse_util_metrics_block_db_sched_duration_seconds                   |
+| synapse_background_process_start_count_total                                 | synapse_background_process_start_count                                 |
+| synapse_background_process_ru_utime_seconds_total                            | synapse_background_process_ru_utime_seconds                            |
+| synapse_background_process_ru_stime_seconds_total                            | synapse_background_process_ru_stime_seconds                            |
+| synapse_background_process_db_txn_count_total                                | synapse_background_process_db_txn_count                                |
+| synapse_background_process_db_txn_duration_seconds_total                     | synapse_background_process_db_txn_duration_seconds                     |
+| synapse_background_process_db_sched_duration_seconds_total                   | synapse_background_process_db_sched_duration_seconds                   |
+| synapse_storage_events_persisted_events_total                                | synapse_storage_events_persisted_events                                |
+| synapse_storage_events_persisted_events_sep_total                            | synapse_storage_events_persisted_events_sep                            |
+| synapse_storage_events_state_delta_total                                     | synapse_storage_events_state_delta                                     |
+| synapse_storage_events_state_delta_single_event_total                        | synapse_storage_events_state_delta_single_event                        |
+| synapse_storage_events_state_delta_reuse_delta_total                         | synapse_storage_events_state_delta_reuse_delta                         |
+| synapse_federation_server_received_pdus_total                                | synapse_federation_server_received_pdus                                |
+| synapse_federation_server_received_edus_total                                | synapse_federation_server_received_edus                                |
+| synapse_handler_presence_notified_presence_total                             | synapse_handler_presence_notified_presence                             |
+| synapse_handler_presence_federation_presence_out_total                       | synapse_handler_presence_federation_presence_out                       |
+| synapse_handler_presence_presence_updates_total                              | synapse_handler_presence_presence_updates                              |
+| synapse_handler_presence_timers_fired_total                                  | synapse_handler_presence_timers_fired                                  |
+| synapse_handler_presence_federation_presence_total                           | synapse_handler_presence_federation_presence                           |
+| synapse_handler_presence_bump_active_time_total                              | synapse_handler_presence_bump_active_time                              |
+| synapse_federation_client_sent_edus_total                                    | synapse_federation_client_sent_edus                                    |
+| synapse_federation_client_sent_pdu_destinations_count_total                  | synapse_federation_client_sent_pdu_destinations:count                  |
+| synapse_federation_client_sent_pdu_destinations_total                        | synapse_federation_client_sent_pdu_destinations:total                  |
+| synapse_handlers_appservice_events_processed_total                           | synapse_handlers_appservice_events_processed                           |
+| synapse_notifier_notified_events_total                                       | synapse_notifier_notified_events                                       |
+| synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter_total  | synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter  |
+| synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter_total    | synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter    |
+| synapse_http_httppusher_http_pushes_processed_total                          | synapse_http_httppusher_http_pushes_processed                          |
+| synapse_http_httppusher_http_pushes_failed_total                             | synapse_http_httppusher_http_pushes_failed                             |
+| synapse_http_httppusher_badge_updates_processed_total                        | synapse_http_httppusher_badge_updates_processed                        |
+| synapse_http_httppusher_badge_updates_failed_total                           | synapse_http_httppusher_badge_updates_failed                           |
+
+Removal of deprecated metrics & time based counters becoming histograms in 0.31.0
+---------------------------------------------------------------------------------
+
+The duplicated metrics deprecated in Synapse 0.27.0 have been removed.
+
+All time duration-based metrics have been changed to be seconds. This
+affects:
+
+| msec -> sec metrics                    |
+| -------------------------------------- |
+| python_gc_time                         |
+| python_twisted_reactor_tick_time       |
+| synapse_storage_query_time             |
+| synapse_storage_schedule_time          |
+| synapse_storage_transaction_time       |
+
+Several metrics have been changed to be histograms, which sort entries
+into buckets and allow better analysis. The following metrics are now
+histograms:
+
+| Altered metrics                                  |
+| ------------------------------------------------ |
+| python_gc_time                                   |
+| python_twisted_reactor_pending_calls             |
+| python_twisted_reactor_tick_time                 |
+| synapse_http_server_response_time_seconds        |
+| synapse_storage_query_time                       |
+| synapse_storage_schedule_time                    |
+| synapse_storage_transaction_time                 |
+
+Block and response metrics renamed for 0.27.0
+---------------------------------------------
+
+Synapse 0.27.0 begins the process of rationalising the duplicate
+`*:count` metrics reported for the resource tracking for code blocks and
+HTTP requests.
+
+At the same time, the corresponding `*:total` metrics are being renamed,
+as the `:total` suffix no longer makes sense in the absence of a
+corresponding `:count` metric.
+
+To enable a graceful migration path, this release just adds new names
+for the metrics being renamed. A future release will remove the old
+ones.
+
+The following table shows the new metrics, and the old metrics which
+they are replacing.
+
+| New name                                                      | Old name                                                   |
+| ------------------------------------------------------------- | ---------------------------------------------------------- |
+| synapse_util_metrics_block_count                              | synapse_util_metrics_block_timer:count                     |
+| synapse_util_metrics_block_count                              | synapse_util_metrics_block_ru_utime:count                  |
+| synapse_util_metrics_block_count                              | synapse_util_metrics_block_ru_stime:count                  |
+| synapse_util_metrics_block_count                              | synapse_util_metrics_block_db_txn_count:count              |
+| synapse_util_metrics_block_count                              | synapse_util_metrics_block_db_txn_duration:count           |
+| synapse_util_metrics_block_time_seconds                       | synapse_util_metrics_block_timer:total                     |
+| synapse_util_metrics_block_ru_utime_seconds                   | synapse_util_metrics_block_ru_utime:total                  |
+| synapse_util_metrics_block_ru_stime_seconds                   | synapse_util_metrics_block_ru_stime:total                  |
+| synapse_util_metrics_block_db_txn_count                       | synapse_util_metrics_block_db_txn_count:total              |
+| synapse_util_metrics_block_db_txn_duration_seconds            | synapse_util_metrics_block_db_txn_duration:total           |
+| synapse_http_server_response_count                            | synapse_http_server_requests                               |
+| synapse_http_server_response_count                            | synapse_http_server_response_time:count                    |
+| synapse_http_server_response_count                            | synapse_http_server_response_ru_utime:count                |
+| synapse_http_server_response_count                            | synapse_http_server_response_ru_stime:count                |
+| synapse_http_server_response_count                            | synapse_http_server_response_db_txn_count:count            |
+| synapse_http_server_response_count                            | synapse_http_server_response_db_txn_duration:count         |
+| synapse_http_server_response_time_seconds                     | synapse_http_server_response_time:total                    |
+| synapse_http_server_response_ru_utime_seconds                 | synapse_http_server_response_ru_utime:total                |
+| synapse_http_server_response_ru_stime_seconds                 | synapse_http_server_response_ru_stime:total                |
+| synapse_http_server_response_db_txn_count                     | synapse_http_server_response_db_txn_count:total            |
+| synapse_http_server_response_db_txn_duration_seconds          | synapse_http_server_response_db_txn_duration:total         |
+
+Standard Metric Names
+---------------------
+
+As of synapse version 0.18.2, the format of the process-wide metrics has
+been changed to fit prometheus standard naming conventions. Additionally
+the units have been changed to seconds, from miliseconds.
+
+| New name                                 | Old name                          |
+| ---------------------------------------- | --------------------------------- |
+| process_cpu_user_seconds_total           | process_resource_utime / 1000     |
+| process_cpu_system_seconds_total         | process_resource_stime / 1000     |
+| process_open_fds (no \'type\' label)     | process_fds                       |
+
+The python-specific counts of garbage collector performance have been
+renamed.
+
+| New name                         | Old name                   |
+| -------------------------------- | -------------------------- |
+| python_gc_time                   | reactor_gc_time            |
+| python_gc_unreachable_total      | reactor_gc_unreachable     |
+| python_gc_counts                 | reactor_gc_counts          |
+
+The twisted-specific reactor metrics have been renamed.
+
+| New name                               | Old name                |
+| -------------------------------------- | ----------------------- |
+| python_twisted_reactor_pending_calls   | reactor_pending_calls   |
+| python_twisted_reactor_tick_time       | reactor_tick_time       |
diff --git a/docs/metrics-howto.rst b/docs/metrics-howto.rst
deleted file mode 100644
index 973641f3dc..0000000000
--- a/docs/metrics-howto.rst
+++ /dev/null
@@ -1,285 +0,0 @@
-How to monitor Synapse metrics using Prometheus
-===============================================
-
-1. Install Prometheus:
-
-   Follow instructions at http://prometheus.io/docs/introduction/install/
-
-2. Enable Synapse metrics:
-
-   There are two methods of enabling metrics in Synapse.
-
-   The first serves the metrics as a part of the usual web server and can be
-   enabled by adding the "metrics" resource to the existing listener as such::
-
-     resources:
-       - names:
-         - client
-         - metrics
-
-   This provides a simple way of adding metrics to your Synapse installation,
-   and serves under ``/_synapse/metrics``. If you do not wish your metrics be
-   publicly exposed, you will need to either filter it out at your load
-   balancer, or use the second method.
-
-   The second method runs the metrics server on a different port, in a
-   different thread to Synapse. This can make it more resilient to heavy load
-   meaning metrics cannot be retrieved, and can be exposed to just internal
-   networks easier. The served metrics are available over HTTP only, and will
-   be available at ``/``.
-
-   Add a new listener to homeserver.yaml::
-
-     listeners:
-       - type: metrics
-         port: 9000
-         bind_addresses:
-           - '0.0.0.0'
-
-   For both options, you will need to ensure that ``enable_metrics`` is set to
-   ``True``.
-
-   Restart Synapse.
-
-3. Add a Prometheus target for Synapse.
-
-   It needs to set the ``metrics_path`` to a non-default value (under ``scrape_configs``)::
-
-    - job_name: "synapse"
-      metrics_path: "/_synapse/metrics"
-      static_configs:
-        - targets: ["my.server.here:port"]
-
-   where ``my.server.here`` is the IP address of Synapse, and ``port`` is the listener port
-   configured with the ``metrics`` resource.
-
-   If your prometheus is older than 1.5.2, you will need to replace
-   ``static_configs`` in the above with ``target_groups``.
-
-   Restart Prometheus.
-
-
-Renaming of metrics & deprecation of old names in 1.2
------------------------------------------------------
-
-Synapse 1.2 updates the Prometheus metrics to match the naming convention of the
-upstream ``prometheus_client``. The old names are considered deprecated and will
-be removed in a future version of Synapse.
-
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-|                                  New Name                                   |                               Old Name                                |
-+=============================================================================+=======================================================================+
-| python_gc_objects_collected_total                                           | python_gc_objects_collected                                           |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| python_gc_objects_uncollectable_total                                       | python_gc_objects_uncollectable                                       |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| python_gc_collections_total                                                 | python_gc_collections                                                 |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| process_cpu_seconds_total                                                   | process_cpu_seconds                                                   |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_federation_client_sent_transactions_total                           | synapse_federation_client_sent_transactions                           |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_federation_client_events_processed_total                            | synapse_federation_client_events_processed                            |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_event_processing_loop_count_total                                   | synapse_event_processing_loop_count                                   |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_event_processing_loop_room_count_total                              | synapse_event_processing_loop_room_count                              |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_util_metrics_block_count_total                                      | synapse_util_metrics_block_count                                      |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_util_metrics_block_time_seconds_total                               | synapse_util_metrics_block_time_seconds                               |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_util_metrics_block_ru_utime_seconds_total                           | synapse_util_metrics_block_ru_utime_seconds                           |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_util_metrics_block_ru_stime_seconds_total                           | synapse_util_metrics_block_ru_stime_seconds                           |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_util_metrics_block_db_txn_count_total                               | synapse_util_metrics_block_db_txn_count                               |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_util_metrics_block_db_txn_duration_seconds_total                    | synapse_util_metrics_block_db_txn_duration_seconds                    |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_util_metrics_block_db_sched_duration_seconds_total                  | synapse_util_metrics_block_db_sched_duration_seconds                  |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_background_process_start_count_total                                | synapse_background_process_start_count                                |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_background_process_ru_utime_seconds_total                           | synapse_background_process_ru_utime_seconds                           |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_background_process_ru_stime_seconds_total                           | synapse_background_process_ru_stime_seconds                           |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_background_process_db_txn_count_total                               | synapse_background_process_db_txn_count                               |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_background_process_db_txn_duration_seconds_total                    | synapse_background_process_db_txn_duration_seconds                    |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_background_process_db_sched_duration_seconds_total                  | synapse_background_process_db_sched_duration_seconds                  |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_storage_events_persisted_events_total                               | synapse_storage_events_persisted_events                               |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_storage_events_persisted_events_sep_total                           | synapse_storage_events_persisted_events_sep                           |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_storage_events_state_delta_total                                    | synapse_storage_events_state_delta                                    |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_storage_events_state_delta_single_event_total                       | synapse_storage_events_state_delta_single_event                       |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_storage_events_state_delta_reuse_delta_total                        | synapse_storage_events_state_delta_reuse_delta                        |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_federation_server_received_pdus_total                               | synapse_federation_server_received_pdus                               |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_federation_server_received_edus_total                               | synapse_federation_server_received_edus                               |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_handler_presence_notified_presence_total                            | synapse_handler_presence_notified_presence                            |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_handler_presence_federation_presence_out_total                      | synapse_handler_presence_federation_presence_out                      |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_handler_presence_presence_updates_total                             | synapse_handler_presence_presence_updates                             |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_handler_presence_timers_fired_total                                 | synapse_handler_presence_timers_fired                                 |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_handler_presence_federation_presence_total                          | synapse_handler_presence_federation_presence                          |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_handler_presence_bump_active_time_total                             | synapse_handler_presence_bump_active_time                             |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_federation_client_sent_edus_total                                   | synapse_federation_client_sent_edus                                   |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_federation_client_sent_pdu_destinations_count_total                 | synapse_federation_client_sent_pdu_destinations:count                 |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_federation_client_sent_pdu_destinations_total                       | synapse_federation_client_sent_pdu_destinations:total                 |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_handlers_appservice_events_processed_total                          | synapse_handlers_appservice_events_processed                          |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_notifier_notified_events_total                                      | synapse_notifier_notified_events                                      |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter_total | synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter_total   | synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter   |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_http_httppusher_http_pushes_processed_total                         | synapse_http_httppusher_http_pushes_processed                         |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_http_httppusher_http_pushes_failed_total                            | synapse_http_httppusher_http_pushes_failed                            |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_http_httppusher_badge_updates_processed_total                       | synapse_http_httppusher_badge_updates_processed                       |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-| synapse_http_httppusher_badge_updates_failed_total                          | synapse_http_httppusher_badge_updates_failed                          |
-+-----------------------------------------------------------------------------+-----------------------------------------------------------------------+
-
-
-Removal of deprecated metrics & time based counters becoming histograms in 0.31.0
----------------------------------------------------------------------------------
-
-The duplicated metrics deprecated in Synapse 0.27.0 have been removed.
-
-All time duration-based metrics have been changed to be seconds. This affects:
-
-+----------------------------------+
-| msec -> sec metrics              |
-+==================================+
-| python_gc_time                   |
-+----------------------------------+
-| python_twisted_reactor_tick_time |
-+----------------------------------+
-| synapse_storage_query_time       |
-+----------------------------------+
-| synapse_storage_schedule_time    |
-+----------------------------------+
-| synapse_storage_transaction_time |
-+----------------------------------+
-
-Several metrics have been changed to be histograms, which sort entries into
-buckets and allow better analysis. The following metrics are now histograms:
-
-+-------------------------------------------+
-| Altered metrics                           |
-+===========================================+
-| python_gc_time                            |
-+-------------------------------------------+
-| python_twisted_reactor_pending_calls      |
-+-------------------------------------------+
-| python_twisted_reactor_tick_time          |
-+-------------------------------------------+
-| synapse_http_server_response_time_seconds |
-+-------------------------------------------+
-| synapse_storage_query_time                |
-+-------------------------------------------+
-| synapse_storage_schedule_time             |
-+-------------------------------------------+
-| synapse_storage_transaction_time          |
-+-------------------------------------------+
-
-
-Block and response metrics renamed for 0.27.0
----------------------------------------------
-
-Synapse 0.27.0 begins the process of rationalising the duplicate ``*:count``
-metrics reported for the resource tracking for code blocks and HTTP requests.
-
-At the same time, the corresponding ``*:total`` metrics are being renamed, as
-the ``:total`` suffix no longer makes sense in the absence of a corresponding
-``:count`` metric.
-
-To enable a graceful migration path, this release just adds new names for the
-metrics being renamed. A future release will remove the old ones.
-
-The following table shows the new metrics, and the old metrics which they are
-replacing.
-
-==================================================== ===================================================
-New name                                             Old name
-==================================================== ===================================================
-synapse_util_metrics_block_count                     synapse_util_metrics_block_timer:count
-synapse_util_metrics_block_count                     synapse_util_metrics_block_ru_utime:count
-synapse_util_metrics_block_count                     synapse_util_metrics_block_ru_stime:count
-synapse_util_metrics_block_count                     synapse_util_metrics_block_db_txn_count:count
-synapse_util_metrics_block_count                     synapse_util_metrics_block_db_txn_duration:count
-
-synapse_util_metrics_block_time_seconds              synapse_util_metrics_block_timer:total
-synapse_util_metrics_block_ru_utime_seconds          synapse_util_metrics_block_ru_utime:total
-synapse_util_metrics_block_ru_stime_seconds          synapse_util_metrics_block_ru_stime:total
-synapse_util_metrics_block_db_txn_count              synapse_util_metrics_block_db_txn_count:total
-synapse_util_metrics_block_db_txn_duration_seconds   synapse_util_metrics_block_db_txn_duration:total
-
-synapse_http_server_response_count                   synapse_http_server_requests
-synapse_http_server_response_count                   synapse_http_server_response_time:count
-synapse_http_server_response_count                   synapse_http_server_response_ru_utime:count
-synapse_http_server_response_count                   synapse_http_server_response_ru_stime:count
-synapse_http_server_response_count                   synapse_http_server_response_db_txn_count:count
-synapse_http_server_response_count                   synapse_http_server_response_db_txn_duration:count
-
-synapse_http_server_response_time_seconds            synapse_http_server_response_time:total
-synapse_http_server_response_ru_utime_seconds        synapse_http_server_response_ru_utime:total
-synapse_http_server_response_ru_stime_seconds        synapse_http_server_response_ru_stime:total
-synapse_http_server_response_db_txn_count            synapse_http_server_response_db_txn_count:total
-synapse_http_server_response_db_txn_duration_seconds synapse_http_server_response_db_txn_duration:total
-==================================================== ===================================================
-
-
-Standard Metric Names
----------------------
-
-As of synapse version 0.18.2, the format of the process-wide metrics has been
-changed to fit prometheus standard naming conventions. Additionally the units
-have been changed to seconds, from miliseconds.
-
-================================== =============================
-New name                           Old name
-================================== =============================
-process_cpu_user_seconds_total     process_resource_utime / 1000
-process_cpu_system_seconds_total   process_resource_stime / 1000
-process_open_fds (no 'type' label) process_fds
-================================== =============================
-
-The python-specific counts of garbage collector performance have been renamed.
-
-=========================== ======================
-New name                    Old name
-=========================== ======================
-python_gc_time              reactor_gc_time
-python_gc_unreachable_total reactor_gc_unreachable
-python_gc_counts            reactor_gc_counts
-=========================== ======================
-
-The twisted-specific reactor metrics have been renamed.
-
-==================================== =====================
-New name                             Old name
-==================================== =====================
-python_twisted_reactor_pending_calls reactor_pending_calls
-python_twisted_reactor_tick_time     reactor_tick_time
-==================================== =====================
diff --git a/docs/opentracing.md b/docs/opentracing.md
new file mode 100644
index 0000000000..4c7a56a5d7
--- /dev/null
+++ b/docs/opentracing.md
@@ -0,0 +1,93 @@
+# OpenTracing
+
+## Background
+
+OpenTracing is a semi-standard being adopted by a number of distributed
+tracing platforms. It is a common api for facilitating vendor-agnostic
+tracing instrumentation. That is, we can use the OpenTracing api and
+select one of a number of tracer implementations to do the heavy lifting
+in the background. Our current selected implementation is Jaeger.
+
+OpenTracing is a tool which gives an insight into the causal
+relationship of work done in and between servers. The servers each track
+events and report them to a centralised server - in Synapse's case:
+Jaeger. The basic unit used to represent events is the span. The span
+roughly represents a single piece of work that was done and the time at
+which it occurred. A span can have child spans, meaning that the work of
+the child had to be completed for the parent span to complete, or it can
+have follow-on spans which represent work that is undertaken as a result
+of the parent but is not depended on by the parent to in order to
+finish.
+
+Since this is undertaken in a distributed environment a request to
+another server, such as an RPC or a simple GET, can be considered a span
+(a unit or work) for the local server. This causal link is what
+OpenTracing aims to capture and visualise. In order to do this metadata
+about the local server's span, i.e the 'span context', needs to be
+included with the request to the remote.
+
+It is up to the remote server to decide what it does with the spans it
+creates. This is called the sampling policy and it can be configured
+through Jaeger's settings.
+
+For OpenTracing concepts see
+.
+
+For more information about Jaeger's implementation see
+
+
+## Setting up OpenTracing
+
+To receive OpenTracing spans, start up a Jaeger server. This can be done
+using docker like so:
+
+```sh
+docker run -d --name jaeger
+  -p 6831:6831/udp \
+  -p 6832:6832/udp \
+  -p 5778:5778 \
+  -p 16686:16686 \
+  -p 14268:14268 \
+  jaegertracing/all-in-one:1.13
+```
+
+Latest documentation is probably at
+
+
+## Enable OpenTracing in Synapse
+
+OpenTracing is not enabled by default. It must be enabled in the
+homeserver config by uncommenting the config options under `opentracing`
+as shown in the [sample config](./sample_config.yaml). For example:
+
+```yaml
+opentracing:
+  tracer_enabled: true
+  homeserver_whitelist:
+    - "mytrustedhomeserver.org"
+    - "*.myotherhomeservers.com"
+```
+
+## Homeserver whitelisting
+
+The homeserver whitelist is configured using regular expressions. A list
+of regular expressions can be given and their union will be compared
+when propagating any spans contexts to another homeserver.
+
+Though it's mostly safe to send and receive span contexts to and from
+untrusted users since span contexts are usually opaque ids it can lead
+to two problems, namely:
+
+-   If the span context is marked as sampled by the sending homeserver
+    the receiver will sample it. Therefore two homeservers with wildly
+    different sampling policies could incur higher sampling counts than
+    intended.
+-   Sending servers can attach arbitrary data to spans, known as
+    'baggage'. For safety this has been disabled in Synapse but that
+    doesn't prevent another server sending you baggage which will be
+    logged to OpenTracing's logs.
+
+## Configuring Jaeger
+
+Sampling strategies can be set as in this document:
+
diff --git a/docs/opentracing.rst b/docs/opentracing.rst
deleted file mode 100644
index 6e98ab56ba..0000000000
--- a/docs/opentracing.rst
+++ /dev/null
@@ -1,123 +0,0 @@
-===========
-OpenTracing
-===========
-
-Background
-----------
-
-OpenTracing is a semi-standard being adopted by a number of distributed tracing
-platforms. It is a common api for facilitating vendor-agnostic tracing
-instrumentation. That is, we can use the OpenTracing api and select one of a
-number of tracer implementations to do the heavy lifting in the background.
-Our current selected implementation is Jaeger.
-
-OpenTracing is a tool which gives an insight into the causal relationship of
-work done in and between servers. The servers each track events and report them
-to a centralised server - in Synapse's case: Jaeger. The basic unit used to
-represent events is the span. The span roughly represents a single piece of work
-that was done and the time at which it occurred. A span can have child spans,
-meaning that the work of the child had to be completed for the parent span to
-complete, or it can have follow-on spans which represent work that is undertaken
-as a result of the parent but is not depended on by the parent to in order to
-finish.
-
-Since this is undertaken in a distributed environment a request to another
-server, such as an RPC or a simple GET, can be considered a span (a unit or
-work) for the local server. This causal link is what OpenTracing aims to
-capture and visualise. In order to do this metadata about the local server's
-span, i.e the 'span context', needs to be included with the request to the
-remote.
-
-It is up to the remote server to decide what it does with the spans
-it creates. This is called the sampling policy and it can be configured
-through Jaeger's settings.
-
-For OpenTracing concepts see
-https://opentracing.io/docs/overview/what-is-tracing/.
-
-For more information about Jaeger's implementation see
-https://www.jaegertracing.io/docs/
-
-=====================
-Seting up OpenTracing
-=====================
-
-To receive OpenTracing spans, start up a Jaeger server. This can be done
-using docker like so:
-
-.. code-block:: bash
-
-   docker run -d --name jaeger
-     -p 6831:6831/udp \
-     -p 6832:6832/udp \
-     -p 5778:5778 \
-     -p 16686:16686 \
-     -p 14268:14268 \
-     jaegertracing/all-in-one:1.13
-
-Latest documentation is probably at
-https://www.jaegertracing.io/docs/1.13/getting-started/
-
-
-Enable OpenTracing in Synapse
------------------------------
-
-OpenTracing is not enabled by default. It must be enabled in the homeserver
-config by uncommenting the config options under ``opentracing`` as shown in
-the `sample config <./sample_config.yaml>`_. For example:
-
-.. code-block:: yaml
-
-  opentracing:
-    tracer_enabled: true
-    homeserver_whitelist:
-      - "mytrustedhomeserver.org"
-      - "*.myotherhomeservers.com"
-
-Homeserver whitelisting
------------------------
-
-The homeserver whitelist is configured using regular expressions. A list of regular
-expressions can be given and their union will be compared when propagating any
-spans contexts to another homeserver.
-
-Though it's mostly safe to send and receive span contexts to and from
-untrusted users since span contexts are usually opaque ids it can lead to
-two problems, namely:
-
-- If the span context is marked as sampled by the sending homeserver the receiver will
-  sample it. Therefore two homeservers with wildly different sampling policies
-  could incur higher sampling counts than intended.
-- Sending servers can attach arbitrary data to spans, known as 'baggage'. For safety this has been disabled in Synapse
-  but that doesn't prevent another server sending you baggage which will be logged
-  to OpenTracing's logs.
-
-==========
-EDU FORMAT
-==========
-
-EDUs can contain tracing data in their content. This is not specced but
-it could be of interest for other homeservers.
-
-EDU format (if you're using jaeger):
-
-.. code-block:: json
-
-   {
-     "edu_type": "type",
-     "content": {
-       "org.matrix.opentracing_context": {
-         "uber-trace-id": "fe57cf3e65083289"
-       }
-     }
-   }
-
-Though you don't have to use jaeger you must inject the span context into
-`org.matrix.opentracing_context` using the opentracing `Format.TEXT_MAP` inject method.
-
-==================
-Configuring Jaeger
-==================
-
-Sampling strategies can be set as in this document:
-https://www.jaegertracing.io/docs/1.13/sampling/
diff --git a/docs/password_auth_providers.md b/docs/password_auth_providers.md
new file mode 100644
index 0000000000..0db1a3804a
--- /dev/null
+++ b/docs/password_auth_providers.md
@@ -0,0 +1,116 @@
+# Password auth provider modules
+
+Password auth providers offer a way for server administrators to
+integrate their Synapse installation with an existing authentication
+system.
+
+A password auth provider is a Python class which is dynamically loaded
+into Synapse, and provides a number of methods by which it can integrate
+with the authentication system.
+
+This document serves as a reference for those looking to implement their
+own password auth providers.
+
+## Required methods
+
+Password auth provider classes must provide the following methods:
+
+*class* `SomeProvider.parse_config`(*config*)
+
+> This method is passed the `config` object for this module from the
+> homeserver configuration file.
+>
+> It should perform any appropriate sanity checks on the provided
+> configuration, and return an object which is then passed into
+> `__init__`.
+
+*class* `SomeProvider`(*config*, *account_handler*)
+
+> The constructor is passed the config object returned by
+> `parse_config`, and a `synapse.module_api.ModuleApi` object which
+> allows the password provider to check if accounts exist and/or create
+> new ones.
+
+## Optional methods
+
+Password auth provider classes may optionally provide the following
+methods.
+
+*class* `SomeProvider.get_db_schema_files`()
+
+> This method, if implemented, should return an Iterable of
+> `(name, stream)` pairs of database schema files. Each file is applied
+> in turn at initialisation, and a record is then made in the database
+> so that it is not re-applied on the next start.
+
+`someprovider.get_supported_login_types`()
+
+> This method, if implemented, should return a `dict` mapping from a
+> login type identifier (such as `m.login.password`) to an iterable
+> giving the fields which must be provided by the user in the submission
+> to the `/login` api. These fields are passed in the `login_dict`
+> dictionary to `check_auth`.
+>
+> For example, if a password auth provider wants to implement a custom
+> login type of `com.example.custom_login`, where the client is expected
+> to pass the fields `secret1` and `secret2`, the provider should
+> implement this method and return the following dict:
+>
+>     {"com.example.custom_login": ("secret1", "secret2")}
+
+`someprovider.check_auth`(*username*, *login_type*, *login_dict*)
+
+> This method is the one that does the real work. If implemented, it
+> will be called for each login attempt where the login type matches one
+> of the keys returned by `get_supported_login_types`.
+>
+> It is passed the (possibly UNqualified) `user` provided by the client,
+> the login type, and a dictionary of login secrets passed by the
+> client.
+>
+> The method should return a Twisted `Deferred` object, which resolves
+> to the canonical `@localpart:domain` user id if authentication is
+> successful, and `None` if not.
+>
+> Alternatively, the `Deferred` can resolve to a `(str, func)` tuple, in
+> which case the second field is a callback which will be called with
+> the result from the `/login` call (including `access_token`,
+> `device_id`, etc.)
+
+`someprovider.check_3pid_auth`(*medium*, *address*, *password*)
+
+> This method, if implemented, is called when a user attempts to
+> register or log in with a third party identifier, such as email. It is
+> passed the medium (ex. "email"), an address (ex.
+> "") and the user's password.
+>
+> The method should return a Twisted `Deferred` object, which resolves
+> to a `str` containing the user's (canonical) User ID if
+> authentication was successful, and `None` if not.
+>
+> As with `check_auth`, the `Deferred` may alternatively resolve to a
+> `(user_id, callback)` tuple.
+
+`someprovider.check_password`(*user_id*, *password*)
+
+> This method provides a simpler interface than
+> `get_supported_login_types` and `check_auth` for password auth
+> providers that just want to provide a mechanism for validating
+> `m.login.password` logins.
+>
+> Iif implemented, it will be called to check logins with an
+> `m.login.password` login type. It is passed a qualified
+> `@localpart:domain` user id, and the password provided by the user.
+>
+> The method should return a Twisted `Deferred` object, which resolves
+> to `True` if authentication is successful, and `False` if not.
+
+`someprovider.on_logged_out`(*user_id*, *device_id*, *access_token*)
+
+> This method, if implemented, is called when a user logs out. It is
+> passed the qualified user ID, the ID of the deactivated device (if
+> any: access tokens are occasionally created without an associated
+> device ID), and the (now deactivated) access token.
+>
+> It may return a Twisted `Deferred` object; the logout request will
+> wait for the deferred to complete but the result is ignored.
diff --git a/docs/password_auth_providers.rst b/docs/password_auth_providers.rst
deleted file mode 100644
index 6149ba7458..0000000000
--- a/docs/password_auth_providers.rst
+++ /dev/null
@@ -1,113 +0,0 @@
-Password auth provider modules
-==============================
-
-Password auth providers offer a way for server administrators to integrate
-their Synapse installation with an existing authentication system.
-
-A password auth provider is a Python class which is dynamically loaded into
-Synapse, and provides a number of methods by which it can integrate with the
-authentication system.
-
-This document serves as a reference for those looking to implement their own
-password auth providers.
-
-Required methods
-----------------
-
-Password auth provider classes must provide the following methods:
-
-*class* ``SomeProvider.parse_config``\(*config*)
-
-    This method is passed the ``config`` object for this module from the
-    homeserver configuration file.
-
-    It should perform any appropriate sanity checks on the provided
-    configuration, and return an object which is then passed into ``__init__``.
-
-*class* ``SomeProvider``\(*config*, *account_handler*)
-
-    The constructor is passed the config object returned by ``parse_config``,
-    and a ``synapse.module_api.ModuleApi`` object which allows the
-    password provider to check if accounts exist and/or create new ones.
-
-Optional methods
-----------------
-
-Password auth provider classes may optionally provide the following methods.
-
-*class* ``SomeProvider.get_db_schema_files``\()
-
-    This method, if implemented, should return an Iterable of ``(name,
-    stream)`` pairs of database schema files. Each file is applied in turn at
-    initialisation, and a record is then made in the database so that it is
-    not re-applied on the next start.
-
-``someprovider.get_supported_login_types``\()
-
-    This method, if implemented, should return a ``dict`` mapping from a login
-    type identifier (such as ``m.login.password``) to an iterable giving the
-    fields which must be provided by the user in the submission to the
-    ``/login`` api. These fields are passed in the ``login_dict`` dictionary
-    to ``check_auth``.
-
-    For example, if a password auth provider wants to implement a custom login
-    type of ``com.example.custom_login``, where the client is expected to pass
-    the fields ``secret1`` and ``secret2``, the provider should implement this
-    method and return the following dict::
-
-      {"com.example.custom_login": ("secret1", "secret2")}
-
-``someprovider.check_auth``\(*username*, *login_type*, *login_dict*)
-
-    This method is the one that does the real work. If implemented, it will be
-    called for each login attempt where the login type matches one of the keys
-    returned by ``get_supported_login_types``.
-
-    It is passed the (possibly UNqualified) ``user`` provided by the client,
-    the login type, and a dictionary of login secrets passed by the client.
-
-    The method should return a Twisted ``Deferred`` object, which resolves to
-    the canonical ``@localpart:domain`` user id if authentication is successful,
-    and ``None`` if not.
-
-    Alternatively, the ``Deferred`` can resolve to a ``(str, func)`` tuple, in
-    which case the second field is a callback which will be called with the
-    result from the ``/login`` call (including ``access_token``, ``device_id``,
-    etc.)
-
-``someprovider.check_3pid_auth``\(*medium*, *address*, *password*)
-
-    This method, if implemented, is called when a user attempts to register or
-    log in with a third party identifier, such as email. It is passed the
-    medium (ex. "email"), an address (ex. "jdoe@example.com") and the user's
-    password.
-
-    The method should return a Twisted ``Deferred`` object, which resolves to
-    a ``str`` containing the user's (canonical) User ID if authentication was
-    successful, and ``None`` if not.
-
-    As with ``check_auth``, the ``Deferred`` may alternatively resolve to a
-    ``(user_id, callback)`` tuple.
-
-``someprovider.check_password``\(*user_id*, *password*)
-
-    This method provides a simpler interface than ``get_supported_login_types``
-    and ``check_auth`` for password auth providers that just want to provide a
-    mechanism for validating ``m.login.password`` logins.
-
-    Iif implemented, it will be called to check logins with an
-    ``m.login.password`` login type. It is passed a qualified
-    ``@localpart:domain`` user id, and the password provided by the user.
-
-    The method should return a Twisted ``Deferred`` object, which resolves to
-    ``True`` if authentication is successful, and ``False`` if not.
-
-``someprovider.on_logged_out``\(*user_id*, *device_id*, *access_token*)
-
-    This method, if implemented, is called when a user logs out. It is passed
-    the qualified user ID, the ID of the deactivated device (if any: access
-    tokens are occasionally created without an associated device ID), and the
-    (now deactivated) access token.
-
-    It may return a Twisted ``Deferred`` object; the logout request will wait
-    for the deferred to complete but the result is ignored.
diff --git a/docs/postgres.md b/docs/postgres.md
new file mode 100644
index 0000000000..7cb1ad18d4
--- /dev/null
+++ b/docs/postgres.md
@@ -0,0 +1,168 @@
+# Using Postgres
+
+Postgres version 9.5 or later is known to work.
+
+## Install postgres client libraries
+
+Synapse will require the python postgres client library in order to
+connect to a postgres database.
+
+-   If you are using the [matrix.org debian/ubuntu
+    packages](../INSTALL.md#matrixorg-packages), the necessary python
+    library will already be installed, but you will need to ensure the
+    low-level postgres library is installed, which you can do with
+    `apt install libpq5`.
+-   For other pre-built packages, please consult the documentation from
+    the relevant package.
+-   If you installed synapse [in a
+    virtualenv](../INSTALL.md#installing-from-source), you can install
+    the library with:
+
+        ~/synapse/env/bin/pip install matrix-synapse[postgres]
+
+    (substituting the path to your virtualenv for `~/synapse/env`, if
+    you used a different path). You will require the postgres
+    development files. These are in the `libpq-dev` package on
+    Debian-derived distributions.
+
+## Set up database
+
+Assuming your PostgreSQL database user is called `postgres`, first authenticate as the database user with:
+
+    su - postgres
+    # Or, if your system uses sudo to get administrative rights
+    sudo -u postgres bash
+  
+Then, create a user ``synapse_user`` with:
+
+    createuser --pwprompt synapse_user
+
+Before you can authenticate with the `synapse_user`, you must create a
+database that it can access. To create a database, first connect to the
+database with your database user:
+
+    su - postgres # Or: sudo -u postgres bash
+    psql
+
+and then run:
+
+    CREATE DATABASE synapse
+     ENCODING 'UTF8'
+     LC_COLLATE='C'
+     LC_CTYPE='C'
+     template=template0
+     OWNER synapse_user;
+
+This would create an appropriate database named `synapse` owned by the
+`synapse_user` user (which must already have been created as above).
+
+Note that the PostgreSQL database *must* have the correct encoding set
+(as shown above), otherwise it will not be able to store UTF8 strings.
+
+You may need to enable password authentication so `synapse_user` can
+connect to the database. See
+.
+
+## Tuning Postgres
+
+The default settings should be fine for most deployments. For larger
+scale deployments tuning some of the settings is recommended, details of
+which can be found at
+.
+
+In particular, we've found tuning the following values helpful for
+performance:
+
+-   `shared_buffers`
+-   `effective_cache_size`
+-   `work_mem`
+-   `maintenance_work_mem`
+-   `autovacuum_work_mem`
+
+Note that the appropriate values for those fields depend on the amount
+of free memory the database host has available.
+
+## Synapse config
+
+When you are ready to start using PostgreSQL, edit the `database`
+section in your config file to match the following lines:
+
+    database:
+        name: psycopg2
+        args:
+            user: 
+            password: 
+            database: 
+            host: 
+            cp_min: 5
+            cp_max: 10
+
+All key, values in `args` are passed to the `psycopg2.connect(..)`
+function, except keys beginning with `cp_`, which are consumed by the
+twisted adbapi connection pool.
+
+## Porting from SQLite
+
+### Overview
+
+The script `synapse_port_db` allows porting an existing synapse server
+backed by SQLite to using PostgreSQL. This is done in as a two phase
+process:
+
+1.  Copy the existing SQLite database to a separate location (while the
+    server is down) and running the port script against that offline
+    database.
+2.  Shut down the server. Rerun the port script to port any data that
+    has come in since taking the first snapshot. Restart server against
+    the PostgreSQL database.
+
+The port script is designed to be run repeatedly against newer snapshots
+of the SQLite database file. This makes it safe to repeat step 1 if
+there was a delay between taking the previous snapshot and being ready
+to do step 2.
+
+It is safe to at any time kill the port script and restart it.
+
+### Using the port script
+
+Firstly, shut down the currently running synapse server and copy its
+database file (typically `homeserver.db`) to another location. Once the
+copy is complete, restart synapse. For instance:
+
+    ./synctl stop
+    cp homeserver.db homeserver.db.snapshot
+    ./synctl start
+
+Copy the old config file into a new config file:
+
+    cp homeserver.yaml homeserver-postgres.yaml
+
+Edit the database section as described in the section *Synapse config*
+above and with the SQLite snapshot located at `homeserver.db.snapshot`
+simply run:
+
+    synapse_port_db --sqlite-database homeserver.db.snapshot \
+        --postgres-config homeserver-postgres.yaml
+
+The flag `--curses` displays a coloured curses progress UI.
+
+If the script took a long time to complete, or time has otherwise passed
+since the original snapshot was taken, repeat the previous steps with a
+newer snapshot.
+
+To complete the conversion shut down the synapse server and run the port
+script one last time, e.g. if the SQLite database is at `homeserver.db`
+run:
+
+    synapse_port_db --sqlite-database homeserver.db \
+        --postgres-config homeserver-postgres.yaml
+
+Once that has completed, change the synapse config to point at the
+PostgreSQL database configuration file `homeserver-postgres.yaml`:
+
+    ./synctl stop
+    mv homeserver.yaml homeserver-old-sqlite.yaml
+    mv homeserver-postgres.yaml homeserver.yaml
+    ./synctl start
+
+Synapse should now be running against PostgreSQL.
diff --git a/docs/postgres.rst b/docs/postgres.rst
deleted file mode 100644
index e08a5116b9..0000000000
--- a/docs/postgres.rst
+++ /dev/null
@@ -1,166 +0,0 @@
-Using Postgres
---------------
-
-Postgres version 9.5 or later is known to work.
-
-Install postgres client libraries
-=================================
-
-Synapse will require the python postgres client library in order to connect to
-a postgres database.
-
-* If you are using the `matrix.org debian/ubuntu
-  packages <../INSTALL.md#matrixorg-packages>`_,
-  the necessary python library will already be installed, but you will need to
-  ensure the low-level postgres library is installed, which you can do with
-  ``apt install libpq5``.
-
-* For other pre-built packages, please consult the documentation from the
-  relevant package.
-
-* If you installed synapse `in a virtualenv
-  <../INSTALL.md#installing-from-source>`_, you can install the library with::
-
-      ~/synapse/env/bin/pip install matrix-synapse[postgres]
-
-  (substituting the path to your virtualenv for ``~/synapse/env``, if you used a
-  different path). You will require the postgres development files. These are in
-  the ``libpq-dev`` package on Debian-derived distributions.
-
-Set up database
-===============
-
-Assuming your PostgreSQL database user is called ``postgres``, create a user
-``synapse_user`` with::
-
-   su - postgres
-   createuser --pwprompt synapse_user
-
-Before you can authenticate with the ``synapse_user``, you must create a
-database that it can access. To create a database, first connect to the database
-with your database user::
-
-   su - postgres
-   psql
-
-and then run::
-
-   CREATE DATABASE synapse
-    ENCODING 'UTF8'
-    LC_COLLATE='C'
-    LC_CTYPE='C'
-    template=template0
-    OWNER synapse_user;
-
-This would create an appropriate database named ``synapse`` owned by the
-``synapse_user`` user (which must already have been created as above).
-
-Note that the PostgreSQL database *must* have the correct encoding set (as
-shown above), otherwise it will not be able to store UTF8 strings.
-
-You may need to enable password authentication so ``synapse_user`` can connect
-to the database. See https://www.postgresql.org/docs/11/auth-pg-hba-conf.html.
-
-Tuning Postgres
-===============
-
-The default settings should be fine for most deployments. For larger scale
-deployments tuning some of the settings is recommended, details of which can be
-found at https://wiki.postgresql.org/wiki/Tuning_Your_PostgreSQL_Server.
-
-In particular, we've found tuning the following values helpful for performance:
-
-- ``shared_buffers``
-- ``effective_cache_size``
-- ``work_mem``
-- ``maintenance_work_mem``
-- ``autovacuum_work_mem``
-
-Note that the appropriate values for those fields depend on the amount of free
-memory the database host has available.
-
-Synapse config
-==============
-
-When you are ready to start using PostgreSQL, edit the ``database`` section in
-your config file to match the following lines::
-
-    database:
-        name: psycopg2
-        args:
-            user: 
-            password: 
-            database: 
-            host: 
-            cp_min: 5
-            cp_max: 10
-
-All key, values in ``args`` are passed to the ``psycopg2.connect(..)``
-function, except keys beginning with ``cp_``, which are consumed by the twisted
-adbapi connection pool.
-
-
-Porting from SQLite
-===================
-
-Overview
-~~~~~~~~
-
-The script ``synapse_port_db`` allows porting an existing synapse server
-backed by SQLite to using PostgreSQL. This is done in as a two phase process:
-
-1. Copy the existing SQLite database to a separate location (while the server
-   is down) and running the port script against that offline database.
-2. Shut down the server. Rerun the port script to port any data that has come
-   in since taking the first snapshot. Restart server against the PostgreSQL
-   database.
-
-The port script is designed to be run repeatedly against newer snapshots of the
-SQLite database file. This makes it safe to repeat step 1 if there was a delay
-between taking the previous snapshot and being ready to do step 2.
-
-It is safe to at any time kill the port script and restart it.
-
-Using the port script
-~~~~~~~~~~~~~~~~~~~~~
-
-Firstly, shut down the currently running synapse server and copy its database
-file (typically ``homeserver.db``) to another location. Once the copy is
-complete, restart synapse.  For instance::
-
-    ./synctl stop
-    cp homeserver.db homeserver.db.snapshot
-    ./synctl start
-
-Copy the old config file into a new config file::
-
-    cp homeserver.yaml homeserver-postgres.yaml
-
-Edit the database section as described in the section *Synapse config* above
-and with the SQLite snapshot located at ``homeserver.db.snapshot`` simply run::
-
-    synapse_port_db --sqlite-database homeserver.db.snapshot \
-        --postgres-config homeserver-postgres.yaml
-
-The flag ``--curses`` displays a coloured curses progress UI.
-
-If the script took a long time to complete, or time has otherwise passed since
-the original snapshot was taken, repeat the previous steps with a newer
-snapshot.
-
-To complete the conversion shut down the synapse server and run the port
-script one last time, e.g. if the SQLite database is at  ``homeserver.db``
-run::
-
-    synapse_port_db --sqlite-database homeserver.db \
-        --postgres-config homeserver-postgres.yaml
-
-Once that has completed, change the synapse config to point at the PostgreSQL
-database configuration file ``homeserver-postgres.yaml``::
-
-    ./synctl stop
-    mv homeserver.yaml homeserver-old-sqlite.yaml
-    mv homeserver-postgres.yaml homeserver.yaml
-    ./synctl start
-
-Synapse should now be running against PostgreSQL.
diff --git a/docs/replication.md b/docs/replication.md
new file mode 100644
index 0000000000..ed88233157
--- /dev/null
+++ b/docs/replication.md
@@ -0,0 +1,37 @@
+# Replication Architecture
+
+## Motivation
+
+We'd like to be able to split some of the work that synapse does into
+multiple python processes. In theory multiple synapse processes could
+share a single postgresql database and we\'d scale up by running more
+synapse processes. However much of synapse assumes that only one process
+is interacting with the database, both for assigning unique identifiers
+when inserting into tables, notifying components about new updates, and
+for invalidating its caches.
+
+So running multiple copies of the current code isn't an option. One way
+to run multiple processes would be to have a single writer process and
+multiple reader processes connected to the same database. In order to do
+this we'd need a way for the reader process to invalidate its in-memory
+caches when an update happens on the writer. One way to do this is for
+the writer to present an append-only log of updates which the readers
+can consume to invalidate their caches and to push updates to listening
+clients or pushers.
+
+Synapse already stores much of its data as an append-only log so that it
+can correctly respond to `/sync` requests so the amount of code changes
+needed to expose the append-only log to the readers should be fairly
+minimal.
+
+## Architecture
+
+### The Replication Protocol
+
+See [tcp_replication.md](tcp_replication.md)
+
+### The Slaved DataStore
+
+There are read-only version of the synapse storage layer in
+`synapse/replication/slave/storage` that use the response of the
+replication API to invalidate their caches.
diff --git a/docs/replication.rst b/docs/replication.rst
deleted file mode 100644
index 310abb3488..0000000000
--- a/docs/replication.rst
+++ /dev/null
@@ -1,40 +0,0 @@
-Replication Architecture
-========================
-
-Motivation
-----------
-
-We'd like to be able to split some of the work that synapse does into multiple
-python processes. In theory multiple synapse processes could share a single
-postgresql database and we'd scale up by running more synapse processes.
-However much of synapse assumes that only one process is interacting with the
-database, both for assigning unique identifiers when inserting into tables,
-notifying components about new updates, and for invalidating its caches.
-
-So running multiple copies of the current code isn't an option. One way to
-run multiple processes would be to have a single writer process and multiple
-reader processes connected to the same database. In order to do this we'd need
-a way for the reader process to invalidate its in-memory caches when an update
-happens on the writer. One way to do this is for the writer to present an
-append-only log of updates which the readers can consume to invalidate their
-caches and to push updates to listening clients or pushers.
-
-Synapse already stores much of its data as an append-only log so that it can
-correctly respond to /sync requests so the amount of code changes needed to
-expose the append-only log to the readers should be fairly minimal.
-
-Architecture
-------------
-
-The Replication Protocol
-~~~~~~~~~~~~~~~~~~~~~~~~
-
-See ``tcp_replication.rst``
-
-
-The Slaved DataStore
-~~~~~~~~~~~~~~~~~~~~
-
-There are read-only version of the synapse storage layer in
-``synapse/replication/slave/storage`` that use the response of the replication
-API to invalidate their caches.
diff --git a/docs/reverse_proxy.md b/docs/reverse_proxy.md
new file mode 100644
index 0000000000..dcfc5c64aa
--- /dev/null
+++ b/docs/reverse_proxy.md
@@ -0,0 +1,123 @@
+# Using a reverse proxy with Synapse
+
+It is recommended to put a reverse proxy such as
+[nginx](https://nginx.org/en/docs/http/ngx_http_proxy_module.html),
+[Apache](https://httpd.apache.org/docs/current/mod/mod_proxy_http.html),
+[Caddy](https://caddyserver.com/docs/proxy) or
+[HAProxy](https://www.haproxy.org/) in front of Synapse. One advantage
+of doing so is that it means that you can expose the default https port
+(443) to Matrix clients without needing to run Synapse with root
+privileges.
+
+> **NOTE**: Your reverse proxy must not `canonicalise` or `normalise`
+the requested URI in any way (for example, by decoding `%xx` escapes).
+Beware that Apache *will* canonicalise URIs unless you specifify
+`nocanon`.
+
+When setting up a reverse proxy, remember that Matrix clients and other
+Matrix servers do not necessarily need to connect to your server via the
+same server name or port. Indeed, clients will use port 443 by default,
+whereas servers default to port 8448. Where these are different, we
+refer to the 'client port' and the \'federation port\'. See [Setting
+up federation](federate.md) for more details of the algorithm used for
+federation connections.
+
+Let's assume that we expect clients to connect to our server at
+`https://matrix.example.com`, and other servers to connect at
+`https://example.com:8448`.  The following sections detail the configuration of
+the reverse proxy and the homeserver.
+
+## Webserver configuration examples
+
+> **NOTE**: You only need one of these.
+
+### nginx
+
+        server {
+            listen 443 ssl;
+            listen [::]:443 ssl;
+            server_name matrix.example.com;
+
+            location /_matrix {
+                proxy_pass http://localhost:8008;
+                proxy_set_header X-Forwarded-For $remote_addr;
+            }
+        }
+
+        server {
+            listen 8448 ssl default_server;
+            listen [::]:8448 ssl default_server;
+            server_name example.com;
+
+            location / {
+                proxy_pass http://localhost:8008;
+                proxy_set_header X-Forwarded-For $remote_addr;
+            }
+        }
+
+> **NOTE**: Do not add a `/` after the port in `proxy_pass`, otherwise nginx will
+canonicalise/normalise the URI.
+
+### Caddy
+
+        matrix.example.com {
+          proxy /_matrix http://localhost:8008 {
+            transparent
+          }
+        }
+
+        example.com:8448 {
+          proxy / http://localhost:8008 {
+            transparent
+          }
+        }
+
+### Apache
+
+        
+            SSLEngine on
+            ServerName matrix.example.com;
+
+            AllowEncodedSlashes NoDecode
+            ProxyPass /_matrix http://127.0.0.1:8008/_matrix nocanon
+            ProxyPassReverse /_matrix http://127.0.0.1:8008/_matrix
+        
+
+        
+            SSLEngine on
+            ServerName example.com;
+
+            AllowEncodedSlashes NoDecode
+            ProxyPass /_matrix http://127.0.0.1:8008/_matrix nocanon
+            ProxyPassReverse /_matrix http://127.0.0.1:8008/_matrix
+        
+
+> **NOTE**: ensure the  `nocanon` options are included.
+
+### HAProxy
+
+        frontend https
+          bind :::443 v4v6 ssl crt /etc/ssl/haproxy/ strict-sni alpn h2,http/1.1
+
+          # Matrix client traffic
+          acl matrix-host hdr(host) -i matrix.example.com
+          acl matrix-path path_beg /_matrix
+
+          use_backend matrix if matrix-host matrix-path
+
+        frontend matrix-federation
+          bind :::8448 v4v6 ssl crt /etc/ssl/haproxy/synapse.pem alpn h2,http/1.1
+          default_backend matrix
+
+        backend matrix
+          server matrix 127.0.0.1:8008
+
+## Homeserver Configuration
+
+You will also want to set `bind_addresses: ['127.0.0.1']` and
+`x_forwarded: true` for port 8008 in `homeserver.yaml` to ensure that
+client IP addresses are recorded correctly.
+
+Having done so, you can then use `https://matrix.example.com` (instead
+of `https://matrix.example.com:8448`) as the "Custom server" when
+connecting to Synapse from a client.
diff --git a/docs/reverse_proxy.rst b/docs/reverse_proxy.rst
deleted file mode 100644
index 4b640ffc4f..0000000000
--- a/docs/reverse_proxy.rst
+++ /dev/null
@@ -1,112 +0,0 @@
-Using a reverse proxy with Synapse
-==================================
-
-It is recommended to put a reverse proxy such as
-`nginx `_,
-`Apache `_,
-`Caddy `_ or
-`HAProxy `_ in front of Synapse. One advantage of
-doing so is that it means that you can expose the default https port (443) to
-Matrix clients without needing to run Synapse with root privileges.
-
-**NOTE**: Your reverse proxy must not 'canonicalise' or 'normalise' the
-requested URI in any way (for example, by decoding ``%xx`` escapes). Beware
-that Apache *will* canonicalise URIs unless you specifify ``nocanon``.
-
-When setting up a reverse proxy, remember that Matrix clients and other Matrix
-servers do not necessarily need to connect to your server via the same server
-name or port. Indeed, clients will use port 443 by default, whereas servers
-default to port 8448. Where these are different, we refer to the 'client port'
-and the 'federation port'. See `Setting up federation
-`_ for more details of the algorithm used for
-federation connections.
-
-Let's assume that we expect clients to connect to our server at
-``https://matrix.example.com``, and other servers to connect at
-``https://example.com:8448``. Here are some example configurations:
-
-* nginx::
-
-      server {
-          listen 443 ssl;
-          listen [::]:443 ssl;
-          server_name matrix.example.com;
-
-          location /_matrix {
-              proxy_pass http://localhost:8008;
-              proxy_set_header X-Forwarded-For $remote_addr;
-          }
-      }
-
-      server {
-          listen 8448 ssl default_server;
-          listen [::]:8448 ssl default_server;
-          server_name example.com;
-
-          location / {
-              proxy_pass http://localhost:8008;
-              proxy_set_header X-Forwarded-For $remote_addr;
-          }
-      }
-      
-  Do not add a `/` after the port in `proxy_pass`, otherwise nginx will canonicalise/normalise the URI.
-
-* Caddy::
-
-      matrix.example.com {
-        proxy /_matrix http://localhost:8008 {
-          transparent
-        }
-      }
-
-      example.com:8448 {
-        proxy / http://localhost:8008 {
-          transparent
-        }
-      }
-
-* Apache (note the ``nocanon`` options here!)::
-
-      
-          SSLEngine on
-          ServerName matrix.example.com;
-
-          AllowEncodedSlashes NoDecode
-          ProxyPass /_matrix http://127.0.0.1:8008/_matrix nocanon
-          ProxyPassReverse /_matrix http://127.0.0.1:8008/_matrix
-      
-
-      
-          SSLEngine on
-          ServerName example.com;
-          
-          AllowEncodedSlashes NoDecode
-          ProxyPass /_matrix http://127.0.0.1:8008/_matrix nocanon
-          ProxyPassReverse /_matrix http://127.0.0.1:8008/_matrix
-      
-
-* HAProxy::
-
-      frontend https
-        bind :::443 v4v6 ssl crt /etc/ssl/haproxy/ strict-sni alpn h2,http/1.1
-
-        # Matrix client traffic
-        acl matrix-host hdr(host) -i matrix.example.com
-        acl matrix-path path_beg /_matrix
-
-        use_backend matrix if matrix-host matrix-path
-
-      frontend matrix-federation
-        bind :::8448 v4v6 ssl crt /etc/ssl/haproxy/synapse.pem alpn h2,http/1.1
-        default_backend matrix
-
-      backend matrix
-        server matrix 127.0.0.1:8008
-
-You will also want to set ``bind_addresses: ['127.0.0.1']`` and ``x_forwarded: true``
-for port 8008 in ``homeserver.yaml`` to ensure that client IP addresses are
-recorded correctly.
-
-Having done so, you can then use ``https://matrix.example.com`` (instead of
-``https://matrix.example.com:8448``) as the "Custom server" when connecting to
-Synapse from a client.
diff --git a/docs/room_and_user_statistics.md b/docs/room_and_user_statistics.md
new file mode 100644
index 0000000000..e1facb38d4
--- /dev/null
+++ b/docs/room_and_user_statistics.md
@@ -0,0 +1,62 @@
+Room and User Statistics
+========================
+
+Synapse maintains room and user statistics (as well as a cache of room state),
+in various tables. These can be used for administrative purposes but are also
+used when generating the public room directory.
+
+
+# Synapse Developer Documentation
+
+## High-Level Concepts
+
+### Definitions
+
+* **subject**: Something we are tracking stats about – currently a room or user.
+* **current row**: An entry for a subject in the appropriate current statistics
+    table. Each subject can have only one.
+* **historical row**: An entry for a subject in the appropriate historical
+    statistics table. Each subject can have any number of these.
+
+### Overview
+
+Stats are maintained as time series. There are two kinds of column:
+
+* absolute columns – where the value is correct for the time given by `end_ts`
+    in the stats row. (Imagine a line graph for these values)
+    * They can also be thought of as 'gauges' in Prometheus, if you are familiar.
+* per-slice columns – where the value corresponds to how many of the occurrences
+    occurred within the time slice given by `(end_ts − bucket_size)…end_ts`
+    or `start_ts…end_ts`. (Imagine a histogram for these values)
+
+Stats are maintained in two tables (for each type): current and historical.
+
+Current stats correspond to the present values. Each subject can only have one
+entry.
+
+Historical stats correspond to values in the past. Subjects may have multiple
+entries.
+
+## Concepts around the management of stats
+
+### Current rows
+
+Current rows contain the most up-to-date statistics for a room.
+They only contain absolute columns
+
+### Historical rows
+
+Historical rows can always be considered to be valid for the time slice and
+end time specified.
+
+* historical rows will not exist for every time slice – they will be omitted
+    if there were no changes. In this case, the following assumptions can be
+    made to interpolate/recreate missing rows:
+    - absolute fields have the same values as in the preceding row
+    - per-slice fields are zero (`0`)
+* historical rows will not be retained forever – rows older than a configurable
+    time will be purged.
+
+#### Purge
+
+The purging of historical rows is not yet implemented.
diff --git a/docs/sample_config.yaml b/docs/sample_config.yaml
index 43969bbb70..d2f4aff826 100644
--- a/docs/sample_config.yaml
+++ b/docs/sample_config.yaml
@@ -86,7 +86,7 @@ pid_file: DATADIR/homeserver.pid
 # Whether room invites to users on this server should be blocked
 # (except those sent by local server admins). The default is False.
 #
-#block_non_admin_invites: True
+#block_non_admin_invites: true
 
 # Room searching
 #
@@ -110,6 +110,9 @@ pid_file: DATADIR/homeserver.pid
 # blacklist IP address CIDR ranges. If this option is not specified, or
 # specified with an empty list, no ip range blacklist will be enforced.
 #
+# As of Synapse v1.4.0 this option also affects any outbound requests to identity
+# servers provided by user input.
+#
 # (0.0.0.0 and :: are always blacklisted, whether or not they are explicitly
 # listed here, since they correspond to unroutable addresses.)
 #
@@ -136,8 +139,8 @@ federation_ip_range_blacklist:
 #
 #   type: the type of listener. Normally 'http', but other valid options are:
 #       'manhole' (see docs/manhole.md),
-#       'metrics' (see docs/metrics-howto.rst),
-#       'replication' (see docs/workers.rst).
+#       'metrics' (see docs/metrics-howto.md),
+#       'replication' (see docs/workers.md).
 #
 #   tls: set to true to enable TLS for this listener. Will use the TLS
 #       key/cert specified in tls_private_key_path / tls_certificate_path.
@@ -172,12 +175,12 @@ federation_ip_range_blacklist:
 #
 #   media: the media API (/_matrix/media).
 #
-#   metrics: the metrics interface. See docs/metrics-howto.rst.
+#   metrics: the metrics interface. See docs/metrics-howto.md.
 #
 #   openid: OpenID authentication.
 #
 #   replication: the HTTP replication API (/_synapse/replication). See
-#       docs/workers.rst.
+#       docs/workers.md.
 #
 #   static: static resources under synapse/static (/_matrix/static). (Mostly
 #       useful for 'fallback authentication'.)
@@ -201,7 +204,7 @@ listeners:
   # that unwraps TLS.
   #
   # If you plan to use a reverse proxy, please see
-  # https://github.com/matrix-org/synapse/blob/master/docs/reverse_proxy.rst.
+  # https://github.com/matrix-org/synapse/blob/master/docs/reverse_proxy.md.
   #
   - port: 8008
     tls: false
@@ -236,9 +239,8 @@ listeners:
 
 # Global blocking
 #
-#hs_disabled: False
+#hs_disabled: false
 #hs_disabled_message: 'Human readable reason for why the HS is blocked'
-#hs_disabled_limit_type: 'error code(str), to help clients decode reason'
 
 # Monthly Active User Blocking
 #
@@ -258,15 +260,22 @@ listeners:
 # sign up in a short space of time never to return after their initial
 # session.
 #
-#limit_usage_by_mau: False
+# 'mau_limit_alerting' is a means of limiting client side alerting
+# should the mau limit be reached. This is useful for small instances
+# where the admin has 5 mau seats (say) for 5 specific people and no
+# interest increasing the mau limit further. Defaults to True, which
+# means that alerting is enabled
+#
+#limit_usage_by_mau: false
 #max_mau_value: 50
 #mau_trial_days: 2
+#mau_limit_alerting: false
 
 # If enabled, the metrics for the number of monthly active users will
 # be populated, however no one will be limited. If limit_usage_by_mau
 # is true, this is implied to be true.
 #
-#mau_stats_only: False
+#mau_stats_only: false
 
 # Sometimes the server admin will want to ensure certain accounts are
 # never blocked by mau checking. These accounts are specified here.
@@ -291,7 +300,7 @@ listeners:
 #
 # Uncomment the below lines to enable:
 #limit_remote_rooms:
-#  enabled: True
+#  enabled: true
 #  complexity: 1.0
 #  complexity_error: "This room is too complex."
 
@@ -306,6 +315,19 @@ listeners:
 #
 #allow_per_room_profiles: false
 
+# How long to keep redacted events in unredacted form in the database. After
+# this period redacted events get replaced with their redacted form in the DB.
+#
+# Defaults to `7d`. Set to `null` to disable.
+#
+#redaction_retention_period: 28d
+
+# How long to track users' last seen time and IPs in the database.
+#
+# Defaults to `28d`. Set to `null` to disable clearing out of old rows.
+#
+#user_ips_max_age: 14d
+
 
 ## TLS ##
 
@@ -395,7 +417,7 @@ acme:
     # ACME support is disabled by default. Set this to `true` and uncomment
     # tls_certificate_path and tls_private_key_path above to enable it.
     #
-    enabled: False
+    enabled: false
 
     # Endpoint to use to request certificates. If you only want to test,
     # use Let's Encrypt's staging url:
@@ -511,6 +533,9 @@ log_config: "CONFDIR/SERVERNAME.log.config"
 #   - one for login that ratelimits login requests based on the account the
 #     client is attempting to log into, based on the amount of failed login
 #     attempts for this account.
+#   - one for ratelimiting redactions by room admins. If this is not explicitly
+#     set then it uses the same ratelimiting as per rc_message. This is useful
+#     to allow room admins to deal with abuse quickly.
 #
 # The defaults are as shown below.
 #
@@ -532,6 +557,10 @@ log_config: "CONFDIR/SERVERNAME.log.config"
 #  failed_attempts:
 #    per_second: 0.17
 #    burst_count: 3
+#
+#rc_admin_redaction:
+#  per_second: 1
+#  burst_count: 50
 
 
 # Ratelimiting settings for incoming federation
@@ -763,7 +792,7 @@ uploads_path: "DATADIR/uploads"
 # connect to arbitrary endpoints without having first signed up for a
 # valid account (e.g. by passing a CAPTCHA).
 #
-#turn_allow_guests: True
+#turn_allow_guests: true
 
 
 ## Registration ##
@@ -806,7 +835,7 @@ uploads_path: "DATADIR/uploads"
 # where d is equal to 10% of the validity period.
 #
 #account_validity:
-#  enabled: True
+#  enabled: true
 #  period: 6w
 #  renew_at: 1w
 #  renew_email_subject: "Renew your %(app)s account"
@@ -891,10 +920,44 @@ uploads_path: "DATADIR/uploads"
 # Also defines the ID server which will be called when an account is
 # deactivated (one will be picked arbitrarily).
 #
+# Note: This option is deprecated. Since v0.99.4, Synapse has tracked which identity
+# server a 3PID has been bound to. For 3PIDs bound before then, Synapse runs a
+# background migration script, informing itself that the identity server all of its
+# 3PIDs have been bound to is likely one of the below.
+#
+# As of Synapse v1.4.0, all other functionality of this option has been deprecated, and
+# it is now solely used for the purposes of the background migration script, and can be
+# removed once it has run.
 #trusted_third_party_id_servers:
 #  - matrix.org
 #  - vector.im
 
+# Handle threepid (email/phone etc) registration and password resets through a set of
+# *trusted* identity servers. Note that this allows the configured identity server to
+# reset passwords for accounts!
+#
+# Be aware that if `email` is not set, and SMTP options have not been
+# configured in the email config block, registration and user password resets via
+# email will be globally disabled.
+#
+# Additionally, if `msisdn` is not set, registration and password resets via msisdn
+# will be disabled regardless. This is due to Synapse currently not supporting any
+# method of sending SMS messages on its own.
+#
+# To enable using an identity server for operations regarding a particular third-party
+# identifier type, set the value to the URL of that identity server as shown in the
+# examples below.
+#
+# Servers handling the these requests must answer the `/requestToken` endpoints defined
+# by the Matrix Identity Service API specification:
+# https://matrix.org/docs/spec/identity_service/latest
+#
+# If a delegate is specified, the config option public_baseurl must also be filled out.
+#
+account_threepid_delegates:
+    #email: https://example.com     # Delegate email sending to example.com
+    #msisdn: http://localhost:8090  # Delegate SMS sending to this local process
+
 # Users who register on this homeserver will automatically be joined
 # to these rooms
 #
@@ -914,7 +977,7 @@ uploads_path: "DATADIR/uploads"
 
 # Enable collection and rendering of performance metrics
 #
-#enable_metrics: False
+#enable_metrics: false
 
 # Enable sentry integration
 # NOTE: While attempts are made to ensure that the logs don't contain
@@ -926,9 +989,24 @@ uploads_path: "DATADIR/uploads"
 #sentry:
 #    dsn: "..."
 
+# Flags to enable Prometheus metrics which are not suitable to be
+# enabled by default, either for performance reasons or limited use.
+#
+metrics_flags:
+    # Publish synapse_federation_known_servers, a g auge of the number of
+    # servers this homeserver knows about, including itself. May cause
+    # performance problems on large homeservers.
+    #
+    #known_servers: true
+
 # Whether or not to report anonymized homeserver usage statistics.
 # report_stats: true|false
 
+# The endpoint to report the anonymized homeserver usage statistics to.
+# Defaults to https://matrix.org/report-usage-stats/push
+#
+#report_stats_endpoint: https://example.com/report-usage-stats/push
+
 
 ## API Configuration ##
 
@@ -951,7 +1029,7 @@ uploads_path: "DATADIR/uploads"
 # Uncomment to enable tracking of application service IP addresses. Implicitly
 # enables MAU tracking for application service users.
 #
-#track_appservice_user_ips: True
+#track_appservice_user_ips: true
 
 
 # a secret which is used to sign access tokens. If none is specified,
@@ -1000,6 +1078,10 @@ signing_key_path: "CONFDIR/SERVERNAME.signing.key"
 # This setting supercedes an older setting named `perspectives`. The old format
 # is still supported for backwards-compatibility, but it is deprecated.
 #
+# 'trusted_key_servers' defaults to matrix.org, but using it will generate a
+# warning on start-up. To suppress this warning, set
+# 'suppress_key_server_warning' to true.
+#
 # Options for each entry in the list include:
 #
 #    server_name: the name of the server. required.
@@ -1024,11 +1106,13 @@ signing_key_path: "CONFDIR/SERVERNAME.signing.key"
 #      "ed25519:auto": "abcdefghijklmnopqrstuvwxyzabcdefghijklmopqr"
 #  - server_name: "my_other_trusted_server.example.com"
 #
-# The default configuration is:
-#
-#trusted_key_servers:
-#  - server_name: "matrix.org"
+trusted_key_servers:
+  - server_name: "matrix.org"
+
+# Uncomment the following to disable the warning that is emitted when the
+# trusted_key_servers include 'matrix.org'. See above.
 #
+#suppress_key_server_warning: true
 
 # The signing keys to use when acting as a trusted key server. If not specified
 # defaults to the server signing key.
@@ -1040,12 +1124,13 @@ signing_key_path: "CONFDIR/SERVERNAME.signing.key"
 
 # Enable SAML2 for registration and login. Uses pysaml2.
 #
-# `sp_config` is the configuration for the pysaml2 Service Provider.
-# See pysaml2 docs for format of config.
+# At least one of `sp_config` or `config_path` must be set in this section to
+# enable SAML login.
 #
-# Default values will be used for the 'entityid' and 'service' settings,
-# so it is not normally necessary to specify them unless you need to
-# override them.
+# (You will probably also want to set the following options to `false` to
+# disable the regular login/registration flows:
+#   * enable_registration
+#   * password_config.enabled
 #
 # Once SAML support is enabled, a metadata file will be exposed at
 # https://:/_matrix/saml2/metadata.xml, which you may be able to
@@ -1053,52 +1138,85 @@ signing_key_path: "CONFDIR/SERVERNAME.signing.key"
 # the IdP to use an ACS location of
 # https://:/_matrix/saml2/authn_response.
 #
-#saml2_config:
-#  sp_config:
-#    # point this to the IdP's metadata. You can use either a local file or
-#    # (preferably) a URL.
-#    metadata:
-#      #local: ["saml2/idp.xml"]
-#      remote:
-#        - url: https://our_idp/metadata.xml
-#
-#    # By default, the user has to go to our login page first. If you'd like to
-#    # allow IdP-initiated login, set 'allow_unsolicited: True' in a
-#    # 'service.sp' section:
-#    #
-#    #service:
-#    #  sp:
-#    #    allow_unsolicited: True
-#
-#    # The examples below are just used to generate our metadata xml, and you
-#    # may well not need it, depending on your setup. Alternatively you
-#    # may need a whole lot more detail - see the pysaml2 docs!
-#
-#    description: ["My awesome SP", "en"]
-#    name: ["Test SP", "en"]
-#
-#    organization:
-#      name: Example com
-#      display_name:
-#        - ["Example co", "en"]
-#      url: "http://example.com"
-#
-#    contact_person:
-#      - given_name: Bob
-#        sur_name: "the Sysadmin"
-#        email_address": ["admin@example.com"]
-#        contact_type": technical
-#
-#  # Instead of putting the config inline as above, you can specify a
-#  # separate pysaml2 configuration file:
-#  #
-#  config_path: "CONFDIR/sp_conf.py"
-#
-#  # the lifetime of a SAML session. This defines how long a user has to
-#  # complete the authentication process, if allow_unsolicited is unset.
-#  # The default is 5 minutes.
-#  #
-#  # saml_session_lifetime: 5m
+saml2_config:
+  # `sp_config` is the configuration for the pysaml2 Service Provider.
+  # See pysaml2 docs for format of config.
+  #
+  # Default values will be used for the 'entityid' and 'service' settings,
+  # so it is not normally necessary to specify them unless you need to
+  # override them.
+  #
+  #sp_config:
+  #  # point this to the IdP's metadata. You can use either a local file or
+  #  # (preferably) a URL.
+  #  metadata:
+  #    #local: ["saml2/idp.xml"]
+  #    remote:
+  #      - url: https://our_idp/metadata.xml
+  #
+  #    # By default, the user has to go to our login page first. If you'd like
+  #    # to allow IdP-initiated login, set 'allow_unsolicited: true' in a
+  #    # 'service.sp' section:
+  #    #
+  #    #service:
+  #    #  sp:
+  #    #    allow_unsolicited: true
+  #
+  #    # The examples below are just used to generate our metadata xml, and you
+  #    # may well not need them, depending on your setup. Alternatively you
+  #    # may need a whole lot more detail - see the pysaml2 docs!
+  #
+  #    description: ["My awesome SP", "en"]
+  #    name: ["Test SP", "en"]
+  #
+  #    organization:
+  #      name: Example com
+  #      display_name:
+  #        - ["Example co", "en"]
+  #      url: "http://example.com"
+  #
+  #    contact_person:
+  #      - given_name: Bob
+  #        sur_name: "the Sysadmin"
+  #        email_address": ["admin@example.com"]
+  #        contact_type": technical
+
+  # Instead of putting the config inline as above, you can specify a
+  # separate pysaml2 configuration file:
+  #
+  #config_path: "CONFDIR/sp_conf.py"
+
+  # the lifetime of a SAML session. This defines how long a user has to
+  # complete the authentication process, if allow_unsolicited is unset.
+  # The default is 5 minutes.
+  #
+  #saml_session_lifetime: 5m
+
+  # The SAML attribute (after mapping via the attribute maps) to use to derive
+  # the Matrix ID from. 'uid' by default.
+  #
+  #mxid_source_attribute: displayName
+
+  # The mapping system to use for mapping the saml attribute onto a matrix ID.
+  # Options include:
+  #  * 'hexencode' (which maps unpermitted characters to '=xx')
+  #  * 'dotreplace' (which replaces unpermitted characters with '.').
+  # The default is 'hexencode'.
+  #
+  #mxid_mapping: dotreplace
+
+  # In previous versions of synapse, the mapping from SAML attribute to MXID was
+  # always calculated dynamically rather than stored in a table. For backwards-
+  # compatibility, we will look for user_ids matching such a pattern before
+  # creating a new account.
+  #
+  # This setting controls the SAML attribute which will be used for this
+  # backwards-compatibility lookup. Typically it should be 'uid', but if the
+  # attribute maps are changed, it may be necessary to change it.
+  #
+  # The default is 'uid'.
+  #
+  #grandfathered_mxid_source_attribute: upn
 
 
 
@@ -1108,6 +1226,7 @@ signing_key_path: "CONFDIR/SERVERNAME.signing.key"
 #   enabled: true
 #   server_url: "https://cas-server.com"
 #   service_url: "https://homeserver.domain.com:8448"
+#   #displayname_attribute: name
 #   #required_attributes:
 #   #    name: value
 
@@ -1150,13 +1269,13 @@ password_config:
 #   smtp_port: 25 # SSL: 465, STARTTLS: 587
 #   smtp_user: "exampleusername"
 #   smtp_pass: "examplepassword"
-#   require_transport_security: False
+#   require_transport_security: false
 #   notif_from: "Your Friendly %(app)s Home Server "
 #   app_name: Matrix
 #
 #   # Enable email notifications by default
 #   #
-#   notif_for_new_users: True
+#   notif_for_new_users: true
 #
 #   # Defining a custom URL for Riot is only needed if email notifications
 #   # should contain links to a self-hosted installation of Riot; when set
@@ -1164,19 +1283,6 @@ password_config:
 #   #
 #   riot_base_url: "http://localhost/riot"
 #
-#   # Enable sending password reset emails via the configured, trusted
-#   # identity servers
-#   #
-#   # IMPORTANT! This will give a malicious or overtaken identity server
-#   # the ability to reset passwords for your users! Make absolutely sure
-#   # that you want to do this! It is strongly recommended that password
-#   # reset emails be sent by the homeserver instead
-#   #
-#   # If this option is set to false and SMTP options have not been
-#   # configured, resetting user passwords via email will be disabled
-#   #
-#   #trust_identity_server_for_password_resets: false
-#
 #   # Configure the time that a validation email or text message code
 #   # will expire after sending
 #   #
@@ -1208,11 +1314,34 @@ password_config:
 #   #password_reset_template_html: password_reset.html
 #   #password_reset_template_text: password_reset.txt
 #
+#   # Templates for registration emails sent by the homeserver
+#   #
+#   #registration_template_html: registration.html
+#   #registration_template_text: registration.txt
+#
+#   # Templates for validation emails sent by the homeserver when adding an email to
+#   # your user account
+#   #
+#   #add_threepid_template_html: add_threepid.html
+#   #add_threepid_template_text: add_threepid.txt
+#
 #   # Templates for password reset success and failure pages that a user
 #   # will see after attempting to reset their password
 #   #
 #   #password_reset_template_success_html: password_reset_success.html
 #   #password_reset_template_failure_html: password_reset_failure.html
+#
+#   # Templates for registration success and failure pages that a user
+#   # will see after attempting to register using an email or phone
+#   #
+#   #registration_template_success_html: registration_success.html
+#   #registration_template_failure_html: registration_failure.html
+#
+#   # Templates for success and failure pages that a user will see after attempting
+#   # to add an email or phone to their account
+#   #
+#   #add_threepid_success_html: add_threepid_success.html
+#   #add_threepid_failure_html: add_threepid_failure.html
 
 
 #password_providers:
@@ -1324,11 +1453,11 @@ password_config:
 #    body: >-
 #      To continue using this homeserver you must review and agree to the
 #      terms and conditions at %(consent_uri)s
-#  send_server_notice_to_guests: True
+#  send_server_notice_to_guests: true
 #  block_events_error: >-
 #    To continue using this homeserver you must review and agree to the
 #    terms and conditions at %(consent_uri)s
-#  require_at_registration: False
+#  require_at_registration: false
 #  policy_name: Privacy Policy
 #
 
diff --git a/docs/tcp_replication.md b/docs/tcp_replication.md
new file mode 100644
index 0000000000..ba9e874d07
--- /dev/null
+++ b/docs/tcp_replication.md
@@ -0,0 +1,262 @@
+# TCP Replication
+
+## Motivation
+
+Previously the workers used an HTTP long poll mechanism to get updates
+from the master, which had the problem of causing a lot of duplicate
+work on the server. This TCP protocol replaces those APIs with the aim
+of increased efficiency.
+
+## Overview
+
+The protocol is based on fire and forget, line based commands. An
+example flow would be (where '>' indicates master to worker and
+'<' worker to master flows):
+
+    > SERVER example.com
+    < REPLICATE events 53
+    > RDATA events 54 ["$foo1:bar.com", ...]
+    > RDATA events 55 ["$foo4:bar.com", ...]
+
+The example shows the server accepting a new connection and sending its
+identity with the `SERVER` command, followed by the client asking to
+subscribe to the `events` stream from the token `53`. The server then
+periodically sends `RDATA` commands which have the format
+`RDATA   `, where the format of `` is
+defined by the individual streams.
+
+Error reporting happens by either the client or server sending an ERROR
+command, and usually the connection will be closed.
+
+Since the protocol is a simple line based, its possible to manually
+connect to the server using a tool like netcat. A few things should be
+noted when manually using the protocol:
+
+-   When subscribing to a stream using `REPLICATE`, the special token
+    `NOW` can be used to get all future updates. The special stream name
+    `ALL` can be used with `NOW` to subscribe to all available streams.
+-   The federation stream is only available if federation sending has
+    been disabled on the main process.
+-   The server will only time connections out that have sent a `PING`
+    command. If a ping is sent then the connection will be closed if no
+    further commands are receieved within 15s. Both the client and
+    server protocol implementations will send an initial PING on
+    connection and ensure at least one command every 5s is sent (not
+    necessarily `PING`).
+-   `RDATA` commands *usually* include a numeric token, however if the
+    stream has multiple rows to replicate per token the server will send
+    multiple `RDATA` commands, with all but the last having a token of
+    `batch`. See the documentation on `commands.RdataCommand` for
+    further details.
+
+## Architecture
+
+The basic structure of the protocol is line based, where the initial
+word of each line specifies the command. The rest of the line is parsed
+based on the command. For example, the RDATA command is defined as:
+
+    RDATA   
+
+(Note that  may contains spaces, but cannot contain
+newlines.)
+
+Blank lines are ignored.
+
+### Keep alives
+
+Both sides are expected to send at least one command every 5s or so, and
+should send a `PING` command if necessary. If either side do not receive
+a command within e.g. 15s then the connection should be closed.
+
+Because the server may be connected to manually using e.g. netcat, the
+timeouts aren't enabled until an initial `PING` command is seen. Both
+the client and server implementations below send a `PING` command
+immediately on connection to ensure the timeouts are enabled.
+
+This ensures that both sides can quickly realize if the tcp connection
+has gone and handle the situation appropriately.
+
+### Start up
+
+When a new connection is made, the server:
+
+-   Sends a `SERVER` command, which includes the identity of the server,
+    allowing the client to detect if its connected to the expected
+    server
+-   Sends a `PING` command as above, to enable the client to time out
+    connections promptly.
+
+The client:
+
+-   Sends a `NAME` command, allowing the server to associate a human
+    friendly name with the connection. This is optional.
+-   Sends a `PING` as above
+-   For each stream the client wishes to subscribe to it sends a
+    `REPLICATE` with the `stream_name` and token it wants to subscribe
+    from.
+-   On receipt of a `SERVER` command, checks that the server name
+    matches the expected server name.
+
+### Error handling
+
+If either side detects an error it can send an `ERROR` command and close
+the connection.
+
+If the client side loses the connection to the server it should
+reconnect, following the steps above.
+
+### Congestion
+
+If the server sends messages faster than the client can consume them the
+server will first buffer a (fairly large) number of commands and then
+disconnect the client. This ensures that we don't queue up an unbounded
+number of commands in memory and gives us a potential oppurtunity to
+squawk loudly. When/if the client recovers it can reconnect to the
+server and ask for missed messages.
+
+### Reliability
+
+In general the replication stream should be considered an unreliable
+transport since e.g. commands are not resent if the connection
+disappears.
+
+The exception to that are the replication streams, i.e. RDATA commands,
+since these include tokens which can be used to restart the stream on
+connection errors.
+
+The client should keep track of the token in the last RDATA command
+received for each stream so that on reconneciton it can start streaming
+from the correct place. Note: not all RDATA have valid tokens due to
+batching. See `RdataCommand` for more details.
+
+### Example
+
+An example iteraction is shown below. Each line is prefixed with '>'
+or '<' to indicate which side is sending, these are *not* included on
+the wire:
+
+    * connection established *
+    > SERVER localhost:8823
+    > PING 1490197665618
+    < NAME synapse.app.appservice
+    < PING 1490197665618
+    < REPLICATE events 1
+    < REPLICATE backfill 1
+    < REPLICATE caches 1
+    > POSITION events 1
+    > POSITION backfill 1
+    > POSITION caches 1
+    > RDATA caches 2 ["get_user_by_id",["@01register-user:localhost:8823"],1490197670513]
+    > RDATA events 14 ["$149019767112vOHxz:localhost:8823",
+        "!AFDCvgApUmpdfVjIXm:localhost:8823","m.room.guest_access","",null]
+    < PING 1490197675618
+    > ERROR server stopping
+    * connection closed by server *
+
+The `POSITION` command sent by the server is used to set the clients
+position without needing to send data with the `RDATA` command.
+
+An example of a batched set of `RDATA` is:
+
+    > RDATA caches batch ["get_user_by_id",["@test:localhost:8823"],1490197670513]
+    > RDATA caches batch ["get_user_by_id",["@test2:localhost:8823"],1490197670513]
+    > RDATA caches batch ["get_user_by_id",["@test3:localhost:8823"],1490197670513]
+    > RDATA caches 54 ["get_user_by_id",["@test4:localhost:8823"],1490197670513]
+
+In this case the client shouldn't advance their caches token until it
+sees the the last `RDATA`.
+
+### List of commands
+
+The list of valid commands, with which side can send it: server (S) or
+client (C):
+
+#### SERVER (S)
+
+   Sent at the start to identify which server the client is talking to
+
+#### RDATA (S)
+
+   A single update in a stream
+
+#### POSITION (S)
+
+   The position of the stream has been updated. Sent to the client
+    after all missing updates for a stream have been sent to the client
+    and they're now up to date.
+
+#### ERROR (S, C)
+
+   There was an error
+
+#### PING (S, C)
+
+   Sent periodically to ensure the connection is still alive
+
+#### NAME (C)
+
+   Sent at the start by client to inform the server who they are
+
+#### REPLICATE (C)
+
+Asks the server to replicate a given stream. The syntax is:
+
+```
+    REPLICATE  
+```
+
+Where `` may be either:
+ * a numeric stream_id to stream updates since (exclusive)
+ * `NOW` to stream all subsequent updates.
+
+The `` is the name of a replication stream to subscribe 
+to (see [here](../synapse/replication/tcp/streams/_base.py) for a list
+of streams). It can also be `ALL` to subscribe to all known streams,
+in which case the `` must be set to `NOW`.
+
+#### USER_SYNC (C)
+
+   A user has started or stopped syncing
+
+#### FEDERATION_ACK (C)
+
+   Acknowledge receipt of some federation data
+
+#### REMOVE_PUSHER (C)
+
+   Inform the server a pusher should be removed
+
+#### INVALIDATE_CACHE (C)
+
+   Inform the server a cache should be invalidated
+
+#### SYNC (S, C)
+
+   Used exclusively in tests
+
+See `synapse/replication/tcp/commands.py` for a detailed description and
+the format of each command.
+
+### Cache Invalidation Stream
+
+The cache invalidation stream is used to inform workers when they need
+to invalidate any of their caches in the data store. This is done by
+streaming all cache invalidations done on master down to the workers,
+assuming that any caches on the workers also exist on the master.
+
+Each individual cache invalidation results in a row being sent down
+replication, which includes the cache name (the name of the function)
+and they key to invalidate. For example:
+
+    > RDATA caches 550953771 ["get_user_by_id", ["@bob:example.com"], 1550574873251]
+
+However, there are times when a number of caches need to be invalidated
+at the same time with the same key. To reduce traffic we batch those
+invalidations into a single poke by defining a special cache name that
+workers understand to mean to expand to invalidate the correct caches.
+
+Currently the special cache names are declared in
+`synapse/storage/_base.py` and are:
+
+1.  `cs_cache_fake` ─ invalidates caches that depend on the current
+    state
diff --git a/docs/tcp_replication.rst b/docs/tcp_replication.rst
deleted file mode 100644
index 75e723484c..0000000000
--- a/docs/tcp_replication.rst
+++ /dev/null
@@ -1,249 +0,0 @@
-TCP Replication
-===============
-
-Motivation
-----------
-
-Previously the workers used an HTTP long poll mechanism to get updates from the
-master, which had the problem of causing a lot of duplicate work on the server.
-This TCP protocol replaces those APIs with the aim of increased efficiency.
-
-
-
-Overview
---------
-
-The protocol is based on fire and forget, line based commands. An example flow
-would be (where '>' indicates master to worker and '<' worker to master flows)::
-
-    > SERVER example.com
-    < REPLICATE events 53
-    > RDATA events 54 ["$foo1:bar.com", ...]
-    > RDATA events 55 ["$foo4:bar.com", ...]
-
-The example shows the server accepting a new connection and sending its identity
-with the ``SERVER`` command, followed by the client asking to subscribe to the
-``events`` stream from the token ``53``. The server then periodically sends ``RDATA``
-commands which have the format ``RDATA   ``, where the
-format of ```` is defined by the individual streams.
-
-Error reporting happens by either the client or server sending an `ERROR`
-command, and usually the connection will be closed.
-
-
-Since the protocol is a simple line based, its possible to manually connect to
-the server using a tool like netcat. A few things should be noted when manually
-using the protocol:
-
-* When subscribing to a stream using ``REPLICATE``, the special token ``NOW`` can
-  be used to get all future updates. The special stream name ``ALL`` can be used
-  with ``NOW`` to subscribe to all available streams.
-* The federation stream is only available if federation sending has been
-  disabled on the main process.
-* The server will only time connections out that have sent a ``PING`` command.
-  If a ping is sent then the connection will be closed if no further commands
-  are receieved within 15s. Both the client and server protocol implementations
-  will send an initial PING on connection and ensure at least one command every
-  5s is sent (not necessarily ``PING``).
-* ``RDATA`` commands *usually* include a numeric token, however if the stream
-  has multiple rows to replicate per token the server will send multiple
-  ``RDATA`` commands, with all but the last having a token of ``batch``. See
-  the documentation on ``commands.RdataCommand`` for further details.
-
-
-Architecture
-------------
-
-The basic structure of the protocol is line based, where the initial word of
-each line specifies the command. The rest of the line is parsed based on the
-command. For example, the `RDATA` command is defined as::
-
-    RDATA   
-
-(Note that `` may contains spaces, but cannot contain newlines.)
-
-Blank lines are ignored.
-
-
-Keep alives
-~~~~~~~~~~~
-
-Both sides are expected to send at least one command every 5s or so, and
-should send a ``PING`` command if necessary. If either side do not receive a
-command within e.g. 15s then the connection should be closed.
-
-Because the server may be connected to manually using e.g. netcat, the timeouts
-aren't enabled until an initial ``PING`` command is seen. Both the client and
-server implementations below send a ``PING`` command immediately on connection to
-ensure the timeouts are enabled.
-
-This ensures that both sides can quickly realize if the tcp connection has gone
-and handle the situation appropriately.
-
-
-Start up
-~~~~~~~~
-
-When a new connection is made, the server:
-
-* Sends a ``SERVER`` command, which includes the identity of the server, allowing
-  the client to detect if its connected to the expected server
-* Sends a ``PING`` command as above, to enable the client to time out connections
-  promptly.
-
-The client:
-
-* Sends a ``NAME`` command, allowing the server to associate a human friendly
-  name with the connection. This is optional.
-* Sends a ``PING`` as above
-* For each stream the client wishes to subscribe to it sends a ``REPLICATE``
-  with the stream_name and token it wants to subscribe from.
-* On receipt of a ``SERVER`` command, checks that the server name matches the
-  expected server name.
-
-
-Error handling
-~~~~~~~~~~~~~~
-
-If either side detects an error it can send an ``ERROR`` command and close the
-connection.
-
-If the client side loses the connection to the server it should reconnect,
-following the steps above.
-
-
-Congestion
-~~~~~~~~~~
-
-If the server sends messages faster than the client can consume them the server
-will first buffer a (fairly large) number of commands and then disconnect the
-client. This ensures that we don't queue up an unbounded number of commands in
-memory and gives us a potential oppurtunity to squawk loudly. When/if the client
-recovers it can reconnect to the server and ask for missed messages.
-
-
-Reliability
-~~~~~~~~~~~
-
-In general the replication stream should be considered an unreliable transport
-since e.g. commands are not resent if the connection disappears.
-
-The exception to that are the replication streams, i.e. RDATA commands, since
-these include tokens which can be used to restart the stream on connection
-errors.
-
-The client should keep track of the token in the last RDATA command received
-for each stream so that on reconneciton it can start streaming from the correct
-place. Note: not all RDATA have valid tokens due to batching. See
-``RdataCommand`` for more details.
-
-Example
-~~~~~~~
-
-An example iteraction is shown below. Each line is prefixed with '>' or '<' to
-indicate which side is sending, these are *not* included on the wire::
-
-    * connection established *
-    > SERVER localhost:8823
-    > PING 1490197665618
-    < NAME synapse.app.appservice
-    < PING 1490197665618
-    < REPLICATE events 1
-    < REPLICATE backfill 1
-    < REPLICATE caches 1
-    > POSITION events 1
-    > POSITION backfill 1
-    > POSITION caches 1
-    > RDATA caches 2 ["get_user_by_id",["@01register-user:localhost:8823"],1490197670513]
-    > RDATA events 14 ["$149019767112vOHxz:localhost:8823",
-        "!AFDCvgApUmpdfVjIXm:localhost:8823","m.room.guest_access","",null]
-    < PING 1490197675618
-    > ERROR server stopping
-    * connection closed by server *
-
-The ``POSITION`` command sent by the server is used to set the clients position
-without needing to send data with the ``RDATA`` command.
-
-
-An example of a batched set of ``RDATA`` is::
-
-    > RDATA caches batch ["get_user_by_id",["@test:localhost:8823"],1490197670513]
-    > RDATA caches batch ["get_user_by_id",["@test2:localhost:8823"],1490197670513]
-    > RDATA caches batch ["get_user_by_id",["@test3:localhost:8823"],1490197670513]
-    > RDATA caches 54 ["get_user_by_id",["@test4:localhost:8823"],1490197670513]
-
-In this case the client shouldn't advance their caches token until it sees the
-the last ``RDATA``.
-
-
-List of commands
-~~~~~~~~~~~~~~~~
-
-The list of valid commands, with which side can send it: server (S) or client (C):
-
-SERVER (S)
-    Sent at the start to identify which server the client is talking to
-
-RDATA (S)
-    A single update in a stream
-
-POSITION (S)
-    The position of the stream has been updated. Sent to the client after all
-    missing updates for a stream have been sent to the client and they're now
-    up to date.
-
-ERROR (S, C)
-    There was an error
-
-PING (S, C)
-    Sent periodically to ensure the connection is still alive
-
-NAME (C)
-    Sent at the start by client to inform the server who they are
-
-REPLICATE (C)
-    Asks the server to replicate a given stream
-
-USER_SYNC (C)
-    A user has started or stopped syncing
-
-FEDERATION_ACK (C)
-    Acknowledge receipt of some federation data
-
-REMOVE_PUSHER (C)
-    Inform the server a pusher should be removed
-
-INVALIDATE_CACHE (C)
-    Inform the server a cache should be invalidated
-
-SYNC (S, C)
-    Used exclusively in tests
-
-
-See ``synapse/replication/tcp/commands.py`` for a detailed description and the
-format of each command.
-
-
-Cache Invalidation Stream
-~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The cache invalidation stream is used to inform workers when they need to
-invalidate any of their caches in the data store. This is done by streaming all
-cache invalidations done on master down to the workers, assuming that any caches
-on the workers also exist on the master.
-
-Each individual cache invalidation results in a row being sent down replication,
-which includes the cache name (the name of the function) and they key to
-invalidate. For example::
-
-    > RDATA caches 550953771 ["get_user_by_id", ["@bob:example.com"], 1550574873251]
-
-However, there are times when a number of caches need to be invalidated at the
-same time with the same key. To reduce traffic we batch those invalidations into
-a single poke by defining a special cache name that workers understand to mean
-to expand to invalidate the correct caches.
-
-Currently the special cache names are declared in ``synapse/storage/_base.py``
-and are:
-
-1. ``cs_cache_fake`` ─ invalidates caches that depend on the current state
diff --git a/docs/turn-howto.md b/docs/turn-howto.md
new file mode 100644
index 0000000000..4a983621e5
--- /dev/null
+++ b/docs/turn-howto.md
@@ -0,0 +1,123 @@
+# Overview
+
+This document explains how to enable VoIP relaying on your Home Server with
+TURN.
+
+The synapse Matrix Home Server supports integration with TURN server via the
+[TURN server REST API](). This
+allows the Home Server to generate credentials that are valid for use on the
+TURN server through the use of a secret shared between the Home Server and the
+TURN server.
+
+The following sections describe how to install [coturn]() (which implements the TURN REST API) and integrate it with synapse.
+
+## `coturn` Setup
+
+### Initial installation
+
+The TURN daemon `coturn` is available from a variety of sources such as native package managers, or installation from source.
+
+#### Debian installation
+
+    # apt install coturn
+
+#### Source installation
+
+1. Download the [latest release](https://github.com/coturn/coturn/releases/latest) from github.  Unpack it and `cd` into the directory.
+
+1.  Configure it:
+
+        ./configure
+
+    > You may need to install `libevent2`: if so, you should do so in
+    > the way recommended by your operating system. You can ignore
+    > warnings about lack of database support: a database is unnecessary
+    > for this purpose.
+
+1.  Build and install it:
+
+        make
+        make install
+
+1.  Create or edit the config file in `/etc/turnserver.conf`. The relevant
+    lines, with example values, are:
+
+        use-auth-secret
+        static-auth-secret=[your secret key here]
+        realm=turn.myserver.org
+
+    See `turnserver.conf` for explanations of the options. One way to generate
+    the `static-auth-secret` is with `pwgen`:
+
+        pwgen -s 64 1
+
+1.  Consider your security settings. TURN lets users request a relay which will
+    connect to arbitrary IP addresses and ports. The following configuration is
+    suggested as a minimum starting point:
+    
+        # VoIP traffic is all UDP. There is no reason to let users connect to arbitrary TCP endpoints via the relay.
+        no-tcp-relay
+        
+        # don't let the relay ever try to connect to private IP address ranges within your network (if any)
+        # given the turn server is likely behind your firewall, remember to include any privileged public IPs too.
+        denied-peer-ip=10.0.0.0-10.255.255.255
+        denied-peer-ip=192.168.0.0-192.168.255.255
+        denied-peer-ip=172.16.0.0-172.31.255.255
+        
+        # special case the turn server itself so that client->TURN->TURN->client flows work
+        allowed-peer-ip=10.0.0.1
+        
+        # consider whether you want to limit the quota of relayed streams per user (or total) to avoid risk of DoS.
+        user-quota=12 # 4 streams per video call, so 12 streams = 3 simultaneous relayed calls per user.
+        total-quota=1200
+
+    Ideally coturn should refuse to relay traffic which isn't SRTP; see
+    
+
+1.  Ensure your firewall allows traffic into the TURN server on the ports
+    you've configured it to listen on (remember to allow both TCP and UDP TURN
+    traffic)
+
+1.  If you've configured coturn to support TLS/DTLS, generate or import your
+    private key and certificate.
+
+1.  Start the turn server:
+
+         bin/turnserver -o
+
+## synapse Setup
+
+Your home server configuration file needs the following extra keys:
+
+1.  "`turn_uris`": This needs to be a yaml list of public-facing URIs
+    for your TURN server to be given out to your clients. Add separate
+    entries for each transport your TURN server supports.
+2.  "`turn_shared_secret`": This is the secret shared between your
+    Home server and your TURN server, so you should set it to the same
+    string you used in turnserver.conf.
+3.  "`turn_user_lifetime`": This is the amount of time credentials
+    generated by your Home Server are valid for (in milliseconds).
+    Shorter times offer less potential for abuse at the expense of
+    increased traffic between web clients and your home server to
+    refresh credentials. The TURN REST API specification recommends
+    one day (86400000).
+4.  "`turn_allow_guests`": Whether to allow guest users to use the
+    TURN server. This is enabled by default, as otherwise VoIP will
+    not work reliably for guests. However, it does introduce a
+    security risk as it lets guests connect to arbitrary endpoints
+    without having gone through a CAPTCHA or similar to register a
+    real account.
+
+As an example, here is the relevant section of the config file for matrix.org:
+
+    turn_uris: [ "turn:turn.matrix.org:3478?transport=udp", "turn:turn.matrix.org:3478?transport=tcp" ]
+    turn_shared_secret: n0t4ctuAllymatr1Xd0TorgSshar3d5ecret4obvIousreAsons
+    turn_user_lifetime: 86400000
+    turn_allow_guests: True
+
+After updating the homeserver configuration, you must restart synapse:
+
+    cd /where/you/run/synapse
+    ./synctl restart
+
+..and your Home Server now supports VoIP relaying!
diff --git a/docs/turn-howto.rst b/docs/turn-howto.rst
deleted file mode 100644
index a2fc5c8820..0000000000
--- a/docs/turn-howto.rst
+++ /dev/null
@@ -1,127 +0,0 @@
-How to enable VoIP relaying on your Home Server with TURN
-
-Overview
---------
-The synapse Matrix Home Server supports integration with TURN server via the
-TURN server REST API
-(http://tools.ietf.org/html/draft-uberti-behave-turn-rest-00). This allows
-the Home Server to generate credentials that are valid for use on the TURN
-server through the use of a secret shared between the Home Server and the
-TURN server.
-
-This document describes how to install coturn
-(https://github.com/coturn/coturn) which also supports the TURN REST API,
-and integrate it with synapse.
-
-coturn Setup
-============
-
-You may be able to setup coturn via your package manager,  or set it up manually using the usual ``configure, make, make install`` process.  
-
- 1. Check out coturn::
- 
-      git clone https://github.com/coturn/coturn.git coturn
-      cd coturn
-
- 2. Configure it::
- 
-      ./configure
-
-    You may need to install ``libevent2``: if so, you should do so
-    in the way recommended by your operating system.
-    You can ignore warnings about lack of database support: a
-    database is unnecessary for this purpose.
-
- 3. Build and install it::
- 
-      make
-      make install
-
- 4. Create or edit the config file in ``/etc/turnserver.conf``. The relevant
-    lines, with example values, are::
-
-      use-auth-secret
-      static-auth-secret=[your secret key here]
-      realm=turn.myserver.org
-
-    See turnserver.conf for explanations of the options.
-    One way to generate the static-auth-secret is with pwgen::
-
-       pwgen -s 64 1
-
- 5. Consider your security settings.  TURN lets users request a relay
-    which will connect to arbitrary IP addresses and ports.  At the least
-    we recommend::
-
-       # VoIP traffic is all UDP. There is no reason to let users connect to arbitrary TCP endpoints via the relay.
-       no-tcp-relay
-
-       # don't let the relay ever try to connect to private IP address ranges within your network (if any)
-       # given the turn server is likely behind your firewall, remember to include any privileged public IPs too.
-       denied-peer-ip=10.0.0.0-10.255.255.255
-       denied-peer-ip=192.168.0.0-192.168.255.255
-       denied-peer-ip=172.16.0.0-172.31.255.255
-
-       # special case the turn server itself so that client->TURN->TURN->client flows work
-       allowed-peer-ip=10.0.0.1
-
-       # consider whether you want to limit the quota of relayed streams per user (or total) to avoid risk of DoS.
-       user-quota=12 # 4 streams per video call, so 12 streams = 3 simultaneous relayed calls per user.
-       total-quota=1200
-
-    Ideally coturn should refuse to relay traffic which isn't SRTP;
-    see https://github.com/matrix-org/synapse/issues/2009
-
- 6. Ensure your firewall allows traffic into the TURN server on
-    the ports you've configured it to listen on (remember to allow
-    both TCP and UDP TURN traffic)
-
- 7. If you've configured coturn to support TLS/DTLS, generate or
-    import your private key and certificate.
-
- 8. Start the turn server::
- 
-       bin/turnserver -o
-
-
-synapse Setup
-=============
-
-Your home server configuration file needs the following extra keys:
-
- 1. "turn_uris": This needs to be a yaml list
-    of public-facing URIs for your TURN server to be given out 
-    to your clients. Add separate entries for each transport your
-    TURN server supports.
-
- 2. "turn_shared_secret": This is the secret shared between your Home
-    server and your TURN server, so you should set it to the same
-    string you used in turnserver.conf.
-
- 3. "turn_user_lifetime": This is the amount of time credentials
-    generated by your Home Server are valid for (in milliseconds).
-    Shorter times offer less potential for abuse at the expense
-    of increased traffic between web clients and your home server
-    to refresh credentials. The TURN REST API specification recommends
-    one day (86400000).
-
- 4. "turn_allow_guests": Whether to allow guest users to use the TURN
-    server.  This is enabled by default, as otherwise VoIP will not
-    work reliably for guests.  However, it does introduce a security risk
-    as it lets guests connect to arbitrary endpoints without having gone
-    through a CAPTCHA or similar to register a real account.
-
-As an example, here is the relevant section of the config file for
-matrix.org::
-
-    turn_uris: [ "turn:turn.matrix.org:3478?transport=udp", "turn:turn.matrix.org:3478?transport=tcp" ]
-    turn_shared_secret: n0t4ctuAllymatr1Xd0TorgSshar3d5ecret4obvIousreAsons
-    turn_user_lifetime: 86400000
-    turn_allow_guests: True
-
-Now, restart synapse::
-
-    cd /where/you/run/synapse
-    ./synctl restart
-
-...and your Home Server now supports VoIP relaying!
diff --git a/docs/workers.rst b/docs/workers.md
similarity index 78%
rename from docs/workers.rst
rename to docs/workers.md
index e11e117418..4bd60ba0a0 100644
--- a/docs/workers.rst
+++ b/docs/workers.md
@@ -1,5 +1,4 @@
-Scaling synapse via workers
-===========================
+# Scaling synapse via workers
 
 Synapse has experimental support for splitting out functionality into
 multiple separate python processes, helping greatly with scalability.  These
@@ -20,17 +19,16 @@ TCP protocol called 'replication' - analogous to MySQL or Postgres style
 database replication; feeding a stream of relevant data to the workers so they
 can be kept in sync with the main synapse process and database state.
 
-Configuration
--------------
+## Configuration
 
 To make effective use of the workers, you will need to configure an HTTP
 reverse-proxy such as nginx or haproxy, which will direct incoming requests to
 the correct worker, or to the main synapse instance. Note that this includes
-requests made to the federation port. See ``_ for
-information on setting up a reverse proxy.
+requests made to the federation port. See [reverse_proxy.md](reverse_proxy.md)
+for information on setting up a reverse proxy.
 
 To enable workers, you need to add two replication listeners to the master
-synapse, e.g.::
+synapse, e.g.:
 
     listeners:
       # The TCP replication port
@@ -56,7 +54,7 @@ You then create a set of configs for the various worker processes.  These
 should be worker configuration files, and should be stored in a dedicated
 subdirectory, to allow synctl to manipulate them. An additional configuration
 for the master synapse process will need to be created because the process will
-not be started automatically. That configuration should look like this::
+not be started automatically. That configuration should look like this:
 
     worker_app: synapse.app.homeserver
     daemonize: true
@@ -66,17 +64,17 @@ configuration file.  You can then override configuration specific to that worker
 e.g. the HTTP listener that it provides (if any); logging configuration; etc.
 You should minimise the number of overrides though to maintain a usable config.
 
-You must specify the type of worker application (``worker_app``). The currently
+You must specify the type of worker application (`worker_app`). The currently
 available worker applications are listed below. You must also specify the
 replication endpoints that it's talking to on the main synapse process.
-``worker_replication_host`` should specify the host of the main synapse,
-``worker_replication_port`` should point to the TCP replication listener port and
-``worker_replication_http_port`` should point to the HTTP replication port.
+`worker_replication_host` should specify the host of the main synapse,
+`worker_replication_port` should point to the TCP replication listener port and
+`worker_replication_http_port` should point to the HTTP replication port.
 
-Currently, the ``event_creator`` and ``federation_reader`` workers require specifying
-``worker_replication_http_port``.
+Currently, the `event_creator` and `federation_reader` workers require specifying
+`worker_replication_http_port`.
 
-For instance::
+For instance:
 
     worker_app: synapse.app.synchrotron
 
@@ -97,15 +95,15 @@ For instance::
     worker_log_config: /home/matrix/synapse/config/synchrotron_log_config.yaml
 
 ...is a full configuration for a synchrotron worker instance, which will expose a
-plain HTTP ``/sync`` endpoint on port 8083 separately from the ``/sync`` endpoint provided
+plain HTTP `/sync` endpoint on port 8083 separately from the `/sync` endpoint provided
 by the main synapse.
 
 Obviously you should configure your reverse-proxy to route the relevant
-endpoints to the worker (``localhost:8083`` in the above example).
+endpoints to the worker (`localhost:8083` in the above example).
 
 Finally, to actually run your worker-based synapse, you must pass synctl the -a
 commandline option to tell it to operate on all the worker configurations found
-in the given directory, e.g.::
+in the given directory, e.g.:
 
     synctl -a $CONFIG/workers start
 
@@ -114,28 +112,24 @@ synapse, unless you explicitly know it's safe not to.  For instance, restarting
 synapse without restarting all the synchrotrons may result in broken typing
 notifications.
 
-To manipulate a specific worker, you pass the -w option to synctl::
+To manipulate a specific worker, you pass the -w option to synctl:
 
     synctl -w $CONFIG/workers/synchrotron.yaml restart
 
+## Available worker applications
 
-Available worker applications
------------------------------
-
-``synapse.app.pusher``
-~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.pusher`
 
 Handles sending push notifications to sygnal and email. Doesn't handle any
-REST endpoints itself, but you should set ``start_pushers: False`` in the
+REST endpoints itself, but you should set `start_pushers: False` in the
 shared configuration file to stop the main synapse sending these notifications.
 
 Note this worker cannot be load-balanced: only one instance should be active.
 
-``synapse.app.synchrotron``
-~~~~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.synchrotron`
 
-The synchrotron handles ``sync`` requests from clients. In particular, it can
-handle REST endpoints matching the following regular expressions::
+The synchrotron handles `sync` requests from clients. In particular, it can
+handle REST endpoints matching the following regular expressions:
 
     ^/_matrix/client/(v2_alpha|r0)/sync$
     ^/_matrix/client/(api/v1|v2_alpha|r0)/events$
@@ -151,20 +145,18 @@ load-balance across the instances, though it will be more efficient if all
 requests from a particular user are routed to a single instance. Extracting
 a userid from the access token is currently left as an exercise for the reader.
 
-``synapse.app.appservice``
-~~~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.appservice`
 
 Handles sending output traffic to Application Services. Doesn't handle any
-REST endpoints itself, but you should set ``notify_appservices: False`` in the
+REST endpoints itself, but you should set `notify_appservices: False` in the
 shared configuration file to stop the main synapse sending these notifications.
 
 Note this worker cannot be load-balanced: only one instance should be active.
 
-``synapse.app.federation_reader``
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.federation_reader`
 
 Handles a subset of federation endpoints. In particular, it can handle REST
-endpoints matching the following regular expressions::
+endpoints matching the following regular expressions:
 
     ^/_matrix/federation/v1/event/
     ^/_matrix/federation/v1/state/
@@ -190,40 +182,36 @@ reverse-proxy configuration.
 The `^/_matrix/federation/v1/send/` endpoint must only be handled by a single
 instance.
 
-``synapse.app.federation_sender``
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.federation_sender`
 
 Handles sending federation traffic to other servers. Doesn't handle any
-REST endpoints itself, but you should set ``send_federation: False`` in the
+REST endpoints itself, but you should set `send_federation: False` in the
 shared configuration file to stop the main synapse sending this traffic.
 
 Note this worker cannot be load-balanced: only one instance should be active.
 
-``synapse.app.media_repository``
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.media_repository`
 
-Handles the media repository. It can handle all endpoints starting with::
+Handles the media repository. It can handle all endpoints starting with:
 
     /_matrix/media/
 
-And the following regular expressions matching media-specific administration
-APIs::
+And the following regular expressions matching media-specific administration APIs:
 
     ^/_synapse/admin/v1/purge_media_cache$
     ^/_synapse/admin/v1/room/.*/media$
     ^/_synapse/admin/v1/quarantine_media/.*$
 
-You should also set ``enable_media_repo: False`` in the shared configuration
+You should also set `enable_media_repo: False` in the shared configuration
 file to stop the main synapse running background jobs related to managing the
 media repository.
 
 Note this worker cannot be load-balanced: only one instance should be active.
 
-``synapse.app.client_reader``
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.client_reader`
 
 Handles client API endpoints. It can handle REST endpoints matching the
-following regular expressions::
+following regular expressions:
 
     ^/_matrix/client/(api/v1|r0|unstable)/publicRooms$
     ^/_matrix/client/(api/v1|r0|unstable)/rooms/.*/joined_members$
@@ -237,60 +225,55 @@ following regular expressions::
     ^/_matrix/client/versions$
     ^/_matrix/client/(api/v1|r0|unstable)/voip/turnServer$
 
-Additionally, the following REST endpoints can be handled for GET requests::
+Additionally, the following REST endpoints can be handled for GET requests:
 
     ^/_matrix/client/(api/v1|r0|unstable)/pushrules/.*$
 
 Additionally, the following REST endpoints can be handled, but all requests must
-be routed to the same instance::
+be routed to the same instance:
 
     ^/_matrix/client/(r0|unstable)/register$
 
 Pagination requests can also be handled, but all requests with the same path
 room must be routed to the same instance. Additionally, care must be taken to
 ensure that the purge history admin API is not used while pagination requests
-for the room are in flight::
+for the room are in flight:
 
     ^/_matrix/client/(api/v1|r0|unstable)/rooms/.*/messages$
 
-
-``synapse.app.user_dir``
-~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.user_dir`
 
 Handles searches in the user directory. It can handle REST endpoints matching
-the following regular expressions::
+the following regular expressions:
 
     ^/_matrix/client/(api/v1|r0|unstable)/user_directory/search$
 
-``synapse.app.frontend_proxy``
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+### `synapse.app.frontend_proxy`
 
 Proxies some frequently-requested client endpoints to add caching and remove
 load from the main synapse. It can handle REST endpoints matching the following
-regular expressions::
+regular expressions:
 
     ^/_matrix/client/(api/v1|r0|unstable)/keys/upload
 
-If ``use_presence`` is False in the homeserver config, it can also handle REST
-endpoints matching the following regular expressions::
+If `use_presence` is False in the homeserver config, it can also handle REST
+endpoints matching the following regular expressions:
 
     ^/_matrix/client/(api/v1|r0|unstable)/presence/[^/]+/status
 
-This "stub" presence handler will pass through ``GET`` request but make the
-``PUT`` effectively a no-op.
+This "stub" presence handler will pass through `GET` request but make the
+`PUT` effectively a no-op.
 
 It will proxy any requests it cannot handle to the main synapse instance. It
 must therefore be configured with the location of the main instance, via
-the ``worker_main_http_uri`` setting in the frontend_proxy worker configuration
-file. For example::
+the `worker_main_http_uri` setting in the `frontend_proxy` worker configuration
+file. For example:
 
     worker_main_http_uri: http://127.0.0.1:8008
 
+### `synapse.app.event_creator`
 
-``synapse.app.event_creator``
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-Handles some event creation. It can handle REST endpoints matching::
+Handles some event creation. It can handle REST endpoints matching:
 
     ^/_matrix/client/(api/v1|r0|unstable)/rooms/.*/send
     ^/_matrix/client/(api/v1|r0|unstable)/rooms/.*/(join|invite|leave|ban|unban|kick)$
diff --git a/mypy.ini b/mypy.ini
new file mode 100644
index 0000000000..1d77c0ecc8
--- /dev/null
+++ b/mypy.ini
@@ -0,0 +1,65 @@
+[mypy]
+namespace_packages = True
+plugins = mypy_zope:plugin
+follow_imports = normal
+check_untyped_defs = True
+show_error_codes = True
+show_traceback = True
+mypy_path = stubs
+
+[mypy-zope]
+ignore_missing_imports = True
+
+[mypy-constantly]
+ignore_missing_imports = True
+
+[mypy-twisted.*]
+ignore_missing_imports = True
+
+[mypy-treq.*]
+ignore_missing_imports = True
+
+[mypy-hyperlink]
+ignore_missing_imports = True
+
+[mypy-h11]
+ignore_missing_imports = True
+
+[mypy-opentracing]
+ignore_missing_imports = True
+
+[mypy-OpenSSL]
+ignore_missing_imports = True
+
+[mypy-netaddr]
+ignore_missing_imports = True
+
+[mypy-saml2.*]
+ignore_missing_imports = True
+
+[mypy-unpaddedbase64]
+ignore_missing_imports = True
+
+[mypy-canonicaljson]
+ignore_missing_imports = True
+
+[mypy-jaeger_client]
+ignore_missing_imports = True
+
+[mypy-jsonschema]
+ignore_missing_imports = True
+
+[mypy-signedjson.*]
+ignore_missing_imports = True
+
+[mypy-prometheus_client.*]
+ignore_missing_imports = True
+
+[mypy-service_identity.*]
+ignore_missing_imports = True
+
+[mypy-daemonize]
+ignore_missing_imports = True
+
+[mypy-sentry_sdk]
+ignore_missing_imports = True
diff --git a/scripts-dev/build_debian_packages b/scripts-dev/build_debian_packages
index 93305ee9b1..84eaec6a95 100755
--- a/scripts-dev/build_debian_packages
+++ b/scripts-dev/build_debian_packages
@@ -20,11 +20,13 @@ from concurrent.futures import ThreadPoolExecutor
 DISTS = (
     "debian:stretch",
     "debian:buster",
+    "debian:bullseye",
     "debian:sid",
     "ubuntu:xenial",
     "ubuntu:bionic",
     "ubuntu:cosmic",
     "ubuntu:disco",
+    "ubuntu:eoan",
 )
 
 DESC = '''\
diff --git a/scripts-dev/check_auth.py b/scripts-dev/check_auth.py
deleted file mode 100644
index 2a1c5f39d4..0000000000
--- a/scripts-dev/check_auth.py
+++ /dev/null
@@ -1,58 +0,0 @@
-from __future__ import print_function
-
-import argparse
-import itertools
-import json
-import sys
-
-from mock import Mock
-
-from synapse.api.auth import Auth
-from synapse.events import FrozenEvent
-
-
-def check_auth(auth, auth_chain, events):
-    auth_chain.sort(key=lambda e: e.depth)
-
-    auth_map = {e.event_id: e for e in auth_chain}
-
-    create_events = {}
-    for e in auth_chain:
-        if e.type == "m.room.create":
-            create_events[e.room_id] = e
-
-    for e in itertools.chain(auth_chain, events):
-        auth_events_list = [auth_map[i] for i, _ in e.auth_events]
-
-        auth_events = {(e.type, e.state_key): e for e in auth_events_list}
-
-        auth_events[("m.room.create", "")] = create_events[e.room_id]
-
-        try:
-            auth.check(e, auth_events=auth_events)
-        except Exception as ex:
-            print("Failed:", e.event_id, e.type, e.state_key)
-            print("Auth_events:", auth_events)
-            print(ex)
-            print(json.dumps(e.get_dict(), sort_keys=True, indent=4))
-            # raise
-        print("Success:", e.event_id, e.type, e.state_key)
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-
-    parser.add_argument(
-        "json", nargs="?", type=argparse.FileType("r"), default=sys.stdin
-    )
-
-    args = parser.parse_args()
-
-    js = json.load(args.json)
-
-    auth = Auth(Mock())
-    check_auth(
-        auth,
-        [FrozenEvent(d) for d in js["auth_chain"]],
-        [FrozenEvent(d) for d in js.get("pdus", [])],
-    )
diff --git a/scripts-dev/config-lint.sh b/scripts-dev/config-lint.sh
new file mode 100755
index 0000000000..677a854c85
--- /dev/null
+++ b/scripts-dev/config-lint.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+# Find linting errors in Synapse's default config file.
+# Exits with 0 if there are no problems, or another code otherwise.
+
+# Fix non-lowercase true/false values
+sed -i -E "s/: +True/: true/g; s/: +False/: false/g;" docs/sample_config.yaml
+
+# Check if anything changed
+git diff --exit-code docs/sample_config.yaml
diff --git a/scripts-dev/lint.sh b/scripts-dev/lint.sh
index ebb4d69f86..34c4854e11 100755
--- a/scripts-dev/lint.sh
+++ b/scripts-dev/lint.sh
@@ -7,6 +7,15 @@
 
 set -e
 
-isort -y -rc synapse tests scripts-dev scripts
-flake8 synapse tests
-python3 -m black synapse tests scripts-dev scripts
+if [ $# -ge 1 ]
+then
+  files=$*
+else
+  files="synapse tests scripts-dev scripts"
+fi
+
+echo "Linting these locations: $files"
+isort -y -rc $files
+flake8 $files
+python3 -m black $files
+./scripts-dev/config-lint.sh
diff --git a/scripts-dev/update_database b/scripts-dev/update_database
new file mode 100755
index 0000000000..27a1ad1e7e
--- /dev/null
+++ b/scripts-dev/update_database
@@ -0,0 +1,124 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import logging
+import sys
+
+import yaml
+
+from twisted.internet import defer, reactor
+
+from synapse.config.homeserver import HomeServerConfig
+from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.server import HomeServer
+from synapse.storage import DataStore
+from synapse.storage.engines import create_engine
+from synapse.storage.prepare_database import prepare_database
+
+logger = logging.getLogger("update_database")
+
+
+class MockHomeserver(HomeServer):
+    DATASTORE_CLASS = DataStore
+
+    def __init__(self, config, database_engine, db_conn, **kwargs):
+        super(MockHomeserver, self).__init__(
+            config.server_name,
+            reactor=reactor,
+            config=config,
+            database_engine=database_engine,
+            **kwargs
+        )
+
+        self.database_engine = database_engine
+        self.db_conn = db_conn
+
+    def get_db_conn(self):
+        return self.db_conn
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description=(
+            "Updates a synapse database to the latest schema and runs background updates"
+            " on it."
+        )
+    )
+    parser.add_argument("-v", action='store_true')
+    parser.add_argument(
+        "--database-config",
+        type=argparse.FileType('r'),
+        required=True,
+        help="A database config file for either a SQLite3 database or a PostgreSQL one.",
+    )
+
+    args = parser.parse_args()
+
+    logging_config = {
+        "level": logging.DEBUG if args.v else logging.INFO,
+        "format": "%(asctime)s - %(name)s - %(lineno)d - %(levelname)s - %(message)s",
+    }
+
+    logging.basicConfig(**logging_config)
+
+    # Load, process and sanity-check the config.
+    hs_config = yaml.safe_load(args.database_config)
+
+    if "database" not in hs_config:
+        sys.stderr.write("The configuration file must have a 'database' section.\n")
+        sys.exit(4)
+
+    config = HomeServerConfig()
+    config.parse_config_dict(hs_config, "", "")
+
+    # Create the database engine and a connection to it.
+    database_engine = create_engine(config.database_config)
+    db_conn = database_engine.module.connect(
+        **{
+            k: v
+            for k, v in config.database_config.get("args", {}).items()
+            if not k.startswith("cp_")
+        }
+    )
+
+    # Update the database to the latest schema.
+    prepare_database(db_conn, database_engine, config=config)
+    db_conn.commit()
+
+    # Instantiate and initialise the homeserver object.
+    hs = MockHomeserver(
+        config,
+        database_engine,
+        db_conn,
+        db_config=config.database_config,
+    )
+    # setup instantiates the store within the homeserver object.
+    hs.setup()
+    store = hs.get_datastore()
+
+    @defer.inlineCallbacks
+    def run_background_updates():
+        yield store.run_background_updates(sleep=False)
+        # Stop the reactor to exit the script once every background update is run.
+        reactor.stop()
+
+    # Apply all background updates on the database.
+    reactor.callWhenRunning(lambda: run_as_background_process(
+        "background_updates", run_background_updates
+    ))
+
+    reactor.run()
diff --git a/scripts/move_remote_media_to_new_store.py b/scripts/move_remote_media_to_new_store.py
index 12747c6024..b5b63933ab 100755
--- a/scripts/move_remote_media_to_new_store.py
+++ b/scripts/move_remote_media_to_new_store.py
@@ -72,7 +72,7 @@ def move_media(origin_server, file_id, src_paths, dest_paths):
     # check that the original exists
     original_file = src_paths.remote_media_filepath(origin_server, file_id)
     if not os.path.exists(original_file):
-        logger.warn(
+        logger.warning(
             "Original for %s/%s (%s) does not exist",
             origin_server,
             file_id,
diff --git a/scripts/synapse_port_db b/scripts/synapse_port_db
index b6ba19c776..0d3321682c 100755
--- a/scripts/synapse_port_db
+++ b/scripts/synapse_port_db
@@ -2,6 +2,7 @@
 # -*- coding: utf-8 -*-
 # Copyright 2015, 2016 OpenMarket Ltd
 # Copyright 2018 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -29,9 +30,33 @@ import yaml
 from twisted.enterprise import adbapi
 from twisted.internet import defer, reactor
 
-from synapse.storage._base import LoggingTransaction, SQLBaseStore
+from synapse.config.homeserver import HomeServerConfig
+from synapse.logging.context import PreserveLoggingContext
+from synapse.storage._base import LoggingTransaction
+from synapse.storage.data_stores.main.client_ips import ClientIpBackgroundUpdateStore
+from synapse.storage.data_stores.main.deviceinbox import (
+    DeviceInboxBackgroundUpdateStore,
+)
+from synapse.storage.data_stores.main.devices import DeviceBackgroundUpdateStore
+from synapse.storage.data_stores.main.events_bg_updates import (
+    EventsBackgroundUpdatesStore,
+)
+from synapse.storage.data_stores.main.media_repository import (
+    MediaRepositoryBackgroundUpdateStore,
+)
+from synapse.storage.data_stores.main.registration import (
+    RegistrationBackgroundUpdateStore,
+)
+from synapse.storage.data_stores.main.roommember import RoomMemberBackgroundUpdateStore
+from synapse.storage.data_stores.main.search import SearchBackgroundUpdateStore
+from synapse.storage.data_stores.main.state import StateBackgroundUpdateStore
+from synapse.storage.data_stores.main.stats import StatsStore
+from synapse.storage.data_stores.main.user_directory import (
+    UserDirectoryBackgroundUpdateStore,
+)
 from synapse.storage.engines import create_engine
 from synapse.storage.prepare_database import prepare_database
+from synapse.util import Clock
 
 logger = logging.getLogger("synapse_port_db")
 
@@ -43,6 +68,7 @@ BOOLEAN_COLUMNS = {
     "presence_list": ["accepted"],
     "presence_stream": ["currently_active"],
     "public_room_list_stream": ["visibility"],
+    "devices": ["hidden"],
     "device_lists_outbound_pokes": ["sent"],
     "users_who_share_rooms": ["share_private"],
     "groups": ["is_public"],
@@ -55,6 +81,8 @@ BOOLEAN_COLUMNS = {
     "local_group_membership": ["is_publicised", "is_admin"],
     "e2e_room_keys": ["is_verified"],
     "account_validity": ["email_sent"],
+    "redactions": ["have_censored"],
+    "room_stats_state": ["is_federatable"],
 }
 
 
@@ -96,33 +124,24 @@ APPEND_ONLY_TABLES = [
 end_error_exec_info = None
 
 
-class Store(object):
-    """This object is used to pull out some of the convenience API from the
-    Storage layer.
-
-    *All* database interactions should go through this object.
-    """
-
-    def __init__(self, db_pool, engine):
-        self.db_pool = db_pool
-        self.database_engine = engine
-
-    _simple_insert_txn = SQLBaseStore.__dict__["_simple_insert_txn"]
-    _simple_insert = SQLBaseStore.__dict__["_simple_insert"]
-
-    _simple_select_onecol_txn = SQLBaseStore.__dict__["_simple_select_onecol_txn"]
-    _simple_select_onecol = SQLBaseStore.__dict__["_simple_select_onecol"]
-    _simple_select_one = SQLBaseStore.__dict__["_simple_select_one"]
-    _simple_select_one_txn = SQLBaseStore.__dict__["_simple_select_one_txn"]
-    _simple_select_one_onecol = SQLBaseStore.__dict__["_simple_select_one_onecol"]
-    _simple_select_one_onecol_txn = SQLBaseStore.__dict__[
-        "_simple_select_one_onecol_txn"
-    ]
-
-    _simple_update_one = SQLBaseStore.__dict__["_simple_update_one"]
-    _simple_update_one_txn = SQLBaseStore.__dict__["_simple_update_one_txn"]
-    _simple_update_txn = SQLBaseStore.__dict__["_simple_update_txn"]
+class Store(
+    ClientIpBackgroundUpdateStore,
+    DeviceInboxBackgroundUpdateStore,
+    DeviceBackgroundUpdateStore,
+    EventsBackgroundUpdatesStore,
+    MediaRepositoryBackgroundUpdateStore,
+    RegistrationBackgroundUpdateStore,
+    RoomMemberBackgroundUpdateStore,
+    SearchBackgroundUpdateStore,
+    StateBackgroundUpdateStore,
+    UserDirectoryBackgroundUpdateStore,
+    StatsStore,
+):
+    def __init__(self, db_conn, hs):
+        super().__init__(db_conn, hs)
+        self.db_pool = hs.get_db_pool()
 
+    @defer.inlineCallbacks
     def runInteraction(self, desc, func, *args, **kwargs):
         def r(conn):
             try:
@@ -138,7 +157,7 @@ class Store(object):
                         )
                     except self.database_engine.module.DatabaseError as e:
                         if self.database_engine.is_deadlock(e):
-                            logger.warn("[TXN DEADLOCK] {%s} %d/%d", desc, i, N)
+                            logger.warning("[TXN DEADLOCK] {%s} %d/%d", desc, i, N)
                             if i < N:
                                 i += 1
                                 conn.rollback()
@@ -148,7 +167,8 @@ class Store(object):
                 logger.debug("[TXN FAIL] {%s} %s", desc, e)
                 raise
 
-        return self.db_pool.runWithConnection(r)
+        with PreserveLoggingContext():
+            return (yield self.db_pool.runWithConnection(r))
 
     def execute(self, f, *args, **kwargs):
         return self.runInteraction(f.__name__, f, *args, **kwargs)
@@ -174,6 +194,25 @@ class Store(object):
             raise
 
 
+class MockHomeserver:
+    def __init__(self, config, database_engine, db_conn, db_pool):
+        self.database_engine = database_engine
+        self.db_conn = db_conn
+        self.db_pool = db_pool
+        self.clock = Clock(reactor)
+        self.config = config
+        self.hostname = config.server_name
+
+    def get_db_conn(self):
+        return self.db_conn
+
+    def get_db_pool(self):
+        return self.db_pool
+
+    def get_clock(self):
+        return self.clock
+
+
 class Porter(object):
     def __init__(self, **kwargs):
         self.__dict__.update(kwargs)
@@ -393,7 +432,7 @@ class Porter(object):
                     for row in rows:
                         d = dict(zip(headers, row))
                         if "\0" in d['value']:
-                            logger.warn('dropping search row %s', d)
+                            logger.warning('dropping search row %s', d)
                         else:
                             rows_dict.append(d)
 
@@ -445,31 +484,75 @@ class Porter(object):
 
         db_conn.commit()
 
+        return db_conn
+
+    @defer.inlineCallbacks
+    def build_db_store(self, config):
+        """Builds and returns a database store using the provided configuration.
+
+        Args:
+            config: The database configuration, i.e. a dict following the structure of
+                the "database" section of Synapse's configuration file.
+
+        Returns:
+            The built Store object.
+        """
+        engine = create_engine(config)
+
+        self.progress.set_state("Preparing %s" % config["name"])
+        conn = self.setup_db(config, engine)
+
+        db_pool = adbapi.ConnectionPool(
+            config["name"], **config["args"]
+        )
+
+        hs = MockHomeserver(self.hs_config, engine, conn, db_pool)
+
+        store = Store(conn, hs)
+
+        yield store.runInteraction(
+            "%s_engine.check_database" % config["name"],
+            engine.check_database,
+        )
+
+        return store
+
+    @defer.inlineCallbacks
+    def run_background_updates_on_postgres(self):
+        # Manually apply all background updates on the PostgreSQL database.
+        postgres_ready = yield self.postgres_store.has_completed_background_updates()
+
+        if not postgres_ready:
+            # Only say that we're running background updates when there are background
+            # updates to run.
+            self.progress.set_state("Running background updates on PostgreSQL")
+
+        while not postgres_ready:
+            yield self.postgres_store.do_next_background_update(100)
+            postgres_ready = yield (
+                self.postgres_store.has_completed_background_updates()
+            )
+
     @defer.inlineCallbacks
     def run(self):
         try:
-            sqlite_db_pool = adbapi.ConnectionPool(
-                self.sqlite_config["name"], **self.sqlite_config["args"]
+            self.sqlite_store = yield self.build_db_store(self.sqlite_config)
+
+            # Check if all background updates are done, abort if not.
+            updates_complete = yield self.sqlite_store.has_completed_background_updates()
+            if not updates_complete:
+                sys.stderr.write(
+                    "Pending background updates exist in the SQLite3 database."
+                    " Please start Synapse again and wait until every update has finished"
+                    " before running this script.\n"
+                )
+                defer.returnValue(None)
+
+            self.postgres_store = yield self.build_db_store(
+                self.hs_config.database_config
             )
 
-            postgres_db_pool = adbapi.ConnectionPool(
-                self.postgres_config["name"], **self.postgres_config["args"]
-            )
-
-            sqlite_engine = create_engine(sqlite_config)
-            postgres_engine = create_engine(postgres_config)
-
-            self.sqlite_store = Store(sqlite_db_pool, sqlite_engine)
-            self.postgres_store = Store(postgres_db_pool, postgres_engine)
-
-            yield self.postgres_store.execute(postgres_engine.check_database)
-
-            # Step 1. Set up databases.
-            self.progress.set_state("Preparing SQLite3")
-            self.setup_db(sqlite_config, sqlite_engine)
-
-            self.progress.set_state("Preparing PostgreSQL")
-            self.setup_db(postgres_config, postgres_engine)
+            yield self.run_background_updates_on_postgres()
 
             self.progress.set_state("Creating port tables")
 
@@ -561,8 +644,10 @@ class Porter(object):
         def conv(j, col):
             if j in bool_cols:
                 return bool(col)
+            if isinstance(col, bytes):
+                return bytearray(col)
             elif isinstance(col, string_types) and "\0" in col:
-                logger.warn(
+                logger.warning(
                     "DROPPING ROW: NUL value in table %s col %s: %r",
                     table,
                     headers[j],
@@ -924,18 +1009,24 @@ if __name__ == "__main__":
         },
     }
 
-    postgres_config = yaml.safe_load(args.postgres_config)
+    hs_config = yaml.safe_load(args.postgres_config)
 
-    if "database" in postgres_config:
-        postgres_config = postgres_config["database"]
+    if "database" not in hs_config:
+        sys.stderr.write("The configuration file must have a 'database' section.\n")
+        sys.exit(4)
+
+    postgres_config = hs_config["database"]
 
     if "name" not in postgres_config:
-        sys.stderr.write("Malformed database config: no 'name'")
+        sys.stderr.write("Malformed database config: no 'name'\n")
         sys.exit(2)
     if postgres_config["name"] != "psycopg2":
-        sys.stderr.write("Database must use 'psycopg2' connector.")
+        sys.stderr.write("Database must use the 'psycopg2' connector.\n")
         sys.exit(3)
 
+    config = HomeServerConfig()
+    config.parse_config_dict(hs_config, "", "")
+
     def start(stdscr=None):
         if stdscr:
             progress = CursesProgress(stdscr)
@@ -944,9 +1035,9 @@ if __name__ == "__main__":
 
         porter = Porter(
             sqlite_config=sqlite_config,
-            postgres_config=postgres_config,
             progress=progress,
             batch_size=args.batch_size,
+            hs_config=config,
         )
 
         reactor.callWhenRunning(porter.run)
diff --git a/snap/snapcraft.yaml b/snap/snapcraft.yaml
new file mode 100644
index 0000000000..1f7df71db2
--- /dev/null
+++ b/snap/snapcraft.yaml
@@ -0,0 +1,22 @@
+name: matrix-synapse
+base: core18
+version: git 
+summary: Reference Matrix homeserver
+description: |
+  Synapse is the reference Matrix homeserver.
+  Matrix is a federated and decentralised instant messaging and VoIP system.
+
+grade: stable 
+confinement: strict 
+
+apps:
+  matrix-synapse: 
+    command: synctl --no-daemonize start $SNAP_COMMON/homeserver.yaml
+    stop-command: synctl -c $SNAP_COMMON stop
+    plugs: [network-bind, network]
+    daemon: simple 
+parts:
+  matrix-synapse:
+    source: .
+    plugin: python
+    python-version: python3
diff --git a/synapse/__init__.py b/synapse/__init__.py
index 6766ef445c..ec16f54a49 100644
--- a/synapse/__init__.py
+++ b/synapse/__init__.py
@@ -17,6 +17,7 @@
 """ This is a reference implementation of a Matrix home server.
 """
 
+import os
 import sys
 
 # Check that we're not running on an unsupported Python version.
@@ -35,4 +36,11 @@ try:
 except ImportError:
     pass
 
-__version__ = "1.3.1"
+__version__ = "1.5.1"
+
+if bool(os.environ.get("SYNAPSE_TEST_PATCH_LOG_CONTEXTS", False)):
+    # We import here so that we don't have to install a bunch of deps when
+    # running the packaging tox test.
+    from synapse.util.patch_inline_callbacks import do_patch
+
+    do_patch()
diff --git a/synapse/api/auth.py b/synapse/api/auth.py
index fd3cdf50b0..5d0b7d2801 100644
--- a/synapse/api/auth.py
+++ b/synapse/api/auth.py
@@ -25,7 +25,13 @@ from twisted.internet import defer
 import synapse.logging.opentracing as opentracing
 import synapse.types
 from synapse import event_auth
-from synapse.api.constants import EventTypes, JoinRules, Membership
+from synapse.api.constants import (
+    EventTypes,
+    JoinRules,
+    LimitBlockingTypes,
+    Membership,
+    UserTypes,
+)
 from synapse.api.errors import (
     AuthError,
     Codes,
@@ -84,27 +90,10 @@ class Auth(object):
         )
         auth_events = yield self.store.get_events(auth_events_ids)
         auth_events = {(e.type, e.state_key): e for e in itervalues(auth_events)}
-        self.check(
+        event_auth.check(
             room_version, event, auth_events=auth_events, do_sig_check=do_sig_check
         )
 
-    def check(self, room_version, event, auth_events, do_sig_check=True):
-        """ Checks if this event is correctly authed.
-
-        Args:
-            room_version (str): version of the room
-            event: the event being checked.
-            auth_events (dict: event-key -> event): the existing room state.
-
-
-        Returns:
-            True if the auth checks pass.
-        """
-        with Measure(self.clock, "auth.check"):
-            event_auth.check(
-                room_version, event, auth_events, do_sig_check=do_sig_check
-            )
-
     @defer.inlineCallbacks
     def check_joined_room(self, room_id, user_id, current_state=None):
         """Check if the user is currently joined in the room
@@ -179,7 +168,6 @@ class Auth(object):
     def get_public_keys(self, invite_event):
         return event_auth.get_public_keys(invite_event)
 
-    @opentracing.trace
     @defer.inlineCallbacks
     def get_user_by_req(
         self, request, allow_guest=False, rights="access", allow_expired=False
@@ -212,6 +200,7 @@ class Auth(object):
             if user_id:
                 request.authenticated_entity = user_id
                 opentracing.set_tag("authenticated_entity", user_id)
+                opentracing.set_tag("appservice_id", app_service.id)
 
                 if ip_addr and self.hs.config.track_appservice_user_ips:
                     yield self.store.insert_client_ip(
@@ -263,6 +252,8 @@ class Auth(object):
 
             request.authenticated_entity = user.to_string()
             opentracing.set_tag("authenticated_entity", user.to_string())
+            if device_id:
+                opentracing.set_tag("device_id", device_id)
 
             return synapse.types.create_requester(
                 user, token_id, is_guest, device_id, app_service=app_service
@@ -506,7 +497,7 @@ class Auth(object):
         token = self.get_access_token_from_request(request)
         service = self.store.get_app_service_by_token(token)
         if not service:
-            logger.warn("Unrecognised appservice access token.")
+            logger.warning("Unrecognised appservice access token.")
             raise InvalidClientTokenError()
         request.authenticated_entity = service.sender
         return defer.succeed(service)
@@ -704,13 +695,12 @@ class Auth(object):
                 and visibility.content["history_visibility"] == "world_readable"
             ):
                 return Membership.JOIN, None
-                return
             raise AuthError(
                 403, "Guest access not allowed", errcode=Codes.GUEST_ACCESS_FORBIDDEN
             )
 
     @defer.inlineCallbacks
-    def check_auth_blocking(self, user_id=None, threepid=None):
+    def check_auth_blocking(self, user_id=None, threepid=None, user_type=None):
         """Checks if the user should be rejected for some external reason,
         such as monthly active user limiting or global disable flag
 
@@ -723,6 +713,9 @@ class Auth(object):
                 with a MAU blocked server, normally they would be rejected but their
                 threepid is on the reserved list. user_id and
                 threepid should never be set at the same time.
+
+            user_type(str|None): If present, is used to decide whether to check against
+                certain blocking reasons like MAU.
         """
 
         # Never fail an auth check for the server notices users or support user
@@ -739,7 +732,7 @@ class Auth(object):
                 self.hs.config.hs_disabled_message,
                 errcode=Codes.RESOURCE_LIMIT_EXCEEDED,
                 admin_contact=self.hs.config.admin_contact,
-                limit_type=self.hs.config.hs_disabled_limit_type,
+                limit_type=LimitBlockingTypes.HS_DISABLED,
             )
         if self.hs.config.limit_usage_by_mau is True:
             assert not (user_id and threepid)
@@ -760,6 +753,10 @@ class Auth(object):
                     self.hs.config.mau_limits_reserved_threepids, threepid
                 ):
                     return
+            elif user_type == UserTypes.SUPPORT:
+                # If the user does not exist yet and is of type "support",
+                # allow registration. Support users are excluded from MAU checks.
+                return
             # Else if there is no room in the MAU bucket, bail
             current_mau = yield self.store.get_monthly_active_count()
             if current_mau >= self.hs.config.max_mau_value:
@@ -768,5 +765,5 @@ class Auth(object):
                     "Monthly Active User Limit Exceeded",
                     admin_contact=self.hs.config.admin_contact,
                     errcode=Codes.RESOURCE_LIMIT_EXCEEDED,
-                    limit_type="monthly_active_user",
+                    limit_type=LimitBlockingTypes.MONTHLY_ACTIVE_USER,
                 )
diff --git a/synapse/api/constants.py b/synapse/api/constants.py
index f29bce560c..49c4b85054 100644
--- a/synapse/api/constants.py
+++ b/synapse/api/constants.py
@@ -97,8 +97,6 @@ class EventTypes(object):
 
 class RejectedReason(object):
     AUTH_ERROR = "auth_error"
-    REPLACED = "replaced"
-    NOT_ANCESTOR = "not_ancestor"
 
 
 class RoomCreationPreset(object):
@@ -133,3 +131,17 @@ class RelationTypes(object):
     ANNOTATION = "m.annotation"
     REPLACE = "m.replace"
     REFERENCE = "m.reference"
+
+
+class LimitBlockingTypes(object):
+    """Reasons that a server may be blocked"""
+
+    MONTHLY_ACTIVE_USER = "monthly_active_user"
+    HS_DISABLED = "hs_disabled"
+
+
+class EventContentFields(object):
+    """Fields found in events' content, regardless of type."""
+
+    # Labels for the event, cf https://github.com/matrix-org/matrix-doc/pull/2326
+    LABELS = "org.matrix.labels"
diff --git a/synapse/api/errors.py b/synapse/api/errors.py
index cf1ebf1af2..cca92c34ba 100644
--- a/synapse/api/errors.py
+++ b/synapse/api/errors.py
@@ -17,6 +17,7 @@
 """Contains exceptions and error codes."""
 
 import logging
+from typing import Dict
 
 from six import iteritems
 from six.moves import http_client
@@ -61,6 +62,7 @@ class Codes(object):
     INCOMPATIBLE_ROOM_VERSION = "M_INCOMPATIBLE_ROOM_VERSION"
     WRONG_ROOM_KEYS_VERSION = "M_WRONG_ROOM_KEYS_VERSION"
     EXPIRED_ACCOUNT = "ORG_MATRIX_EXPIRED_ACCOUNT"
+    INVALID_SIGNATURE = "M_INVALID_SIGNATURE"
     USER_DEACTIVATED = "M_USER_DEACTIVATED"
 
 
@@ -111,7 +113,7 @@ class ProxiedRequestError(SynapseError):
     def __init__(self, code, msg, errcode=Codes.UNKNOWN, additional_fields=None):
         super(ProxiedRequestError, self).__init__(code, msg, errcode)
         if additional_fields is None:
-            self._additional_fields = {}
+            self._additional_fields = {}  # type: Dict
         else:
             self._additional_fields = dict(additional_fields)
 
diff --git a/synapse/api/filtering.py b/synapse/api/filtering.py
index 9f06556bd2..bec13f08d8 100644
--- a/synapse/api/filtering.py
+++ b/synapse/api/filtering.py
@@ -20,6 +20,7 @@ from jsonschema import FormatChecker
 
 from twisted.internet import defer
 
+from synapse.api.constants import EventContentFields
 from synapse.api.errors import SynapseError
 from synapse.storage.presence import UserPresenceState
 from synapse.types import RoomID, UserID
@@ -66,6 +67,10 @@ ROOM_EVENT_FILTER_SCHEMA = {
         "contains_url": {"type": "boolean"},
         "lazy_load_members": {"type": "boolean"},
         "include_redundant_members": {"type": "boolean"},
+        # Include or exclude events with the provided labels.
+        # cf https://github.com/matrix-org/matrix-doc/pull/2326
+        "org.matrix.labels": {"type": "array", "items": {"type": "string"}},
+        "org.matrix.not_labels": {"type": "array", "items": {"type": "string"}},
     },
 }
 
@@ -259,6 +264,9 @@ class Filter(object):
 
         self.contains_url = self.filter_json.get("contains_url", None)
 
+        self.labels = self.filter_json.get("org.matrix.labels", None)
+        self.not_labels = self.filter_json.get("org.matrix.not_labels", [])
+
     def filters_all_types(self):
         return "*" in self.not_types
 
@@ -282,6 +290,7 @@ class Filter(object):
             room_id = None
             ev_type = "m.presence"
             contains_url = False
+            labels = []
         else:
             sender = event.get("sender", None)
             if not sender:
@@ -300,10 +309,11 @@ class Filter(object):
             content = event.get("content", {})
             # check if there is a string url field in the content for filtering purposes
             contains_url = isinstance(content.get("url"), text_type)
+            labels = content.get(EventContentFields.LABELS, [])
 
-        return self.check_fields(room_id, sender, ev_type, contains_url)
+        return self.check_fields(room_id, sender, ev_type, labels, contains_url)
 
-    def check_fields(self, room_id, sender, event_type, contains_url):
+    def check_fields(self, room_id, sender, event_type, labels, contains_url):
         """Checks whether the filter matches the given event fields.
 
         Returns:
@@ -313,6 +323,7 @@ class Filter(object):
             "rooms": lambda v: room_id == v,
             "senders": lambda v: sender == v,
             "types": lambda v: _matches_wildcard(event_type, v),
+            "labels": lambda v: v in labels,
         }
 
         for name, match_func in literal_keys.items():
diff --git a/synapse/api/room_versions.py b/synapse/api/room_versions.py
index 95292b7dec..c6f50fd7b9 100644
--- a/synapse/api/room_versions.py
+++ b/synapse/api/room_versions.py
@@ -12,6 +12,9 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
+from typing import Dict
+
 import attr
 
 
@@ -102,4 +105,4 @@ KNOWN_ROOM_VERSIONS = {
         RoomVersions.V4,
         RoomVersions.V5,
     )
-}  # type: dict[str, RoomVersion]
+}  # type: Dict[str, RoomVersion]
diff --git a/synapse/app/__init__.py b/synapse/app/__init__.py
index d877c77834..a01bac2997 100644
--- a/synapse/app/__init__.py
+++ b/synapse/app/__init__.py
@@ -44,6 +44,8 @@ def check_bind_error(e, address, bind_addresses):
         bind_addresses (list): Addresses on which the service listens.
     """
     if address == "0.0.0.0" and "::" in bind_addresses:
-        logger.warn("Failed to listen on 0.0.0.0, continuing because listening on [::]")
+        logger.warning(
+            "Failed to listen on 0.0.0.0, continuing because listening on [::]"
+        )
     else:
         raise e
diff --git a/synapse/app/_base.py b/synapse/app/_base.py
index c30fdeee9a..2ac7d5c064 100644
--- a/synapse/app/_base.py
+++ b/synapse/app/_base.py
@@ -263,7 +263,9 @@ def start(hs, listeners=None):
         refresh_certificate(hs)
 
         # Start the tracer
-        synapse.logging.opentracing.init_tracer(hs.config)
+        synapse.logging.opentracing.init_tracer(  # type: ignore[attr-defined] # noqa
+            hs.config
+        )
 
         # It is now safe to start your Synapse.
         hs.start_listening(listeners)
diff --git a/synapse/app/appservice.py b/synapse/app/appservice.py
index 767b87d2db..02b900f382 100644
--- a/synapse/app/appservice.py
+++ b/synapse/app/appservice.py
@@ -94,7 +94,7 @@ class AppserviceServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -103,7 +103,7 @@ class AppserviceServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/client_reader.py b/synapse/app/client_reader.py
index 86193d35a8..dadb487d5f 100644
--- a/synapse/app/client_reader.py
+++ b/synapse/app/client_reader.py
@@ -119,7 +119,7 @@ class ClientReaderServer(HomeServer):
                     KeyChangesServlet(self).register(resource)
                     VoipRestServlet(self).register(resource)
                     PushRuleRestServlet(self).register(resource)
-                    VersionsRestServlet().register(resource)
+                    VersionsRestServlet(self).register(resource)
 
                     resources.update({"/_matrix/client": resource})
 
@@ -153,7 +153,7 @@ class ClientReaderServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -162,7 +162,7 @@ class ClientReaderServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/event_creator.py b/synapse/app/event_creator.py
index c67fe69a50..d110599a35 100644
--- a/synapse/app/event_creator.py
+++ b/synapse/app/event_creator.py
@@ -56,8 +56,8 @@ from synapse.rest.client.v1.room import (
     RoomStateEventRestServlet,
 )
 from synapse.server import HomeServer
+from synapse.storage.data_stores.main.user_directory import UserDirectoryStore
 from synapse.storage.engines import create_engine
-from synapse.storage.user_directory import UserDirectoryStore
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.manhole import manhole
 from synapse.util.versionstring import get_version_string
@@ -147,7 +147,7 @@ class EventCreatorServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -156,7 +156,7 @@ class EventCreatorServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/federation_reader.py b/synapse/app/federation_reader.py
index 1ef027a88c..418c086254 100644
--- a/synapse/app/federation_reader.py
+++ b/synapse/app/federation_reader.py
@@ -132,7 +132,7 @@ class FederationReaderServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -141,7 +141,7 @@ class FederationReaderServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/federation_sender.py b/synapse/app/federation_sender.py
index 04fbb407af..139221ad34 100644
--- a/synapse/app/federation_sender.py
+++ b/synapse/app/federation_sender.py
@@ -123,7 +123,7 @@ class FederationSenderServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -132,7 +132,7 @@ class FederationSenderServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/frontend_proxy.py b/synapse/app/frontend_proxy.py
index 9504bfbc70..e647459d0e 100644
--- a/synapse/app/frontend_proxy.py
+++ b/synapse/app/frontend_proxy.py
@@ -204,7 +204,7 @@ class FrontendProxyServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -213,7 +213,7 @@ class FrontendProxyServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/homeserver.py b/synapse/app/homeserver.py
index 04f1ed14f3..73e2c29d06 100644
--- a/synapse/app/homeserver.py
+++ b/synapse/app/homeserver.py
@@ -19,12 +19,13 @@ from __future__ import print_function
 
 import gc
 import logging
+import math
 import os
+import resource
 import sys
 
 from six import iteritems
 
-import psutil
 from prometheus_client import Gauge
 
 from twisted.application import service
@@ -282,7 +283,7 @@ class SynapseHomeServer(HomeServer):
                     reactor.addSystemEventTrigger("before", "shutdown", s.stopListening)
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -291,7 +292,7 @@ class SynapseHomeServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
     def run_startup_checks(self, db_conn, database_engine):
         all_users_native = are_all_users_on_domain(
@@ -471,6 +472,87 @@ class SynapseService(service.Service):
         return self._port.stopListening()
 
 
+# Contains the list of processes we will be monitoring
+# currently either 0 or 1
+_stats_process = []
+
+
+@defer.inlineCallbacks
+def phone_stats_home(hs, stats, stats_process=_stats_process):
+    logger.info("Gathering stats for reporting")
+    now = int(hs.get_clock().time())
+    uptime = int(now - hs.start_time)
+    if uptime < 0:
+        uptime = 0
+
+    stats["homeserver"] = hs.config.server_name
+    stats["server_context"] = hs.config.server_context
+    stats["timestamp"] = now
+    stats["uptime_seconds"] = uptime
+    version = sys.version_info
+    stats["python_version"] = "{}.{}.{}".format(
+        version.major, version.minor, version.micro
+    )
+    stats["total_users"] = yield hs.get_datastore().count_all_users()
+
+    total_nonbridged_users = yield hs.get_datastore().count_nonbridged_users()
+    stats["total_nonbridged_users"] = total_nonbridged_users
+
+    daily_user_type_results = yield hs.get_datastore().count_daily_user_type()
+    for name, count in iteritems(daily_user_type_results):
+        stats["daily_user_type_" + name] = count
+
+    room_count = yield hs.get_datastore().get_room_count()
+    stats["total_room_count"] = room_count
+
+    stats["daily_active_users"] = yield hs.get_datastore().count_daily_users()
+    stats["monthly_active_users"] = yield hs.get_datastore().count_monthly_users()
+    stats["daily_active_rooms"] = yield hs.get_datastore().count_daily_active_rooms()
+    stats["daily_messages"] = yield hs.get_datastore().count_daily_messages()
+
+    r30_results = yield hs.get_datastore().count_r30_users()
+    for name, count in iteritems(r30_results):
+        stats["r30_users_" + name] = count
+
+    daily_sent_messages = yield hs.get_datastore().count_daily_sent_messages()
+    stats["daily_sent_messages"] = daily_sent_messages
+    stats["cache_factor"] = CACHE_SIZE_FACTOR
+    stats["event_cache_size"] = hs.config.event_cache_size
+
+    #
+    # Performance statistics
+    #
+    old = stats_process[0]
+    new = (now, resource.getrusage(resource.RUSAGE_SELF))
+    stats_process[0] = new
+
+    # Get RSS in bytes
+    stats["memory_rss"] = new[1].ru_maxrss
+
+    # Get CPU time in % of a single core, not % of all cores
+    used_cpu_time = (new[1].ru_utime + new[1].ru_stime) - (
+        old[1].ru_utime + old[1].ru_stime
+    )
+    if used_cpu_time == 0 or new[0] == old[0]:
+        stats["cpu_average"] = 0
+    else:
+        stats["cpu_average"] = math.floor(used_cpu_time / (new[0] - old[0]) * 100)
+
+    #
+    # Database version
+    #
+
+    stats["database_engine"] = hs.get_datastore().database_engine_name
+    stats["database_server_version"] = hs.get_datastore().get_server_version()
+    logger.info("Reporting stats to %s: %s" % (hs.config.report_stats_endpoint, stats))
+    try:
+        yield hs.get_proxied_http_client().put_json(
+            hs.config.report_stats_endpoint, stats
+        )
+    except Exception as e:
+        logger.warning("Error reporting stats: %s", e)
+
+
 def run(hs):
     PROFILE_SYNAPSE = False
     if PROFILE_SYNAPSE:
@@ -497,89 +579,19 @@ def run(hs):
         reactor.run = profile(reactor.run)
 
     clock = hs.get_clock()
-    start_time = clock.time()
 
     stats = {}
 
-    # Contains the list of processes we will be monitoring
-    # currently either 0 or 1
-    stats_process = []
+    def performance_stats_init():
+        _stats_process.clear()
+        _stats_process.append(
+            (int(hs.get_clock().time(), resource.getrusage(resource.RUSAGE_SELF)))
+        )
 
     def start_phone_stats_home():
-        return run_as_background_process("phone_stats_home", phone_stats_home)
-
-    @defer.inlineCallbacks
-    def phone_stats_home():
-        logger.info("Gathering stats for reporting")
-        now = int(hs.get_clock().time())
-        uptime = int(now - start_time)
-        if uptime < 0:
-            uptime = 0
-
-        stats["homeserver"] = hs.config.server_name
-        stats["server_context"] = hs.config.server_context
-        stats["timestamp"] = now
-        stats["uptime_seconds"] = uptime
-        version = sys.version_info
-        stats["python_version"] = "{}.{}.{}".format(
-            version.major, version.minor, version.micro
+        return run_as_background_process(
+            "phone_stats_home", phone_stats_home, hs, stats
         )
-        stats["total_users"] = yield hs.get_datastore().count_all_users()
-
-        total_nonbridged_users = yield hs.get_datastore().count_nonbridged_users()
-        stats["total_nonbridged_users"] = total_nonbridged_users
-
-        daily_user_type_results = yield hs.get_datastore().count_daily_user_type()
-        for name, count in iteritems(daily_user_type_results):
-            stats["daily_user_type_" + name] = count
-
-        room_count = yield hs.get_datastore().get_room_count()
-        stats["total_room_count"] = room_count
-
-        stats["daily_active_users"] = yield hs.get_datastore().count_daily_users()
-        stats["monthly_active_users"] = yield hs.get_datastore().count_monthly_users()
-        stats[
-            "daily_active_rooms"
-        ] = yield hs.get_datastore().count_daily_active_rooms()
-        stats["daily_messages"] = yield hs.get_datastore().count_daily_messages()
-
-        r30_results = yield hs.get_datastore().count_r30_users()
-        for name, count in iteritems(r30_results):
-            stats["r30_users_" + name] = count
-
-        daily_sent_messages = yield hs.get_datastore().count_daily_sent_messages()
-        stats["daily_sent_messages"] = daily_sent_messages
-        stats["cache_factor"] = CACHE_SIZE_FACTOR
-        stats["event_cache_size"] = hs.config.event_cache_size
-
-        if len(stats_process) > 0:
-            stats["memory_rss"] = 0
-            stats["cpu_average"] = 0
-            for process in stats_process:
-                stats["memory_rss"] += process.memory_info().rss
-                stats["cpu_average"] += int(process.cpu_percent(interval=None))
-
-        stats["database_engine"] = hs.get_datastore().database_engine_name
-        stats["database_server_version"] = hs.get_datastore().get_server_version()
-        logger.info("Reporting stats to matrix.org: %s" % (stats,))
-        try:
-            yield hs.get_simple_http_client().put_json(
-                "https://matrix.org/report-usage-stats/push", stats
-            )
-        except Exception as e:
-            logger.warn("Error reporting stats: %s", e)
-
-    def performance_stats_init():
-        try:
-            process = psutil.Process()
-            # Ensure we can fetch both, and make the initial request for cpu_percent
-            # so the next request will use this as the initial point.
-            process.memory_info().rss
-            process.cpu_percent(interval=None)
-            logger.info("report_stats can use psutil")
-            stats_process.append(process)
-        except (AttributeError):
-            logger.warning("Unable to read memory/cpu stats. Disabling reporting.")
 
     def generate_user_daily_visit_stats():
         return run_as_background_process(
@@ -603,13 +615,13 @@ def run(hs):
     @defer.inlineCallbacks
     def generate_monthly_active_users():
         current_mau_count = 0
-        reserved_count = 0
+        reserved_users = ()
         store = hs.get_datastore()
         if hs.config.limit_usage_by_mau or hs.config.mau_stats_only:
             current_mau_count = yield store.get_monthly_active_count()
-            reserved_count = yield store.get_registered_reserved_users_count()
+            reserved_users = yield store.get_registered_reserved_users()
         current_mau_gauge.set(float(current_mau_count))
-        registered_reserved_users_mau_gauge.set(float(reserved_count))
+        registered_reserved_users_mau_gauge.set(float(len(reserved_users)))
         max_mau_gauge.set(float(hs.config.max_mau_value))
 
     def start_generate_monthly_active_users():
diff --git a/synapse/app/media_repository.py b/synapse/app/media_repository.py
index 2ac783ffa3..2c6dd3ef02 100644
--- a/synapse/app/media_repository.py
+++ b/synapse/app/media_repository.py
@@ -39,8 +39,8 @@ from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.rest.admin import register_servlets_for_media_repo
 from synapse.rest.media.v0.content_repository import ContentRepoResource
 from synapse.server import HomeServer
+from synapse.storage.data_stores.main.media_repository import MediaRepositoryStore
 from synapse.storage.engines import create_engine
-from synapse.storage.media_repository import MediaRepositoryStore
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.manhole import manhole
 from synapse.util.versionstring import get_version_string
@@ -120,7 +120,7 @@ class MediaRepositoryServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -129,7 +129,7 @@ class MediaRepositoryServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/pusher.py b/synapse/app/pusher.py
index d84732ee3c..01a5ffc363 100644
--- a/synapse/app/pusher.py
+++ b/synapse/app/pusher.py
@@ -114,7 +114,7 @@ class PusherServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -123,7 +123,7 @@ class PusherServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index 473026fce5..b14da09f47 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -54,8 +54,8 @@ from synapse.rest.client.v1.initial_sync import InitialSyncRestServlet
 from synapse.rest.client.v1.room import RoomInitialSyncRestServlet
 from synapse.rest.client.v2_alpha import sync
 from synapse.server import HomeServer
+from synapse.storage.data_stores.main.presence import UserPresenceState
 from synapse.storage.engines import create_engine
-from synapse.storage.presence import UserPresenceState
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.manhole import manhole
 from synapse.util.stringutils import random_string
@@ -326,7 +326,7 @@ class SynchrotronServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -335,7 +335,7 @@ class SynchrotronServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/app/user_dir.py b/synapse/app/user_dir.py
index e01afb39f2..6cb100319f 100644
--- a/synapse/app/user_dir.py
+++ b/synapse/app/user_dir.py
@@ -42,8 +42,8 @@ from synapse.replication.tcp.streams.events import (
 )
 from synapse.rest.client.v2_alpha import user_directory
 from synapse.server import HomeServer
+from synapse.storage.data_stores.main.user_directory import UserDirectoryStore
 from synapse.storage.engines import create_engine
-from synapse.storage.user_directory import UserDirectoryStore
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.manhole import manhole
@@ -150,7 +150,7 @@ class UserDirectoryServer(HomeServer):
                 )
             elif listener["type"] == "metrics":
                 if not self.get_config().enable_metrics:
-                    logger.warn(
+                    logger.warning(
                         (
                             "Metrics listener configured, but "
                             "enable_metrics is not True!"
@@ -159,7 +159,7 @@ class UserDirectoryServer(HomeServer):
                 else:
                     _base.listen_metrics(listener["bind_addresses"], listener["port"])
             else:
-                logger.warn("Unrecognized listener type: %s", listener["type"])
+                logger.warning("Unrecognized listener type: %s", listener["type"])
 
         self.get_tcp_replication().start_replication(self)
 
diff --git a/synapse/appservice/__init__.py b/synapse/appservice/__init__.py
index 33b3579425..aea3985a5f 100644
--- a/synapse/appservice/__init__.py
+++ b/synapse/appservice/__init__.py
@@ -94,7 +94,9 @@ class ApplicationService(object):
         ip_range_whitelist=None,
     ):
         self.token = token
-        self.url = url
+        self.url = (
+            url.rstrip("/") if isinstance(url, str) else None
+        )  # url must not end with a slash
         self.hs_token = hs_token
         self.sender = sender
         self.server_name = hostname
diff --git a/synapse/appservice/api.py b/synapse/appservice/api.py
index 007ca75a94..3e25bf5747 100644
--- a/synapse/appservice/api.py
+++ b/synapse/appservice/api.py
@@ -107,7 +107,6 @@ class ApplicationServiceApi(SimpleHttpClient):
         except CodeMessageException as e:
             if e.code == 404:
                 return False
-                return
             logger.warning("query_user to %s received %s", uri, e.code)
         except Exception as ex:
             logger.warning("query_user to %s threw exception %s", uri, ex)
@@ -127,7 +126,6 @@ class ApplicationServiceApi(SimpleHttpClient):
             logger.warning("query_alias to %s received %s", uri, e.code)
             if e.code == 404:
                 return False
-                return
         except Exception as ex:
             logger.warning("query_alias to %s threw exception %s", uri, ex)
         return False
@@ -230,7 +228,6 @@ class ApplicationServiceApi(SimpleHttpClient):
             sent_transactions_counter.labels(service.id).inc()
             sent_events_counter.labels(service.id).inc(len(events))
             return True
-            return
         except CodeMessageException as e:
             logger.warning("push_bulk to %s received %s", uri, e.code)
         except Exception as ex:
diff --git a/synapse/config/_base.py b/synapse/config/_base.py
index 31f6530978..08619404bb 100644
--- a/synapse/config/_base.py
+++ b/synapse/config/_base.py
@@ -18,7 +18,9 @@
 import argparse
 import errno
 import os
+from collections import OrderedDict
 from textwrap import dedent
+from typing import Any, MutableMapping, Optional
 
 from six import integer_types
 
@@ -51,7 +53,56 @@ Missing mandatory `server_name` config option.
 """
 
 
+def path_exists(file_path):
+    """Check if a file exists
+
+    Unlike os.path.exists, this throws an exception if there is an error
+    checking if the file exists (for example, if there is a perms error on
+    the parent dir).
+
+    Returns:
+        bool: True if the file exists; False if not.
+    """
+    try:
+        os.stat(file_path)
+        return True
+    except OSError as e:
+        if e.errno != errno.ENOENT:
+            raise e
+        return False
+
+
 class Config(object):
+    """
+    A configuration section, containing configuration keys and values.
+
+    Attributes:
+        section (str): The section title of this config object, such as
+            "tls" or "logger". This is used to refer to it on the root
+            logger (for example, `config.tls.some_option`). Must be
+            defined in subclasses.
+    """
+
+    section = None
+
+    def __init__(self, root_config=None):
+        self.root = root_config
+
+    def __getattr__(self, item: str) -> Any:
+        """
+        Try and fetch a configuration option that does not exist on this class.
+
+        This is so that existing configs that rely on `self.value`, where value
+        is actually from a different config section, continue to work.
+        """
+        if item in ["generate_config_section", "read_config"]:
+            raise AttributeError(item)
+
+        if self.root is None:
+            raise AttributeError(item)
+        else:
+            return self.root._get_unclassed_config(self.section, item)
+
     @staticmethod
     def parse_size(value):
         if isinstance(value, integer_types):
@@ -88,22 +139,7 @@ class Config(object):
 
     @classmethod
     def path_exists(cls, file_path):
-        """Check if a file exists
-
-        Unlike os.path.exists, this throws an exception if there is an error
-        checking if the file exists (for example, if there is a perms error on
-        the parent dir).
-
-        Returns:
-            bool: True if the file exists; False if not.
-        """
-        try:
-            os.stat(file_path)
-            return True
-        except OSError as e:
-            if e.errno != errno.ENOENT:
-                raise e
-            return False
+        return path_exists(file_path)
 
     @classmethod
     def check_file(cls, file_path, config_name):
@@ -136,42 +172,106 @@ class Config(object):
         with open(file_path) as file_stream:
             return file_stream.read()
 
-    def invoke_all(self, name, *args, **kargs):
-        """Invoke all instance methods with the given name and arguments in the
-        class's MRO.
+
+class RootConfig(object):
+    """
+    Holder of an application's configuration.
+
+    What configuration this object holds is defined by `config_classes`, a list
+    of Config classes that will be instantiated and given the contents of a
+    configuration file to read. They can then be accessed on this class by their
+    section name, defined in the Config or dynamically set to be the name of the
+    class, lower-cased and with "Config" removed.
+    """
+
+    config_classes = []
+
+    def __init__(self):
+        self._configs = OrderedDict()
+
+        for config_class in self.config_classes:
+            if config_class.section is None:
+                raise ValueError("%r requires a section name" % (config_class,))
+
+            try:
+                conf = config_class(self)
+            except Exception as e:
+                raise Exception("Failed making %s: %r" % (config_class.section, e))
+            self._configs[config_class.section] = conf
+
+    def __getattr__(self, item: str) -> Any:
+        """
+        Redirect lookups on this object either to config objects, or values on
+        config objects, so that `config.tls.blah` works, as well as legacy uses
+        of things like `config.server_name`. It will first look up the config
+        section name, and then values on those config classes.
+        """
+        if item in self._configs.keys():
+            return self._configs[item]
+
+        return self._get_unclassed_config(None, item)
+
+    def _get_unclassed_config(self, asking_section: Optional[str], item: str):
+        """
+        Fetch a config value from one of the instantiated config classes that
+        has not been fetched directly.
 
         Args:
-            name (str): Name of function to invoke
+            asking_section: If this check is coming from a Config child, which
+                one? This section will not be asked if it has the value.
+            item: The configuration value key.
+
+        Raises:
+            AttributeError if no config classes have the config key. The body
+                will contain what sections were checked.
+        """
+        for key, val in self._configs.items():
+            if key == asking_section:
+                continue
+
+            if item in dir(val):
+                return getattr(val, item)
+
+        raise AttributeError(item, "not found in %s" % (list(self._configs.keys()),))
+
+    def invoke_all(self, func_name: str, *args, **kwargs) -> MutableMapping[str, Any]:
+        """
+        Invoke a function on all instantiated config objects this RootConfig is
+        configured to use.
+
+        Args:
+            func_name: Name of function to invoke
             *args
             **kwargs
-
         Returns:
-            list: The list of the return values from each method called
+            ordered dictionary of config section name and the result of the
+            function from it.
         """
-        results = []
-        for cls in type(self).mro():
-            if name in cls.__dict__:
-                results.append(getattr(cls, name)(self, *args, **kargs))
-        return results
+        res = OrderedDict()
+
+        for name, config in self._configs.items():
+            if hasattr(config, func_name):
+                res[name] = getattr(config, func_name)(*args, **kwargs)
+
+        return res
 
     @classmethod
-    def invoke_all_static(cls, name, *args, **kargs):
-        """Invoke all static methods with the given name and arguments in the
-        class's MRO.
+    def invoke_all_static(cls, func_name: str, *args, **kwargs):
+        """
+        Invoke a static function on config objects this RootConfig is
+        configured to use.
 
         Args:
-            name (str): Name of function to invoke
+            func_name: Name of function to invoke
             *args
             **kwargs
-
         Returns:
-            list: The list of the return values from each method called
+            ordered dictionary of config section name and the result of the
+            function from it.
         """
-        results = []
-        for c in cls.mro():
-            if name in c.__dict__:
-                results.append(getattr(c, name)(*args, **kargs))
-        return results
+        for config in cls.config_classes:
+            if hasattr(config, func_name):
+                getattr(config, func_name)(*args, **kwargs)
 
     def generate_config(
         self,
@@ -187,7 +287,8 @@ class Config(object):
         tls_private_key_path=None,
         acme_domain=None,
     ):
-        """Build a default configuration file
+        """
+        Build a default configuration file
 
         This is used when the user explicitly asks us to generate a config file
         (eg with --generate_config).
@@ -242,6 +343,7 @@ class Config(object):
         Returns:
             str: the yaml config file
         """
+
         return "\n\n".join(
             dedent(conf)
             for conf in self.invoke_all(
@@ -257,7 +359,7 @@ class Config(object):
                 tls_certificate_path=tls_certificate_path,
                 tls_private_key_path=tls_private_key_path,
                 acme_domain=acme_domain,
-            )
+            ).values()
         )
 
     @classmethod
@@ -444,7 +546,7 @@ class Config(object):
                 )
 
             (config_path,) = config_files
-            if not cls.path_exists(config_path):
+            if not path_exists(config_path):
                 print("Generating config file %s" % (config_path,))
 
                 if config_args.data_directory:
@@ -469,7 +571,7 @@ class Config(object):
                     open_private_ports=config_args.open_private_ports,
                 )
 
-                if not cls.path_exists(config_dir_path):
+                if not path_exists(config_dir_path):
                     os.makedirs(config_dir_path)
                 with open(config_path, "w") as config_file:
                     config_file.write("# vim:ft=yaml\n\n")
@@ -518,7 +620,7 @@ class Config(object):
 
         return obj
 
-    def parse_config_dict(self, config_dict, config_dir_path, data_dir_path):
+    def parse_config_dict(self, config_dict, config_dir_path=None, data_dir_path=None):
         """Read the information from the config dict into this Config object.
 
         Args:
@@ -607,3 +709,6 @@ def find_config_files(search_paths):
             else:
                 config_files.append(config_path)
     return config_files
+
+
+__all__ = ["Config", "RootConfig"]
diff --git a/synapse/config/_base.pyi b/synapse/config/_base.pyi
new file mode 100644
index 0000000000..86bc965ee4
--- /dev/null
+++ b/synapse/config/_base.pyi
@@ -0,0 +1,135 @@
+from typing import Any, List, Optional
+
+from synapse.config import (
+    api,
+    appservice,
+    captcha,
+    cas,
+    consent_config,
+    database,
+    emailconfig,
+    groups,
+    jwt_config,
+    key,
+    logger,
+    metrics,
+    password,
+    password_auth_providers,
+    push,
+    ratelimiting,
+    registration,
+    repository,
+    room_directory,
+    saml2_config,
+    server,
+    server_notices_config,
+    spam_checker,
+    stats,
+    third_party_event_rules,
+    tls,
+    tracer,
+    user_directory,
+    voip,
+    workers,
+)
+
+class ConfigError(Exception): ...
+
+MISSING_REPORT_STATS_CONFIG_INSTRUCTIONS: str
+MISSING_REPORT_STATS_SPIEL: str
+MISSING_SERVER_NAME: str
+
+def path_exists(file_path: str): ...
+
+class RootConfig:
+    server: server.ServerConfig
+    tls: tls.TlsConfig
+    database: database.DatabaseConfig
+    logging: logger.LoggingConfig
+    ratelimit: ratelimiting.RatelimitConfig
+    media: repository.ContentRepositoryConfig
+    captcha: captcha.CaptchaConfig
+    voip: voip.VoipConfig
+    registration: registration.RegistrationConfig
+    metrics: metrics.MetricsConfig
+    api: api.ApiConfig
+    appservice: appservice.AppServiceConfig
+    key: key.KeyConfig
+    saml2: saml2_config.SAML2Config
+    cas: cas.CasConfig
+    jwt: jwt_config.JWTConfig
+    password: password.PasswordConfig
+    email: emailconfig.EmailConfig
+    worker: workers.WorkerConfig
+    authproviders: password_auth_providers.PasswordAuthProviderConfig
+    push: push.PushConfig
+    spamchecker: spam_checker.SpamCheckerConfig
+    groups: groups.GroupsConfig
+    userdirectory: user_directory.UserDirectoryConfig
+    consent: consent_config.ConsentConfig
+    stats: stats.StatsConfig
+    servernotices: server_notices_config.ServerNoticesConfig
+    roomdirectory: room_directory.RoomDirectoryConfig
+    thirdpartyrules: third_party_event_rules.ThirdPartyRulesConfig
+    tracer: tracer.TracerConfig
+
+    config_classes: List = ...
+    def __init__(self) -> None: ...
+    def invoke_all(self, func_name: str, *args: Any, **kwargs: Any): ...
+    @classmethod
+    def invoke_all_static(cls, func_name: str, *args: Any, **kwargs: Any) -> None: ...
+    def __getattr__(self, item: str): ...
+    def parse_config_dict(
+        self,
+        config_dict: Any,
+        config_dir_path: Optional[Any] = ...,
+        data_dir_path: Optional[Any] = ...,
+    ) -> None: ...
+    read_config: Any = ...
+    def generate_config(
+        self,
+        config_dir_path: str,
+        data_dir_path: str,
+        server_name: str,
+        generate_secrets: bool = ...,
+        report_stats: Optional[str] = ...,
+        open_private_ports: bool = ...,
+        listeners: Optional[Any] = ...,
+        database_conf: Optional[Any] = ...,
+        tls_certificate_path: Optional[str] = ...,
+        tls_private_key_path: Optional[str] = ...,
+        acme_domain: Optional[str] = ...,
+    ): ...
+    @classmethod
+    def load_or_generate_config(cls, description: Any, argv: Any): ...
+    @classmethod
+    def load_config(cls, description: Any, argv: Any): ...
+    @classmethod
+    def add_arguments_to_parser(cls, config_parser: Any) -> None: ...
+    @classmethod
+    def load_config_with_parser(cls, parser: Any, argv: Any): ...
+    def generate_missing_files(
+        self, config_dict: dict, config_dir_path: str
+    ) -> None: ...
+
+class Config:
+    root: RootConfig
+    def __init__(self, root_config: Optional[RootConfig] = ...) -> None: ...
+    def __getattr__(self, item: str, from_root: bool = ...): ...
+    @staticmethod
+    def parse_size(value: Any): ...
+    @staticmethod
+    def parse_duration(value: Any): ...
+    @staticmethod
+    def abspath(file_path: Optional[str]): ...
+    @classmethod
+    def path_exists(cls, file_path: str): ...
+    @classmethod
+    def check_file(cls, file_path: str, config_name: str): ...
+    @classmethod
+    def ensure_directory(cls, dir_path: str): ...
+    @classmethod
+    def read_file(cls, file_path: str, config_name: str): ...
+
+def read_config_files(config_files: List[str]): ...
+def find_config_files(search_paths: List[str]): ...
diff --git a/synapse/config/api.py b/synapse/config/api.py
index dddea79a8a..74cd53a8ed 100644
--- a/synapse/config/api.py
+++ b/synapse/config/api.py
@@ -18,6 +18,8 @@ from ._base import Config
 
 
 class ApiConfig(Config):
+    section = "api"
+
     def read_config(self, config, **kwargs):
         self.room_invite_state_types = config.get(
             "room_invite_state_types",
diff --git a/synapse/config/appservice.py b/synapse/config/appservice.py
index 8387ff6805..e77d3387ff 100644
--- a/synapse/config/appservice.py
+++ b/synapse/config/appservice.py
@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import logging
+from typing import Dict
 
 from six import string_types
 from six.moves.urllib import parse as urlparse
@@ -29,6 +30,8 @@ logger = logging.getLogger(__name__)
 
 
 class AppServiceConfig(Config):
+    section = "appservice"
+
     def read_config(self, config, **kwargs):
         self.app_service_config_files = config.get("app_service_config_files", [])
         self.notify_appservices = config.get("notify_appservices", True)
@@ -45,7 +48,7 @@ class AppServiceConfig(Config):
         # Uncomment to enable tracking of application service IP addresses. Implicitly
         # enables MAU tracking for application service users.
         #
-        #track_appservice_user_ips: True
+        #track_appservice_user_ips: true
         """
 
 
@@ -56,8 +59,8 @@ def load_appservices(hostname, config_files):
         return []
 
     # Dicts of value -> filename
-    seen_as_tokens = {}
-    seen_ids = {}
+    seen_as_tokens = {}  # type: Dict[str, str]
+    seen_ids = {}  # type: Dict[str, str]
 
     appservices = []
 
diff --git a/synapse/config/captcha.py b/synapse/config/captcha.py
index 8dac8152cf..44bd5c6799 100644
--- a/synapse/config/captcha.py
+++ b/synapse/config/captcha.py
@@ -16,6 +16,8 @@ from ._base import Config
 
 
 class CaptchaConfig(Config):
+    section = "captcha"
+
     def read_config(self, config, **kwargs):
         self.recaptcha_private_key = config.get("recaptcha_private_key")
         self.recaptcha_public_key = config.get("recaptcha_public_key")
diff --git a/synapse/config/cas.py b/synapse/config/cas.py
index ebe34d933b..4526c1a67b 100644
--- a/synapse/config/cas.py
+++ b/synapse/config/cas.py
@@ -22,17 +22,21 @@ class CasConfig(Config):
     cas_server_url: URL of CAS server
     """
 
+    section = "cas"
+
     def read_config(self, config, **kwargs):
         cas_config = config.get("cas_config", None)
         if cas_config:
             self.cas_enabled = cas_config.get("enabled", True)
             self.cas_server_url = cas_config["server_url"]
             self.cas_service_url = cas_config["service_url"]
+            self.cas_displayname_attribute = cas_config.get("displayname_attribute")
             self.cas_required_attributes = cas_config.get("required_attributes", {})
         else:
             self.cas_enabled = False
             self.cas_server_url = None
             self.cas_service_url = None
+            self.cas_displayname_attribute = None
             self.cas_required_attributes = {}
 
     def generate_config_section(self, config_dir_path, server_name, **kwargs):
@@ -43,6 +47,7 @@ class CasConfig(Config):
         #   enabled: true
         #   server_url: "https://cas-server.com"
         #   service_url: "https://homeserver.domain.com:8448"
+        #   #displayname_attribute: name
         #   #required_attributes:
         #   #    name: value
         """
diff --git a/synapse/config/consent_config.py b/synapse/config/consent_config.py
index 94916f3a49..aec9c4bbce 100644
--- a/synapse/config/consent_config.py
+++ b/synapse/config/consent_config.py
@@ -62,19 +62,22 @@ DEFAULT_CONFIG = """\
 #    body: >-
 #      To continue using this homeserver you must review and agree to the
 #      terms and conditions at %(consent_uri)s
-#  send_server_notice_to_guests: True
+#  send_server_notice_to_guests: true
 #  block_events_error: >-
 #    To continue using this homeserver you must review and agree to the
 #    terms and conditions at %(consent_uri)s
-#  require_at_registration: False
+#  require_at_registration: false
 #  policy_name: Privacy Policy
 #
 """
 
 
 class ConsentConfig(Config):
-    def __init__(self):
-        super(ConsentConfig, self).__init__()
+
+    section = "consent"
+
+    def __init__(self, *args):
+        super(ConsentConfig, self).__init__(*args)
 
         self.user_consent_version = None
         self.user_consent_template_dir = None
diff --git a/synapse/config/database.py b/synapse/config/database.py
index 118aafbd4a..0e2509f0b1 100644
--- a/synapse/config/database.py
+++ b/synapse/config/database.py
@@ -21,6 +21,8 @@ from ._base import Config
 
 
 class DatabaseConfig(Config):
+    section = "database"
+
     def read_config(self, config, **kwargs):
         self.event_cache_size = self.parse_size(config.get("event_cache_size", "10K"))
 
diff --git a/synapse/config/emailconfig.py b/synapse/config/emailconfig.py
index f83c05df44..39e7a1dddb 100644
--- a/synapse/config/emailconfig.py
+++ b/synapse/config/emailconfig.py
@@ -20,6 +20,7 @@ from __future__ import print_function
 # This file can't be called email.py because if it is, we cannot:
 import email.utils
 import os
+from enum import Enum
 
 import pkg_resources
 
@@ -27,6 +28,8 @@ from ._base import Config, ConfigError
 
 
 class EmailConfig(Config):
+    section = "email"
+
     def read_config(self, config, **kwargs):
         # TODO: We should separate better the email configuration from the notification
         # and account validity config.
@@ -74,19 +77,48 @@ class EmailConfig(Config):
             "renew_at"
         )
 
-        email_trust_identity_server_for_password_resets = email_config.get(
-            "trust_identity_server_for_password_resets", False
+        self.threepid_behaviour_email = (
+            # Have Synapse handle the email sending if account_threepid_delegates.email
+            # is not defined
+            # msisdn is currently always remote while Synapse does not support any method of
+            # sending SMS messages
+            ThreepidBehaviour.REMOTE
+            if self.account_threepid_delegate_email
+            else ThreepidBehaviour.LOCAL
         )
-        self.email_password_reset_behaviour = (
-            "remote" if email_trust_identity_server_for_password_resets else "local"
-        )
-        self.password_resets_were_disabled_due_to_email_config = False
-        if self.email_password_reset_behaviour == "local" and email_config == {}:
+        # Prior to Synapse v1.4.0, there was another option that defined whether Synapse would
+        # use an identity server to password reset tokens on its behalf. We now warn the user
+        # if they have this set and tell them to use the updated option, while using a default
+        # identity server in the process.
+        self.using_identity_server_from_trusted_list = False
+        if (
+            not self.account_threepid_delegate_email
+            and config.get("trust_identity_server_for_password_resets", False) is True
+        ):
+            # Use the first entry in self.trusted_third_party_id_servers instead
+            if self.trusted_third_party_id_servers:
+                # XXX: It's a little confusing that account_threepid_delegate_email is modified
+                # both in RegistrationConfig and here. We should factor this bit out
+                self.account_threepid_delegate_email = self.trusted_third_party_id_servers[
+                    0
+                ]
+                self.using_identity_server_from_trusted_list = True
+            else:
+                raise ConfigError(
+                    "Attempted to use an identity server from"
+                    '"trusted_third_party_id_servers" but it is empty.'
+                )
+
+        self.local_threepid_handling_disabled_due_to_email_config = False
+        if (
+            self.threepid_behaviour_email == ThreepidBehaviour.LOCAL
+            and email_config == {}
+        ):
             # We cannot warn the user this has happened here
             # Instead do so when a user attempts to reset their password
-            self.password_resets_were_disabled_due_to_email_config = True
+            self.local_threepid_handling_disabled_due_to_email_config = True
 
-            self.email_password_reset_behaviour = "off"
+            self.threepid_behaviour_email = ThreepidBehaviour.OFF
 
         # Get lifetime of a validation token in milliseconds
         self.email_validation_token_lifetime = self.parse_duration(
@@ -96,7 +128,7 @@ class EmailConfig(Config):
         if (
             self.email_enable_notifs
             or account_validity_renewal_enabled
-            or self.email_password_reset_behaviour == "local"
+            or self.threepid_behaviour_email == ThreepidBehaviour.LOCAL
         ):
             # make sure we can import the required deps
             import jinja2
@@ -106,7 +138,7 @@ class EmailConfig(Config):
             jinja2
             bleach
 
-        if self.email_password_reset_behaviour == "local":
+        if self.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
             required = ["smtp_host", "smtp_port", "notif_from"]
 
             missing = []
@@ -125,28 +157,63 @@ class EmailConfig(Config):
                     % (", ".join(missing),)
                 )
 
-            # Templates for password reset emails
+            # These email templates have placeholders in them, and thus must be
+            # parsed using a templating engine during a request
             self.email_password_reset_template_html = email_config.get(
                 "password_reset_template_html", "password_reset.html"
             )
             self.email_password_reset_template_text = email_config.get(
                 "password_reset_template_text", "password_reset.txt"
             )
+            self.email_registration_template_html = email_config.get(
+                "registration_template_html", "registration.html"
+            )
+            self.email_registration_template_text = email_config.get(
+                "registration_template_text", "registration.txt"
+            )
+            self.email_add_threepid_template_html = email_config.get(
+                "add_threepid_template_html", "add_threepid.html"
+            )
+            self.email_add_threepid_template_text = email_config.get(
+                "add_threepid_template_text", "add_threepid.txt"
+            )
+
             self.email_password_reset_template_failure_html = email_config.get(
                 "password_reset_template_failure_html", "password_reset_failure.html"
             )
-            # This template does not support any replaceable variables, so we will
-            # read it from the disk once during setup
+            self.email_registration_template_failure_html = email_config.get(
+                "registration_template_failure_html", "registration_failure.html"
+            )
+            self.email_add_threepid_template_failure_html = email_config.get(
+                "add_threepid_template_failure_html", "add_threepid_failure.html"
+            )
+
+            # These templates do not support any placeholder variables, so we
+            # will read them from disk once during setup
             email_password_reset_template_success_html = email_config.get(
                 "password_reset_template_success_html", "password_reset_success.html"
             )
+            email_registration_template_success_html = email_config.get(
+                "registration_template_success_html", "registration_success.html"
+            )
+            email_add_threepid_template_success_html = email_config.get(
+                "add_threepid_template_success_html", "add_threepid_success.html"
+            )
 
             # Check templates exist
             for f in [
                 self.email_password_reset_template_html,
                 self.email_password_reset_template_text,
+                self.email_registration_template_html,
+                self.email_registration_template_text,
+                self.email_add_threepid_template_html,
+                self.email_add_threepid_template_text,
                 self.email_password_reset_template_failure_html,
+                self.email_registration_template_failure_html,
+                self.email_add_threepid_template_failure_html,
                 email_password_reset_template_success_html,
+                email_registration_template_success_html,
+                email_add_threepid_template_success_html,
             ]:
                 p = os.path.join(self.email_template_dir, f)
                 if not os.path.isfile(p):
@@ -156,9 +223,21 @@ class EmailConfig(Config):
             filepath = os.path.join(
                 self.email_template_dir, email_password_reset_template_success_html
             )
-            self.email_password_reset_template_success_html_content = self.read_file(
+            self.email_password_reset_template_success_html = self.read_file(
                 filepath, "email.password_reset_template_success_html"
             )
+            filepath = os.path.join(
+                self.email_template_dir, email_registration_template_success_html
+            )
+            self.email_registration_template_success_html_content = self.read_file(
+                filepath, "email.registration_template_success_html"
+            )
+            filepath = os.path.join(
+                self.email_template_dir, email_add_threepid_template_success_html
+            )
+            self.email_add_threepid_template_success_html_content = self.read_file(
+                filepath, "email.add_threepid_template_success_html"
+            )
 
         if self.email_enable_notifs:
             required = [
@@ -225,13 +304,13 @@ class EmailConfig(Config):
         #   smtp_port: 25 # SSL: 465, STARTTLS: 587
         #   smtp_user: "exampleusername"
         #   smtp_pass: "examplepassword"
-        #   require_transport_security: False
+        #   require_transport_security: false
         #   notif_from: "Your Friendly %(app)s Home Server "
         #   app_name: Matrix
         #
         #   # Enable email notifications by default
         #   #
-        #   notif_for_new_users: True
+        #   notif_for_new_users: true
         #
         #   # Defining a custom URL for Riot is only needed if email notifications
         #   # should contain links to a self-hosted installation of Riot; when set
@@ -239,19 +318,6 @@ class EmailConfig(Config):
         #   #
         #   riot_base_url: "http://localhost/riot"
         #
-        #   # Enable sending password reset emails via the configured, trusted
-        #   # identity servers
-        #   #
-        #   # IMPORTANT! This will give a malicious or overtaken identity server
-        #   # the ability to reset passwords for your users! Make absolutely sure
-        #   # that you want to do this! It is strongly recommended that password
-        #   # reset emails be sent by the homeserver instead
-        #   #
-        #   # If this option is set to false and SMTP options have not been
-        #   # configured, resetting user passwords via email will be disabled
-        #   #
-        #   #trust_identity_server_for_password_resets: false
-        #
         #   # Configure the time that a validation email or text message code
         #   # will expire after sending
         #   #
@@ -283,9 +349,47 @@ class EmailConfig(Config):
         #   #password_reset_template_html: password_reset.html
         #   #password_reset_template_text: password_reset.txt
         #
+        #   # Templates for registration emails sent by the homeserver
+        #   #
+        #   #registration_template_html: registration.html
+        #   #registration_template_text: registration.txt
+        #
+        #   # Templates for validation emails sent by the homeserver when adding an email to
+        #   # your user account
+        #   #
+        #   #add_threepid_template_html: add_threepid.html
+        #   #add_threepid_template_text: add_threepid.txt
+        #
         #   # Templates for password reset success and failure pages that a user
         #   # will see after attempting to reset their password
         #   #
         #   #password_reset_template_success_html: password_reset_success.html
         #   #password_reset_template_failure_html: password_reset_failure.html
+        #
+        #   # Templates for registration success and failure pages that a user
+        #   # will see after attempting to register using an email or phone
+        #   #
+        #   #registration_template_success_html: registration_success.html
+        #   #registration_template_failure_html: registration_failure.html
+        #
+        #   # Templates for success and failure pages that a user will see after attempting
+        #   # to add an email or phone to their account
+        #   #
+        #   #add_threepid_success_html: add_threepid_success.html
+        #   #add_threepid_failure_html: add_threepid_failure.html
         """
+
+
+class ThreepidBehaviour(Enum):
+    """
+    Enum to define the behaviour of Synapse with regards to when it contacts an identity
+    server for 3pid registration and password resets
+
+    REMOTE = use an external server to send tokens
+    LOCAL = send tokens ourselves
+    OFF = disable registration via 3pid and password resets
+    """
+
+    REMOTE = "remote"
+    LOCAL = "local"
+    OFF = "off"
diff --git a/synapse/config/groups.py b/synapse/config/groups.py
index 2a522b5f44..d6862d9a64 100644
--- a/synapse/config/groups.py
+++ b/synapse/config/groups.py
@@ -17,6 +17,8 @@ from ._base import Config
 
 
 class GroupsConfig(Config):
+    section = "groups"
+
     def read_config(self, config, **kwargs):
         self.enable_group_creation = config.get("enable_group_creation", False)
         self.group_creation_prefix = config.get("group_creation_prefix", "")
diff --git a/synapse/config/homeserver.py b/synapse/config/homeserver.py
index 72acad4f18..6e348671c7 100644
--- a/synapse/config/homeserver.py
+++ b/synapse/config/homeserver.py
@@ -14,6 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from ._base import RootConfig
 from .api import ApiConfig
 from .appservice import AppServiceConfig
 from .captcha import CaptchaConfig
@@ -46,36 +47,37 @@ from .voip import VoipConfig
 from .workers import WorkerConfig
 
 
-class HomeServerConfig(
-    ServerConfig,
-    TlsConfig,
-    DatabaseConfig,
-    LoggingConfig,
-    RatelimitConfig,
-    ContentRepositoryConfig,
-    CaptchaConfig,
-    VoipConfig,
-    RegistrationConfig,
-    MetricsConfig,
-    ApiConfig,
-    AppServiceConfig,
-    KeyConfig,
-    SAML2Config,
-    CasConfig,
-    JWTConfig,
-    PasswordConfig,
-    EmailConfig,
-    WorkerConfig,
-    PasswordAuthProviderConfig,
-    PushConfig,
-    SpamCheckerConfig,
-    GroupsConfig,
-    UserDirectoryConfig,
-    ConsentConfig,
-    StatsConfig,
-    ServerNoticesConfig,
-    RoomDirectoryConfig,
-    ThirdPartyRulesConfig,
-    TracerConfig,
-):
-    pass
+class HomeServerConfig(RootConfig):
+
+    config_classes = [
+        ServerConfig,
+        TlsConfig,
+        DatabaseConfig,
+        LoggingConfig,
+        RatelimitConfig,
+        ContentRepositoryConfig,
+        CaptchaConfig,
+        VoipConfig,
+        RegistrationConfig,
+        MetricsConfig,
+        ApiConfig,
+        AppServiceConfig,
+        KeyConfig,
+        SAML2Config,
+        CasConfig,
+        JWTConfig,
+        PasswordConfig,
+        EmailConfig,
+        WorkerConfig,
+        PasswordAuthProviderConfig,
+        PushConfig,
+        SpamCheckerConfig,
+        GroupsConfig,
+        UserDirectoryConfig,
+        ConsentConfig,
+        StatsConfig,
+        ServerNoticesConfig,
+        RoomDirectoryConfig,
+        ThirdPartyRulesConfig,
+        TracerConfig,
+    ]
diff --git a/synapse/config/jwt_config.py b/synapse/config/jwt_config.py
index 36d87cef03..a568726985 100644
--- a/synapse/config/jwt_config.py
+++ b/synapse/config/jwt_config.py
@@ -23,6 +23,8 @@ MISSING_JWT = """Missing jwt library. This is required for jwt login.
 
 
 class JWTConfig(Config):
+    section = "jwt"
+
     def read_config(self, config, **kwargs):
         jwt_config = config.get("jwt_config", None)
         if jwt_config:
diff --git a/synapse/config/key.py b/synapse/config/key.py
index ba2199bceb..52ff1b2621 100644
--- a/synapse/config/key.py
+++ b/synapse/config/key.py
@@ -50,6 +50,33 @@ and you should enable 'federation_verify_certificates' in your configuration.
 If you are *sure* you want to do this, set 'accept_keys_insecurely' on the
 trusted_key_server configuration."""
 
+TRUSTED_KEY_SERVER_NOT_CONFIGURED_WARN = """\
+Synapse requires that a list of trusted key servers are specified in order to
+provide signing keys for other servers in the federation.
+
+This homeserver does not have a trusted key server configured in
+homeserver.yaml and will fall back to the default of 'matrix.org'.
+
+Trusted key servers should be long-lived and stable which makes matrix.org a
+good choice for many admins, but some admins may wish to choose another. To
+suppress this warning, the admin should set 'trusted_key_servers' in
+homeserver.yaml to their desired key server and 'suppress_key_server_warning'
+to 'true'.
+
+In a future release the software-defined default will be removed entirely and
+the trusted key server will be defined exclusively by the value of
+'trusted_key_servers'.
+--------------------------------------------------------------------------------"""
+
+TRUSTED_KEY_SERVER_CONFIGURED_AS_M_ORG_WARN = """\
+This server is configured to use 'matrix.org' as its trusted key server via the
+'trusted_key_servers' config option. 'matrix.org' is a good choice for a key
+server since it is long-lived, stable and trusted. However, some admins may
+wish to use another server for this purpose.
+
+To suppress this warning and continue using 'matrix.org', admins should set
+'suppress_key_server_warning' to 'true' in homeserver.yaml.
+--------------------------------------------------------------------------------"""
 
 logger = logging.getLogger(__name__)
 
@@ -65,6 +92,8 @@ class TrustedKeyServer(object):
 
 
 class KeyConfig(Config):
+    section = "key"
+
     def read_config(self, config, config_dir_path, **kwargs):
         # the signing key can be specified inline or in a separate file
         if "signing_key" in config:
@@ -85,6 +114,7 @@ class KeyConfig(Config):
             config.get("key_refresh_interval", "1d")
         )
 
+        suppress_key_server_warning = config.get("suppress_key_server_warning", False)
         key_server_signing_keys_path = config.get("key_server_signing_keys_path")
         if key_server_signing_keys_path:
             self.key_server_signing_keys = self.read_signing_keys(
@@ -95,6 +125,7 @@ class KeyConfig(Config):
 
         # if neither trusted_key_servers nor perspectives are given, use the default.
         if "perspectives" not in config and "trusted_key_servers" not in config:
+            logger.warning(TRUSTED_KEY_SERVER_NOT_CONFIGURED_WARN)
             key_servers = [{"server_name": "matrix.org"}]
         else:
             key_servers = config.get("trusted_key_servers", [])
@@ -108,6 +139,11 @@ class KeyConfig(Config):
             # merge the 'perspectives' config into the 'trusted_key_servers' config.
             key_servers.extend(_perspectives_to_key_servers(config))
 
+            if not suppress_key_server_warning and "matrix.org" in (
+                s["server_name"] for s in key_servers
+            ):
+                logger.warning(TRUSTED_KEY_SERVER_CONFIGURED_AS_M_ORG_WARN)
+
         # list of TrustedKeyServer objects
         self.key_servers = list(
             _parse_key_servers(key_servers, self.federation_verify_certificates)
@@ -120,7 +156,7 @@ class KeyConfig(Config):
         if not self.macaroon_secret_key:
             # Unfortunately, there are people out there that don't have this
             # set. Lets just be "nice" and derive one from their secret key.
-            logger.warn("Config is missing macaroon_secret_key")
+            logger.warning("Config is missing macaroon_secret_key")
             seed = bytes(self.signing_key[0])
             self.macaroon_secret_key = hashlib.sha256(seed).digest()
 
@@ -190,6 +226,10 @@ class KeyConfig(Config):
         # This setting supercedes an older setting named `perspectives`. The old format
         # is still supported for backwards-compatibility, but it is deprecated.
         #
+        # 'trusted_key_servers' defaults to matrix.org, but using it will generate a
+        # warning on start-up. To suppress this warning, set
+        # 'suppress_key_server_warning' to true.
+        #
         # Options for each entry in the list include:
         #
         #    server_name: the name of the server. required.
@@ -214,11 +254,13 @@ class KeyConfig(Config):
         #      "ed25519:auto": "abcdefghijklmnopqrstuvwxyzabcdefghijklmopqr"
         #  - server_name: "my_other_trusted_server.example.com"
         #
-        # The default configuration is:
-        #
-        #trusted_key_servers:
-        #  - server_name: "matrix.org"
+        trusted_key_servers:
+          - server_name: "matrix.org"
+
+        # Uncomment the following to disable the warning that is emitted when the
+        # trusted_key_servers include 'matrix.org'. See above.
         #
+        #suppress_key_server_warning: true
 
         # The signing keys to use when acting as a trusted key server. If not specified
         # defaults to the server signing key.
diff --git a/synapse/config/logger.py b/synapse/config/logger.py
index 2704c18720..75bb904718 100644
--- a/synapse/config/logger.py
+++ b/synapse/config/logger.py
@@ -21,7 +21,12 @@ from string import Template
 
 import yaml
 
-from twisted.logger import STDLibLogObserver, globalLogBeginner
+from twisted.logger import (
+    ILogObserver,
+    LogBeginner,
+    STDLibLogObserver,
+    globalLogBeginner,
+)
 
 import synapse
 from synapse.app import _base as appbase
@@ -63,9 +68,6 @@ handlers:
         filters: [context]
 
 loggers:
-    synapse:
-        level: INFO
-
     synapse.storage.SQL:
         # beware: increasing this to DEBUG will make synapse log sensitive
         # information such as access tokens.
@@ -74,11 +76,15 @@ loggers:
 root:
     level: INFO
     handlers: [file, console]
+
+disable_existing_loggers: false
 """
 )
 
 
 class LoggingConfig(Config):
+    section = "logging"
+
     def read_config(self, config, **kwargs):
         self.log_config = self.abspath(config.get("log_config"))
         self.no_redirect_stdio = config.get("no_redirect_stdio", False)
@@ -124,7 +130,7 @@ class LoggingConfig(Config):
                 log_config_file.write(DEFAULT_LOG_CONFIG.substitute(log_file=log_file))
 
 
-def _setup_stdlib_logging(config, log_config):
+def _setup_stdlib_logging(config, log_config, logBeginner: LogBeginner):
     """
     Set up Python stdlib logging.
     """
@@ -165,23 +171,25 @@ def _setup_stdlib_logging(config, log_config):
 
         return observer(event)
 
-    globalLogBeginner.beginLoggingTo(
-        [_log], redirectStandardIO=not config.no_redirect_stdio
-    )
+    logBeginner.beginLoggingTo([_log], redirectStandardIO=not config.no_redirect_stdio)
     if not config.no_redirect_stdio:
         print("Redirected stdout/stderr to logs")
 
+    return observer
+
 
 def _reload_stdlib_logging(*args, log_config=None):
     logger = logging.getLogger("")
 
     if not log_config:
-        logger.warn("Reloaded a blank config?")
+        logger.warning("Reloaded a blank config?")
 
     logging.config.dictConfig(log_config)
 
 
-def setup_logging(hs, config, use_worker_options=False):
+def setup_logging(
+    hs, config, use_worker_options=False, logBeginner: LogBeginner = globalLogBeginner
+) -> ILogObserver:
     """
     Set up the logging subsystem.
 
@@ -191,6 +199,12 @@ def setup_logging(hs, config, use_worker_options=False):
 
         use_worker_options (bool): True to use the 'worker_log_config' option
             instead of 'log_config'.
+
+        logBeginner: The Twisted logBeginner to use.
+
+    Returns:
+        The "root" Twisted Logger observer, suitable for sending logs to from a
+        Logger instance.
     """
     log_config = config.worker_log_config if use_worker_options else config.log_config
 
@@ -210,14 +224,18 @@ def setup_logging(hs, config, use_worker_options=False):
     log_config_body = read_config()
 
     if log_config_body and log_config_body.get("structured") is True:
-        setup_structured_logging(hs, config, log_config_body)
+        logger = setup_structured_logging(
+            hs, config, log_config_body, logBeginner=logBeginner
+        )
         appbase.register_sighup(read_config, callback=reload_structured_logging)
     else:
-        _setup_stdlib_logging(config, log_config_body)
+        logger = _setup_stdlib_logging(config, log_config_body, logBeginner=logBeginner)
         appbase.register_sighup(read_config, callback=_reload_stdlib_logging)
 
     # make sure that the first thing we log is a thing we can grep backwards
     # for
-    logging.warn("***** STARTING SERVER *****")
-    logging.warn("Server %s version %s", sys.argv[0], get_version_string(synapse))
+    logging.warning("***** STARTING SERVER *****")
+    logging.warning("Server %s version %s", sys.argv[0], get_version_string(synapse))
     logging.info("Server hostname: %s", config.server_name)
+
+    return logger
diff --git a/synapse/config/metrics.py b/synapse/config/metrics.py
index 3698441963..22538153e1 100644
--- a/synapse/config/metrics.py
+++ b/synapse/config/metrics.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2015, 2016 OpenMarket Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,26 +14,49 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import attr
+
+from synapse.python_dependencies import DependencyException, check_requirements
+
 from ._base import Config, ConfigError
 
-MISSING_SENTRY = """Missing sentry-sdk library. This is required to enable sentry
-    integration.
-    """
+
+@attr.s
+class MetricsFlags(object):
+    known_servers = attr.ib(default=False, validator=attr.validators.instance_of(bool))
+
+    @classmethod
+    def all_off(cls):
+        """
+        Instantiate the flags with all options set to off.
+        """
+        return cls(**{x.name: False for x in attr.fields(cls)})
 
 
 class MetricsConfig(Config):
+    section = "metrics"
+
     def read_config(self, config, **kwargs):
         self.enable_metrics = config.get("enable_metrics", False)
         self.report_stats = config.get("report_stats", None)
+        self.report_stats_endpoint = config.get(
+            "report_stats_endpoint", "https://matrix.org/report-usage-stats/push"
+        )
         self.metrics_port = config.get("metrics_port")
         self.metrics_bind_host = config.get("metrics_bind_host", "127.0.0.1")
 
+        if self.enable_metrics:
+            _metrics_config = config.get("metrics_flags") or {}
+            self.metrics_flags = MetricsFlags(**_metrics_config)
+        else:
+            self.metrics_flags = MetricsFlags.all_off()
+
         self.sentry_enabled = "sentry" in config
         if self.sentry_enabled:
             try:
-                import sentry_sdk  # noqa F401
-            except ImportError:
-                raise ConfigError(MISSING_SENTRY)
+                check_requirements("sentry")
+            except DependencyException as e:
+                raise ConfigError(e.message)
 
             self.sentry_dsn = config["sentry"].get("dsn")
             if not self.sentry_dsn:
@@ -46,7 +70,7 @@ class MetricsConfig(Config):
 
         # Enable collection and rendering of performance metrics
         #
-        #enable_metrics: False
+        #enable_metrics: false
 
         # Enable sentry integration
         # NOTE: While attempts are made to ensure that the logs don't contain
@@ -58,6 +82,16 @@ class MetricsConfig(Config):
         #sentry:
         #    dsn: "..."
 
+        # Flags to enable Prometheus metrics which are not suitable to be
+        # enabled by default, either for performance reasons or limited use.
+        #
+        metrics_flags:
+            # Publish synapse_federation_known_servers, a g auge of the number of
+            # servers this homeserver knows about, including itself. May cause
+            # performance problems on large homeservers.
+            #
+            #known_servers: true
+
         # Whether or not to report anonymized homeserver usage statistics.
         """
 
@@ -66,4 +100,10 @@ class MetricsConfig(Config):
         else:
             res += "report_stats: %s\n" % ("true" if report_stats else "false")
 
+        res += """
+        # The endpoint to report the anonymized homeserver usage statistics to.
+        # Defaults to https://matrix.org/report-usage-stats/push
+        #
+        #report_stats_endpoint: https://example.com/report-usage-stats/push
+        """
         return res
diff --git a/synapse/config/password.py b/synapse/config/password.py
index d5b5953f2f..2a634ac751 100644
--- a/synapse/config/password.py
+++ b/synapse/config/password.py
@@ -20,6 +20,8 @@ class PasswordConfig(Config):
     """Password login configuration
     """
 
+    section = "password"
+
     def read_config(self, config, **kwargs):
         password_config = config.get("password_config", {})
         if password_config is None:
diff --git a/synapse/config/password_auth_providers.py b/synapse/config/password_auth_providers.py
index 788c39c9fb..9746bbc681 100644
--- a/synapse/config/password_auth_providers.py
+++ b/synapse/config/password_auth_providers.py
@@ -13,6 +13,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from typing import Any, List
+
 from synapse.util.module_loader import load_module
 
 from ._base import Config
@@ -21,8 +23,10 @@ LDAP_PROVIDER = "ldap_auth_provider.LdapAuthProvider"
 
 
 class PasswordAuthProviderConfig(Config):
+    section = "authproviders"
+
     def read_config(self, config, **kwargs):
-        self.password_providers = []
+        self.password_providers = []  # type: List[Any]
         providers = []
 
         # We want to be backwards compatible with the old `ldap_config`
diff --git a/synapse/config/push.py b/synapse/config/push.py
index 1b932722a5..0910958649 100644
--- a/synapse/config/push.py
+++ b/synapse/config/push.py
@@ -18,6 +18,8 @@ from ._base import Config
 
 
 class PushConfig(Config):
+    section = "push"
+
     def read_config(self, config, **kwargs):
         push_config = config.get("push", {})
         self.push_include_content = push_config.get("include_content", True)
diff --git a/synapse/config/ratelimiting.py b/synapse/config/ratelimiting.py
index 33f31cf213..947f653e03 100644
--- a/synapse/config/ratelimiting.py
+++ b/synapse/config/ratelimiting.py
@@ -36,6 +36,8 @@ class FederationRateLimitConfig(object):
 
 
 class RatelimitConfig(Config):
+    section = "ratelimiting"
+
     def read_config(self, config, **kwargs):
 
         # Load the new-style messages config if it exists. Otherwise fall back
@@ -80,6 +82,12 @@ class RatelimitConfig(Config):
             "federation_rr_transactions_per_room_per_second", 50
         )
 
+        rc_admin_redaction = config.get("rc_admin_redaction")
+        if rc_admin_redaction:
+            self.rc_admin_redaction = RateLimitConfig(rc_admin_redaction)
+        else:
+            self.rc_admin_redaction = None
+
     def generate_config_section(self, **kwargs):
         return """\
         ## Ratelimiting ##
@@ -102,6 +110,9 @@ class RatelimitConfig(Config):
         #   - one for login that ratelimits login requests based on the account the
         #     client is attempting to log into, based on the amount of failed login
         #     attempts for this account.
+        #   - one for ratelimiting redactions by room admins. If this is not explicitly
+        #     set then it uses the same ratelimiting as per rc_message. This is useful
+        #     to allow room admins to deal with abuse quickly.
         #
         # The defaults are as shown below.
         #
@@ -123,6 +134,10 @@ class RatelimitConfig(Config):
         #  failed_attempts:
         #    per_second: 0.17
         #    burst_count: 3
+        #
+        #rc_admin_redaction:
+        #  per_second: 1
+        #  burst_count: 50
 
 
         # Ratelimiting settings for incoming federation
diff --git a/synapse/config/registration.py b/synapse/config/registration.py
index e2bee3c116..1f6dac69da 100644
--- a/synapse/config/registration.py
+++ b/synapse/config/registration.py
@@ -24,6 +24,8 @@ from synapse.util.stringutils import random_string_with_symbols
 
 
 class AccountValidityConfig(Config):
+    section = "accountvalidity"
+
     def __init__(self, config, synapse_config):
         self.enabled = config.get("enabled", False)
         self.renew_by_email_enabled = "renew_at" in config
@@ -77,6 +79,8 @@ class AccountValidityConfig(Config):
 
 
 class RegistrationConfig(Config):
+    section = "registration"
+
     def read_config(self, config, **kwargs):
         self.enable_registration = bool(
             strtobool(str(config.get("enable_registration", False)))
@@ -99,6 +103,10 @@ class RegistrationConfig(Config):
         self.trusted_third_party_id_servers = config.get(
             "trusted_third_party_id_servers", ["matrix.org", "vector.im"]
         )
+        account_threepid_delegates = config.get("account_threepid_delegates") or {}
+        self.account_threepid_delegate_email = account_threepid_delegates.get("email")
+        self.account_threepid_delegate_msisdn = account_threepid_delegates.get("msisdn")
+
         self.default_identity_server = config.get("default_identity_server")
         self.allow_guest_access = config.get("allow_guest_access", False)
 
@@ -172,7 +180,7 @@ class RegistrationConfig(Config):
         # where d is equal to 10%% of the validity period.
         #
         #account_validity:
-        #  enabled: True
+        #  enabled: true
         #  period: 6w
         #  renew_at: 1w
         #  renew_email_subject: "Renew your %%(app)s account"
@@ -257,10 +265,44 @@ class RegistrationConfig(Config):
         # Also defines the ID server which will be called when an account is
         # deactivated (one will be picked arbitrarily).
         #
+        # Note: This option is deprecated. Since v0.99.4, Synapse has tracked which identity
+        # server a 3PID has been bound to. For 3PIDs bound before then, Synapse runs a
+        # background migration script, informing itself that the identity server all of its
+        # 3PIDs have been bound to is likely one of the below.
+        #
+        # As of Synapse v1.4.0, all other functionality of this option has been deprecated, and
+        # it is now solely used for the purposes of the background migration script, and can be
+        # removed once it has run.
         #trusted_third_party_id_servers:
         #  - matrix.org
         #  - vector.im
 
+        # Handle threepid (email/phone etc) registration and password resets through a set of
+        # *trusted* identity servers. Note that this allows the configured identity server to
+        # reset passwords for accounts!
+        #
+        # Be aware that if `email` is not set, and SMTP options have not been
+        # configured in the email config block, registration and user password resets via
+        # email will be globally disabled.
+        #
+        # Additionally, if `msisdn` is not set, registration and password resets via msisdn
+        # will be disabled regardless. This is due to Synapse currently not supporting any
+        # method of sending SMS messages on its own.
+        #
+        # To enable using an identity server for operations regarding a particular third-party
+        # identifier type, set the value to the URL of that identity server as shown in the
+        # examples below.
+        #
+        # Servers handling the these requests must answer the `/requestToken` endpoints defined
+        # by the Matrix Identity Service API specification:
+        # https://matrix.org/docs/spec/identity_service/latest
+        #
+        # If a delegate is specified, the config option public_baseurl must also be filled out.
+        #
+        account_threepid_delegates:
+            #email: https://example.com     # Delegate email sending to example.com
+            #msisdn: http://localhost:8090  # Delegate SMS sending to this local process
+
         # Users who register on this homeserver will automatically be joined
         # to these rooms
         #
diff --git a/synapse/config/repository.py b/synapse/config/repository.py
index fdb1f246d0..d0205e14b9 100644
--- a/synapse/config/repository.py
+++ b/synapse/config/repository.py
@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-# Copyright 2014, 2015 matrix.org
+# Copyright 2014, 2015 OpenMarket Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -15,7 +15,9 @@
 
 import os
 from collections import namedtuple
+from typing import Dict, List
 
+from synapse.python_dependencies import DependencyException, check_requirements
 from synapse.util.module_loader import load_module
 
 from ._base import Config, ConfigError
@@ -34,17 +36,6 @@ THUMBNAIL_SIZE_YAML = """\
         #    method: %(method)s
 """
 
-MISSING_NETADDR = "Missing netaddr library. This is required for URL preview API."
-
-MISSING_LXML = """Missing lxml library. This is required for URL preview API.
-
-    Install by running:
-        pip install lxml
-
-    Requires libxslt1-dev system package.
-    """
-
-
 ThumbnailRequirement = namedtuple(
     "ThumbnailRequirement", ["width", "height", "method", "media_type"]
 )
@@ -71,7 +62,7 @@ def parse_thumbnail_requirements(thumbnail_sizes):
         Dictionary mapping from media type string to list of
         ThumbnailRequirement tuples.
     """
-    requirements = {}
+    requirements = {}  # type: Dict[str, List]
     for size in thumbnail_sizes:
         width = size["width"]
         height = size["height"]
@@ -87,6 +78,8 @@ def parse_thumbnail_requirements(thumbnail_sizes):
 
 
 class ContentRepositoryConfig(Config):
+    section = "media"
+
     def read_config(self, config, **kwargs):
 
         # Only enable the media repo if either the media repo is enabled or the
@@ -140,7 +133,7 @@ class ContentRepositoryConfig(Config):
         #
         # We don't create the storage providers here as not all workers need
         # them to be started.
-        self.media_storage_providers = []
+        self.media_storage_providers = []  # type: List[tuple]
 
         for provider_config in storage_providers:
             # We special case the module "file_system" so as not to need to
@@ -171,16 +164,10 @@ class ContentRepositoryConfig(Config):
         self.url_preview_enabled = config.get("url_preview_enabled", False)
         if self.url_preview_enabled:
             try:
-                import lxml
+                check_requirements("url_preview")
 
-                lxml  # To stop unused lint.
-            except ImportError:
-                raise ConfigError(MISSING_LXML)
-
-            try:
-                from netaddr import IPSet
-            except ImportError:
-                raise ConfigError(MISSING_NETADDR)
+            except DependencyException as e:
+                raise ConfigError(e.message)
 
             if "url_preview_ip_range_blacklist" not in config:
                 raise ConfigError(
@@ -189,6 +176,9 @@ class ContentRepositoryConfig(Config):
                     "to work"
                 )
 
+            # netaddr is a dependency for url_preview
+            from netaddr import IPSet
+
             self.url_preview_ip_range_blacklist = IPSet(
                 config["url_preview_ip_range_blacklist"]
             )
diff --git a/synapse/config/room_directory.py b/synapse/config/room_directory.py
index a92693017b..7c9f05bde4 100644
--- a/synapse/config/room_directory.py
+++ b/synapse/config/room_directory.py
@@ -19,6 +19,8 @@ from ._base import Config, ConfigError
 
 
 class RoomDirectoryConfig(Config):
+    section = "roomdirectory"
+
     def read_config(self, config, **kwargs):
         self.enable_room_list_search = config.get("enable_room_list_search", True)
 
diff --git a/synapse/config/saml2_config.py b/synapse/config/saml2_config.py
index 6a8161547a..c5ea2d43a1 100644
--- a/synapse/config/saml2_config.py
+++ b/synapse/config/saml2_config.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2018 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -12,12 +13,50 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
+import re
+
 from synapse.python_dependencies import DependencyException, check_requirements
+from synapse.types import (
+    map_username_to_mxid_localpart,
+    mxid_localpart_allowed_characters,
+)
+from synapse.util.module_loader import load_python_module
 
 from ._base import Config, ConfigError
 
 
+def _dict_merge(merge_dict, into_dict):
+    """Do a deep merge of two dicts
+
+    Recursively merges `merge_dict` into `into_dict`:
+      * For keys where both `merge_dict` and `into_dict` have a dict value, the values
+        are recursively merged
+      * For all other keys, the values in `into_dict` (if any) are overwritten with
+        the value from `merge_dict`.
+
+    Args:
+        merge_dict (dict): dict to merge
+        into_dict (dict): target dict
+    """
+    for k, v in merge_dict.items():
+        if k not in into_dict:
+            into_dict[k] = v
+            continue
+
+        current_val = into_dict[k]
+
+        if isinstance(v, dict) and isinstance(current_val, dict):
+            _dict_merge(v, current_val)
+            continue
+
+        # otherwise we just overwrite
+        into_dict[k] = v
+
+
 class SAML2Config(Config):
+    section = "saml2"
+
     def read_config(self, config, **kwargs):
         self.saml2_enabled = False
 
@@ -26,6 +65,9 @@ class SAML2Config(Config):
         if not saml2_config or not saml2_config.get("enabled", True):
             return
 
+        if not saml2_config.get("sp_config") and not saml2_config.get("config_path"):
+            return
+
         try:
             check_requirements("saml2")
         except DependencyException as e:
@@ -33,21 +75,40 @@ class SAML2Config(Config):
 
         self.saml2_enabled = True
 
-        import saml2.config
+        self.saml2_mxid_source_attribute = saml2_config.get(
+            "mxid_source_attribute", "uid"
+        )
 
-        self.saml2_sp_config = saml2.config.SPConfig()
-        self.saml2_sp_config.load(self._default_saml_config_dict())
-        self.saml2_sp_config.load(saml2_config.get("sp_config", {}))
+        self.saml2_grandfathered_mxid_source_attribute = saml2_config.get(
+            "grandfathered_mxid_source_attribute", "uid"
+        )
+
+        saml2_config_dict = self._default_saml_config_dict()
+        _dict_merge(
+            merge_dict=saml2_config.get("sp_config", {}), into_dict=saml2_config_dict
+        )
 
         config_path = saml2_config.get("config_path", None)
         if config_path is not None:
-            self.saml2_sp_config.load_file(config_path)
+            mod = load_python_module(config_path)
+            _dict_merge(merge_dict=mod.CONFIG, into_dict=saml2_config_dict)
+
+        import saml2.config
+
+        self.saml2_sp_config = saml2.config.SPConfig()
+        self.saml2_sp_config.load(saml2_config_dict)
 
         # session lifetime: in milliseconds
         self.saml2_session_lifetime = self.parse_duration(
             saml2_config.get("saml_session_lifetime", "5m")
         )
 
+        mapping = saml2_config.get("mxid_mapping", "hexencode")
+        try:
+            self.saml2_mxid_mapper = MXID_MAPPER_MAP[mapping]
+        except KeyError:
+            raise ConfigError("%s is not a known mxid_mapping" % (mapping,))
+
     def _default_saml_config_dict(self):
         import saml2
 
@@ -55,6 +116,13 @@ class SAML2Config(Config):
         if public_baseurl is None:
             raise ConfigError("saml2_config requires a public_baseurl to be set")
 
+        required_attributes = {"uid", self.saml2_mxid_source_attribute}
+
+        optional_attributes = {"displayName"}
+        if self.saml2_grandfathered_mxid_source_attribute:
+            optional_attributes.add(self.saml2_grandfathered_mxid_source_attribute)
+        optional_attributes -= required_attributes
+
         metadata_url = public_baseurl + "_matrix/saml2/metadata.xml"
         response_url = public_baseurl + "_matrix/saml2/authn_response"
         return {
@@ -66,8 +134,9 @@ class SAML2Config(Config):
                             (response_url, saml2.BINDING_HTTP_POST)
                         ]
                     },
-                    "required_attributes": ["uid"],
-                    "optional_attributes": ["mail", "surname", "givenname"],
+                    "required_attributes": list(required_attributes),
+                    "optional_attributes": list(optional_attributes),
+                    # "name_id_format": saml2.saml.NAMEID_FORMAT_PERSISTENT,
                 }
             },
         }
@@ -76,12 +145,13 @@ class SAML2Config(Config):
         return """\
         # Enable SAML2 for registration and login. Uses pysaml2.
         #
-        # `sp_config` is the configuration for the pysaml2 Service Provider.
-        # See pysaml2 docs for format of config.
+        # At least one of `sp_config` or `config_path` must be set in this section to
+        # enable SAML login.
         #
-        # Default values will be used for the 'entityid' and 'service' settings,
-        # so it is not normally necessary to specify them unless you need to
-        # override them.
+        # (You will probably also want to set the following options to `false` to
+        # disable the regular login/registration flows:
+        #   * enable_registration
+        #   * password_config.enabled
         #
         # Once SAML support is enabled, a metadata file will be exposed at
         # https://:/_matrix/saml2/metadata.xml, which you may be able to
@@ -89,52 +159,105 @@ class SAML2Config(Config):
         # the IdP to use an ACS location of
         # https://:/_matrix/saml2/authn_response.
         #
-        #saml2_config:
-        #  sp_config:
-        #    # point this to the IdP's metadata. You can use either a local file or
-        #    # (preferably) a URL.
-        #    metadata:
-        #      #local: ["saml2/idp.xml"]
-        #      remote:
-        #        - url: https://our_idp/metadata.xml
-        #
-        #    # By default, the user has to go to our login page first. If you'd like to
-        #    # allow IdP-initiated login, set 'allow_unsolicited: True' in a
-        #    # 'service.sp' section:
-        #    #
-        #    #service:
-        #    #  sp:
-        #    #    allow_unsolicited: True
-        #
-        #    # The examples below are just used to generate our metadata xml, and you
-        #    # may well not need it, depending on your setup. Alternatively you
-        #    # may need a whole lot more detail - see the pysaml2 docs!
-        #
-        #    description: ["My awesome SP", "en"]
-        #    name: ["Test SP", "en"]
-        #
-        #    organization:
-        #      name: Example com
-        #      display_name:
-        #        - ["Example co", "en"]
-        #      url: "http://example.com"
-        #
-        #    contact_person:
-        #      - given_name: Bob
-        #        sur_name: "the Sysadmin"
-        #        email_address": ["admin@example.com"]
-        #        contact_type": technical
-        #
-        #  # Instead of putting the config inline as above, you can specify a
-        #  # separate pysaml2 configuration file:
-        #  #
-        #  config_path: "%(config_dir_path)s/sp_conf.py"
-        #
-        #  # the lifetime of a SAML session. This defines how long a user has to
-        #  # complete the authentication process, if allow_unsolicited is unset.
-        #  # The default is 5 minutes.
-        #  #
-        #  # saml_session_lifetime: 5m
+        saml2_config:
+          # `sp_config` is the configuration for the pysaml2 Service Provider.
+          # See pysaml2 docs for format of config.
+          #
+          # Default values will be used for the 'entityid' and 'service' settings,
+          # so it is not normally necessary to specify them unless you need to
+          # override them.
+          #
+          #sp_config:
+          #  # point this to the IdP's metadata. You can use either a local file or
+          #  # (preferably) a URL.
+          #  metadata:
+          #    #local: ["saml2/idp.xml"]
+          #    remote:
+          #      - url: https://our_idp/metadata.xml
+          #
+          #    # By default, the user has to go to our login page first. If you'd like
+          #    # to allow IdP-initiated login, set 'allow_unsolicited: true' in a
+          #    # 'service.sp' section:
+          #    #
+          #    #service:
+          #    #  sp:
+          #    #    allow_unsolicited: true
+          #
+          #    # The examples below are just used to generate our metadata xml, and you
+          #    # may well not need them, depending on your setup. Alternatively you
+          #    # may need a whole lot more detail - see the pysaml2 docs!
+          #
+          #    description: ["My awesome SP", "en"]
+          #    name: ["Test SP", "en"]
+          #
+          #    organization:
+          #      name: Example com
+          #      display_name:
+          #        - ["Example co", "en"]
+          #      url: "http://example.com"
+          #
+          #    contact_person:
+          #      - given_name: Bob
+          #        sur_name: "the Sysadmin"
+          #        email_address": ["admin@example.com"]
+          #        contact_type": technical
+
+          # Instead of putting the config inline as above, you can specify a
+          # separate pysaml2 configuration file:
+          #
+          #config_path: "%(config_dir_path)s/sp_conf.py"
+
+          # the lifetime of a SAML session. This defines how long a user has to
+          # complete the authentication process, if allow_unsolicited is unset.
+          # The default is 5 minutes.
+          #
+          #saml_session_lifetime: 5m
+
+          # The SAML attribute (after mapping via the attribute maps) to use to derive
+          # the Matrix ID from. 'uid' by default.
+          #
+          #mxid_source_attribute: displayName
+
+          # The mapping system to use for mapping the saml attribute onto a matrix ID.
+          # Options include:
+          #  * 'hexencode' (which maps unpermitted characters to '=xx')
+          #  * 'dotreplace' (which replaces unpermitted characters with '.').
+          # The default is 'hexencode'.
+          #
+          #mxid_mapping: dotreplace
+
+          # In previous versions of synapse, the mapping from SAML attribute to MXID was
+          # always calculated dynamically rather than stored in a table. For backwards-
+          # compatibility, we will look for user_ids matching such a pattern before
+          # creating a new account.
+          #
+          # This setting controls the SAML attribute which will be used for this
+          # backwards-compatibility lookup. Typically it should be 'uid', but if the
+          # attribute maps are changed, it may be necessary to change it.
+          #
+          # The default is 'uid'.
+          #
+          #grandfathered_mxid_source_attribute: upn
         """ % {
             "config_dir_path": config_dir_path
         }
+
+
+DOT_REPLACE_PATTERN = re.compile(
+    ("[^%s]" % (re.escape("".join(mxid_localpart_allowed_characters)),))
+)
+
+
+def dot_replace_for_mxid(username: str) -> str:
+    username = username.lower()
+    username = DOT_REPLACE_PATTERN.sub(".", username)
+
+    # regular mxids aren't allowed to start with an underscore either
+    username = re.sub("^_", "", username)
+    return username
+
+
+MXID_MAPPER_MAP = {
+    "hexencode": map_username_to_mxid_localpart,
+    "dotreplace": dot_replace_for_mxid,
+}
diff --git a/synapse/config/server.py b/synapse/config/server.py
index 2abdef0971..d556df308d 100644
--- a/synapse/config/server.py
+++ b/synapse/config/server.py
@@ -19,6 +19,7 @@ import logging
 import os.path
 import re
 from textwrap import indent
+from typing import List
 
 import attr
 import yaml
@@ -48,8 +49,17 @@ ROOM_COMPLEXITY_TOO_GREAT = (
     "to join this room."
 )
 
+METRICS_PORT_WARNING = """\
+The metrics_port configuration option is deprecated in Synapse 0.31 in favour of
+a listener. Please see
+https://github.com/matrix-org/synapse/blob/master/docs/metrics-howto.md
+on how to configure the new listener.
+--------------------------------------------------------------------------------"""
+
 
 class ServerConfig(Config):
+    section = "server"
+
     def read_config(self, config, **kwargs):
         self.server_name = config["server_name"]
         self.server_context = config.get("server_context", None)
@@ -161,11 +171,28 @@ class ServerConfig(Config):
         )
 
         self.mau_trial_days = config.get("mau_trial_days", 0)
+        self.mau_limit_alerting = config.get("mau_limit_alerting", True)
+
+        # How long to keep redacted events in the database in unredacted form
+        # before redacting them.
+        redaction_retention_period = config.get("redaction_retention_period", "7d")
+        if redaction_retention_period is not None:
+            self.redaction_retention_period = self.parse_duration(
+                redaction_retention_period
+            )
+        else:
+            self.redaction_retention_period = None
+
+        # How long to keep entries in the `users_ips` table.
+        user_ips_max_age = config.get("user_ips_max_age", "28d")
+        if user_ips_max_age is not None:
+            self.user_ips_max_age = self.parse_duration(user_ips_max_age)
+        else:
+            self.user_ips_max_age = None
 
         # Options to disable HS
         self.hs_disabled = config.get("hs_disabled", False)
         self.hs_disabled_message = config.get("hs_disabled_message", "")
-        self.hs_disabled_limit_type = config.get("hs_disabled_limit_type", "")
 
         # Admin uri to direct users at should their instance become blocked
         # due to resource constraints
@@ -219,7 +246,7 @@ class ServerConfig(Config):
         # events with profile information that differ from the target's global profile.
         self.allow_per_room_profiles = config.get("allow_per_room_profiles", True)
 
-        self.listeners = []
+        self.listeners = []  # type: List[dict]
         for listener in config.get("listeners", []):
             if not isinstance(listener.get("port", None), int):
                 raise ConfigError(
@@ -263,7 +290,10 @@ class ServerConfig(Config):
                 validator=attr.validators.instance_of(bool), default=False
             )
             complexity = attr.ib(
-                validator=attr.validators.instance_of((int, float)), default=1.0
+                validator=attr.validators.instance_of(
+                    (float, int)  # type: ignore[arg-type] # noqa
+                ),
+                default=1.0,
             )
             complexity_error = attr.ib(
                 validator=attr.validators.instance_of(str),
@@ -324,14 +354,7 @@ class ServerConfig(Config):
 
         metrics_port = config.get("metrics_port")
         if metrics_port:
-            logger.warn(
-                (
-                    "The metrics_port configuration option is deprecated in Synapse 0.31 "
-                    "in favour of a listener. Please see "
-                    "http://github.com/matrix-org/synapse/blob/master/docs/metrics-howto.rst"
-                    " on how to configure the new listener."
-                )
-            )
+            logger.warning(METRICS_PORT_WARNING)
 
             self.listeners.append(
                 {
@@ -345,13 +368,11 @@ class ServerConfig(Config):
 
         _check_resource_config(self.listeners)
 
-        # An experimental option to try and periodically clean up extremities
-        # by sending dummy events.
         self.cleanup_extremities_with_dummy_events = config.get(
-            "cleanup_extremities_with_dummy_events", False
+            "cleanup_extremities_with_dummy_events", True
         )
 
-    def has_tls_listener(self):
+    def has_tls_listener(self) -> bool:
         return any(l["tls"] for l in self.listeners)
 
     def generate_config_section(
@@ -511,7 +532,7 @@ class ServerConfig(Config):
         # Whether room invites to users on this server should be blocked
         # (except those sent by local server admins). The default is False.
         #
-        #block_non_admin_invites: True
+        #block_non_admin_invites: true
 
         # Room searching
         #
@@ -535,6 +556,9 @@ class ServerConfig(Config):
         # blacklist IP address CIDR ranges. If this option is not specified, or
         # specified with an empty list, no ip range blacklist will be enforced.
         #
+        # As of Synapse v1.4.0 this option also affects any outbound requests to identity
+        # servers provided by user input.
+        #
         # (0.0.0.0 and :: are always blacklisted, whether or not they are explicitly
         # listed here, since they correspond to unroutable addresses.)
         #
@@ -561,8 +585,8 @@ class ServerConfig(Config):
         #
         #   type: the type of listener. Normally 'http', but other valid options are:
         #       'manhole' (see docs/manhole.md),
-        #       'metrics' (see docs/metrics-howto.rst),
-        #       'replication' (see docs/workers.rst).
+        #       'metrics' (see docs/metrics-howto.md),
+        #       'replication' (see docs/workers.md).
         #
         #   tls: set to true to enable TLS for this listener. Will use the TLS
         #       key/cert specified in tls_private_key_path / tls_certificate_path.
@@ -597,12 +621,12 @@ class ServerConfig(Config):
         #
         #   media: the media API (/_matrix/media).
         #
-        #   metrics: the metrics interface. See docs/metrics-howto.rst.
+        #   metrics: the metrics interface. See docs/metrics-howto.md.
         #
         #   openid: OpenID authentication.
         #
         #   replication: the HTTP replication API (/_synapse/replication). See
-        #       docs/workers.rst.
+        #       docs/workers.md.
         #
         #   static: static resources under synapse/static (/_matrix/static). (Mostly
         #       useful for 'fallback authentication'.)
@@ -622,7 +646,7 @@ class ServerConfig(Config):
           # that unwraps TLS.
           #
           # If you plan to use a reverse proxy, please see
-          # https://github.com/matrix-org/synapse/blob/master/docs/reverse_proxy.rst.
+          # https://github.com/matrix-org/synapse/blob/master/docs/reverse_proxy.md.
           #
           %(unsecure_http_bindings)s
 
@@ -649,9 +673,8 @@ class ServerConfig(Config):
 
         # Global blocking
         #
-        #hs_disabled: False
+        #hs_disabled: false
         #hs_disabled_message: 'Human readable reason for why the HS is blocked'
-        #hs_disabled_limit_type: 'error code(str), to help clients decode reason'
 
         # Monthly Active User Blocking
         #
@@ -671,15 +694,22 @@ class ServerConfig(Config):
         # sign up in a short space of time never to return after their initial
         # session.
         #
-        #limit_usage_by_mau: False
+        # 'mau_limit_alerting' is a means of limiting client side alerting
+        # should the mau limit be reached. This is useful for small instances
+        # where the admin has 5 mau seats (say) for 5 specific people and no
+        # interest increasing the mau limit further. Defaults to True, which
+        # means that alerting is enabled
+        #
+        #limit_usage_by_mau: false
         #max_mau_value: 50
         #mau_trial_days: 2
+        #mau_limit_alerting: false
 
         # If enabled, the metrics for the number of monthly active users will
         # be populated, however no one will be limited. If limit_usage_by_mau
         # is true, this is implied to be true.
         #
-        #mau_stats_only: False
+        #mau_stats_only: false
 
         # Sometimes the server admin will want to ensure certain accounts are
         # never blocked by mau checking. These accounts are specified here.
@@ -704,7 +734,7 @@ class ServerConfig(Config):
         #
         # Uncomment the below lines to enable:
         #limit_remote_rooms:
-        #  enabled: True
+        #  enabled: true
         #  complexity: 1.0
         #  complexity_error: "This room is too complex."
 
@@ -718,6 +748,19 @@ class ServerConfig(Config):
         # Defaults to 'true'.
         #
         #allow_per_room_profiles: false
+
+        # How long to keep redacted events in unredacted form in the database. After
+        # this period redacted events get replaced with their redacted form in the DB.
+        #
+        # Defaults to `7d`. Set to `null` to disable.
+        #
+        #redaction_retention_period: 28d
+
+        # How long to track users' last seen time and IPs in the database.
+        #
+        # Defaults to `28d`. Set to `null` to disable clearing out of old rows.
+        #
+        #user_ips_max_age: 14d
         """
             % locals()
         )
diff --git a/synapse/config/server_notices_config.py b/synapse/config/server_notices_config.py
index eaac3d73bc..6ea2ea8869 100644
--- a/synapse/config/server_notices_config.py
+++ b/synapse/config/server_notices_config.py
@@ -59,8 +59,10 @@ class ServerNoticesConfig(Config):
             None if server notices are not enabled.
     """
 
-    def __init__(self):
-        super(ServerNoticesConfig, self).__init__()
+    section = "servernotices"
+
+    def __init__(self, *args):
+        super(ServerNoticesConfig, self).__init__(*args)
         self.server_notices_mxid = None
         self.server_notices_mxid_display_name = None
         self.server_notices_mxid_avatar_url = None
diff --git a/synapse/config/spam_checker.py b/synapse/config/spam_checker.py
index e40797ab50..36e0ddab5c 100644
--- a/synapse/config/spam_checker.py
+++ b/synapse/config/spam_checker.py
@@ -19,6 +19,8 @@ from ._base import Config
 
 
 class SpamCheckerConfig(Config):
+    section = "spamchecker"
+
     def read_config(self, config, **kwargs):
         self.spam_checker = None
 
diff --git a/synapse/config/stats.py b/synapse/config/stats.py
index b518a3ed9c..62485189ea 100644
--- a/synapse/config/stats.py
+++ b/synapse/config/stats.py
@@ -25,21 +25,20 @@ class StatsConfig(Config):
     Configuration for the behaviour of synapse's stats engine
     """
 
+    section = "stats"
+
     def read_config(self, config, **kwargs):
         self.stats_enabled = True
-        self.stats_bucket_size = 86400
+        self.stats_bucket_size = 86400 * 1000
         self.stats_retention = sys.maxsize
         stats_config = config.get("stats", None)
         if stats_config:
             self.stats_enabled = stats_config.get("enabled", self.stats_enabled)
-            self.stats_bucket_size = (
-                self.parse_duration(stats_config.get("bucket_size", "1d")) / 1000
+            self.stats_bucket_size = self.parse_duration(
+                stats_config.get("bucket_size", "1d")
             )
-            self.stats_retention = (
-                self.parse_duration(
-                    stats_config.get("retention", "%ds" % (sys.maxsize,))
-                )
-                / 1000
+            self.stats_retention = self.parse_duration(
+                stats_config.get("retention", "%ds" % (sys.maxsize,))
             )
 
     def generate_config_section(self, config_dir_path, server_name, **kwargs):
diff --git a/synapse/config/third_party_event_rules.py b/synapse/config/third_party_event_rules.py
index b3431441b9..10a99c792e 100644
--- a/synapse/config/third_party_event_rules.py
+++ b/synapse/config/third_party_event_rules.py
@@ -19,6 +19,8 @@ from ._base import Config
 
 
 class ThirdPartyRulesConfig(Config):
+    section = "thirdpartyrules"
+
     def read_config(self, config, **kwargs):
         self.third_party_event_rules = None
 
diff --git a/synapse/config/tls.py b/synapse/config/tls.py
index c0148aa95c..2e9e478a2a 100644
--- a/synapse/config/tls.py
+++ b/synapse/config/tls.py
@@ -18,6 +18,7 @@ import os
 import warnings
 from datetime import datetime
 from hashlib import sha256
+from typing import List
 
 import six
 
@@ -33,7 +34,9 @@ logger = logging.getLogger(__name__)
 
 
 class TlsConfig(Config):
-    def read_config(self, config, config_dir_path, **kwargs):
+    section = "tls"
+
+    def read_config(self, config: dict, config_dir_path: str, **kwargs):
 
         acme_config = config.get("acme", None)
         if acme_config is None:
@@ -57,7 +60,7 @@ class TlsConfig(Config):
         self.tls_certificate_file = self.abspath(config.get("tls_certificate_path"))
         self.tls_private_key_file = self.abspath(config.get("tls_private_key_path"))
 
-        if self.has_tls_listener():
+        if self.root.server.has_tls_listener():
             if not self.tls_certificate_file:
                 raise ConfigError(
                     "tls_certificate_path must be specified if TLS-enabled listeners are "
@@ -108,10 +111,17 @@ class TlsConfig(Config):
         )
 
         # Support globs (*) in whitelist values
-        self.federation_certificate_verification_whitelist = []
+        self.federation_certificate_verification_whitelist = []  # type: List[str]
         for entry in fed_whitelist_entries:
+            try:
+                entry_regex = glob_to_regex(entry.encode("ascii").decode("ascii"))
+            except UnicodeEncodeError:
+                raise ConfigError(
+                    "IDNA domain names are not allowed in the "
+                    "federation_certificate_verification_whitelist: %s" % (entry,)
+                )
+
             # Convert globs to regex
-            entry_regex = glob_to_regex(entry)
             self.federation_certificate_verification_whitelist.append(entry_regex)
 
         # List of custom certificate authorities for federation traffic validation
@@ -279,6 +289,9 @@ class TlsConfig(Config):
             "http://localhost:8009/.well-known/acme-challenge"
         )
 
+        # flake8 doesn't recognise that variables are used in the below string
+        _ = tls_enabled, proxypassline, acme_enabled, default_acme_account_file
+
         return (
             """\
         ## TLS ##
@@ -441,7 +454,11 @@ class TlsConfig(Config):
         #tls_fingerprints: [{"sha256": ""}]
 
         """
-            % locals()
+            # Lowercase the string representation of boolean values
+            % {
+                x[0]: str(x[1]).lower() if isinstance(x[1], bool) else x[1]
+                for x in locals().items()
+            }
         )
 
     def read_tls_certificate(self):
diff --git a/synapse/config/tracer.py b/synapse/config/tracer.py
index 95e7ccb3a3..8be1346113 100644
--- a/synapse/config/tracer.py
+++ b/synapse/config/tracer.py
@@ -13,10 +13,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from synapse.python_dependencies import DependencyException, check_requirements
+
 from ._base import Config, ConfigError
 
 
 class TracerConfig(Config):
+    section = "tracing"
+
     def read_config(self, config, **kwargs):
         opentracing_config = config.get("opentracing")
         if opentracing_config is None:
@@ -32,6 +36,11 @@ class TracerConfig(Config):
         if not self.opentracer_enabled:
             return
 
+        try:
+            check_requirements("opentracing")
+        except DependencyException as e:
+            raise ConfigError(e.message)
+
         # The tracer is enabled so sanitize the config
 
         self.opentracer_whitelist = opentracing_config.get("homeserver_whitelist", [])
diff --git a/synapse/config/user_directory.py b/synapse/config/user_directory.py
index f6313e17d4..c8d19c5d6b 100644
--- a/synapse/config/user_directory.py
+++ b/synapse/config/user_directory.py
@@ -21,6 +21,8 @@ class UserDirectoryConfig(Config):
     Configuration for the behaviour of the /user_directory API
     """
 
+    section = "userdirectory"
+
     def read_config(self, config, **kwargs):
         self.user_directory_search_enabled = True
         self.user_directory_search_all_users = False
diff --git a/synapse/config/voip.py b/synapse/config/voip.py
index 2ca0e1cf70..b313bff140 100644
--- a/synapse/config/voip.py
+++ b/synapse/config/voip.py
@@ -16,6 +16,8 @@ from ._base import Config
 
 
 class VoipConfig(Config):
+    section = "voip"
+
     def read_config(self, config, **kwargs):
         self.turn_uris = config.get("turn_uris", [])
         self.turn_shared_secret = config.get("turn_shared_secret")
@@ -54,5 +56,5 @@ class VoipConfig(Config):
         # connect to arbitrary endpoints without having first signed up for a
         # valid account (e.g. by passing a CAPTCHA).
         #
-        #turn_allow_guests: True
+        #turn_allow_guests: true
         """
diff --git a/synapse/config/workers.py b/synapse/config/workers.py
index bc0fc165e3..fef72ed974 100644
--- a/synapse/config/workers.py
+++ b/synapse/config/workers.py
@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-# Copyright 2016 matrix.org
+# Copyright 2016 OpenMarket Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -21,6 +21,8 @@ class WorkerConfig(Config):
     They have their own pid_file and listener configuration. They use the
     replication_url to talk to the main synapse process."""
 
+    section = "worker"
+
     def read_config(self, config, **kwargs):
         self.worker_app = config.get("worker_app")
 
diff --git a/synapse/crypto/context_factory.py b/synapse/crypto/context_factory.py
index 06e63a96b5..e93f0b3705 100644
--- a/synapse/crypto/context_factory.py
+++ b/synapse/crypto/context_factory.py
@@ -15,7 +15,6 @@
 
 import logging
 
-import idna
 from service_identity import VerificationError
 from service_identity.pyopenssl import verify_hostname, verify_ip_address
 from zope.interface import implementer
@@ -114,14 +113,20 @@ class ClientTLSOptionsFactory(object):
         self._no_verify_ssl_context = self._no_verify_ssl.getContext()
         self._no_verify_ssl_context.set_info_callback(self._context_info_cb)
 
-    def get_options(self, host):
+    def get_options(self, host: bytes):
+
+        # IPolicyForHTTPS.get_options takes bytes, but we want to compare
+        # against the str whitelist. The hostnames in the whitelist are already
+        # IDNA-encoded like the hosts will be here.
+        ascii_host = host.decode("ascii")
+
         # Check if certificate verification has been enabled
         should_verify = self._config.federation_verify_certificates
 
         # Check if we've disabled certificate verification for this host
         if should_verify:
             for regex in self._config.federation_certificate_verification_whitelist:
-                if regex.match(host):
+                if regex.match(ascii_host):
                     should_verify = False
                     break
 
@@ -162,7 +167,7 @@ class SSLClientConnectionCreator(object):
     Replaces twisted.internet.ssl.ClientTLSOptions
     """
 
-    def __init__(self, hostname, ctx, verify_certs):
+    def __init__(self, hostname: bytes, ctx, verify_certs: bool):
         self._ctx = ctx
         self._verifier = ConnectionVerifier(hostname, verify_certs)
 
@@ -190,21 +195,16 @@ class ConnectionVerifier(object):
 
     # This code is based on twisted.internet.ssl.ClientTLSOptions.
 
-    def __init__(self, hostname, verify_certs):
+    def __init__(self, hostname: bytes, verify_certs):
         self._verify_certs = verify_certs
 
-        if isIPAddress(hostname) or isIPv6Address(hostname):
-            self._hostnameBytes = hostname.encode("ascii")
+        _decoded = hostname.decode("ascii")
+        if isIPAddress(_decoded) or isIPv6Address(_decoded):
             self._is_ip_address = True
         else:
-            # twisted's ClientTLSOptions falls back to the stdlib impl here if
-            # idna is not installed, but points out that lacks support for
-            # IDNA2008 (http://bugs.python.org/issue17305).
-            #
-            # We can rely on having idna.
-            self._hostnameBytes = idna.encode(hostname)
             self._is_ip_address = False
 
+        self._hostnameBytes = hostname
         self._hostnameASCII = self._hostnameBytes.decode("ascii")
 
     def verify_context_info_cb(self, ssl_connection, where):
diff --git a/synapse/crypto/event_signing.py b/synapse/crypto/event_signing.py
index 694fb2c816..ccaa8a9920 100644
--- a/synapse/crypto/event_signing.py
+++ b/synapse/crypto/event_signing.py
@@ -125,9 +125,11 @@ def compute_event_signature(event_dict, signature_name, signing_key):
     redact_json = prune_event_dict(event_dict)
     redact_json.pop("age_ts", None)
     redact_json.pop("unsigned", None)
-    logger.debug("Signing event: %s", encode_canonical_json(redact_json))
+    if logger.isEnabledFor(logging.DEBUG):
+        logger.debug("Signing event: %s", encode_canonical_json(redact_json))
     redact_json = sign_json(redact_json, signature_name, signing_key)
-    logger.debug("Signed event: %s", encode_canonical_json(redact_json))
+    if logger.isEnabledFor(logging.DEBUG):
+        logger.debug("Signed event: %s", encode_canonical_json(redact_json))
     return redact_json["signatures"]
 
 
diff --git a/synapse/event_auth.py b/synapse/event_auth.py
index cd52e3f867..ec3243b27b 100644
--- a/synapse/event_auth.py
+++ b/synapse/event_auth.py
@@ -77,7 +77,7 @@ def check(room_version, event, auth_events, do_sig_check=True, do_size_check=Tru
     if auth_events is None:
         # Oh, we don't know what the state of the room was, so we
         # are trusting that this is allowed (at least for now)
-        logger.warn("Trusting event: %s", event.event_id)
+        logger.warning("Trusting event: %s", event.event_id)
         return
 
     if event.type == EventTypes.Create:
@@ -493,8 +493,7 @@ def _check_power_levels(event, auth_events):
         new_level_too_big = new_level is not None and new_level > user_level
         if old_level_too_big or new_level_too_big:
             raise AuthError(
-                403,
-                "You don't have permission to add ops level greater " "than your own",
+                403, "You don't have permission to add ops level greater than your own"
             )
 
 
@@ -637,11 +636,11 @@ def auth_types_for_event(event):
     if event.type == EventTypes.Create:
         return []
 
-    auth_types = []
-
-    auth_types.append((EventTypes.PowerLevels, ""))
-    auth_types.append((EventTypes.Member, event.sender))
-    auth_types.append((EventTypes.Create, ""))
+    auth_types = [
+        (EventTypes.PowerLevels, ""),
+        (EventTypes.Member, event.sender),
+        (EventTypes.Create, ""),
+    ]
 
     if event.type == EventTypes.Member:
         membership = event.content["membership"]
diff --git a/synapse/events/snapshot.py b/synapse/events/snapshot.py
index acbcbeeced..64e898f40c 100644
--- a/synapse/events/snapshot.py
+++ b/synapse/events/snapshot.py
@@ -12,104 +12,125 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+from typing import Dict, Optional, Tuple, Union
 
 from six import iteritems
 
+import attr
 from frozendict import frozendict
 
 from twisted.internet import defer
 
+from synapse.appservice import ApplicationService
 from synapse.logging.context import make_deferred_yieldable, run_in_background
 
 
-class EventContext(object):
+@attr.s(slots=True)
+class EventContext:
     """
+    Holds information relevant to persisting an event
+
     Attributes:
-        state_group (int|None): state group id, if the state has been stored
-            as a state group. This is usually only None if e.g. the event is
-            an outlier.
-        rejected (bool|str): A rejection reason if the event was rejected, else
-            False
+        rejected: A rejection reason if the event was rejected, else False
 
-        push_actions (list[(str, list[object])]): list of (user_id, actions)
-            tuples
+        _state_group: The ID of the state group for this event. Note that state events
+            are persisted with a state group which includes the new event, so this is
+            effectively the state *after* the event in question.
 
-        prev_group (int): Previously persisted state group. ``None`` for an
-            outlier.
-        delta_ids (dict[(str, str), str]): Delta from ``prev_group``.
-            (type, state_key) -> event_id. ``None`` for an outlier.
+            For a *rejected* state event, where the state of the rejected event is
+            ignored, this state_group should never make it into the
+            event_to_state_groups table. Indeed, inspecting this value for a rejected
+            state event is almost certainly incorrect.
 
-        prev_state_events (?): XXX: is this ever set to anything other than
-            the empty list?
+            For an outlier, where we don't have the state at the event, this will be
+            None.
+
+            Note that this is a private attribute: it should be accessed via
+            the ``state_group`` property.
+
+        state_group_before_event: The ID of the state group representing the state
+            of the room before this event.
+
+            If this is a non-state event, this will be the same as ``state_group``. If
+            it's a state event, it will be the same as ``prev_group``.
+
+            If ``state_group`` is None (ie, the event is an outlier),
+            ``state_group_before_event`` will always also be ``None``.
+
+        prev_group: If it is known, ``state_group``'s prev_group. Note that this being
+            None does not necessarily mean that ``state_group`` does not have
+            a prev_group!
+
+            If the event is a state event, this is normally the same as ``prev_group``.
+
+            If ``state_group`` is None (ie, the event is an outlier), ``prev_group``
+            will always also be ``None``.
+
+            Note that this *not* (necessarily) the state group associated with
+            ``_prev_state_ids``.
+
+        delta_ids: If ``prev_group`` is not None, the state delta between ``prev_group``
+            and ``state_group``.
+
+        app_service: If this event is being sent by a (local) application service, that
+            app service.
+
+        _current_state_ids: The room state map, including this event - ie, the state
+            in ``state_group``.
 
-        _current_state_ids (dict[(str, str), str]|None):
-            The current state map including the current event. None if outlier
-            or we haven't fetched the state from DB yet.
             (type, state_key) -> event_id
 
-        _prev_state_ids (dict[(str, str), str]|None):
-            The current state map excluding the current event. None if outlier
-            or we haven't fetched the state from DB yet.
+            FIXME: what is this for an outlier? it seems ill-defined. It seems like
+            it could be either {}, or the state we were given by the remote
+            server, depending on $THINGS
+
+            Note that this is a private attribute: it should be accessed via
+            ``get_current_state_ids``. _AsyncEventContext impl calculates this
+            on-demand: it will be None until that happens.
+
+        _prev_state_ids: The room state map, excluding this event - ie, the state
+            in ``state_group_before_event``. For a non-state
+            event, this will be the same as _current_state_events.
+
+            Note that it is a completely different thing to prev_group!
+
             (type, state_key) -> event_id
 
-        _fetching_state_deferred (Deferred|None): Resolves when *_state_ids have
-            been calculated. None if we haven't started calculating yet
+            FIXME: again, what is this for an outlier?
 
-        _event_type (str): The type of the event the context is associated with.
-            Only set when state has not been fetched yet.
-
-        _event_state_key (str|None): The state_key of the event the context is
-            associated with. Only set when state has not been fetched yet.
-
-        _prev_state_id (str|None): If the event associated with the context is
-            a state event, then `_prev_state_id` is the event_id of the state
-            that was replaced.
-            Only set when state has not been fetched yet.
+            As with _current_state_ids, this is a private attribute. It should be
+            accessed via get_prev_state_ids.
     """
 
-    __slots__ = [
-        "state_group",
-        "rejected",
-        "prev_group",
-        "delta_ids",
-        "prev_state_events",
-        "app_service",
-        "_current_state_ids",
-        "_prev_state_ids",
-        "_prev_state_id",
-        "_event_type",
-        "_event_state_key",
-        "_fetching_state_deferred",
-    ]
+    rejected = attr.ib(default=False, type=Union[bool, str])
+    _state_group = attr.ib(default=None, type=Optional[int])
+    state_group_before_event = attr.ib(default=None, type=Optional[int])
+    prev_group = attr.ib(default=None, type=Optional[int])
+    delta_ids = attr.ib(default=None, type=Optional[Dict[Tuple[str, str], str]])
+    app_service = attr.ib(default=None, type=Optional[ApplicationService])
 
-    def __init__(self):
-        self.prev_state_events = []
-        self.rejected = False
-        self.app_service = None
+    _current_state_ids = attr.ib(
+        default=None, type=Optional[Dict[Tuple[str, str], str]]
+    )
+    _prev_state_ids = attr.ib(default=None, type=Optional[Dict[Tuple[str, str], str]])
 
     @staticmethod
     def with_state(
-        state_group, current_state_ids, prev_state_ids, prev_group=None, delta_ids=None
+        state_group,
+        state_group_before_event,
+        current_state_ids,
+        prev_state_ids,
+        prev_group=None,
+        delta_ids=None,
     ):
-        context = EventContext()
-
-        # The current state including the current event
-        context._current_state_ids = current_state_ids
-        # The current state excluding the current event
-        context._prev_state_ids = prev_state_ids
-        context.state_group = state_group
-
-        context._prev_state_id = None
-        context._event_type = None
-        context._event_state_key = None
-        context._fetching_state_deferred = defer.succeed(None)
-
-        # A previously persisted state group and a delta between that
-        # and this state.
-        context.prev_group = prev_group
-        context.delta_ids = delta_ids
-
-        return context
+        return EventContext(
+            current_state_ids=current_state_ids,
+            prev_state_ids=prev_state_ids,
+            state_group=state_group,
+            state_group_before_event=state_group_before_event,
+            prev_group=prev_group,
+            delta_ids=delta_ids,
+        )
 
     @defer.inlineCallbacks
     def serialize(self, event, store):
@@ -137,11 +158,11 @@ class EventContext(object):
             "prev_state_id": prev_state_id,
             "event_type": event.type,
             "event_state_key": event.state_key if event.is_state() else None,
-            "state_group": self.state_group,
+            "state_group": self._state_group,
+            "state_group_before_event": self.state_group_before_event,
             "rejected": self.rejected,
             "prev_group": self.prev_group,
             "delta_ids": _encode_state_dict(self.delta_ids),
-            "prev_state_events": self.prev_state_events,
             "app_service_id": self.app_service.id if self.app_service else None,
         }
 
@@ -157,24 +178,18 @@ class EventContext(object):
         Returns:
             EventContext
         """
-        context = EventContext()
-
-        # We use the state_group and prev_state_id stuff to pull the
-        # current_state_ids out of the DB and construct prev_state_ids.
-        context._prev_state_id = input["prev_state_id"]
-        context._event_type = input["event_type"]
-        context._event_state_key = input["event_state_key"]
-
-        context._current_state_ids = None
-        context._prev_state_ids = None
-        context._fetching_state_deferred = None
-
-        context.state_group = input["state_group"]
-        context.prev_group = input["prev_group"]
-        context.delta_ids = _decode_state_dict(input["delta_ids"])
-
-        context.rejected = input["rejected"]
-        context.prev_state_events = input["prev_state_events"]
+        context = _AsyncEventContextImpl(
+            # We use the state_group and prev_state_id stuff to pull the
+            # current_state_ids out of the DB and construct prev_state_ids.
+            prev_state_id=input["prev_state_id"],
+            event_type=input["event_type"],
+            event_state_key=input["event_state_key"],
+            state_group=input["state_group"],
+            state_group_before_event=input["state_group_before_event"],
+            prev_group=input["prev_group"],
+            delta_ids=_decode_state_dict(input["delta_ids"]),
+            rejected=input["rejected"],
+        )
 
         app_service_id = input["app_service_id"]
         if app_service_id:
@@ -182,29 +197,52 @@ class EventContext(object):
 
         return context
 
+    @property
+    def state_group(self) -> Optional[int]:
+        """The ID of the state group for this event.
+
+        Note that state events are persisted with a state group which includes the new
+        event, so this is effectively the state *after* the event in question.
+
+        For an outlier, where we don't have the state at the event, this will be None.
+
+        It is an error to access this for a rejected event, since rejected state should
+        not make it into the room state. Accessing this property will raise an exception
+        if ``rejected`` is set.
+        """
+        if self.rejected:
+            raise RuntimeError("Attempt to access state_group of rejected event")
+
+        return self._state_group
+
     @defer.inlineCallbacks
     def get_current_state_ids(self, store):
-        """Gets the current state IDs
+        """
+        Gets the room state map, including this event - ie, the state in ``state_group``
+
+        It is an error to access this for a rejected event, since rejected state should
+        not make it into the room state. This method will raise an exception if
+        ``rejected`` is set.
 
         Returns:
             Deferred[dict[(str, str), str]|None]: Returns None if state_group
                 is None, which happens when the associated event is an outlier.
+
                 Maps a (type, state_key) to the event ID of the state event matching
                 this tuple.
         """
+        if self.rejected:
+            raise RuntimeError("Attempt to access state_ids of rejected event")
 
-        if not self._fetching_state_deferred:
-            self._fetching_state_deferred = run_in_background(
-                self._fill_out_state, store
-            )
-
-        yield make_deferred_yieldable(self._fetching_state_deferred)
-
+        yield self._ensure_fetched(store)
         return self._current_state_ids
 
     @defer.inlineCallbacks
     def get_prev_state_ids(self, store):
-        """Gets the prev state IDs
+        """
+        Gets the room state map, excluding this event.
+
+        For a non-state event, this will be the same as get_current_state_ids().
 
         Returns:
             Deferred[dict[(str, str), str]|None]: Returns None if state_group
@@ -212,27 +250,64 @@ class EventContext(object):
                 Maps a (type, state_key) to the event ID of the state event matching
                 this tuple.
         """
-
-        if not self._fetching_state_deferred:
-            self._fetching_state_deferred = run_in_background(
-                self._fill_out_state, store
-            )
-
-        yield make_deferred_yieldable(self._fetching_state_deferred)
-
+        yield self._ensure_fetched(store)
         return self._prev_state_ids
 
     def get_cached_current_state_ids(self):
         """Gets the current state IDs if we have them already cached.
 
+        It is an error to access this for a rejected event, since rejected state should
+        not make it into the room state. This method will raise an exception if
+        ``rejected`` is set.
+
         Returns:
             dict[(str, str), str]|None: Returns None if we haven't cached the
             state or if state_group is None, which happens when the associated
             event is an outlier.
         """
+        if self.rejected:
+            raise RuntimeError("Attempt to access state_ids of rejected event")
 
         return self._current_state_ids
 
+    def _ensure_fetched(self, store):
+        return defer.succeed(None)
+
+
+@attr.s(slots=True)
+class _AsyncEventContextImpl(EventContext):
+    """
+    An implementation of EventContext which fetches _current_state_ids and
+    _prev_state_ids from the database on demand.
+
+    Attributes:
+
+        _fetching_state_deferred (Deferred|None): Resolves when *_state_ids have
+            been calculated. None if we haven't started calculating yet
+
+        _event_type (str): The type of the event the context is associated with.
+
+        _event_state_key (str): The state_key of the event the context is
+            associated with.
+
+        _prev_state_id (str|None): If the event associated with the context is
+            a state event, then `_prev_state_id` is the event_id of the state
+            that was replaced.
+    """
+
+    _prev_state_id = attr.ib(default=None)
+    _event_type = attr.ib(default=None)
+    _event_state_key = attr.ib(default=None)
+    _fetching_state_deferred = attr.ib(default=None)
+
+    def _ensure_fetched(self, store):
+        if not self._fetching_state_deferred:
+            self._fetching_state_deferred = run_in_background(
+                self._fill_out_state, store
+            )
+
+        return make_deferred_yieldable(self._fetching_state_deferred)
+
     @defer.inlineCallbacks
     def _fill_out_state(self, store):
         """Called to populate the _current_state_ids and _prev_state_ids
@@ -250,27 +325,6 @@ class EventContext(object):
         else:
             self._prev_state_ids = self._current_state_ids
 
-    @defer.inlineCallbacks
-    def update_state(
-        self, state_group, prev_state_ids, current_state_ids, prev_group, delta_ids
-    ):
-        """Replace the state in the context
-        """
-
-        # We need to make sure we wait for any ongoing fetching of state
-        # to complete so that the updated state doesn't get clobbered
-        if self._fetching_state_deferred:
-            yield make_deferred_yieldable(self._fetching_state_deferred)
-
-        self.state_group = state_group
-        self._prev_state_ids = prev_state_ids
-        self.prev_group = prev_group
-        self._current_state_ids = current_state_ids
-        self.delta_ids = delta_ids
-
-        # We need to ensure that that we've marked as having fetched the state
-        self._fetching_state_deferred = defer.succeed(None)
-
 
 def _encode_state_dict(state_dict):
     """Since dicts of (type, state_key) -> event_id cannot be serialized in
diff --git a/synapse/events/spamcheck.py b/synapse/events/spamcheck.py
index 129771f183..5a907718d6 100644
--- a/synapse/events/spamcheck.py
+++ b/synapse/events/spamcheck.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2017 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,6 +14,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import inspect
+
+from synapse.spam_checker_api import SpamCheckerApi
+
 
 class SpamChecker(object):
     def __init__(self, hs):
@@ -26,7 +31,14 @@ class SpamChecker(object):
             pass
 
         if module is not None:
-            self.spam_checker = module(config=config)
+            # Older spam checkers don't accept the `api` argument, so we
+            # try and detect support.
+            spam_args = inspect.getfullargspec(module)
+            if "api" in spam_args.args:
+                api = SpamCheckerApi(hs)
+                self.spam_checker = module(config=config, api=api)
+            else:
+                self.spam_checker = module(config=config)
 
     def check_event_for_spam(self, event):
         """Checks if a given event is considered "spammy" by this server.
diff --git a/synapse/federation/federation_base.py b/synapse/federation/federation_base.py
index 5a1e23a145..0e22183280 100644
--- a/synapse/federation/federation_base.py
+++ b/synapse/federation/federation_base.py
@@ -102,7 +102,7 @@ class FederationBase(object):
                     pass
 
             if not res:
-                logger.warn(
+                logger.warning(
                     "Failed to find copy of %s with valid signature", pdu.event_id
                 )
 
@@ -173,7 +173,7 @@ class FederationBase(object):
                     return redacted_event
 
                 if self.spam_checker.check_event_for_spam(pdu):
-                    logger.warn(
+                    logger.warning(
                         "Event contains spam, redacting %s: %s",
                         pdu.event_id,
                         pdu.get_pdu_json(),
@@ -185,7 +185,7 @@ class FederationBase(object):
         def errback(failure, pdu):
             failure.trap(SynapseError)
             with PreserveLoggingContext(ctx):
-                logger.warn(
+                logger.warning(
                     "Signature check failed for %s: %s",
                     pdu.event_id,
                     failure.getErrorMessage(),
@@ -278,9 +278,7 @@ def _check_sigs_on_pdus(keyring, room_version, pdus):
             pdu_to_check.sender_domain,
             e.getErrorMessage(),
         )
-        # XX not really sure if these are the right codes, but they are what
-        # we've done for ages
-        raise SynapseError(400, errmsg, Codes.UNAUTHORIZED)
+        raise SynapseError(403, errmsg, Codes.FORBIDDEN)
 
     for p, d in zip(pdus_to_check_sender, more_deferreds):
         d.addErrback(sender_err, p)
@@ -314,8 +312,7 @@ def _check_sigs_on_pdus(keyring, room_version, pdus):
                 "event id %s: unable to verify signature for event id domain: %s"
                 % (pdu_to_check.pdu.event_id, e.getErrorMessage())
             )
-            # XX as above: not really sure if these are the right codes
-            raise SynapseError(400, errmsg, Codes.UNAUTHORIZED)
+            raise SynapseError(403, errmsg, Codes.FORBIDDEN)
 
         for p, d in zip(pdus_to_check_event_id, more_deferreds):
             d.addErrback(event_err, p)
diff --git a/synapse/federation/federation_client.py b/synapse/federation/federation_client.py
index 6ee6216660..545d719652 100644
--- a/synapse/federation/federation_client.py
+++ b/synapse/federation/federation_client.py
@@ -196,7 +196,7 @@ class FederationClient(FederationBase):
             dest, room_id, extremities, limit
         )
 
-        logger.debug("backfill transaction_data=%s", repr(transaction_data))
+        logger.debug("backfill transaction_data=%r", transaction_data)
 
         room_version = yield self.store.get_room_version(room_id)
         format_ver = room_version_to_event_format(room_version)
@@ -522,12 +522,12 @@ class FederationClient(FederationBase):
                 res = yield callback(destination)
                 return res
             except InvalidResponseError as e:
-                logger.warn("Failed to %s via %s: %s", description, destination, e)
+                logger.warning("Failed to %s via %s: %s", description, destination, e)
             except HttpResponseException as e:
                 if not 500 <= e.code < 600:
                     raise e.to_synapse_error()
                 else:
-                    logger.warn(
+                    logger.warning(
                         "Failed to %s via %s: %i %s",
                         description,
                         destination,
@@ -535,7 +535,9 @@ class FederationClient(FederationBase):
                         e.args[0],
                     )
             except Exception:
-                logger.warn("Failed to %s via %s", description, destination, exc_info=1)
+                logger.warning(
+                    "Failed to %s via %s", description, destination, exc_info=1
+                )
 
         raise SynapseError(502, "Failed to %s via any server" % (description,))
 
@@ -553,7 +555,7 @@ class FederationClient(FederationBase):
         Note that this does not append any events to any graphs.
 
         Args:
-            destinations (str): Candidate homeservers which are probably
+            destinations (Iterable[str]): Candidate homeservers which are probably
                 participating in the room.
             room_id (str): The room in which the event will happen.
             user_id (str): The user whose membership is being evented.
@@ -878,44 +880,6 @@ class FederationClient(FederationBase):
             third_party_instance_id=third_party_instance_id,
         )
 
-    @defer.inlineCallbacks
-    def query_auth(self, destination, room_id, event_id, local_auth):
-        """
-        Params:
-            destination (str)
-            event_it (str)
-            local_auth (list)
-        """
-        time_now = self._clock.time_msec()
-
-        send_content = {"auth_chain": [e.get_pdu_json(time_now) for e in local_auth]}
-
-        code, content = yield self.transport_layer.send_query_auth(
-            destination=destination,
-            room_id=room_id,
-            event_id=event_id,
-            content=send_content,
-        )
-
-        room_version = yield self.store.get_room_version(room_id)
-        format_ver = room_version_to_event_format(room_version)
-
-        auth_chain = [event_from_pdu_json(e, format_ver) for e in content["auth_chain"]]
-
-        signed_auth = yield self._check_sigs_and_hash_and_fetch(
-            destination, auth_chain, outlier=True, room_version=room_version
-        )
-
-        signed_auth.sort(key=lambda e: e.depth)
-
-        ret = {
-            "auth_chain": signed_auth,
-            "rejects": content.get("rejects", []),
-            "missing": content.get("missing", []),
-        }
-
-        return ret
-
     @defer.inlineCallbacks
     def get_missing_events(
         self,
diff --git a/synapse/federation/federation_server.py b/synapse/federation/federation_server.py
index e5f0b90aec..d942d77a72 100644
--- a/synapse/federation/federation_server.py
+++ b/synapse/federation/federation_server.py
@@ -21,7 +21,6 @@ from six import iteritems
 from canonicaljson import json
 from prometheus_client import Counter
 
-from twisted.internet import defer
 from twisted.internet.abstract import isIPAddress
 from twisted.python import failure
 
@@ -36,7 +35,6 @@ from synapse.api.errors import (
     UnsupportedRoomVersionError,
 )
 from synapse.api.room_versions import KNOWN_ROOM_VERSIONS
-from synapse.crypto.event_signing import compute_event_signature
 from synapse.events import room_version_to_event_format
 from synapse.federation.federation_base import FederationBase, event_from_pdu_json
 from synapse.federation.persistence import TransactionActions
@@ -87,14 +85,12 @@ class FederationServer(FederationBase):
         # come in waves.
         self._state_resp_cache = ResponseCache(hs, "state_resp", timeout_ms=30000)
 
-    @defer.inlineCallbacks
-    @log_function
-    def on_backfill_request(self, origin, room_id, versions, limit):
-        with (yield self._server_linearizer.queue((origin, room_id))):
+    async def on_backfill_request(self, origin, room_id, versions, limit):
+        with (await self._server_linearizer.queue((origin, room_id))):
             origin_host, _ = parse_server_name(origin)
-            yield self.check_server_matches_acl(origin_host, room_id)
+            await self.check_server_matches_acl(origin_host, room_id)
 
-            pdus = yield self.handler.on_backfill_request(
+            pdus = await self.handler.on_backfill_request(
                 origin, room_id, versions, limit
             )
 
@@ -102,9 +98,7 @@ class FederationServer(FederationBase):
 
         return 200, res
 
-    @defer.inlineCallbacks
-    @log_function
-    def on_incoming_transaction(self, origin, transaction_data):
+    async def on_incoming_transaction(self, origin, transaction_data):
         # keep this as early as possible to make the calculated origin ts as
         # accurate as possible.
         request_time = self._clock.time_msec()
@@ -119,18 +113,17 @@ class FederationServer(FederationBase):
         # use a linearizer to ensure that we don't process the same transaction
         # multiple times in parallel.
         with (
-            yield self._transaction_linearizer.queue(
+            await self._transaction_linearizer.queue(
                 (origin, transaction.transaction_id)
             )
         ):
-            result = yield self._handle_incoming_transaction(
+            result = await self._handle_incoming_transaction(
                 origin, transaction, request_time
             )
 
         return result
 
-    @defer.inlineCallbacks
-    def _handle_incoming_transaction(self, origin, transaction, request_time):
+    async def _handle_incoming_transaction(self, origin, transaction, request_time):
         """ Process an incoming transaction and return the HTTP response
 
         Args:
@@ -141,7 +134,7 @@ class FederationServer(FederationBase):
         Returns:
             Deferred[(int, object)]: http response code and body
         """
-        response = yield self.transaction_actions.have_responded(origin, transaction)
+        response = await self.transaction_actions.have_responded(origin, transaction)
 
         if response:
             logger.debug(
@@ -152,7 +145,7 @@ class FederationServer(FederationBase):
 
         logger.debug("[%s] Transaction is new", transaction.transaction_id)
 
-        # Reject if PDU count > 50 and EDU count > 100
+        # Reject if PDU count > 50 or EDU count > 100
         if len(transaction.pdus) > 50 or (
             hasattr(transaction, "edus") and len(transaction.edus) > 100
         ):
@@ -160,7 +153,7 @@ class FederationServer(FederationBase):
             logger.info("Transaction PDU or EDU count too large. Returning 400")
 
             response = {}
-            yield self.transaction_actions.set_response(
+            await self.transaction_actions.set_response(
                 origin, transaction, 400, response
             )
             return 400, response
@@ -196,7 +189,7 @@ class FederationServer(FederationBase):
                 continue
 
             try:
-                room_version = yield self.store.get_room_version(room_id)
+                room_version = await self.store.get_room_version(room_id)
             except NotFoundError:
                 logger.info("Ignoring PDU for unknown room_id: %s", room_id)
                 continue
@@ -222,13 +215,12 @@ class FederationServer(FederationBase):
         # require callouts to other servers to fetch missing events), but
         # impose a limit to avoid going too crazy with ram/cpu.
 
-        @defer.inlineCallbacks
-        def process_pdus_for_room(room_id):
+        async def process_pdus_for_room(room_id):
             logger.debug("Processing PDUs for %s", room_id)
             try:
-                yield self.check_server_matches_acl(origin_host, room_id)
+                await self.check_server_matches_acl(origin_host, room_id)
             except AuthError as e:
-                logger.warn("Ignoring PDUs for room %s from banned server", room_id)
+                logger.warning("Ignoring PDUs for room %s from banned server", room_id)
                 for pdu in pdus_by_room[room_id]:
                     event_id = pdu.event_id
                     pdu_results[event_id] = e.error_dict()
@@ -238,10 +230,10 @@ class FederationServer(FederationBase):
                 event_id = pdu.event_id
                 with nested_logging_context(event_id):
                     try:
-                        yield self._handle_received_pdu(origin, pdu)
+                        await self._handle_received_pdu(origin, pdu)
                         pdu_results[event_id] = {}
                     except FederationError as e:
-                        logger.warn("Error handling PDU %s: %s", event_id, e)
+                        logger.warning("Error handling PDU %s: %s", event_id, e)
                         pdu_results[event_id] = {"error": str(e)}
                     except Exception as e:
                         f = failure.Failure()
@@ -252,36 +244,33 @@ class FederationServer(FederationBase):
                             exc_info=(f.type, f.value, f.getTracebackObject()),
                         )
 
-        yield concurrently_execute(
+        await concurrently_execute(
             process_pdus_for_room, pdus_by_room.keys(), TRANSACTION_CONCURRENCY_LIMIT
         )
 
         if hasattr(transaction, "edus"):
             for edu in (Edu(**x) for x in transaction.edus):
-                yield self.received_edu(origin, edu.edu_type, edu.content)
+                await self.received_edu(origin, edu.edu_type, edu.content)
 
         response = {"pdus": pdu_results}
 
         logger.debug("Returning: %s", str(response))
 
-        yield self.transaction_actions.set_response(origin, transaction, 200, response)
+        await self.transaction_actions.set_response(origin, transaction, 200, response)
         return 200, response
 
-    @defer.inlineCallbacks
-    def received_edu(self, origin, edu_type, content):
+    async def received_edu(self, origin, edu_type, content):
         received_edus_counter.inc()
-        yield self.registry.on_edu(edu_type, origin, content)
+        await self.registry.on_edu(edu_type, origin, content)
 
-    @defer.inlineCallbacks
-    @log_function
-    def on_context_state_request(self, origin, room_id, event_id):
+    async def on_context_state_request(self, origin, room_id, event_id):
         if not event_id:
             raise NotImplementedError("Specify an event")
 
         origin_host, _ = parse_server_name(origin)
-        yield self.check_server_matches_acl(origin_host, room_id)
+        await self.check_server_matches_acl(origin_host, room_id)
 
-        in_room = yield self.auth.check_host_in_room(room_id, origin)
+        in_room = await self.auth.check_host_in_room(room_id, origin)
         if not in_room:
             raise AuthError(403, "Host not in room.")
 
@@ -290,8 +279,8 @@ class FederationServer(FederationBase):
         # in the cache so we could return it without waiting for the linearizer
         # - but that's non-trivial to get right, and anyway somewhat defeats
         # the point of the linearizer.
-        with (yield self._server_linearizer.queue((origin, room_id))):
-            resp = yield self._state_resp_cache.wrap(
+        with (await self._server_linearizer.queue((origin, room_id))):
+            resp = await self._state_resp_cache.wrap(
                 (room_id, event_id),
                 self._on_context_state_request_compute,
                 room_id,
@@ -300,77 +289,60 @@ class FederationServer(FederationBase):
 
         return 200, resp
 
-    @defer.inlineCallbacks
-    def on_state_ids_request(self, origin, room_id, event_id):
+    async def on_state_ids_request(self, origin, room_id, event_id):
         if not event_id:
             raise NotImplementedError("Specify an event")
 
         origin_host, _ = parse_server_name(origin)
-        yield self.check_server_matches_acl(origin_host, room_id)
+        await self.check_server_matches_acl(origin_host, room_id)
 
-        in_room = yield self.auth.check_host_in_room(room_id, origin)
+        in_room = await self.auth.check_host_in_room(room_id, origin)
         if not in_room:
             raise AuthError(403, "Host not in room.")
 
-        state_ids = yield self.handler.get_state_ids_for_pdu(room_id, event_id)
-        auth_chain_ids = yield self.store.get_auth_chain_ids(state_ids)
+        state_ids = await self.handler.get_state_ids_for_pdu(room_id, event_id)
+        auth_chain_ids = await self.store.get_auth_chain_ids(state_ids)
 
         return 200, {"pdu_ids": state_ids, "auth_chain_ids": auth_chain_ids}
 
-    @defer.inlineCallbacks
-    def _on_context_state_request_compute(self, room_id, event_id):
-        pdus = yield self.handler.get_state_for_pdu(room_id, event_id)
-        auth_chain = yield self.store.get_auth_chain([pdu.event_id for pdu in pdus])
-
-        for event in auth_chain:
-            # We sign these again because there was a bug where we
-            # incorrectly signed things the first time round
-            if self.hs.is_mine_id(event.event_id):
-                event.signatures.update(
-                    compute_event_signature(
-                        event.get_pdu_json(),
-                        self.hs.hostname,
-                        self.hs.config.signing_key[0],
-                    )
-                )
+    async def _on_context_state_request_compute(self, room_id, event_id):
+        pdus = await self.handler.get_state_for_pdu(room_id, event_id)
+        auth_chain = await self.store.get_auth_chain([pdu.event_id for pdu in pdus])
 
         return {
             "pdus": [pdu.get_pdu_json() for pdu in pdus],
             "auth_chain": [pdu.get_pdu_json() for pdu in auth_chain],
         }
 
-    @defer.inlineCallbacks
-    @log_function
-    def on_pdu_request(self, origin, event_id):
-        pdu = yield self.handler.get_persisted_pdu(origin, event_id)
+    async def on_pdu_request(self, origin, event_id):
+        pdu = await self.handler.get_persisted_pdu(origin, event_id)
 
         if pdu:
             return 200, self._transaction_from_pdus([pdu]).get_dict()
         else:
             return 404, ""
 
-    @defer.inlineCallbacks
-    def on_query_request(self, query_type, args):
+    async def on_query_request(self, query_type, args):
         received_queries_counter.labels(query_type).inc()
-        resp = yield self.registry.on_query(query_type, args)
+        resp = await self.registry.on_query(query_type, args)
         return 200, resp
 
-    @defer.inlineCallbacks
-    def on_make_join_request(self, origin, room_id, user_id, supported_versions):
+    async def on_make_join_request(self, origin, room_id, user_id, supported_versions):
         origin_host, _ = parse_server_name(origin)
-        yield self.check_server_matches_acl(origin_host, room_id)
+        await self.check_server_matches_acl(origin_host, room_id)
 
-        room_version = yield self.store.get_room_version(room_id)
+        room_version = await self.store.get_room_version(room_id)
         if room_version not in supported_versions:
-            logger.warn("Room version %s not in %s", room_version, supported_versions)
+            logger.warning(
+                "Room version %s not in %s", room_version, supported_versions
+            )
             raise IncompatibleRoomVersionError(room_version=room_version)
 
-        pdu = yield self.handler.on_make_join_request(origin, room_id, user_id)
+        pdu = await self.handler.on_make_join_request(origin, room_id, user_id)
         time_now = self._clock.time_msec()
         return {"event": pdu.get_pdu_json(time_now), "room_version": room_version}
 
-    @defer.inlineCallbacks
-    def on_invite_request(self, origin, content, room_version):
+    async def on_invite_request(self, origin, content, room_version):
         if room_version not in KNOWN_ROOM_VERSIONS:
             raise SynapseError(
                 400,
@@ -382,24 +354,27 @@ class FederationServer(FederationBase):
 
         pdu = event_from_pdu_json(content, format_ver)
         origin_host, _ = parse_server_name(origin)
-        yield self.check_server_matches_acl(origin_host, pdu.room_id)
-        ret_pdu = yield self.handler.on_invite_request(origin, pdu)
+        await self.check_server_matches_acl(origin_host, pdu.room_id)
+        pdu = await self._check_sigs_and_hash(room_version, pdu)
+        ret_pdu = await self.handler.on_invite_request(origin, pdu)
         time_now = self._clock.time_msec()
         return {"event": ret_pdu.get_pdu_json(time_now)}
 
-    @defer.inlineCallbacks
-    def on_send_join_request(self, origin, content, room_id):
+    async def on_send_join_request(self, origin, content, room_id):
         logger.debug("on_send_join_request: content: %s", content)
 
-        room_version = yield self.store.get_room_version(room_id)
+        room_version = await self.store.get_room_version(room_id)
         format_ver = room_version_to_event_format(room_version)
         pdu = event_from_pdu_json(content, format_ver)
 
         origin_host, _ = parse_server_name(origin)
-        yield self.check_server_matches_acl(origin_host, pdu.room_id)
+        await self.check_server_matches_acl(origin_host, pdu.room_id)
 
         logger.debug("on_send_join_request: pdu sigs: %s", pdu.signatures)
-        res_pdus = yield self.handler.on_send_join_request(origin, pdu)
+
+        pdu = await self._check_sigs_and_hash(room_version, pdu)
+
+        res_pdus = await self.handler.on_send_join_request(origin, pdu)
         time_now = self._clock.time_msec()
         return (
             200,
@@ -411,45 +386,44 @@ class FederationServer(FederationBase):
             },
         )
 
-    @defer.inlineCallbacks
-    def on_make_leave_request(self, origin, room_id, user_id):
+    async def on_make_leave_request(self, origin, room_id, user_id):
         origin_host, _ = parse_server_name(origin)
-        yield self.check_server_matches_acl(origin_host, room_id)
-        pdu = yield self.handler.on_make_leave_request(origin, room_id, user_id)
+        await self.check_server_matches_acl(origin_host, room_id)
+        pdu = await self.handler.on_make_leave_request(origin, room_id, user_id)
 
-        room_version = yield self.store.get_room_version(room_id)
+        room_version = await self.store.get_room_version(room_id)
 
         time_now = self._clock.time_msec()
         return {"event": pdu.get_pdu_json(time_now), "room_version": room_version}
 
-    @defer.inlineCallbacks
-    def on_send_leave_request(self, origin, content, room_id):
+    async def on_send_leave_request(self, origin, content, room_id):
         logger.debug("on_send_leave_request: content: %s", content)
 
-        room_version = yield self.store.get_room_version(room_id)
+        room_version = await self.store.get_room_version(room_id)
         format_ver = room_version_to_event_format(room_version)
         pdu = event_from_pdu_json(content, format_ver)
 
         origin_host, _ = parse_server_name(origin)
-        yield self.check_server_matches_acl(origin_host, pdu.room_id)
+        await self.check_server_matches_acl(origin_host, pdu.room_id)
 
         logger.debug("on_send_leave_request: pdu sigs: %s", pdu.signatures)
-        yield self.handler.on_send_leave_request(origin, pdu)
+
+        pdu = await self._check_sigs_and_hash(room_version, pdu)
+
+        await self.handler.on_send_leave_request(origin, pdu)
         return 200, {}
 
-    @defer.inlineCallbacks
-    def on_event_auth(self, origin, room_id, event_id):
-        with (yield self._server_linearizer.queue((origin, room_id))):
+    async def on_event_auth(self, origin, room_id, event_id):
+        with (await self._server_linearizer.queue((origin, room_id))):
             origin_host, _ = parse_server_name(origin)
-            yield self.check_server_matches_acl(origin_host, room_id)
+            await self.check_server_matches_acl(origin_host, room_id)
 
             time_now = self._clock.time_msec()
-            auth_pdus = yield self.handler.on_event_auth(event_id)
+            auth_pdus = await self.handler.on_event_auth(event_id)
             res = {"auth_chain": [a.get_pdu_json(time_now) for a in auth_pdus]}
         return 200, res
 
-    @defer.inlineCallbacks
-    def on_query_auth_request(self, origin, content, room_id, event_id):
+    async def on_query_auth_request(self, origin, content, room_id, event_id):
         """
         Content is a dict with keys::
             auth_chain (list): A list of events that give the auth chain.
@@ -468,22 +442,22 @@ class FederationServer(FederationBase):
         Returns:
             Deferred: Results in `dict` with the same format as `content`
         """
-        with (yield self._server_linearizer.queue((origin, room_id))):
+        with (await self._server_linearizer.queue((origin, room_id))):
             origin_host, _ = parse_server_name(origin)
-            yield self.check_server_matches_acl(origin_host, room_id)
+            await self.check_server_matches_acl(origin_host, room_id)
 
-            room_version = yield self.store.get_room_version(room_id)
+            room_version = await self.store.get_room_version(room_id)
             format_ver = room_version_to_event_format(room_version)
 
             auth_chain = [
                 event_from_pdu_json(e, format_ver) for e in content["auth_chain"]
             ]
 
-            signed_auth = yield self._check_sigs_and_hash_and_fetch(
+            signed_auth = await self._check_sigs_and_hash_and_fetch(
                 origin, auth_chain, outlier=True, room_version=room_version
             )
 
-            ret = yield self.handler.on_query_auth(
+            ret = await self.handler.on_query_auth(
                 origin,
                 event_id,
                 room_id,
@@ -509,16 +483,14 @@ class FederationServer(FederationBase):
         return self.on_query_request("user_devices", user_id)
 
     @trace
-    @defer.inlineCallbacks
-    @log_function
-    def on_claim_client_keys(self, origin, content):
+    async def on_claim_client_keys(self, origin, content):
         query = []
         for user_id, device_keys in content.get("one_time_keys", {}).items():
             for device_id, algorithm in device_keys.items():
                 query.append((user_id, device_id, algorithm))
 
         log_kv({"message": "Claiming one time keys.", "user, device pairs": query})
-        results = yield self.store.claim_e2e_one_time_keys(query)
+        results = await self.store.claim_e2e_one_time_keys(query)
 
         json_result = {}
         for user_id, device_keys in results.items():
@@ -542,14 +514,12 @@ class FederationServer(FederationBase):
 
         return {"one_time_keys": json_result}
 
-    @defer.inlineCallbacks
-    @log_function
-    def on_get_missing_events(
+    async def on_get_missing_events(
         self, origin, room_id, earliest_events, latest_events, limit
     ):
-        with (yield self._server_linearizer.queue((origin, room_id))):
+        with (await self._server_linearizer.queue((origin, room_id))):
             origin_host, _ = parse_server_name(origin)
-            yield self.check_server_matches_acl(origin_host, room_id)
+            await self.check_server_matches_acl(origin_host, room_id)
 
             logger.info(
                 "on_get_missing_events: earliest_events: %r, latest_events: %r,"
@@ -559,7 +529,7 @@ class FederationServer(FederationBase):
                 limit,
             )
 
-            missing_events = yield self.handler.on_get_missing_events(
+            missing_events = await self.handler.on_get_missing_events(
                 origin, room_id, earliest_events, latest_events, limit
             )
 
@@ -592,8 +562,7 @@ class FederationServer(FederationBase):
             destination=None,
         )
 
-    @defer.inlineCallbacks
-    def _handle_received_pdu(self, origin, pdu):
+    async def _handle_received_pdu(self, origin, pdu):
         """ Process a PDU received in a federation /send/ transaction.
 
         If the event is invalid, then this method throws a FederationError.
@@ -646,37 +615,34 @@ class FederationServer(FederationBase):
                 logger.info("Accepting join PDU %s from %s", pdu.event_id, origin)
 
         # We've already checked that we know the room version by this point
-        room_version = yield self.store.get_room_version(pdu.room_id)
+        room_version = await self.store.get_room_version(pdu.room_id)
 
         # Check signature.
         try:
-            pdu = yield self._check_sigs_and_hash(room_version, pdu)
+            pdu = await self._check_sigs_and_hash(room_version, pdu)
         except SynapseError as e:
             raise FederationError("ERROR", e.code, e.msg, affected=pdu.event_id)
 
-        yield self.handler.on_receive_pdu(origin, pdu, sent_to_us_directly=True)
+        await self.handler.on_receive_pdu(origin, pdu, sent_to_us_directly=True)
 
     def __str__(self):
         return "" % self.server_name
 
-    @defer.inlineCallbacks
-    def exchange_third_party_invite(
+    async def exchange_third_party_invite(
         self, sender_user_id, target_user_id, room_id, signed
     ):
-        ret = yield self.handler.exchange_third_party_invite(
+        ret = await self.handler.exchange_third_party_invite(
             sender_user_id, target_user_id, room_id, signed
         )
         return ret
 
-    @defer.inlineCallbacks
-    def on_exchange_third_party_invite_request(self, origin, room_id, event_dict):
-        ret = yield self.handler.on_exchange_third_party_invite_request(
-            origin, room_id, event_dict
+    async def on_exchange_third_party_invite_request(self, room_id, event_dict):
+        ret = await self.handler.on_exchange_third_party_invite_request(
+            room_id, event_dict
         )
         return ret
 
-    @defer.inlineCallbacks
-    def check_server_matches_acl(self, server_name, room_id):
+    async def check_server_matches_acl(self, server_name, room_id):
         """Check if the given server is allowed by the server ACLs in the room
 
         Args:
@@ -686,13 +652,13 @@ class FederationServer(FederationBase):
         Raises:
             AuthError if the server does not match the ACL
         """
-        state_ids = yield self.store.get_current_state_ids(room_id)
+        state_ids = await self.store.get_current_state_ids(room_id)
         acl_event_id = state_ids.get((EventTypes.ServerACL, ""))
 
         if not acl_event_id:
             return
 
-        acl_event = yield self.store.get_event(acl_event_id)
+        acl_event = await self.store.get_event(acl_event_id)
         if server_matches_acl_event(server_name, acl_event):
             return
 
@@ -715,7 +681,7 @@ def server_matches_acl_event(server_name, acl_event):
     # server name is a literal IP
     allow_ip_literals = acl_event.content.get("allow_ip_literals", True)
     if not isinstance(allow_ip_literals, bool):
-        logger.warn("Ignorning non-bool allow_ip_literals flag")
+        logger.warning("Ignorning non-bool allow_ip_literals flag")
         allow_ip_literals = True
     if not allow_ip_literals:
         # check for ipv6 literals. These start with '['.
@@ -729,7 +695,7 @@ def server_matches_acl_event(server_name, acl_event):
     # next,  check the deny list
     deny = acl_event.content.get("deny", [])
     if not isinstance(deny, (list, tuple)):
-        logger.warn("Ignorning non-list deny ACL %s", deny)
+        logger.warning("Ignorning non-list deny ACL %s", deny)
         deny = []
     for e in deny:
         if _acl_entry_matches(server_name, e):
@@ -739,7 +705,7 @@ def server_matches_acl_event(server_name, acl_event):
     # then the allow list.
     allow = acl_event.content.get("allow", [])
     if not isinstance(allow, (list, tuple)):
-        logger.warn("Ignorning non-list allow ACL %s", allow)
+        logger.warning("Ignorning non-list allow ACL %s", allow)
         allow = []
     for e in allow:
         if _acl_entry_matches(server_name, e):
@@ -753,7 +719,7 @@ def server_matches_acl_event(server_name, acl_event):
 
 def _acl_entry_matches(server_name, acl_entry):
     if not isinstance(acl_entry, six.string_types):
-        logger.warn(
+        logger.warning(
             "Ignoring non-str ACL entry '%s' (is %s)", acl_entry, type(acl_entry)
         )
         return False
@@ -805,15 +771,14 @@ class FederationHandlerRegistry(object):
 
         self.query_handlers[query_type] = handler
 
-    @defer.inlineCallbacks
-    def on_edu(self, edu_type, origin, content):
+    async def on_edu(self, edu_type, origin, content):
         handler = self.edu_handlers.get(edu_type)
         if not handler:
-            logger.warn("No handler registered for EDU type %s", edu_type)
+            logger.warning("No handler registered for EDU type %s", edu_type)
 
         with start_active_span_from_edu(content, "handle_edu"):
             try:
-                yield handler(origin, content)
+                await handler(origin, content)
             except SynapseError as e:
                 logger.info("Failed to handle edu %r: %r", edu_type, e)
             except Exception:
@@ -822,7 +787,7 @@ class FederationHandlerRegistry(object):
     def on_query(self, query_type, args):
         handler = self.query_handlers.get(query_type)
         if not handler:
-            logger.warn("No handler registered for query type %s", query_type)
+            logger.warning("No handler registered for query type %s", query_type)
             raise NotFoundError("No handler for Query type '%s'" % (query_type,))
 
         return handler(args)
@@ -846,7 +811,7 @@ class ReplicationFederationHandlerRegistry(FederationHandlerRegistry):
 
         super(ReplicationFederationHandlerRegistry, self).__init__()
 
-    def on_edu(self, edu_type, origin, content):
+    async def on_edu(self, edu_type, origin, content):
         """Overrides FederationHandlerRegistry
         """
         if not self.config.use_presence and edu_type == "m.presence":
@@ -854,17 +819,17 @@ class ReplicationFederationHandlerRegistry(FederationHandlerRegistry):
 
         handler = self.edu_handlers.get(edu_type)
         if handler:
-            return super(ReplicationFederationHandlerRegistry, self).on_edu(
+            return await super(ReplicationFederationHandlerRegistry, self).on_edu(
                 edu_type, origin, content
             )
 
-        return self._send_edu(edu_type=edu_type, origin=origin, content=content)
+        return await self._send_edu(edu_type=edu_type, origin=origin, content=content)
 
-    def on_query(self, query_type, args):
+    async def on_query(self, query_type, args):
         """Overrides FederationHandlerRegistry
         """
         handler = self.query_handlers.get(query_type)
         if handler:
-            return handler(args)
+            return await handler(args)
 
-        return self._get_query_client(query_type=query_type, args=args)
+        return await self._get_query_client(query_type=query_type, args=args)
diff --git a/synapse/federation/send_queue.py b/synapse/federation/send_queue.py
index 454456a52d..ced4925a98 100644
--- a/synapse/federation/send_queue.py
+++ b/synapse/federation/send_queue.py
@@ -36,6 +36,8 @@ from six import iteritems
 
 from sortedcontainers import SortedDict
 
+from twisted.internet import defer
+
 from synapse.metrics import LaterGauge
 from synapse.storage.presence import UserPresenceState
 from synapse.util.metrics import Measure
@@ -212,7 +214,7 @@ class FederationRemoteSendQueue(object):
             receipt (synapse.types.ReadReceipt):
         """
         # nothing to do here: the replication listener will handle it.
-        pass
+        return defer.succeed(None)
 
     def send_presence(self, states):
         """As per FederationSender
diff --git a/synapse/federation/sender/__init__.py b/synapse/federation/sender/__init__.py
index d46f4aaeb1..2b2ee8612a 100644
--- a/synapse/federation/sender/__init__.py
+++ b/synapse/federation/sender/__init__.py
@@ -38,7 +38,7 @@ from synapse.metrics import (
     events_processed_counter,
 )
 from synapse.metrics.background_process_metrics import run_as_background_process
-from synapse.util.metrics import measure_func
+from synapse.util.metrics import Measure, measure_func
 
 logger = logging.getLogger(__name__)
 
@@ -183,8 +183,8 @@ class FederationSender(object):
                         # Otherwise if the last member on a server in a room is
                         # banned then it won't receive the event because it won't
                         # be in the room after the ban.
-                        destinations = yield self.state.get_current_hosts_in_room(
-                            event.room_id, latest_event_ids=event.prev_event_ids()
+                        destinations = yield self.state.get_hosts_in_room_at_events(
+                            event.room_id, event_ids=event.prev_event_ids()
                         )
                     except Exception:
                         logger.exception(
@@ -207,8 +207,9 @@ class FederationSender(object):
 
                 @defer.inlineCallbacks
                 def handle_room_events(events):
-                    for event in events:
-                        yield handle_event(event)
+                    with Measure(self.clock, "handle_room_events"):
+                        for event in events:
+                            yield handle_event(event)
 
                 events_by_room = {}
                 for event in events:
diff --git a/synapse/federation/sender/per_destination_queue.py b/synapse/federation/sender/per_destination_queue.py
index fad980b893..a5b36b1827 100644
--- a/synapse/federation/sender/per_destination_queue.py
+++ b/synapse/federation/sender/per_destination_queue.py
@@ -30,7 +30,7 @@ from synapse.federation.units import Edu
 from synapse.handlers.presence import format_user_presence_state
 from synapse.metrics import sent_transactions_counter
 from synapse.metrics.background_process_metrics import run_as_background_process
-from synapse.storage import UserPresenceState
+from synapse.storage.presence import UserPresenceState
 from synapse.util.retryutils import NotRetryingDestination, get_retry_limiter
 
 # This is defined in the Matrix spec and enforced by the receiver.
@@ -192,15 +192,16 @@ class PerDestinationQueue(object):
                 # We have to keep 2 free slots for presence and rr_edus
                 limit = MAX_EDUS_PER_TRANSACTION - 2
 
-                device_update_edus, dev_list_id = (
-                    yield self._get_device_update_edus(limit)
+                device_update_edus, dev_list_id = yield self._get_device_update_edus(
+                    limit
                 )
 
                 limit -= len(device_update_edus)
 
-                to_device_edus, device_stream_id = (
-                    yield self._get_to_device_message_edus(limit)
-                )
+                (
+                    to_device_edus,
+                    device_stream_id,
+                ) = yield self._get_to_device_message_edus(limit)
 
                 pending_edus = device_update_edus + to_device_edus
 
@@ -359,20 +360,20 @@ class PerDestinationQueue(object):
         last_device_list = self._last_device_list_stream_id
 
         # Retrieve list of new device updates to send to the destination
-        now_stream_id, results = yield self._store.get_devices_by_remote(
+        now_stream_id, results = yield self._store.get_device_updates_by_remote(
             self._destination, last_device_list, limit=limit
         )
         edus = [
             Edu(
                 origin=self._server_name,
                 destination=self._destination,
-                edu_type="m.device_list_update",
+                edu_type=edu_type,
                 content=content,
             )
-            for content in results
+            for (edu_type, content) in results
         ]
 
-        assert len(edus) <= limit, "get_devices_by_remote returned too many EDUs"
+        assert len(edus) <= limit, "get_device_updates_by_remote returned too many EDUs"
 
         return (edus, now_stream_id)
 
diff --git a/synapse/federation/sender/transaction_manager.py b/synapse/federation/sender/transaction_manager.py
index 62ca6a3e87..67b3e1ab6e 100644
--- a/synapse/federation/sender/transaction_manager.py
+++ b/synapse/federation/sender/transaction_manager.py
@@ -26,6 +26,7 @@ from synapse.logging.opentracing import (
     set_tag,
     start_active_span_follows_from,
     tags,
+    whitelisted_homeserver,
 )
 from synapse.util.metrics import measure_func
 
@@ -59,9 +60,15 @@ class TransactionManager(object):
         # The span_contexts is a generator so that it won't be evaluated if
         # opentracing is disabled. (Yay speed!)
 
-        span_contexts = (
-            extract_text_map(json.loads(edu.get_context())) for edu in pending_edus
-        )
+        span_contexts = []
+        keep_destination = whitelisted_homeserver(destination)
+
+        for edu in pending_edus:
+            context = edu.get_context()
+            if context:
+                span_contexts.append(extract_text_map(json.loads(context)))
+            if keep_destination:
+                edu.strip_context()
 
         with start_active_span_follows_from("send_transaction", span_contexts):
 
@@ -139,7 +146,7 @@ class TransactionManager(object):
             if code == 200:
                 for e_id, r in response.get("pdus", {}).items():
                     if "error" in r:
-                        logger.warn(
+                        logger.warning(
                             "TX [%s] {%s} Remote returned error for %s: %s",
                             destination,
                             txn_id,
@@ -148,7 +155,7 @@ class TransactionManager(object):
                         )
             else:
                 for p in pdus:
-                    logger.warn(
+                    logger.warning(
                         "TX [%s] {%s} Failed to send event %s",
                         destination,
                         txn_id,
diff --git a/synapse/federation/transport/client.py b/synapse/federation/transport/client.py
index 482a101c09..920fa86853 100644
--- a/synapse/federation/transport/client.py
+++ b/synapse/federation/transport/client.py
@@ -122,10 +122,10 @@ class TransportLayerClient(object):
             Deferred: Results in a dict received from the remote homeserver.
         """
         logger.debug(
-            "backfill dest=%s, room_id=%s, event_tuples=%s, limit=%s",
+            "backfill dest=%s, room_id=%s, event_tuples=%r, limit=%s",
             destination,
             room_id,
-            repr(event_tuples),
+            event_tuples,
             str(limit),
         )
 
@@ -381,17 +381,6 @@ class TransportLayerClient(object):
 
         return content
 
-    @defer.inlineCallbacks
-    @log_function
-    def send_query_auth(self, destination, room_id, event_id, content):
-        path = _create_v1_path("/query_auth/%s/%s", room_id, event_id)
-
-        content = yield self.client.post_json(
-            destination=destination, path=path, data=content
-        )
-
-        return content
-
     @defer.inlineCallbacks
     @log_function
     def query_client_keys(self, destination, query_content, timeout):
diff --git a/synapse/federation/transport/server.py b/synapse/federation/transport/server.py
index f9930b6460..d6c23f22bd 100644
--- a/synapse/federation/transport/server.py
+++ b/synapse/federation/transport/server.py
@@ -165,7 +165,7 @@ class Authenticator(object):
     async def _reset_retry_timings(self, origin):
         try:
             logger.info("Marking origin %r as up", origin)
-            await self.store.set_destination_retry_timings(origin, 0, 0)
+            await self.store.set_destination_retry_timings(origin, None, 0, 0)
         except Exception:
             logger.exception("Error resetting retry timings on %s", origin)
 
@@ -202,7 +202,7 @@ def _parse_auth_header(header_bytes):
         sig = strip_quotes(param_dict["sig"])
         return origin, key, sig
     except Exception as e:
-        logger.warn(
+        logger.warning(
             "Error parsing auth header '%s': %s",
             header_bytes.decode("ascii", "replace"),
             e,
@@ -287,10 +287,12 @@ class BaseFederationServlet(object):
             except NoAuthenticationError:
                 origin = None
                 if self.REQUIRE_AUTH:
-                    logger.warn("authenticate_request failed: missing authentication")
+                    logger.warning(
+                        "authenticate_request failed: missing authentication"
+                    )
                     raise
             except Exception as e:
-                logger.warn("authenticate_request failed: %s", e)
+                logger.warning("authenticate_request failed: %s", e)
                 raise
 
             request_tags = {
@@ -342,7 +344,11 @@ class BaseFederationServlet(object):
                 continue
 
             server.register_paths(
-                method, (pattern,), self._wrap(code), self.__class__.__name__
+                method,
+                (pattern,),
+                self._wrap(code),
+                self.__class__.__name__,
+                trace=False,
             )
 
 
@@ -571,7 +577,7 @@ class FederationThirdPartyInviteExchangeServlet(BaseFederationServlet):
 
     async def on_PUT(self, origin, content, query, room_id):
         content = await self.handler.on_exchange_third_party_invite_request(
-            origin, room_id, content
+            room_id, content
         )
         return 200, content
 
@@ -761,6 +767,10 @@ class PublicRoomList(BaseFederationServlet):
         else:
             network_tuple = ThirdPartyInstanceID(None, None)
 
+        if limit == 0:
+            # zero is a special value which corresponds to no limit.
+            limit = None
+
         data = await maybeDeferred(
             self.handler.get_local_public_room_list,
             limit,
@@ -796,6 +806,10 @@ class PublicRoomList(BaseFederationServlet):
         if search_filter is None:
             logger.warning("Nonefilter")
 
+        if limit == 0:
+            # zero is a special value which corresponds to no limit.
+            limit = None
+
         data = await self.handler.get_local_public_room_list(
             limit=limit,
             since_token=since_token,
diff --git a/synapse/federation/units.py b/synapse/federation/units.py
index aa84621206..b4d743cde7 100644
--- a/synapse/federation/units.py
+++ b/synapse/federation/units.py
@@ -41,6 +41,9 @@ class Edu(JsonEncodedObject):
     def get_context(self):
         return getattr(self, "content", {}).get("org.matrix.opentracing_context", "{}")
 
+    def strip_context(self):
+        getattr(self, "content", {})["org.matrix.opentracing_context"] = "{}"
+
 
 class Transaction(JsonEncodedObject):
     """ A transaction is a list of Pdus and Edus to be sent to a remote home
diff --git a/synapse/groups/attestations.py b/synapse/groups/attestations.py
index dfd7ae041b..d950a8b246 100644
--- a/synapse/groups/attestations.py
+++ b/synapse/groups/attestations.py
@@ -181,7 +181,7 @@ class GroupAttestionRenewer(object):
                 elif not self.is_mine_id(user_id):
                     destination = get_domain_from_id(user_id)
                 else:
-                    logger.warn(
+                    logger.warning(
                         "Incorrectly trying to do attestations for user: %r in %r",
                         user_id,
                         group_id,
diff --git a/synapse/groups/groups_server.py b/synapse/groups/groups_server.py
index d50e691436..29e8ffc295 100644
--- a/synapse/groups/groups_server.py
+++ b/synapse/groups/groups_server.py
@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
 # Copyright 2017 Vector Creations Ltd
 # Copyright 2018 New Vector Ltd
+# Copyright 2019 Michael Telatynski <7t3chguy@gmail.com>
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -20,16 +21,16 @@ from six import string_types
 
 from twisted.internet import defer
 
-from synapse.api.errors import SynapseError
+from synapse.api.errors import Codes, SynapseError
 from synapse.types import GroupID, RoomID, UserID, get_domain_from_id
 from synapse.util.async_helpers import concurrently_execute
 
 logger = logging.getLogger(__name__)
 
 
-# TODO: Allow users to "knock" or simpkly join depending on rules
+# TODO: Allow users to "knock" or simply join depending on rules
 # TODO: Federation admin APIs
-# TODO: is_priveged flag to users and is_public to users and rooms
+# TODO: is_privileged flag to users and is_public to users and rooms
 # TODO: Audit log for admins (profile updates, membership changes, users who tried
 #       to join but were rejected, etc)
 # TODO: Flairs
@@ -487,7 +488,7 @@ class GroupsServerHandler(object):
                 profile = yield self.profile_handler.get_profile_from_cache(user_id)
                 user_profile.update(profile)
             except Exception as e:
-                logger.warn("Error getting profile for %s: %s", user_id, e)
+                logger.warning("Error getting profile for %s: %s", user_id, e)
             user_profiles.append(user_profile)
 
         return {"chunk": user_profiles, "total_user_count_estimate": len(invited_users)}
@@ -590,7 +591,18 @@ class GroupsServerHandler(object):
         )
 
         # TODO: Check if user knocked
-        # TODO: Check if user is already invited
+
+        invited_users = yield self.store.get_invited_users_in_group(group_id)
+        if user_id in invited_users:
+            raise SynapseError(
+                400, "User already invited to group", errcode=Codes.BAD_STATE
+            )
+
+        user_results = yield self.store.get_users_in_group(
+            group_id, include_private=True
+        )
+        if user_id in [user_result["user_id"] for user_result in user_results]:
+            raise SynapseError(400, "User already in group")
 
         content = {
             "profile": {"name": group["name"], "avatar_url": group["avatar_url"]},
diff --git a/synapse/handlers/_base.py b/synapse/handlers/_base.py
index c29c78bd65..d15c6282fb 100644
--- a/synapse/handlers/_base.py
+++ b/synapse/handlers/_base.py
@@ -45,6 +45,7 @@ class BaseHandler(object):
         self.state_handler = hs.get_state_handler()
         self.distributor = hs.get_distributor()
         self.ratelimiter = hs.get_ratelimiter()
+        self.admin_redaction_ratelimiter = hs.get_admin_redaction_ratelimiter()
         self.clock = hs.get_clock()
         self.hs = hs
 
@@ -53,7 +54,7 @@ class BaseHandler(object):
         self.event_builder_factory = hs.get_event_builder_factory()
 
     @defer.inlineCallbacks
-    def ratelimit(self, requester, update=True):
+    def ratelimit(self, requester, update=True, is_admin_redaction=False):
         """Ratelimits requests.
 
         Args:
@@ -62,6 +63,9 @@ class BaseHandler(object):
                 Set to False when doing multiple checks for one request (e.g.
                 to check up front if we would reject the request), and set to
                 True for the last call for a given request.
+            is_admin_redaction (bool): Whether this is a room admin/moderator
+                redacting an event. If so then we may apply different
+                ratelimits depending on config.
 
         Raises:
             LimitExceededError if the request should be ratelimited
@@ -90,16 +94,33 @@ class BaseHandler(object):
             messages_per_second = override.messages_per_second
             burst_count = override.burst_count
         else:
-            messages_per_second = self.hs.config.rc_message.per_second
-            burst_count = self.hs.config.rc_message.burst_count
+            # We default to different values if this is an admin redaction and
+            # the config is set
+            if is_admin_redaction and self.hs.config.rc_admin_redaction:
+                messages_per_second = self.hs.config.rc_admin_redaction.per_second
+                burst_count = self.hs.config.rc_admin_redaction.burst_count
+            else:
+                messages_per_second = self.hs.config.rc_message.per_second
+                burst_count = self.hs.config.rc_message.burst_count
 
-        allowed, time_allowed = self.ratelimiter.can_do_action(
-            user_id,
-            time_now,
-            rate_hz=messages_per_second,
-            burst_count=burst_count,
-            update=update,
-        )
+        if is_admin_redaction and self.hs.config.rc_admin_redaction:
+            # If we have separate config for admin redactions we use a separate
+            # ratelimiter
+            allowed, time_allowed = self.admin_redaction_ratelimiter.can_do_action(
+                user_id,
+                time_now,
+                rate_hz=messages_per_second,
+                burst_count=burst_count,
+                update=update,
+            )
+        else:
+            allowed, time_allowed = self.ratelimiter.can_do_action(
+                user_id,
+                time_now,
+                rate_hz=messages_per_second,
+                burst_count=burst_count,
+                update=update,
+            )
         if not allowed:
             raise LimitExceededError(
                 retry_after_ms=int(1000 * (time_allowed - time_now))
diff --git a/synapse/handlers/account_data.py b/synapse/handlers/account_data.py
index 38bc67191c..2d7e6df6e4 100644
--- a/synapse/handlers/account_data.py
+++ b/synapse/handlers/account_data.py
@@ -38,9 +38,10 @@ class AccountDataEventSource(object):
                 {"type": "m.tag", "content": {"tags": room_tags}, "room_id": room_id}
             )
 
-        account_data, room_account_data = (
-            yield self.store.get_updated_account_data_for_user(user_id, last_stream_id)
-        )
+        (
+            account_data,
+            room_account_data,
+        ) = yield self.store.get_updated_account_data_for_user(user_id, last_stream_id)
 
         for account_data_type, content in account_data.items():
             results.append({"type": account_data_type, "content": content})
diff --git a/synapse/handlers/account_validity.py b/synapse/handlers/account_validity.py
index 34574f1a12..d04e0fe576 100644
--- a/synapse/handlers/account_validity.py
+++ b/synapse/handlers/account_validity.py
@@ -38,6 +38,7 @@ logger = logging.getLogger(__name__)
 class AccountValidityHandler(object):
     def __init__(self, hs):
         self.hs = hs
+        self.config = hs.config
         self.store = self.hs.get_datastore()
         self.sendmail = self.hs.get_sendmail()
         self.clock = self.hs.get_clock()
@@ -62,9 +63,14 @@ class AccountValidityHandler(object):
             self._raw_from = email.utils.parseaddr(self._from_string)[1]
 
             self._template_html, self._template_text = load_jinja2_templates(
-                config=self.hs.config,
-                template_html_name=self.hs.config.email_expiry_template_html,
-                template_text_name=self.hs.config.email_expiry_template_text,
+                self.config.email_template_dir,
+                [
+                    self.config.email_expiry_template_html,
+                    self.config.email_expiry_template_text,
+                ],
+                apply_format_ts_filter=True,
+                apply_mxc_to_http_filter=True,
+                public_baseurl=self.config.public_baseurl,
             )
 
             # Check the renewal emails to send and send them every 30min.
diff --git a/synapse/handlers/admin.py b/synapse/handlers/admin.py
index 1a87b58838..6407d56f8e 100644
--- a/synapse/handlers/admin.py
+++ b/synapse/handlers/admin.py
@@ -30,6 +30,9 @@ class AdminHandler(BaseHandler):
     def __init__(self, hs):
         super(AdminHandler, self).__init__(hs)
 
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
+
     @defer.inlineCallbacks
     def get_whois(self, user):
         connections = []
@@ -205,7 +208,7 @@ class AdminHandler(BaseHandler):
 
                 from_key = events[-1].internal_metadata.after
 
-                events = yield filter_events_for_client(self.store, user_id, events)
+                events = yield filter_events_for_client(self.storage, user_id, events)
 
                 writer.write_events(room_id, events)
 
@@ -241,7 +244,7 @@ class AdminHandler(BaseHandler):
             for event_id in extremities:
                 if not event_to_unseen_prevs[event_id]:
                     continue
-                state = yield self.store.get_state_for_event(event_id)
+                state = yield self.state_store.get_state_for_event(event_id)
                 writer.write_state(room_id, event_id, state)
 
         return writer.finished()
diff --git a/synapse/handlers/appservice.py b/synapse/handlers/appservice.py
index d1a51df6f9..fe62f78e67 100644
--- a/synapse/handlers/appservice.py
+++ b/synapse/handlers/appservice.py
@@ -73,7 +73,10 @@ class ApplicationServicesHandler(object):
             try:
                 limit = 100
                 while True:
-                    upper_bound, events = yield self.store.get_new_events_for_appservice(
+                    (
+                        upper_bound,
+                        events,
+                    ) = yield self.store.get_new_events_for_appservice(
                         self.current_max, limit
                     )
 
@@ -294,12 +297,10 @@ class ApplicationServicesHandler(object):
             # we don't know if they are unknown or not since it isn't one of our
             # users. We can't poke ASes.
             return False
-            return
 
         user_info = yield self.store.get_user_by_id(user_id)
         if user_info:
             return False
-            return
 
         # user not found; could be the AS though, so check.
         services = self.store.get_app_services()
diff --git a/synapse/handlers/auth.py b/synapse/handlers/auth.py
index f844409d21..7a0f54ca24 100644
--- a/synapse/handlers/auth.py
+++ b/synapse/handlers/auth.py
@@ -21,10 +21,8 @@ import unicodedata
 import attr
 import bcrypt
 import pymacaroons
-from canonicaljson import json
 
 from twisted.internet import defer
-from twisted.web.client import PartialDownloadError
 
 import synapse.util.stringutils as stringutils
 from synapse.api.constants import LoginType
@@ -38,6 +36,8 @@ from synapse.api.errors import (
     UserDeactivatedError,
 )
 from synapse.api.ratelimiting import Ratelimiter
+from synapse.handlers.ui_auth import INTERACTIVE_AUTH_CHECKERS
+from synapse.handlers.ui_auth.checkers import UserInteractiveAuthChecker
 from synapse.logging.context import defer_to_thread
 from synapse.module_api import ModuleApi
 from synapse.types import UserID
@@ -57,13 +57,13 @@ class AuthHandler(BaseHandler):
             hs (synapse.server.HomeServer):
         """
         super(AuthHandler, self).__init__(hs)
-        self.checkers = {
-            LoginType.RECAPTCHA: self._check_recaptcha,
-            LoginType.EMAIL_IDENTITY: self._check_email_identity,
-            LoginType.MSISDN: self._check_msisdn,
-            LoginType.DUMMY: self._check_dummy_auth,
-            LoginType.TERMS: self._check_terms_auth,
-        }
+
+        self.checkers = {}  # type: dict[str, UserInteractiveAuthChecker]
+        for auth_checker_class in INTERACTIVE_AUTH_CHECKERS:
+            inst = auth_checker_class(hs)
+            if inst.is_enabled():
+                self.checkers[inst.AUTH_TYPE] = inst
+
         self.bcrypt_rounds = hs.config.bcrypt_rounds
 
         # This is not a cache per se, but a store of all current sessions that
@@ -157,8 +157,16 @@ class AuthHandler(BaseHandler):
 
         return params
 
+    def get_enabled_auth_types(self):
+        """Return the enabled user-interactive authentication types
+
+        Returns the UI-Auth types which are supported by the homeserver's current
+        config.
+        """
+        return self.checkers.keys()
+
     @defer.inlineCallbacks
-    def check_auth(self, flows, clientdict, clientip, password_servlet=False):
+    def check_auth(self, flows, clientdict, clientip):
         """
         Takes a dictionary sent by the client in the login / registration
         protocol and handles the User-Interactive Auth flow.
@@ -182,16 +190,6 @@ class AuthHandler(BaseHandler):
 
             clientip (str): The IP address of the client.
 
-            password_servlet (bool): Whether the request originated from
-                PasswordRestServlet.
-                XXX: This is a temporary hack to distinguish between checking
-                for threepid validations locally (in the case of password
-                resets) and using the identity server (in the case of binding
-                a 3PID during registration). Once we start using the
-                homeserver for both tasks, this distinction will no longer be
-                necessary.
-
-
         Returns:
             defer.Deferred[dict, dict, str]: a deferred tuple of
                 (creds, params, session_id).
@@ -247,9 +245,7 @@ class AuthHandler(BaseHandler):
         if "type" in authdict:
             login_type = authdict["type"]
             try:
-                result = yield self._check_auth_dict(
-                    authdict, clientip, password_servlet=password_servlet
-                )
+                result = yield self._check_auth_dict(authdict, clientip)
                 if result:
                     creds[login_type] = result
                     self._save_session(session)
@@ -303,7 +299,7 @@ class AuthHandler(BaseHandler):
             sess["creds"] = {}
         creds = sess["creds"]
 
-        result = yield self.checkers[stagetype](authdict, clientip)
+        result = yield self.checkers[stagetype].check_auth(authdict, clientip)
         if result:
             creds[stagetype] = result
             self._save_session(sess)
@@ -356,7 +352,7 @@ class AuthHandler(BaseHandler):
         return sess.setdefault("serverdict", {}).get(key, default)
 
     @defer.inlineCallbacks
-    def _check_auth_dict(self, authdict, clientip, password_servlet=False):
+    def _check_auth_dict(self, authdict, clientip):
         """Attempt to validate the auth dict provided by a client
 
         Args:
@@ -374,11 +370,7 @@ class AuthHandler(BaseHandler):
         login_type = authdict["type"]
         checker = self.checkers.get(login_type)
         if checker is not None:
-            # XXX: Temporary workaround for having Synapse handle password resets
-            # See AuthHandler.check_auth for further details
-            res = yield checker(
-                authdict, clientip=clientip, password_servlet=password_servlet
-            )
+            res = yield checker.check_auth(authdict, clientip=clientip)
             return res
 
         # build a v1-login-style dict out of the authdict and fall back to the
@@ -391,119 +383,6 @@ class AuthHandler(BaseHandler):
         (canonical_id, callback) = yield self.validate_login(user_id, authdict)
         return canonical_id
 
-    @defer.inlineCallbacks
-    def _check_recaptcha(self, authdict, clientip, **kwargs):
-        try:
-            user_response = authdict["response"]
-        except KeyError:
-            # Client tried to provide captcha but didn't give the parameter:
-            # bad request.
-            raise LoginError(
-                400, "Captcha response is required", errcode=Codes.CAPTCHA_NEEDED
-            )
-
-        logger.info(
-            "Submitting recaptcha response %s with remoteip %s", user_response, clientip
-        )
-
-        # TODO: get this from the homeserver rather than creating a new one for
-        # each request
-        try:
-            client = self.hs.get_simple_http_client()
-            resp_body = yield client.post_urlencoded_get_json(
-                self.hs.config.recaptcha_siteverify_api,
-                args={
-                    "secret": self.hs.config.recaptcha_private_key,
-                    "response": user_response,
-                    "remoteip": clientip,
-                },
-            )
-        except PartialDownloadError as pde:
-            # Twisted is silly
-            data = pde.response
-            resp_body = json.loads(data)
-
-        if "success" in resp_body:
-            # Note that we do NOT check the hostname here: we explicitly
-            # intend the CAPTCHA to be presented by whatever client the
-            # user is using, we just care that they have completed a CAPTCHA.
-            logger.info(
-                "%s reCAPTCHA from hostname %s",
-                "Successful" if resp_body["success"] else "Failed",
-                resp_body.get("hostname"),
-            )
-            if resp_body["success"]:
-                return True
-        raise LoginError(401, "", errcode=Codes.UNAUTHORIZED)
-
-    def _check_email_identity(self, authdict, **kwargs):
-        return self._check_threepid("email", authdict, **kwargs)
-
-    def _check_msisdn(self, authdict, **kwargs):
-        return self._check_threepid("msisdn", authdict)
-
-    def _check_dummy_auth(self, authdict, **kwargs):
-        return defer.succeed(True)
-
-    def _check_terms_auth(self, authdict, **kwargs):
-        return defer.succeed(True)
-
-    @defer.inlineCallbacks
-    def _check_threepid(self, medium, authdict, password_servlet=False, **kwargs):
-        if "threepid_creds" not in authdict:
-            raise LoginError(400, "Missing threepid_creds", Codes.MISSING_PARAM)
-
-        threepid_creds = authdict["threepid_creds"]
-
-        identity_handler = self.hs.get_handlers().identity_handler
-
-        logger.info("Getting validated threepid. threepidcreds: %r", (threepid_creds,))
-        if (
-            not password_servlet
-            or self.hs.config.email_password_reset_behaviour == "remote"
-        ):
-            threepid = yield identity_handler.threepid_from_creds(threepid_creds)
-        elif self.hs.config.email_password_reset_behaviour == "local":
-            row = yield self.store.get_threepid_validation_session(
-                medium,
-                threepid_creds["client_secret"],
-                sid=threepid_creds["sid"],
-                validated=True,
-            )
-
-            threepid = (
-                {
-                    "medium": row["medium"],
-                    "address": row["address"],
-                    "validated_at": row["validated_at"],
-                }
-                if row
-                else None
-            )
-
-            if row:
-                # Valid threepid returned, delete from the db
-                yield self.store.delete_threepid_session(threepid_creds["sid"])
-        else:
-            raise SynapseError(
-                400, "Password resets are not enabled on this homeserver"
-            )
-
-        if not threepid:
-            raise LoginError(401, "", errcode=Codes.UNAUTHORIZED)
-
-        if threepid["medium"] != medium:
-            raise LoginError(
-                401,
-                "Expecting threepid of type '%s', got '%s'"
-                % (medium, threepid["medium"]),
-                errcode=Codes.UNAUTHORIZED,
-            )
-
-        threepid["threepid_creds"] = authdict["threepid_creds"]
-
-        return threepid
-
     def _get_params_recaptcha(self):
         return {"public_key": self.hs.config.recaptcha_public_key}
 
@@ -646,7 +525,7 @@ class AuthHandler(BaseHandler):
 
         result = None
         if not user_infos:
-            logger.warn("Attempted to login as %s but they do not exist", user_id)
+            logger.warning("Attempted to login as %s but they do not exist", user_id)
         elif len(user_infos) == 1:
             # a single match (possibly not exact)
             result = user_infos.popitem()
@@ -655,7 +534,7 @@ class AuthHandler(BaseHandler):
             result = (user_id, user_infos[user_id])
         else:
             # multiple matches, none of them exact
-            logger.warn(
+            logger.warning(
                 "Attempted to login as %s but it matches more than one user "
                 "inexactly: %r",
                 user_id,
@@ -849,7 +728,7 @@ class AuthHandler(BaseHandler):
 
         result = yield self.validate_hash(password, password_hash)
         if not result:
-            logger.warn("Failed password login for user %s", user_id)
+            logger.warning("Failed password login for user %s", user_id)
             return None
         return user_id
 
diff --git a/synapse/handlers/deactivate_account.py b/synapse/handlers/deactivate_account.py
index 5f804d1f13..63267a0a4c 100644
--- a/synapse/handlers/deactivate_account.py
+++ b/synapse/handlers/deactivate_account.py
@@ -73,7 +73,9 @@ class DeactivateAccountHandler(BaseHandler):
         # unbinding
         identity_server_supports_unbinding = True
 
-        threepids = yield self.store.user_get_threepids(user_id)
+        # Retrieve the 3PIDs this user has bound to an identity server
+        threepids = yield self.store.user_get_bound_threepids(user_id)
+
         for threepid in threepids:
             try:
                 result = yield self._identity_handler.try_unbind_threepid(
@@ -118,6 +120,10 @@ class DeactivateAccountHandler(BaseHandler):
         # parts users from rooms (if it isn't already running)
         self._start_user_parting()
 
+        # Reject all pending invites for the user, so that the user doesn't show up in the
+        # "invited" section of rooms' members list.
+        yield self._reject_pending_invites_for_user(user_id)
+
         # Remove all information on the user from the account_validity table.
         if self._account_validity_enabled:
             yield self.store.delete_account_validity_for_user(user_id)
@@ -127,6 +133,39 @@ class DeactivateAccountHandler(BaseHandler):
 
         return identity_server_supports_unbinding
 
+    @defer.inlineCallbacks
+    def _reject_pending_invites_for_user(self, user_id):
+        """Reject pending invites addressed to a given user ID.
+
+        Args:
+            user_id (str): The user ID to reject pending invites for.
+        """
+        user = UserID.from_string(user_id)
+        pending_invites = yield self.store.get_invited_rooms_for_user(user_id)
+
+        for room in pending_invites:
+            try:
+                yield self._room_member_handler.update_membership(
+                    create_requester(user),
+                    user,
+                    room.room_id,
+                    "leave",
+                    ratelimit=False,
+                    require_consent=False,
+                )
+                logger.info(
+                    "Rejected invite for deactivated user %r in room %r",
+                    user_id,
+                    room.room_id,
+                )
+            except Exception:
+                logger.exception(
+                    "Failed to reject invite for user %r in room %r:"
+                    " ignoring and continuing",
+                    user_id,
+                    room.room_id,
+                )
+
     def _start_user_parting(self):
         """
         Start the process that goes through the table of users
diff --git a/synapse/handlers/device.py b/synapse/handlers/device.py
index 5c1cf83c9d..26ef5e150c 100644
--- a/synapse/handlers/device.py
+++ b/synapse/handlers/device.py
@@ -1,5 +1,7 @@
 # -*- coding: utf-8 -*-
 # Copyright 2016 OpenMarket Ltd
+# Copyright 2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -25,6 +27,7 @@ from synapse.api.errors import (
     HttpResponseException,
     RequestSendFailed,
 )
+from synapse.logging.opentracing import log_kv, set_tag, trace
 from synapse.types import RoomStreamToken, get_domain_from_id
 from synapse.util import stringutils
 from synapse.util.async_helpers import Linearizer
@@ -43,8 +46,10 @@ class DeviceWorkerHandler(BaseHandler):
 
         self.hs = hs
         self.state = hs.get_state_handler()
+        self.state_store = hs.get_storage().state
         self._auth_handler = hs.get_auth_handler()
 
+    @trace
     @defer.inlineCallbacks
     def get_devices_by_user(self, user_id):
         """
@@ -56,6 +61,7 @@ class DeviceWorkerHandler(BaseHandler):
             defer.Deferred: list[dict[str, X]]: info on each device
         """
 
+        set_tag("user_id", user_id)
         device_map = yield self.store.get_devices_by_user(user_id)
 
         ips = yield self.store.get_last_client_ip_by_device(user_id, device_id=None)
@@ -64,8 +70,10 @@ class DeviceWorkerHandler(BaseHandler):
         for device in devices:
             _update_device_from_client_ips(device, ips)
 
+        log_kv(device_map)
         return devices
 
+    @trace
     @defer.inlineCallbacks
     def get_device(self, user_id, device_id):
         """ Retrieve the given device
@@ -85,9 +93,14 @@ class DeviceWorkerHandler(BaseHandler):
             raise errors.NotFoundError
         ips = yield self.store.get_last_client_ip_by_device(user_id, device_id)
         _update_device_from_client_ips(device, ips)
+
+        set_tag("device", device)
+        set_tag("ips", ips)
+
         return device
 
     @measure_func("device.get_user_ids_changed")
+    @trace
     @defer.inlineCallbacks
     def get_user_ids_changed(self, user_id, from_token):
         """Get list of users that have had the devices updated, or have newly
@@ -97,6 +110,9 @@ class DeviceWorkerHandler(BaseHandler):
             user_id (str)
             from_token (StreamToken)
         """
+
+        set_tag("user_id", user_id)
+        set_tag("from_token", from_token)
         now_room_key = yield self.store.get_room_events_max_id()
 
         room_ids = yield self.store.get_rooms_for_user(user_id)
@@ -148,6 +164,9 @@ class DeviceWorkerHandler(BaseHandler):
             # special-case for an empty prev state: include all members
             # in the changed list
             if not event_ids:
+                log_kv(
+                    {"event": "encountered empty previous state", "room_id": room_id}
+                )
                 for key, event_id in iteritems(current_state_ids):
                     etype, state_key = key
                     if etype != EventTypes.Member:
@@ -160,7 +179,7 @@ class DeviceWorkerHandler(BaseHandler):
                 continue
 
             # mapping from event_id -> state_dict
-            prev_state_ids = yield self.store.get_state_ids_for_events(event_ids)
+            prev_state_ids = yield self.state_store.get_state_ids_for_events(event_ids)
 
             # Check if we've joined the room? If so we just blindly add all the users to
             # the "possibly changed" users.
@@ -200,7 +219,11 @@ class DeviceWorkerHandler(BaseHandler):
             possibly_joined = []
             possibly_left = []
 
-        return {"changed": list(possibly_joined), "left": list(possibly_left)}
+        result = {"changed": list(possibly_joined), "left": list(possibly_left)}
+
+        log_kv(result)
+
+        return result
 
 
 class DeviceHandler(DeviceWorkerHandler):
@@ -267,6 +290,7 @@ class DeviceHandler(DeviceWorkerHandler):
 
         raise errors.StoreError(500, "Couldn't generate a device ID.")
 
+    @trace
     @defer.inlineCallbacks
     def delete_device(self, user_id, device_id):
         """ Delete the given device
@@ -284,6 +308,10 @@ class DeviceHandler(DeviceWorkerHandler):
         except errors.StoreError as e:
             if e.code == 404:
                 # no match
+                set_tag("error", True)
+                log_kv(
+                    {"reason": "User doesn't have device id.", "device_id": device_id}
+                )
                 pass
             else:
                 raise
@@ -296,6 +324,7 @@ class DeviceHandler(DeviceWorkerHandler):
 
         yield self.notify_device_update(user_id, [device_id])
 
+    @trace
     @defer.inlineCallbacks
     def delete_all_devices_for_user(self, user_id, except_device_id=None):
         """Delete all of the user's devices
@@ -331,6 +360,8 @@ class DeviceHandler(DeviceWorkerHandler):
         except errors.StoreError as e:
             if e.code == 404:
                 # no match
+                set_tag("error", True)
+                set_tag("reason", "User doesn't have that device id.")
                 pass
             else:
                 raise
@@ -371,6 +402,7 @@ class DeviceHandler(DeviceWorkerHandler):
             else:
                 raise
 
+    @trace
     @measure_func("notify_device_update")
     @defer.inlineCallbacks
     def notify_device_update(self, user_id, device_ids):
@@ -386,6 +418,8 @@ class DeviceHandler(DeviceWorkerHandler):
             hosts.update(get_domain_from_id(u) for u in users_who_share_room)
             hosts.discard(self.server_name)
 
+        set_tag("target_hosts", hosts)
+
         position = yield self.store.add_device_change_to_streams(
             user_id, device_ids, list(hosts)
         )
@@ -405,11 +439,38 @@ class DeviceHandler(DeviceWorkerHandler):
             )
             for host in hosts:
                 self.federation_sender.send_device_messages(host)
+                log_kv({"message": "sent device update to host", "host": host})
+
+    @defer.inlineCallbacks
+    def notify_user_signature_update(self, from_user_id, user_ids):
+        """Notify a user that they have made new signatures of other users.
+
+        Args:
+            from_user_id (str): the user who made the signature
+            user_ids (list[str]): the users IDs that have new signatures
+        """
+
+        position = yield self.store.add_user_signature_change_to_streams(
+            from_user_id, user_ids
+        )
+
+        self.notifier.on_new_event("device_list_key", position, users=[from_user_id])
 
     @defer.inlineCallbacks
     def on_federation_query_user_devices(self, user_id):
         stream_id, devices = yield self.store.get_devices_with_keys_by_user(user_id)
-        return {"user_id": user_id, "stream_id": stream_id, "devices": devices}
+        master_key = yield self.store.get_e2e_cross_signing_key(user_id, "master")
+        self_signing_key = yield self.store.get_e2e_cross_signing_key(
+            user_id, "self_signing"
+        )
+
+        return {
+            "user_id": user_id,
+            "stream_id": stream_id,
+            "devices": devices,
+            "master_key": master_key,
+            "self_signing_key": self_signing_key,
+        }
 
     @defer.inlineCallbacks
     def user_left_room(self, user, room_id):
@@ -451,12 +512,15 @@ class DeviceListUpdater(object):
             iterable=True,
         )
 
+    @trace
     @defer.inlineCallbacks
     def incoming_device_list_update(self, origin, edu_content):
         """Called on incoming device list update from federation. Responsible
         for parsing the EDU and adding to pending updates list.
         """
 
+        set_tag("origin", origin)
+        set_tag("edu_content", edu_content)
         user_id = edu_content.pop("user_id")
         device_id = edu_content.pop("device_id")
         stream_id = str(edu_content.pop("stream_id"))  # They may come as ints
@@ -471,12 +535,30 @@ class DeviceListUpdater(object):
                 device_id,
                 origin,
             )
+
+            set_tag("error", True)
+            log_kv(
+                {
+                    "message": "Got a device list update edu from a user and "
+                    "device which does not match the origin of the request.",
+                    "user_id": user_id,
+                    "device_id": device_id,
+                }
+            )
             return
 
         room_ids = yield self.store.get_rooms_for_user(user_id)
         if not room_ids:
             # We don't share any rooms with this user. Ignore update, as we
             # probably won't get any further updates.
+            set_tag("error", True)
+            log_kv(
+                {
+                    "message": "Got an update from a user for which "
+                    "we don't share any rooms",
+                    "other user_id": user_id,
+                }
+            )
             logger.warning(
                 "Got device list update edu for %r/%r, but don't share a room",
                 user_id,
@@ -578,6 +660,7 @@ class DeviceListUpdater(object):
             request:
             https://matrix.org/docs/spec/server_server/r0.1.2#get-matrix-federation-v1-user-devices-userid
         """
+        log_kv({"message": "Doing resync to update device list."})
         # Fetch all devices for the user.
         origin = get_domain_from_id(user_id)
         try:
@@ -585,7 +668,7 @@ class DeviceListUpdater(object):
         except (NotRetryingDestination, RequestSendFailed, HttpResponseException):
             # TODO: Remember that we are now out of sync and try again
             # later
-            logger.warn("Failed to handle device list update for %s", user_id)
+            logger.warning("Failed to handle device list update for %s", user_id)
             # We abort on exceptions rather than accepting the update
             # as otherwise synapse will 'forget' that its device list
             # is out of date. If we bail then we will retry the resync
@@ -594,13 +677,20 @@ class DeviceListUpdater(object):
             # eventually become consistent.
             return
         except FederationDeniedError as e:
+            set_tag("error", True)
+            log_kv({"reason": "FederationDeniedError"})
             logger.info(e)
             return
-        except Exception:
+        except Exception as e:
             # TODO: Remember that we are now out of sync and try again
             # later
+            set_tag("error", True)
+            log_kv(
+                {"message": "Exception raised by federation request", "exception": e}
+            )
             logger.exception("Failed to handle device list update for %s", user_id)
             return
+        log_kv({"result": result})
         stream_id = result["stream_id"]
         devices = result["devices"]
 
@@ -616,7 +706,7 @@ class DeviceListUpdater(object):
         # up on storing the total list of devices and only handle the
         # delta instead.
         if len(devices) > 1000:
-            logger.warn(
+            logger.warning(
                 "Ignoring device list snapshot for %s as it has >1K devs (%d)",
                 user_id,
                 len(devices),
diff --git a/synapse/handlers/devicemessage.py b/synapse/handlers/devicemessage.py
index c7d56779b8..73b9e120f5 100644
--- a/synapse/handlers/devicemessage.py
+++ b/synapse/handlers/devicemessage.py
@@ -22,9 +22,9 @@ from twisted.internet import defer
 from synapse.api.errors import SynapseError
 from synapse.logging.opentracing import (
     get_active_span_text_map,
+    log_kv,
     set_tag,
     start_active_span,
-    whitelisted_homeserver,
 )
 from synapse.types import UserID, get_domain_from_id
 from synapse.util.stringutils import random_string
@@ -52,7 +52,7 @@ class DeviceMessageHandler(object):
         local_messages = {}
         sender_user_id = content["sender"]
         if origin != get_domain_from_id(sender_user_id):
-            logger.warn(
+            logger.warning(
                 "Dropping device message from %r with spoofed sender %r",
                 origin,
                 sender_user_id,
@@ -86,7 +86,8 @@ class DeviceMessageHandler(object):
 
     @defer.inlineCallbacks
     def send_device_message(self, sender_user_id, message_type, messages):
-
+        set_tag("number_of_messages", len(messages))
+        set_tag("sender", sender_user_id)
         local_messages = {}
         remote_messages = {}
         for user_id, by_device in messages.items():
@@ -119,11 +120,10 @@ class DeviceMessageHandler(object):
                     "sender": sender_user_id,
                     "type": message_type,
                     "message_id": message_id,
-                    "org.matrix.opentracing_context": json.dumps(context)
-                    if whitelisted_homeserver(destination)
-                    else None,
+                    "org.matrix.opentracing_context": json.dumps(context),
                 }
 
+        log_kv({"local_messages": local_messages})
         stream_id = yield self.store.add_messages_to_device_inbox(
             local_messages, remote_edu_contents
         )
@@ -132,6 +132,7 @@ class DeviceMessageHandler(object):
             "to_device_key", stream_id, users=local_messages.keys()
         )
 
+        log_kv({"remote_messages": remote_messages})
         for destination in remote_messages.keys():
             # Enqueue a new federation transaction to send the new
             # device messages to each remote destination.
diff --git a/synapse/handlers/directory.py b/synapse/handlers/directory.py
index 526379c6f7..c4632f8984 100644
--- a/synapse/handlers/directory.py
+++ b/synapse/handlers/directory.py
@@ -250,7 +250,7 @@ class DirectoryHandler(BaseHandler):
                     ignore_backoff=True,
                 )
             except CodeMessageException as e:
-                logging.warn("Error retrieving alias")
+                logging.warning("Error retrieving alias")
                 if e.code == 404:
                     result = None
                 else:
diff --git a/synapse/handlers/e2e_keys.py b/synapse/handlers/e2e_keys.py
index 056fb97acb..f09a0b73c8 100644
--- a/synapse/handlers/e2e_keys.py
+++ b/synapse/handlers/e2e_keys.py
@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
 # Copyright 2016 OpenMarket Ltd
-# Copyright 2018 New Vector Ltd
+# Copyright 2018-2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -18,15 +19,25 @@ import logging
 
 from six import iteritems
 
+import attr
 from canonicaljson import encode_canonical_json, json
+from signedjson.key import decode_verify_key_bytes
+from signedjson.sign import SignatureVerifyException, verify_signed_json
+from unpaddedbase64 import decode_base64
 
 from twisted.internet import defer
 
-from synapse.api.errors import CodeMessageException, SynapseError
+from synapse.api.errors import CodeMessageException, Codes, NotFoundError, SynapseError
 from synapse.logging.context import make_deferred_yieldable, run_in_background
 from synapse.logging.opentracing import log_kv, set_tag, tag_args, trace
-from synapse.types import UserID, get_domain_from_id
+from synapse.types import (
+    UserID,
+    get_domain_from_id,
+    get_verify_key_from_cross_signing_key,
+)
 from synapse.util import unwrapFirstError
+from synapse.util.async_helpers import Linearizer
+from synapse.util.caches.expiringcache import ExpiringCache
 from synapse.util.retryutils import NotRetryingDestination
 
 logger = logging.getLogger(__name__)
@@ -40,16 +51,25 @@ class E2eKeysHandler(object):
         self.is_mine = hs.is_mine
         self.clock = hs.get_clock()
 
+        self._edu_updater = SigningKeyEduUpdater(hs, self)
+
+        federation_registry = hs.get_federation_registry()
+
+        # FIXME: switch to m.signing_key_update when MSC1756 is merged into the spec
+        federation_registry.register_edu_handler(
+            "org.matrix.signing_key_update",
+            self._edu_updater.incoming_signing_key_update,
+        )
         # doesn't really work as part of the generic query API, because the
         # query request requires an object POST, but we abuse the
         # "query handler" interface.
-        hs.get_federation_registry().register_query_handler(
+        federation_registry.register_query_handler(
             "client_keys", self.on_federation_query_client_keys
         )
 
     @trace
     @defer.inlineCallbacks
-    def query_devices(self, query_body, timeout):
+    def query_devices(self, query_body, timeout, from_user_id):
         """ Handle a device key query from a client
 
         {
@@ -67,6 +87,11 @@ class E2eKeysHandler(object):
                 }
             }
         }
+
+        Args:
+            from_user_id (str): the user making the query.  This is used when
+                adding cross-signing signatures to limit what signatures users
+                can see.
         """
 
         device_keys_query = query_body.get("device_keys", {})
@@ -105,9 +130,10 @@ class E2eKeysHandler(object):
                 else:
                     query_list.append((user_id, None))
 
-            user_ids_not_in_cache, remote_results = (
-                yield self.store.get_user_devices_from_cache(query_list)
-            )
+            (
+                user_ids_not_in_cache,
+                remote_results,
+            ) = yield self.store.get_user_devices_from_cache(query_list)
             for user_id, devices in iteritems(remote_results):
                 user_devices = results.setdefault(user_id, {})
                 for device_id, device in iteritems(devices):
@@ -125,6 +151,11 @@ class E2eKeysHandler(object):
                 r = remote_queries_not_in_cache.setdefault(domain, {})
                 r[user_id] = remote_queries[user_id]
 
+        # Get cached cross-signing keys
+        cross_signing_keys = yield self.get_cross_signing_keys_from_cache(
+            device_keys_query, from_user_id
+        )
+
         # Now fetch any devices that we don't have in our cache
         @trace
         @defer.inlineCallbacks
@@ -188,6 +219,16 @@ class E2eKeysHandler(object):
                     if user_id in destination_query:
                         results[user_id] = keys
 
+                if "master_keys" in remote_result:
+                    for user_id, key in remote_result["master_keys"].items():
+                        if user_id in destination_query:
+                            cross_signing_keys["master_keys"][user_id] = key
+
+                if "self_signing_keys" in remote_result:
+                    for user_id, key in remote_result["self_signing_keys"].items():
+                        if user_id in destination_query:
+                            cross_signing_keys["self_signing_keys"][user_id] = key
+
             except Exception as e:
                 failure = _exception_to_failure(e)
                 failures[destination] = failure
@@ -204,7 +245,61 @@ class E2eKeysHandler(object):
             ).addErrback(unwrapFirstError)
         )
 
-        return {"device_keys": results, "failures": failures}
+        ret = {"device_keys": results, "failures": failures}
+
+        ret.update(cross_signing_keys)
+
+        return ret
+
+    @defer.inlineCallbacks
+    def get_cross_signing_keys_from_cache(self, query, from_user_id):
+        """Get cross-signing keys for users from the database
+
+        Args:
+            query (Iterable[string]) an iterable of user IDs.  A dict whose keys
+                are user IDs satisfies this, so the query format used for
+                query_devices can be used here.
+            from_user_id (str): the user making the query.  This is used when
+                adding cross-signing signatures to limit what signatures users
+                can see.
+
+        Returns:
+            defer.Deferred[dict[str, dict[str, dict]]]: map from
+                (master_keys|self_signing_keys|user_signing_keys) -> user_id -> key
+        """
+        master_keys = {}
+        self_signing_keys = {}
+        user_signing_keys = {}
+
+        for user_id in query:
+            # XXX: consider changing the store functions to allow querying
+            # multiple users simultaneously.
+            key = yield self.store.get_e2e_cross_signing_key(
+                user_id, "master", from_user_id
+            )
+            if key:
+                master_keys[user_id] = key
+
+            key = yield self.store.get_e2e_cross_signing_key(
+                user_id, "self_signing", from_user_id
+            )
+            if key:
+                self_signing_keys[user_id] = key
+
+            # users can see other users' master and self-signing keys, but can
+            # only see their own user-signing keys
+            if from_user_id == user_id:
+                key = yield self.store.get_e2e_cross_signing_key(
+                    user_id, "user_signing", from_user_id
+                )
+                if key:
+                    user_signing_keys[user_id] = key
+
+        return {
+            "master_keys": master_keys,
+            "self_signing_keys": self_signing_keys,
+            "user_signing_keys": user_signing_keys,
+        }
 
     @trace
     @defer.inlineCallbacks
@@ -248,16 +343,10 @@ class E2eKeysHandler(object):
 
         results = yield self.store.get_e2e_device_keys(local_query)
 
-        # Build the result structure, un-jsonify the results, and add the
-        # "unsigned" section
+        # Build the result structure
         for user_id, device_keys in results.items():
             for device_id, device_info in device_keys.items():
-                r = dict(device_info["keys"])
-                r["unsigned"] = {}
-                display_name = device_info["device_display_name"]
-                if display_name is not None:
-                    r["unsigned"]["device_display_name"] = display_name
-                result_dict[user_id][device_id] = r
+                result_dict[user_id][device_id] = device_info
 
         log_kv(results)
         return result_dict
@@ -268,7 +357,16 @@ class E2eKeysHandler(object):
         """
         device_keys_query = query_body.get("device_keys", {})
         res = yield self.query_local_devices(device_keys_query)
-        return {"device_keys": res}
+        ret = {"device_keys": res}
+
+        # add in the cross-signing keys
+        cross_signing_keys = yield self.get_cross_signing_keys_from_cache(
+            device_keys_query, None
+        )
+
+        ret.update(cross_signing_keys)
+
+        return ret
 
     @trace
     @defer.inlineCallbacks
@@ -447,8 +545,503 @@ class E2eKeysHandler(object):
         log_kv({"message": "Inserting new one_time_keys.", "keys": new_keys})
         yield self.store.add_e2e_one_time_keys(user_id, device_id, time_now, new_keys)
 
+    @defer.inlineCallbacks
+    def upload_signing_keys_for_user(self, user_id, keys):
+        """Upload signing keys for cross-signing
+
+        Args:
+            user_id (string): the user uploading the keys
+            keys (dict[string, dict]): the signing keys
+        """
+
+        # if a master key is uploaded, then check it.  Otherwise, load the
+        # stored master key, to check signatures on other keys
+        if "master_key" in keys:
+            master_key = keys["master_key"]
+
+            _check_cross_signing_key(master_key, user_id, "master")
+        else:
+            master_key = yield self.store.get_e2e_cross_signing_key(user_id, "master")
+
+        # if there is no master key, then we can't do anything, because all the
+        # other cross-signing keys need to be signed by the master key
+        if not master_key:
+            raise SynapseError(400, "No master key available", Codes.MISSING_PARAM)
+
+        try:
+            master_key_id, master_verify_key = get_verify_key_from_cross_signing_key(
+                master_key
+            )
+        except ValueError:
+            if "master_key" in keys:
+                # the invalid key came from the request
+                raise SynapseError(400, "Invalid master key", Codes.INVALID_PARAM)
+            else:
+                # the invalid key came from the database
+                logger.error("Invalid master key found for user %s", user_id)
+                raise SynapseError(500, "Invalid master key")
+
+        # for the other cross-signing keys, make sure that they have valid
+        # signatures from the master key
+        if "self_signing_key" in keys:
+            self_signing_key = keys["self_signing_key"]
+
+            _check_cross_signing_key(
+                self_signing_key, user_id, "self_signing", master_verify_key
+            )
+
+        if "user_signing_key" in keys:
+            user_signing_key = keys["user_signing_key"]
+
+            _check_cross_signing_key(
+                user_signing_key, user_id, "user_signing", master_verify_key
+            )
+
+        # if everything checks out, then store the keys and send notifications
+        deviceids = []
+        if "master_key" in keys:
+            yield self.store.set_e2e_cross_signing_key(user_id, "master", master_key)
+            deviceids.append(master_verify_key.version)
+        if "self_signing_key" in keys:
+            yield self.store.set_e2e_cross_signing_key(
+                user_id, "self_signing", self_signing_key
+            )
+            try:
+                deviceids.append(
+                    get_verify_key_from_cross_signing_key(self_signing_key)[1].version
+                )
+            except ValueError:
+                raise SynapseError(400, "Invalid self-signing key", Codes.INVALID_PARAM)
+        if "user_signing_key" in keys:
+            yield self.store.set_e2e_cross_signing_key(
+                user_id, "user_signing", user_signing_key
+            )
+            # the signature stream matches the semantics that we want for
+            # user-signing key updates: only the user themselves is notified of
+            # their own user-signing key updates
+            yield self.device_handler.notify_user_signature_update(user_id, [user_id])
+
+        # master key and self-signing key updates match the semantics of device
+        # list updates: all users who share an encrypted room are notified
+        if len(deviceids):
+            yield self.device_handler.notify_device_update(user_id, deviceids)
+
+        return {}
+
+    @defer.inlineCallbacks
+    def upload_signatures_for_device_keys(self, user_id, signatures):
+        """Upload device signatures for cross-signing
+
+        Args:
+            user_id (string): the user uploading the signatures
+            signatures (dict[string, dict[string, dict]]): map of users to
+                devices to signed keys. This is the submission from the user; an
+                exception will be raised if it is malformed.
+        Returns:
+            dict: response to be sent back to the client.  The response will have
+                a "failures" key, which will be a dict mapping users to devices
+                to errors for the signatures that failed.
+        Raises:
+            SynapseError: if the signatures dict is not valid.
+        """
+        failures = {}
+
+        # signatures to be stored.  Each item will be a SignatureListItem
+        signature_list = []
+
+        # split between checking signatures for own user and signatures for
+        # other users, since we verify them with different keys
+        self_signatures = signatures.get(user_id, {})
+        other_signatures = {k: v for k, v in signatures.items() if k != user_id}
+
+        self_signature_list, self_failures = yield self._process_self_signatures(
+            user_id, self_signatures
+        )
+        signature_list.extend(self_signature_list)
+        failures.update(self_failures)
+
+        other_signature_list, other_failures = yield self._process_other_signatures(
+            user_id, other_signatures
+        )
+        signature_list.extend(other_signature_list)
+        failures.update(other_failures)
+
+        # store the signature, and send the appropriate notifications for sync
+        logger.debug("upload signature failures: %r", failures)
+        yield self.store.store_e2e_cross_signing_signatures(user_id, signature_list)
+
+        self_device_ids = [item.target_device_id for item in self_signature_list]
+        if self_device_ids:
+            yield self.device_handler.notify_device_update(user_id, self_device_ids)
+        signed_users = [item.target_user_id for item in other_signature_list]
+        if signed_users:
+            yield self.device_handler.notify_user_signature_update(
+                user_id, signed_users
+            )
+
+        return {"failures": failures}
+
+    @defer.inlineCallbacks
+    def _process_self_signatures(self, user_id, signatures):
+        """Process uploaded signatures of the user's own keys.
+
+        Signatures of the user's own keys from this API come in two forms:
+        - signatures of the user's devices by the user's self-signing key,
+        - signatures of the user's master key by the user's devices.
+
+        Args:
+            user_id (string): the user uploading the keys
+            signatures (dict[string, dict]): map of devices to signed keys
+
+        Returns:
+            (list[SignatureListItem], dict[string, dict[string, dict]]):
+            a list of signatures to store, and a map of users to devices to failure
+            reasons
+
+        Raises:
+            SynapseError: if the input is malformed
+        """
+        signature_list = []
+        failures = {}
+        if not signatures:
+            return signature_list, failures
+
+        if not isinstance(signatures, dict):
+            raise SynapseError(400, "Invalid parameter", Codes.INVALID_PARAM)
+
+        try:
+            # get our self-signing key to verify the signatures
+            (
+                _,
+                self_signing_key_id,
+                self_signing_verify_key,
+            ) = yield self._get_e2e_cross_signing_verify_key(user_id, "self_signing")
+
+            # get our master key, since we may have received a signature of it.
+            # We need to fetch it here so that we know what its key ID is, so
+            # that we can check if a signature that was sent is a signature of
+            # the master key or of a device
+            (
+                master_key,
+                _,
+                master_verify_key,
+            ) = yield self._get_e2e_cross_signing_verify_key(user_id, "master")
+
+            # fetch our stored devices.  This is used to 1. verify
+            # signatures on the master key, and 2. to compare with what
+            # was sent if the device was signed
+            devices = yield self.store.get_e2e_device_keys([(user_id, None)])
+
+            if user_id not in devices:
+                raise NotFoundError("No device keys found")
+
+            devices = devices[user_id]
+        except SynapseError as e:
+            failure = _exception_to_failure(e)
+            failures[user_id] = {device: failure for device in signatures.keys()}
+            return signature_list, failures
+
+        for device_id, device in signatures.items():
+            # make sure submitted data is in the right form
+            if not isinstance(device, dict):
+                raise SynapseError(400, "Invalid parameter", Codes.INVALID_PARAM)
+
+            try:
+                if "signatures" not in device or user_id not in device["signatures"]:
+                    # no signature was sent
+                    raise SynapseError(
+                        400, "Invalid signature", Codes.INVALID_SIGNATURE
+                    )
+
+                if device_id == master_verify_key.version:
+                    # The signature is of the master key. This needs to be
+                    # handled differently from signatures of normal devices.
+                    master_key_signature_list = self._check_master_key_signature(
+                        user_id, device_id, device, master_key, devices
+                    )
+                    signature_list.extend(master_key_signature_list)
+                    continue
+
+                # at this point, we have a device that should be signed
+                # by the self-signing key
+                if self_signing_key_id not in device["signatures"][user_id]:
+                    # no signature was sent
+                    raise SynapseError(
+                        400, "Invalid signature", Codes.INVALID_SIGNATURE
+                    )
+
+                try:
+                    stored_device = devices[device_id]
+                except KeyError:
+                    raise NotFoundError("Unknown device")
+                if self_signing_key_id in stored_device.get("signatures", {}).get(
+                    user_id, {}
+                ):
+                    # we already have a signature on this device, so we
+                    # can skip it, since it should be exactly the same
+                    continue
+
+                _check_device_signature(
+                    user_id, self_signing_verify_key, device, stored_device
+                )
+
+                signature = device["signatures"][user_id][self_signing_key_id]
+                signature_list.append(
+                    SignatureListItem(
+                        self_signing_key_id, user_id, device_id, signature
+                    )
+                )
+            except SynapseError as e:
+                failures.setdefault(user_id, {})[device_id] = _exception_to_failure(e)
+
+        return signature_list, failures
+
+    def _check_master_key_signature(
+        self, user_id, master_key_id, signed_master_key, stored_master_key, devices
+    ):
+        """Check signatures of a user's master key made by their devices.
+
+        Args:
+            user_id (string): the user whose master key is being checked
+            master_key_id (string): the ID of the user's master key
+            signed_master_key (dict): the user's signed master key that was uploaded
+            stored_master_key (dict): our previously-stored copy of the user's master key
+            devices (iterable(dict)): the user's devices
+
+        Returns:
+            list[SignatureListItem]: a list of signatures to store
+
+        Raises:
+            SynapseError: if a signature is invalid
+        """
+        # for each device that signed the master key, check the signature.
+        master_key_signature_list = []
+        sigs = signed_master_key["signatures"]
+        for signing_key_id, signature in sigs[user_id].items():
+            _, signing_device_id = signing_key_id.split(":", 1)
+            if (
+                signing_device_id not in devices
+                or signing_key_id not in devices[signing_device_id]["keys"]
+            ):
+                # signed by an unknown device, or the
+                # device does not have the key
+                raise SynapseError(400, "Invalid signature", Codes.INVALID_SIGNATURE)
+
+            # get the key and check the signature
+            pubkey = devices[signing_device_id]["keys"][signing_key_id]
+            verify_key = decode_verify_key_bytes(signing_key_id, decode_base64(pubkey))
+            _check_device_signature(
+                user_id, verify_key, signed_master_key, stored_master_key
+            )
+
+            master_key_signature_list.append(
+                SignatureListItem(signing_key_id, user_id, master_key_id, signature)
+            )
+
+        return master_key_signature_list
+
+    @defer.inlineCallbacks
+    def _process_other_signatures(self, user_id, signatures):
+        """Process uploaded signatures of other users' keys.  These will be the
+        target user's master keys, signed by the uploading user's user-signing
+        key.
+
+        Args:
+            user_id (string): the user uploading the keys
+            signatures (dict[string, dict]): map of users to devices to signed keys
+
+        Returns:
+            (list[SignatureListItem], dict[string, dict[string, dict]]):
+            a list of signatures to store, and a map of users to devices to failure
+            reasons
+
+        Raises:
+            SynapseError: if the input is malformed
+        """
+        signature_list = []
+        failures = {}
+        if not signatures:
+            return signature_list, failures
+
+        try:
+            # get our user-signing key to verify the signatures
+            (
+                user_signing_key,
+                user_signing_key_id,
+                user_signing_verify_key,
+            ) = yield self._get_e2e_cross_signing_verify_key(user_id, "user_signing")
+        except SynapseError as e:
+            failure = _exception_to_failure(e)
+            for user, devicemap in signatures.items():
+                failures[user] = {device_id: failure for device_id in devicemap.keys()}
+            return signature_list, failures
+
+        for target_user, devicemap in signatures.items():
+            # make sure submitted data is in the right form
+            if not isinstance(devicemap, dict):
+                raise SynapseError(400, "Invalid parameter", Codes.INVALID_PARAM)
+            for device in devicemap.values():
+                if not isinstance(device, dict):
+                    raise SynapseError(400, "Invalid parameter", Codes.INVALID_PARAM)
+
+            device_id = None
+            try:
+                # get the target user's master key, to make sure it matches
+                # what was sent
+                (
+                    master_key,
+                    master_key_id,
+                    _,
+                ) = yield self._get_e2e_cross_signing_verify_key(
+                    target_user, "master", user_id
+                )
+
+                # make sure that the target user's master key is the one that
+                # was signed (and no others)
+                device_id = master_key_id.split(":", 1)[1]
+                if device_id not in devicemap:
+                    logger.debug(
+                        "upload signature: could not find signature for device %s",
+                        device_id,
+                    )
+                    # set device to None so that the failure gets
+                    # marked on all the signatures
+                    device_id = None
+                    raise NotFoundError("Unknown device")
+                key = devicemap[device_id]
+                other_devices = [k for k in devicemap.keys() if k != device_id]
+                if other_devices:
+                    # other devices were signed -- mark those as failures
+                    logger.debug("upload signature: too many devices specified")
+                    failure = _exception_to_failure(NotFoundError("Unknown device"))
+                    failures[target_user] = {
+                        device: failure for device in other_devices
+                    }
+
+                if user_signing_key_id in master_key.get("signatures", {}).get(
+                    user_id, {}
+                ):
+                    # we already have the signature, so we can skip it
+                    continue
+
+                _check_device_signature(
+                    user_id, user_signing_verify_key, key, master_key
+                )
+
+                signature = key["signatures"][user_id][user_signing_key_id]
+                signature_list.append(
+                    SignatureListItem(
+                        user_signing_key_id, target_user, device_id, signature
+                    )
+                )
+            except SynapseError as e:
+                failure = _exception_to_failure(e)
+                if device_id is None:
+                    failures[target_user] = {
+                        device_id: failure for device_id in devicemap.keys()
+                    }
+                else:
+                    failures.setdefault(target_user, {})[device_id] = failure
+
+        return signature_list, failures
+
+    @defer.inlineCallbacks
+    def _get_e2e_cross_signing_verify_key(self, user_id, key_type, from_user_id=None):
+        """Fetch the cross-signing public key from storage and interpret it.
+
+        Args:
+            user_id (str): the user whose key should be fetched
+            key_type (str): the type of key to fetch
+            from_user_id (str): the user that we are fetching the keys for.
+                This affects what signatures are fetched.
+
+        Returns:
+            dict, str, VerifyKey: the raw key data, the key ID, and the
+                signedjson verify key
+
+        Raises:
+            NotFoundError: if the key is not found
+        """
+        key = yield self.store.get_e2e_cross_signing_key(
+            user_id, key_type, from_user_id
+        )
+        if key is None:
+            logger.debug("no %s key found for %s", key_type, user_id)
+            raise NotFoundError("No %s key found for %s" % (key_type, user_id))
+        key_id, verify_key = get_verify_key_from_cross_signing_key(key)
+        return key, key_id, verify_key
+
+
+def _check_cross_signing_key(key, user_id, key_type, signing_key=None):
+    """Check a cross-signing key uploaded by a user.  Performs some basic sanity
+    checking, and ensures that it is signed, if a signature is required.
+
+    Args:
+        key (dict): the key data to verify
+        user_id (str): the user whose key is being checked
+        key_type (str): the type of key that the key should be
+        signing_key (VerifyKey): (optional) the signing key that the key should
+            be signed with.  If omitted, signatures will not be checked.
+    """
+    if (
+        key.get("user_id") != user_id
+        or key_type not in key.get("usage", [])
+        or len(key.get("keys", {})) != 1
+    ):
+        raise SynapseError(400, ("Invalid %s key" % (key_type,)), Codes.INVALID_PARAM)
+
+    if signing_key:
+        try:
+            verify_signed_json(key, user_id, signing_key)
+        except SignatureVerifyException:
+            raise SynapseError(
+                400, ("Invalid signature on %s key" % key_type), Codes.INVALID_SIGNATURE
+            )
+
+
+def _check_device_signature(user_id, verify_key, signed_device, stored_device):
+    """Check that a signature on a device or cross-signing key is correct and
+    matches the copy of the device/key that we have stored.  Throws an
+    exception if an error is detected.
+
+    Args:
+        user_id (str): the user ID whose signature is being checked
+        verify_key (VerifyKey): the key to verify the device with
+        signed_device (dict): the uploaded signed device data
+        stored_device (dict): our previously stored copy of the device
+
+    Raises:
+        SynapseError: if the signature was invalid or the sent device is not the
+            same as the stored device
+
+    """
+
+    # make sure that the device submitted matches what we have stored
+    stripped_signed_device = {
+        k: v for k, v in signed_device.items() if k not in ["signatures", "unsigned"]
+    }
+    stripped_stored_device = {
+        k: v for k, v in stored_device.items() if k not in ["signatures", "unsigned"]
+    }
+    if stripped_signed_device != stripped_stored_device:
+        logger.debug(
+            "upload signatures: key does not match %s vs %s",
+            signed_device,
+            stored_device,
+        )
+        raise SynapseError(400, "Key does not match")
+
+    try:
+        verify_signed_json(signed_device, user_id, verify_key)
+    except SignatureVerifyException:
+        logger.debug("invalid signature on key")
+        raise SynapseError(400, "Invalid signature", Codes.INVALID_SIGNATURE)
+
 
 def _exception_to_failure(e):
+    if isinstance(e, SynapseError):
+        return {"status": e.code, "errcode": e.errcode, "message": str(e)}
+
     if isinstance(e, CodeMessageException):
         return {"status": e.code, "message": str(e)}
 
@@ -476,3 +1069,111 @@ def _one_time_keys_match(old_key_json, new_key):
     new_key_copy.pop("signatures", None)
 
     return old_key == new_key_copy
+
+
+@attr.s
+class SignatureListItem:
+    """An item in the signature list as used by upload_signatures_for_device_keys.
+    """
+
+    signing_key_id = attr.ib()
+    target_user_id = attr.ib()
+    target_device_id = attr.ib()
+    signature = attr.ib()
+
+
+class SigningKeyEduUpdater(object):
+    """Handles incoming signing key updates from federation and updates the DB"""
+
+    def __init__(self, hs, e2e_keys_handler):
+        self.store = hs.get_datastore()
+        self.federation = hs.get_federation_client()
+        self.clock = hs.get_clock()
+        self.e2e_keys_handler = e2e_keys_handler
+
+        self._remote_edu_linearizer = Linearizer(name="remote_signing_key")
+
+        # user_id -> list of updates waiting to be handled.
+        self._pending_updates = {}
+
+        # Recently seen stream ids. We don't bother keeping these in the DB,
+        # but they're useful to have them about to reduce the number of spurious
+        # resyncs.
+        self._seen_updates = ExpiringCache(
+            cache_name="signing_key_update_edu",
+            clock=self.clock,
+            max_len=10000,
+            expiry_ms=30 * 60 * 1000,
+            iterable=True,
+        )
+
+    @defer.inlineCallbacks
+    def incoming_signing_key_update(self, origin, edu_content):
+        """Called on incoming signing key update from federation. Responsible for
+        parsing the EDU and adding to pending updates list.
+
+        Args:
+            origin (string): the server that sent the EDU
+            edu_content (dict): the contents of the EDU
+        """
+
+        user_id = edu_content.pop("user_id")
+        master_key = edu_content.pop("master_key", None)
+        self_signing_key = edu_content.pop("self_signing_key", None)
+
+        if get_domain_from_id(user_id) != origin:
+            logger.warning("Got signing key update edu for %r from %r", user_id, origin)
+            return
+
+        room_ids = yield self.store.get_rooms_for_user(user_id)
+        if not room_ids:
+            # We don't share any rooms with this user. Ignore update, as we
+            # probably won't get any further updates.
+            return
+
+        self._pending_updates.setdefault(user_id, []).append(
+            (master_key, self_signing_key)
+        )
+
+        yield self._handle_signing_key_updates(user_id)
+
+    @defer.inlineCallbacks
+    def _handle_signing_key_updates(self, user_id):
+        """Actually handle pending updates.
+
+        Args:
+            user_id (string): the user whose updates we are processing
+        """
+
+        device_handler = self.e2e_keys_handler.device_handler
+
+        with (yield self._remote_edu_linearizer.queue(user_id)):
+            pending_updates = self._pending_updates.pop(user_id, [])
+            if not pending_updates:
+                # This can happen since we batch updates
+                return
+
+            device_ids = []
+
+            logger.info("pending updates: %r", pending_updates)
+
+            for master_key, self_signing_key in pending_updates:
+                if master_key:
+                    yield self.store.set_e2e_cross_signing_key(
+                        user_id, "master", master_key
+                    )
+                    _, verify_key = get_verify_key_from_cross_signing_key(master_key)
+                    # verify_key is a VerifyKey from signedjson, which uses
+                    # .version to denote the portion of the key ID after the
+                    # algorithm and colon, which is the device ID
+                    device_ids.append(verify_key.version)
+                if self_signing_key:
+                    yield self.store.set_e2e_cross_signing_key(
+                        user_id, "self_signing", self_signing_key
+                    )
+                    _, verify_key = get_verify_key_from_cross_signing_key(
+                        self_signing_key
+                    )
+                    device_ids.append(verify_key.version)
+
+            yield device_handler.notify_device_update(user_id, device_ids)
diff --git a/synapse/handlers/e2e_room_keys.py b/synapse/handlers/e2e_room_keys.py
index 05836e8149..fec88eac26 100644
--- a/synapse/handlers/e2e_room_keys.py
+++ b/synapse/handlers/e2e_room_keys.py
@@ -377,8 +377,8 @@ class E2eRoomKeysHandler(object):
             A deferred of an empty dict.
         """
         if "version" not in version_info:
-            raise SynapseError(400, "Missing version in body", Codes.MISSING_PARAM)
-        if version_info["version"] != version:
+            version_info["version"] = version
+        elif version_info["version"] != version:
             raise SynapseError(
                 400, "Version in body does not match", Codes.INVALID_PARAM
             )
diff --git a/synapse/handlers/events.py b/synapse/handlers/events.py
index 2f1f10a9af..45fe13c62f 100644
--- a/synapse/handlers/events.py
+++ b/synapse/handlers/events.py
@@ -147,6 +147,10 @@ class EventStreamHandler(BaseHandler):
 
 
 class EventHandler(BaseHandler):
+    def __init__(self, hs):
+        super(EventHandler, self).__init__(hs)
+        self.storage = hs.get_storage()
+
     @defer.inlineCallbacks
     def get_event(self, user, room_id, event_id):
         """Retrieve a single specified event.
@@ -167,13 +171,12 @@ class EventHandler(BaseHandler):
 
         if not event:
             return None
-            return
 
         users = yield self.store.get_users_in_room(event.room_id)
         is_peeking = user.to_string() not in users
 
         filtered = yield filter_events_for_client(
-            self.store, user.to_string(), [event], is_peeking=is_peeking
+            self.storage, user.to_string(), [event], is_peeking=is_peeking
         )
 
         if not filtered:
diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py
index 538b16efd6..05dd8d2671 100644
--- a/synapse/handlers/federation.py
+++ b/synapse/handlers/federation.py
@@ -30,6 +30,7 @@ from unpaddedbase64 import decode_base64
 
 from twisted.internet import defer
 
+from synapse import event_auth
 from synapse.api.constants import EventTypes, Membership, RejectedReason
 from synapse.api.errors import (
     AuthError,
@@ -44,6 +45,7 @@ from synapse.api.errors import (
 from synapse.api.room_versions import KNOWN_ROOM_VERSIONS, RoomVersions
 from synapse.crypto.event_signing import compute_event_signature
 from synapse.event_auth import auth_types_for_event
+from synapse.events.snapshot import EventContext
 from synapse.events.validator import EventValidator
 from synapse.logging.context import (
     make_deferred_yieldable,
@@ -108,6 +110,8 @@ class FederationHandler(BaseHandler):
         self.hs = hs
 
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
         self.federation_client = hs.get_federation_client()
         self.state_handler = hs.get_state_handler()
         self.server_name = hs.hostname
@@ -179,7 +183,7 @@ class FederationHandler(BaseHandler):
         try:
             self._sanity_check_event(pdu)
         except SynapseError as err:
-            logger.warn(
+            logger.warning(
                 "[%s %s] Received event failed sanity checks", room_id, event_id
             )
             raise FederationError("ERROR", err.code, err.msg, affected=pdu.event_id)
@@ -300,7 +304,7 @@ class FederationHandler(BaseHandler):
                 # following.
 
                 if sent_to_us_directly:
-                    logger.warn(
+                    logger.warning(
                         "[%s %s] Rejecting: failed to fetch %d prev events: %s",
                         room_id,
                         event_id,
@@ -323,7 +327,7 @@ class FederationHandler(BaseHandler):
                 event_map = {event_id: pdu}
                 try:
                     # Get the state of the events we know about
-                    ours = yield self.store.get_state_groups_ids(room_id, seen)
+                    ours = yield self.state_store.get_state_groups_ids(room_id, seen)
 
                     # state_maps is a list of mappings from (type, state_key) to event_id
                     state_maps = list(
@@ -349,10 +353,11 @@ class FederationHandler(BaseHandler):
                             # note that if any of the missing prevs share missing state or
                             # auth events, the requests to fetch those events are deduped
                             # by the get_pdu_cache in federation_client.
-                            remote_state, got_auth_chain = (
-                                yield self.federation_client.get_state_for_room(
-                                    origin, room_id, p
-                                )
+                            (
+                                remote_state,
+                                got_auth_chain,
+                            ) = yield self.federation_client.get_state_for_room(
+                                origin, room_id, p
                             )
 
                             # we want the state *after* p; get_state_for_room returns the
@@ -404,7 +409,7 @@ class FederationHandler(BaseHandler):
                     state = [event_map[e] for e in six.itervalues(state_map)]
                     auth_chain = list(auth_chains)
                 except Exception:
-                    logger.warn(
+                    logger.warning(
                         "[%s %s] Error attempting to resolve state at missing "
                         "prev_events",
                         room_id,
@@ -517,7 +522,9 @@ class FederationHandler(BaseHandler):
             # We failed to get the missing events, but since we need to handle
             # the case of `get_missing_events` not returning the necessary
             # events anyway, it is safe to simply log the error and continue.
-            logger.warn("[%s %s]: Failed to get prev_events: %s", room_id, event_id, e)
+            logger.warning(
+                "[%s %s]: Failed to get prev_events: %s", room_id, event_id, e
+            )
             return
 
         logger.info(
@@ -544,7 +551,7 @@ class FederationHandler(BaseHandler):
                     yield self.on_receive_pdu(origin, ev, sent_to_us_directly=False)
                 except FederationError as e:
                     if e.code == 403:
-                        logger.warn(
+                        logger.warning(
                             "[%s %s] Received prev_event %s failed history check.",
                             room_id,
                             event_id,
@@ -887,7 +894,7 @@ class FederationHandler(BaseHandler):
         # We set `check_history_visibility_only` as we might otherwise get false
         # positives from users having been erased.
         filtered_extremities = yield filter_events_for_server(
-            self.store,
+            self.storage,
             self.server_name,
             list(extremities_events.values()),
             redact=False,
@@ -1058,7 +1065,7 @@ class FederationHandler(BaseHandler):
             SynapseError if the event does not pass muster
         """
         if len(ev.prev_event_ids()) > 20:
-            logger.warn(
+            logger.warning(
                 "Rejecting event %s which has %i prev_events",
                 ev.event_id,
                 len(ev.prev_event_ids()),
@@ -1066,7 +1073,7 @@ class FederationHandler(BaseHandler):
             raise SynapseError(http_client.BAD_REQUEST, "Too many prev_events")
 
         if len(ev.auth_event_ids()) > 10:
-            logger.warn(
+            logger.warning(
                 "Rejecting event %s which has %i auth_events",
                 ev.event_id,
                 len(ev.auth_event_ids()),
@@ -1100,7 +1107,7 @@ class FederationHandler(BaseHandler):
     @defer.inlineCallbacks
     def do_invite_join(self, target_hosts, room_id, joinee, content):
         """ Attempts to join the `joinee` to the room `room_id` via the
-        server `target_host`.
+        servers contained in `target_hosts`.
 
         This first triggers a /make_join/ request that returns a partial
         event that we can fill out and sign. This is then sent to the
@@ -1109,6 +1116,15 @@ class FederationHandler(BaseHandler):
 
         We suspend processing of any received events from this room until we
         have finished processing the join.
+
+        Args:
+            target_hosts (Iterable[str]): List of servers to attempt to join the room with.
+
+            room_id (str): The ID of the room to join.
+
+            joinee (str): The User ID of the joining user.
+
+            content (dict): The event content to use for the join event.
         """
         logger.debug("Joining %s to %s", joinee, room_id)
 
@@ -1168,6 +1184,22 @@ class FederationHandler(BaseHandler):
 
             yield self._persist_auth_tree(origin, auth_chain, state, event)
 
+            # Check whether this room is the result of an upgrade of a room we already know
+            # about. If so, migrate over user information
+            predecessor = yield self.store.get_room_predecessor(room_id)
+            if not predecessor:
+                return
+            old_room_id = predecessor["room_id"]
+            logger.debug(
+                "Found predecessor for %s during remote join: %s", room_id, old_room_id
+            )
+
+            # We retrieve the room member handler here as to not cause a cyclic dependency
+            member_handler = self.hs.get_room_member_handler()
+            yield member_handler.transfer_room_state_on_room_upgrade(
+                old_room_id, room_id
+            )
+
             logger.debug("Finished joining %s to %s", joinee, room_id)
         finally:
             room_queue = self.room_queues[room_id]
@@ -1202,7 +1234,7 @@ class FederationHandler(BaseHandler):
                 with nested_logging_context(p.event_id):
                     yield self.on_receive_pdu(origin, p, sent_to_us_directly=True)
             except Exception as e:
-                logger.warn(
+                logger.warning(
                     "Error handling queued PDU %s from %s: %s", p.event_id, origin, e
                 )
 
@@ -1221,7 +1253,6 @@ class FederationHandler(BaseHandler):
         Returns:
             Deferred[FrozenEvent]
         """
-
         if get_domain_from_id(user_id) != origin:
             logger.info(
                 "Got /make_join request for user %r from different origin %s, ignoring",
@@ -1250,7 +1281,7 @@ class FederationHandler(BaseHandler):
                 builder=builder
             )
         except AuthError as e:
-            logger.warn("Failed to create join %r because %s", event, e)
+            logger.warning("Failed to create join to %s because %s", room_id, e)
             raise e
 
         event_allowed = yield self.third_party_event_rules.check_event_allowed(
@@ -1279,11 +1310,20 @@ class FederationHandler(BaseHandler):
         event = pdu
 
         logger.debug(
-            "on_send_join_request: Got event: %s, signatures: %s",
+            "on_send_join_request from %s: Got event: %s, signatures: %s",
+            origin,
             event.event_id,
             event.signatures,
         )
 
+        if get_domain_from_id(event.sender) != origin:
+            logger.info(
+                "Got /send_join request for user %r from different origin %s",
+                event.sender,
+                origin,
+            )
+            raise SynapseError(403, "User not from origin", Codes.FORBIDDEN)
+
         event.internal_metadata.outlier = False
         # Send this event on behalf of the origin server.
         #
@@ -1485,7 +1525,7 @@ class FederationHandler(BaseHandler):
                 room_version, event, context, do_sig_check=False
             )
         except AuthError as e:
-            logger.warn("Failed to create new leave %r because %s", event, e)
+            logger.warning("Failed to create new leave %r because %s", event, e)
             raise e
 
         return event
@@ -1502,6 +1542,14 @@ class FederationHandler(BaseHandler):
             event.signatures,
         )
 
+        if get_domain_from_id(event.sender) != origin:
+            logger.info(
+                "Got /send_leave request for user %r from different origin %s",
+                event.sender,
+                origin,
+            )
+            raise SynapseError(403, "User not from origin", Codes.FORBIDDEN)
+
         event.internal_metadata.outlier = False
 
         context = yield self._handle_new_event(origin, event)
@@ -1532,7 +1580,7 @@ class FederationHandler(BaseHandler):
             event_id, allow_none=False, check_room_id=room_id
         )
 
-        state_groups = yield self.store.get_state_groups(room_id, [event_id])
+        state_groups = yield self.state_store.get_state_groups(room_id, [event_id])
 
         if state_groups:
             _, state = list(iteritems(state_groups)).pop()
@@ -1561,7 +1609,7 @@ class FederationHandler(BaseHandler):
             event_id, allow_none=False, check_room_id=room_id
         )
 
-        state_groups = yield self.store.get_state_groups_ids(room_id, [event_id])
+        state_groups = yield self.state_store.get_state_groups_ids(room_id, [event_id])
 
         if state_groups:
             _, state = list(state_groups.items()).pop()
@@ -1589,7 +1637,7 @@ class FederationHandler(BaseHandler):
 
         events = yield self.store.get_backfill_events(room_id, pdu_list, limit)
 
-        events = yield filter_events_for_server(self.store, origin, events)
+        events = yield filter_events_for_server(self.storage, origin, events)
 
         return events
 
@@ -1619,7 +1667,7 @@ class FederationHandler(BaseHandler):
             if not in_room:
                 raise AuthError(403, "Host not in room.")
 
-            events = yield filter_events_for_server(self.store, origin, [event])
+            events = yield filter_events_for_server(self.storage, origin, [event])
             event = events[0]
             return event
         else:
@@ -1640,7 +1688,11 @@ class FederationHandler(BaseHandler):
         # hack around with a try/finally instead.
         success = False
         try:
-            if not event.internal_metadata.is_outlier() and not backfilled:
+            if (
+                not event.internal_metadata.is_outlier()
+                and not backfilled
+                and not context.rejected
+            ):
                 yield self.action_generator.handle_push_actions_for_event(
                     event, context
                 )
@@ -1763,7 +1815,7 @@ class FederationHandler(BaseHandler):
                 auth_for_e[(EventTypes.Create, "")] = create_event
 
             try:
-                self.auth.check(room_version, e, auth_events=auth_for_e)
+                event_auth.check(room_version, e, auth_events=auth_for_e)
             except SynapseError as err:
                 # we may get SynapseErrors here as well as AuthErrors. For
                 # instance, there are a couple of (ancient) events in some
@@ -1771,7 +1823,7 @@ class FederationHandler(BaseHandler):
                 # cause SynapseErrors in auth.check. We don't want to give up
                 # the attempt to federate altogether in such cases.
 
-                logger.warn("Rejecting %s because %s", e.event_id, err.msg)
+                logger.warning("Rejecting %s because %s", e.event_id, err.msg)
 
                 if e == event:
                     raise
@@ -1824,12 +1876,7 @@ class FederationHandler(BaseHandler):
                 if c and c.type == EventTypes.Create:
                     auth_events[(c.type, c.state_key)] = c
 
-        try:
-            yield self.do_auth(origin, event, context, auth_events=auth_events)
-        except AuthError as e:
-            logger.warn("[%s %s] Rejecting: %s", event.room_id, event.event_id, e.msg)
-
-            context.rejected = RejectedReason.AUTH_ERROR
+        context = yield self.do_auth(origin, event, context, auth_events=auth_events)
 
         if not context.rejected:
             yield self._check_for_soft_fail(event, state, backfilled)
@@ -1885,7 +1932,7 @@ class FederationHandler(BaseHandler):
                 # given state at the event. This should correctly handle cases
                 # like bans, especially with state res v2.
 
-                state_sets = yield self.store.get_state_groups(
+                state_sets = yield self.state_store.get_state_groups(
                     event.room_id, extrem_ids
                 )
                 state_sets = list(state_sets.values())
@@ -1919,9 +1966,9 @@ class FederationHandler(BaseHandler):
             }
 
             try:
-                self.auth.check(room_version, event, auth_events=current_auth_events)
+                event_auth.check(room_version, event, auth_events=current_auth_events)
             except AuthError as e:
-                logger.warn("Soft-failing %r because %s", event, e)
+                logger.warning("Soft-failing %r because %s", event, e)
                 event.internal_metadata.soft_failed = True
 
     @defer.inlineCallbacks
@@ -1976,7 +2023,7 @@ class FederationHandler(BaseHandler):
         )
 
         missing_events = yield filter_events_for_server(
-            self.store, origin, missing_events
+            self.storage, origin, missing_events
         )
 
         return missing_events
@@ -1998,12 +2045,12 @@ class FederationHandler(BaseHandler):
 
                 Also NB that this function adds entries to it.
         Returns:
-            defer.Deferred[None]
+            defer.Deferred[EventContext]: updated context object
         """
         room_version = yield self.store.get_room_version(event.room_id)
 
         try:
-            yield self._update_auth_events_and_context_for_auth(
+            context = yield self._update_auth_events_and_context_for_auth(
                 origin, event, context, auth_events
             )
         except Exception:
@@ -2018,10 +2065,12 @@ class FederationHandler(BaseHandler):
             )
 
         try:
-            self.auth.check(room_version, event, auth_events=auth_events)
+            event_auth.check(room_version, event, auth_events=auth_events)
         except AuthError as e:
-            logger.warn("Failed auth resolution for %r because %s", event, e)
-            raise e
+            logger.warning("Failed auth resolution for %r because %s", event, e)
+            context.rejected = RejectedReason.AUTH_ERROR
+
+        return context
 
     @defer.inlineCallbacks
     def _update_auth_events_and_context_for_auth(
@@ -2045,7 +2094,7 @@ class FederationHandler(BaseHandler):
             auth_events (dict[(str, str)->synapse.events.EventBase]):
 
         Returns:
-            defer.Deferred[None]
+            defer.Deferred[EventContext]: updated context
         """
         event_auth_events = set(event.auth_event_ids())
 
@@ -2084,7 +2133,7 @@ class FederationHandler(BaseHandler):
                     # The other side isn't around or doesn't implement the
                     # endpoint, so lets just bail out.
                     logger.info("Failed to get event auth from remote: %s", e)
-                    return
+                    return context
 
                 seen_remotes = yield self.store.have_seen_events(
                     [e.event_id for e in remote_auth_chain]
@@ -2125,7 +2174,7 @@ class FederationHandler(BaseHandler):
 
         if event.internal_metadata.is_outlier():
             logger.info("Skipping auth_event fetch for outlier")
-            return
+            return context
 
         # FIXME: Assumes we have and stored all the state for all the
         # prev_events
@@ -2134,7 +2183,7 @@ class FederationHandler(BaseHandler):
         )
 
         if not different_auth:
-            return
+            return context
 
         logger.info(
             "auth_events refers to events which are not in our calculated auth "
@@ -2181,102 +2230,11 @@ class FederationHandler(BaseHandler):
 
             auth_events.update(new_state)
 
-            different_auth = event_auth_events.difference(
-                e.event_id for e in auth_events.values()
-            )
-
-            yield self._update_context_for_auth_events(
+            context = yield self._update_context_for_auth_events(
                 event, context, auth_events, event_key
             )
 
-        if not different_auth:
-            # we're done
-            return
-
-        logger.info(
-            "auth_events still refers to events which are not in the calculated auth "
-            "chain after state resolution: %s",
-            different_auth,
-        )
-
-        # Only do auth resolution if we have something new to say.
-        # We can't prove an auth failure.
-        do_resolution = False
-
-        for e_id in different_auth:
-            if e_id in have_events:
-                if have_events[e_id] == RejectedReason.NOT_ANCESTOR:
-                    do_resolution = True
-                    break
-
-        if not do_resolution:
-            logger.info(
-                "Skipping auth resolution due to lack of provable rejection reasons"
-            )
-            return
-
-        logger.info("Doing auth resolution")
-
-        prev_state_ids = yield context.get_prev_state_ids(self.store)
-
-        # 1. Get what we think is the auth chain.
-        auth_ids = yield self.auth.compute_auth_events(event, prev_state_ids)
-        local_auth_chain = yield self.store.get_auth_chain(auth_ids, include_given=True)
-
-        try:
-            # 2. Get remote difference.
-            try:
-                result = yield self.federation_client.query_auth(
-                    origin, event.room_id, event.event_id, local_auth_chain
-                )
-            except RequestSendFailed as e:
-                # The other side isn't around or doesn't implement the
-                # endpoint, so lets just bail out.
-                logger.info("Failed to query auth from remote: %s", e)
-                return
-
-            seen_remotes = yield self.store.have_seen_events(
-                [e.event_id for e in result["auth_chain"]]
-            )
-
-            # 3. Process any remote auth chain events we haven't seen.
-            for ev in result["auth_chain"]:
-                if ev.event_id in seen_remotes:
-                    continue
-
-                if ev.event_id == event.event_id:
-                    continue
-
-                try:
-                    auth_ids = ev.auth_event_ids()
-                    auth = {
-                        (e.type, e.state_key): e
-                        for e in result["auth_chain"]
-                        if e.event_id in auth_ids or event.type == EventTypes.Create
-                    }
-                    ev.internal_metadata.outlier = True
-
-                    logger.debug(
-                        "do_auth %s different_auth: %s", event.event_id, e.event_id
-                    )
-
-                    yield self._handle_new_event(origin, ev, auth_events=auth)
-
-                    if ev.event_id in event_auth_events:
-                        auth_events[(ev.type, ev.state_key)] = ev
-                except AuthError:
-                    pass
-
-        except Exception:
-            # FIXME:
-            logger.exception("Failed to query auth chain")
-
-        # 4. Look at rejects and their proofs.
-        # TODO.
-
-        yield self._update_context_for_auth_events(
-            event, context, auth_events, event_key
-        )
+        return context
 
     @defer.inlineCallbacks
     def _update_context_for_auth_events(self, event, context, auth_events, event_key):
@@ -2286,14 +2244,16 @@ class FederationHandler(BaseHandler):
         Args:
             event (Event): The event we're handling the context for
 
-            context (synapse.events.snapshot.EventContext): event context
-                to be updated
+            context (synapse.events.snapshot.EventContext): initial event context
 
             auth_events (dict[(str, str)->str]): Events to update in the event
                 context.
 
             event_key ((str, str)): (type, state_key) for the current event.
                 this will not be included in the current_state in the context.
+
+        Returns:
+            Deferred[EventContext]: new event context
         """
         state_updates = {
             k: a.event_id for k, a in iteritems(auth_events) if k != event_key
@@ -2310,7 +2270,7 @@ class FederationHandler(BaseHandler):
 
         # create a new state group as a delta from the existing one.
         prev_group = context.state_group
-        state_group = yield self.store.store_state_group(
+        state_group = yield self.state_store.store_state_group(
             event.event_id,
             event.room_id,
             prev_group=prev_group,
@@ -2318,8 +2278,9 @@ class FederationHandler(BaseHandler):
             current_state_ids=current_state_ids,
         )
 
-        yield context.update_state(
+        return EventContext.with_state(
             state_group=state_group,
+            state_group_before_event=context.state_group_before_event,
             current_state_ids=current_state_ids,
             prev_state_ids=prev_state_ids,
             prev_group=prev_group,
@@ -2444,15 +2405,6 @@ class FederationHandler(BaseHandler):
 
             reason_map[e.event_id] = reason
 
-            if reason == RejectedReason.AUTH_ERROR:
-                pass
-            elif reason == RejectedReason.REPLACED:
-                # TODO: Get proof
-                pass
-            elif reason == RejectedReason.NOT_ANCESTOR:
-                # TODO: Get proof.
-                pass
-
         logger.debug("construct_auth_difference returning")
 
         return {
@@ -2516,10 +2468,12 @@ class FederationHandler(BaseHandler):
             try:
                 yield self.auth.check_from_context(room_version, event, context)
             except AuthError as e:
-                logger.warn("Denying new third party invite %r because %s", event, e)
+                logger.warning("Denying new third party invite %r because %s", event, e)
                 raise e
 
             yield self._check_signature(event, context)
+
+            # We retrieve the room member handler here as to not cause a cyclic dependency
             member_handler = self.hs.get_room_member_handler()
             yield member_handler.send_membership_event(None, event, context)
         else:
@@ -2530,12 +2484,17 @@ class FederationHandler(BaseHandler):
 
     @defer.inlineCallbacks
     @log_function
-    def on_exchange_third_party_invite_request(self, origin, room_id, event_dict):
+    def on_exchange_third_party_invite_request(self, room_id, event_dict):
         """Handle an exchange_third_party_invite request from a remote server
 
         The remote server will call this when it wants to turn a 3pid invite
         into a normal m.room.member invite.
 
+        Args:
+            room_id (str): The ID of the room.
+
+            event_dict (dict[str, Any]): Dictionary containing the event body.
+
         Returns:
             Deferred: resolves (to None)
         """
@@ -2565,9 +2524,9 @@ class FederationHandler(BaseHandler):
         )
 
         try:
-            self.auth.check_from_context(room_version, event, context)
+            yield self.auth.check_from_context(room_version, event, context)
         except AuthError as e:
-            logger.warn("Denying third party invite %r because %s", event, e)
+            logger.warning("Denying third party invite %r because %s", event, e)
             raise e
         yield self._check_signature(event, context)
 
@@ -2575,6 +2534,7 @@ class FederationHandler(BaseHandler):
         # though the sender isn't a local user.
         event.internal_metadata.send_on_behalf_of = get_domain_from_id(event.sender)
 
+        # We retrieve the room member handler here as to not cause a cyclic dependency
         member_handler = self.hs.get_room_member_handler()
         yield member_handler.send_membership_event(None, event, context)
 
@@ -2594,7 +2554,12 @@ class FederationHandler(BaseHandler):
                 original_invite_id, allow_none=True
             )
         if original_invite:
-            display_name = original_invite.content["display_name"]
+            # If the m.room.third_party_invite event's content is empty, it means the
+            # invite has been revoked. In this case, we don't have to raise an error here
+            # because the auth check will fail on the invite (because it's not able to
+            # fetch public keys from the m.room.third_party_invite event's content, which
+            # is empty).
+            display_name = original_invite.content.get("display_name")
             event_dict["content"]["third_party_invite"]["display_name"] = display_name
         else:
             logger.info(
@@ -2739,7 +2704,7 @@ class FederationHandler(BaseHandler):
                 backfilled=backfilled,
             )
         else:
-            max_stream_id = yield self.store.persist_events(
+            max_stream_id = yield self.storage.persistence.persist_events(
                 event_and_contexts, backfilled=backfilled
             )
 
diff --git a/synapse/handlers/groups_local.py b/synapse/handlers/groups_local.py
index 46eb9ee88b..92fecbfc44 100644
--- a/synapse/handlers/groups_local.py
+++ b/synapse/handlers/groups_local.py
@@ -392,7 +392,7 @@ class GroupsLocalHandler(object):
         try:
             user_profile = yield self.profile_handler.get_profile(user_id)
         except Exception as e:
-            logger.warn("No profile for user %s: %s", user_id, e)
+            logger.warning("No profile for user %s: %s", user_id, e)
             user_profile = {}
 
         return {"state": "invite", "user_profile": user_profile}
diff --git a/synapse/handlers/identity.py b/synapse/handlers/identity.py
index d199521b58..000fbf090f 100644
--- a/synapse/handlers/identity.py
+++ b/synapse/handlers/identity.py
@@ -18,111 +18,150 @@
 """Utilities for interacting with Identity Servers"""
 
 import logging
+import urllib
 
 from canonicaljson import json
+from signedjson.key import decode_verify_key_bytes
+from signedjson.sign import verify_signed_json
+from unpaddedbase64 import decode_base64
 
 from twisted.internet import defer
+from twisted.internet.error import TimeoutError
 
 from synapse.api.errors import (
+    AuthError,
     CodeMessageException,
     Codes,
     HttpResponseException,
     SynapseError,
 )
+from synapse.config.emailconfig import ThreepidBehaviour
+from synapse.http.client import SimpleHttpClient
+from synapse.util.hash import sha256_and_url_safe_base64
+from synapse.util.stringutils import random_string
 
 from ._base import BaseHandler
 
 logger = logging.getLogger(__name__)
 
+id_server_scheme = "https://"
+
 
 class IdentityHandler(BaseHandler):
     def __init__(self, hs):
         super(IdentityHandler, self).__init__(hs)
 
-        self.http_client = hs.get_simple_http_client()
-        self.federation_http_client = hs.get_http_client()
-
-        self.trusted_id_servers = set(hs.config.trusted_third_party_id_servers)
-        self.trust_any_id_server_just_for_testing_do_not_use = (
-            hs.config.use_insecure_ssl_client_just_for_testing_do_not_use
+        self.http_client = SimpleHttpClient(hs)
+        # We create a blacklisting instance of SimpleHttpClient for contacting identity
+        # servers specified by clients
+        self.blacklisting_http_client = SimpleHttpClient(
+            hs, ip_blacklist=hs.config.federation_ip_range_blacklist
         )
-
-    def _should_trust_id_server(self, id_server):
-        if id_server not in self.trusted_id_servers:
-            if self.trust_any_id_server_just_for_testing_do_not_use:
-                logger.warn(
-                    "Trusting untrustworthy ID server %r even though it isn't"
-                    " in the trusted id list for testing because"
-                    " 'use_insecure_ssl_client_just_for_testing_do_not_use'"
-                    " is set in the config",
-                    id_server,
-                )
-            else:
-                return False
-        return True
+        self.federation_http_client = hs.get_http_client()
+        self.hs = hs
 
     @defer.inlineCallbacks
-    def threepid_from_creds(self, creds):
-        if "id_server" in creds:
-            id_server = creds["id_server"]
-        elif "idServer" in creds:
-            id_server = creds["idServer"]
-        else:
-            raise SynapseError(400, "No id_server in creds")
+    def threepid_from_creds(self, id_server, creds):
+        """
+        Retrieve and validate a threepid identifier from a "credentials" dictionary against a
+        given identity server
 
-        if "client_secret" in creds:
-            client_secret = creds["client_secret"]
-        elif "clientSecret" in creds:
-            client_secret = creds["clientSecret"]
-        else:
-            raise SynapseError(400, "No client_secret in creds")
+        Args:
+            id_server (str): The identity server to validate 3PIDs against. Must be a
+                complete URL including the protocol (http(s)://)
 
-        if not self._should_trust_id_server(id_server):
-            logger.warn(
-                "%s is not a trusted ID server: rejecting 3pid " + "credentials",
+            creds (dict[str, str]): Dictionary containing the following keys:
+                * client_secret|clientSecret: A unique secret str provided by the client
+                * sid: The ID of the validation session
+
+        Returns:
+            Deferred[dict[str,str|int]|None]: A dictionary consisting of response params to
+                the /getValidated3pid endpoint of the Identity Service API, or None if the
+                threepid was not found
+        """
+        client_secret = creds.get("client_secret") or creds.get("clientSecret")
+        if not client_secret:
+            raise SynapseError(
+                400, "Missing param client_secret in creds", errcode=Codes.MISSING_PARAM
+            )
+        session_id = creds.get("sid")
+        if not session_id:
+            raise SynapseError(
+                400, "Missing param session_id in creds", errcode=Codes.MISSING_PARAM
+            )
+
+        query_params = {"sid": session_id, "client_secret": client_secret}
+
+        url = id_server + "/_matrix/identity/api/v1/3pid/getValidated3pid"
+
+        try:
+            data = yield self.http_client.get_json(url, query_params)
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
+        except HttpResponseException as e:
+            logger.info(
+                "%s returned %i for threepid validation for: %s",
                 id_server,
+                e.code,
+                creds,
             )
             return None
 
-        try:
-            data = yield self.http_client.get_json(
-                "https://%s%s"
-                % (id_server, "/_matrix/identity/api/v1/3pid/getValidated3pid"),
-                {"sid": creds["sid"], "client_secret": client_secret},
-            )
-        except HttpResponseException as e:
-            logger.info("getValidated3pid failed with Matrix error: %r", e)
-            raise e.to_synapse_error()
-
+        # Old versions of Sydent return a 200 http code even on a failed validation
+        # check. Thus, in addition to the HttpResponseException check above (which
+        # checks for non-200 errors), we need to make sure validation_session isn't
+        # actually an error, identified by the absence of a "medium" key
+        # See https://github.com/matrix-org/sydent/issues/215 for details
         if "medium" in data:
             return data
+
+        logger.info("%s reported non-validated threepid: %s", id_server, creds)
         return None
 
     @defer.inlineCallbacks
-    def bind_threepid(self, creds, mxid):
-        logger.debug("binding threepid %r to %s", creds, mxid)
-        data = None
+    def bind_threepid(
+        self, client_secret, sid, mxid, id_server, id_access_token=None, use_v2=True
+    ):
+        """Bind a 3PID to an identity server
 
-        if "id_server" in creds:
-            id_server = creds["id_server"]
-        elif "idServer" in creds:
-            id_server = creds["idServer"]
-        else:
-            raise SynapseError(400, "No id_server in creds")
+        Args:
+            client_secret (str): A unique secret provided by the client
 
-        if "client_secret" in creds:
-            client_secret = creds["client_secret"]
-        elif "clientSecret" in creds:
-            client_secret = creds["clientSecret"]
+            sid (str): The ID of the validation session
+
+            mxid (str): The MXID to bind the 3PID to
+
+            id_server (str): The domain of the identity server to query
+
+            id_access_token (str): The access token to authenticate to the identity
+                server with, if necessary. Required if use_v2 is true
+
+            use_v2 (bool): Whether to use v2 Identity Service API endpoints. Defaults to True
+
+        Returns:
+            Deferred[dict]: The response from the identity server
+        """
+        logger.debug("Proxying threepid bind request for %s to %s", mxid, id_server)
+
+        # If an id_access_token is not supplied, force usage of v1
+        if id_access_token is None:
+            use_v2 = False
+
+        # Decide which API endpoint URLs to use
+        headers = {}
+        bind_data = {"sid": sid, "client_secret": client_secret, "mxid": mxid}
+        if use_v2:
+            bind_url = "https://%s/_matrix/identity/v2/3pid/bind" % (id_server,)
+            headers["Authorization"] = create_id_access_token_header(id_access_token)
         else:
-            raise SynapseError(400, "No client_secret in creds")
+            bind_url = "https://%s/_matrix/identity/api/v1/3pid/bind" % (id_server,)
 
         try:
-            data = yield self.http_client.post_json_get_json(
-                "https://%s%s" % (id_server, "/_matrix/identity/api/v1/3pid/bind"),
-                {"sid": creds["sid"], "client_secret": client_secret, "mxid": mxid},
+            # Use the blacklisting http client as this call is only to identity servers
+            # provided by a client
+            data = yield self.blacklisting_http_client.post_json_get_json(
+                bind_url, bind_data, headers=headers
             )
-            logger.debug("bound threepid %r to %s", creds, mxid)
 
             # Remember where we bound the threepid
             yield self.store.add_user_bound_threepid(
@@ -131,13 +170,28 @@ class IdentityHandler(BaseHandler):
                 address=data["address"],
                 id_server=id_server,
             )
+
+            return data
+        except HttpResponseException as e:
+            if e.code != 404 or not use_v2:
+                logger.error("3PID bind failed with Matrix error: %r", e)
+                raise e.to_synapse_error()
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
         except CodeMessageException as e:
             data = json.loads(e.msg)  # XXX WAT?
-        return data
+            return data
+
+        logger.info("Got 404 when POSTing JSON %s, falling back to v1 URL", bind_url)
+        res = yield self.bind_threepid(
+            client_secret, sid, mxid, id_server, id_access_token, use_v2=False
+        )
+        return res
 
     @defer.inlineCallbacks
     def try_unbind_threepid(self, mxid, threepid):
-        """Removes a binding from an identity server
+        """Attempt to remove a 3PID from an identity server, or if one is not provided, all
+        identity servers we're aware the binding is present on
 
         Args:
             mxid (str): Matrix user ID of binding to be removed
@@ -188,6 +242,8 @@ class IdentityHandler(BaseHandler):
             server doesn't support unbinding
         """
         url = "https://%s/_matrix/identity/api/v1/3pid/unbind" % (id_server,)
+        url_bytes = "/_matrix/identity/api/v1/3pid/unbind".encode("ascii")
+
         content = {
             "mxid": mxid,
             "threepid": {"medium": threepid["medium"], "address": threepid["address"]},
@@ -199,23 +255,29 @@ class IdentityHandler(BaseHandler):
         auth_headers = self.federation_http_client.build_auth_headers(
             destination=None,
             method="POST",
-            url_bytes="/_matrix/identity/api/v1/3pid/unbind".encode("ascii"),
+            url_bytes=url_bytes,
             content=content,
             destination_is=id_server,
         )
         headers = {b"Authorization": auth_headers}
 
         try:
-            yield self.http_client.post_json_get_json(url, content, headers)
+            # Use the blacklisting http client as this call is only to identity servers
+            # provided by a client
+            yield self.blacklisting_http_client.post_json_get_json(
+                url, content, headers
+            )
             changed = True
         except HttpResponseException as e:
             changed = False
             if e.code in (400, 404, 501):
                 # The remote server probably doesn't support unbinding (yet)
-                logger.warn("Received %d response while unbinding threepid", e.code)
+                logger.warning("Received %d response while unbinding threepid", e.code)
             else:
                 logger.error("Failed to unbind threepid on identity server: %s", e)
-                raise SynapseError(502, "Failed to contact identity server")
+                raise SynapseError(500, "Failed to contact identity server")
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
 
         yield self.store.remove_user_bound_threepid(
             user_id=mxid,
@@ -226,59 +288,657 @@ class IdentityHandler(BaseHandler):
 
         return changed
 
+    @defer.inlineCallbacks
+    def send_threepid_validation(
+        self,
+        email_address,
+        client_secret,
+        send_attempt,
+        send_email_func,
+        next_link=None,
+    ):
+        """Send a threepid validation email for password reset or
+        registration purposes
+
+        Args:
+            email_address (str): The user's email address
+            client_secret (str): The provided client secret
+            send_attempt (int): Which send attempt this is
+            send_email_func (func): A function that takes an email address, token,
+                                    client_secret and session_id, sends an email
+                                    and returns a Deferred.
+            next_link (str|None): The URL to redirect the user to after validation
+
+        Returns:
+            The new session_id upon success
+
+        Raises:
+            SynapseError is an error occurred when sending the email
+        """
+        # Check that this email/client_secret/send_attempt combo is new or
+        # greater than what we've seen previously
+        session = yield self.store.get_threepid_validation_session(
+            "email", client_secret, address=email_address, validated=False
+        )
+
+        # Check to see if a session already exists and that it is not yet
+        # marked as validated
+        if session and session.get("validated_at") is None:
+            session_id = session["session_id"]
+            last_send_attempt = session["last_send_attempt"]
+
+            # Check that the send_attempt is higher than previous attempts
+            if send_attempt <= last_send_attempt:
+                # If not, just return a success without sending an email
+                return session_id
+        else:
+            # An non-validated session does not exist yet.
+            # Generate a session id
+            session_id = random_string(16)
+
+        if next_link:
+            # Manipulate the next_link to add the sid, because the caller won't get
+            # it until we send a response, by which time we've sent the mail.
+            if "?" in next_link:
+                next_link += "&"
+            else:
+                next_link += "?"
+            next_link += "sid=" + urllib.parse.quote(session_id)
+
+        # Generate a new validation token
+        token = random_string(32)
+
+        # Send the mail with the link containing the token, client_secret
+        # and session_id
+        try:
+            yield send_email_func(email_address, token, client_secret, session_id)
+        except Exception:
+            logger.exception(
+                "Error sending threepid validation email to %s", email_address
+            )
+            raise SynapseError(500, "An error was encountered when sending the email")
+
+        token_expires = (
+            self.hs.clock.time_msec() + self.hs.config.email_validation_token_lifetime
+        )
+
+        yield self.store.start_or_continue_validation_session(
+            "email",
+            email_address,
+            session_id,
+            client_secret,
+            send_attempt,
+            next_link,
+            token,
+            token_expires,
+        )
+
+        return session_id
+
     @defer.inlineCallbacks
     def requestEmailToken(
         self, id_server, email, client_secret, send_attempt, next_link=None
     ):
-        if not self._should_trust_id_server(id_server):
-            raise SynapseError(
-                400, "Untrusted ID server '%s'" % id_server, Codes.SERVER_NOT_TRUSTED
-            )
+        """
+        Request an external server send an email on our behalf for the purposes of threepid
+        validation.
 
+        Args:
+            id_server (str): The identity server to proxy to
+            email (str): The email to send the message to
+            client_secret (str): The unique client_secret sends by the user
+            send_attempt (int): Which attempt this is
+            next_link: A link to redirect the user to once they submit the token
+
+        Returns:
+            The json response body from the server
+        """
         params = {
             "email": email,
             "client_secret": client_secret,
             "send_attempt": send_attempt,
         }
-
         if next_link:
-            params.update({"next_link": next_link})
+            params["next_link"] = next_link
+
+        if self.hs.config.using_identity_server_from_trusted_list:
+            # Warn that a deprecated config option is in use
+            logger.warning(
+                'The config option "trust_identity_server_for_password_resets" '
+                'has been replaced by "account_threepid_delegate". '
+                "Please consult the sample config at docs/sample_config.yaml for "
+                "details and update your config file."
+            )
 
         try:
             data = yield self.http_client.post_json_get_json(
-                "https://%s%s"
-                % (id_server, "/_matrix/identity/api/v1/validate/email/requestToken"),
+                id_server + "/_matrix/identity/api/v1/validate/email/requestToken",
                 params,
             )
             return data
         except HttpResponseException as e:
             logger.info("Proxied requestToken failed: %r", e)
             raise e.to_synapse_error()
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
 
     @defer.inlineCallbacks
     def requestMsisdnToken(
-        self, id_server, country, phone_number, client_secret, send_attempt, **kwargs
+        self,
+        id_server,
+        country,
+        phone_number,
+        client_secret,
+        send_attempt,
+        next_link=None,
     ):
-        if not self._should_trust_id_server(id_server):
-            raise SynapseError(
-                400, "Untrusted ID server '%s'" % id_server, Codes.SERVER_NOT_TRUSTED
-            )
+        """
+        Request an external server send an SMS message on our behalf for the purposes of
+        threepid validation.
+        Args:
+            id_server (str): The identity server to proxy to
+            country (str): The country code of the phone number
+            phone_number (str): The number to send the message to
+            client_secret (str): The unique client_secret sends by the user
+            send_attempt (int): Which attempt this is
+            next_link: A link to redirect the user to once they submit the token
 
+        Returns:
+            The json response body from the server
+        """
         params = {
             "country": country,
             "phone_number": phone_number,
             "client_secret": client_secret,
             "send_attempt": send_attempt,
         }
-        params.update(kwargs)
+        if next_link:
+            params["next_link"] = next_link
+
+        if self.hs.config.using_identity_server_from_trusted_list:
+            # Warn that a deprecated config option is in use
+            logger.warning(
+                'The config option "trust_identity_server_for_password_resets" '
+                'has been replaced by "account_threepid_delegate". '
+                "Please consult the sample config at docs/sample_config.yaml for "
+                "details and update your config file."
+            )
 
         try:
             data = yield self.http_client.post_json_get_json(
-                "https://%s%s"
-                % (id_server, "/_matrix/identity/api/v1/validate/msisdn/requestToken"),
+                id_server + "/_matrix/identity/api/v1/validate/msisdn/requestToken",
                 params,
             )
-            return data
         except HttpResponseException as e:
             logger.info("Proxied requestToken failed: %r", e)
             raise e.to_synapse_error()
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
+
+        assert self.hs.config.public_baseurl
+
+        # we need to tell the client to send the token back to us, since it doesn't
+        # otherwise know where to send it, so add submit_url response parameter
+        # (see also MSC2078)
+        data["submit_url"] = (
+            self.hs.config.public_baseurl
+            + "_matrix/client/unstable/add_threepid/msisdn/submit_token"
+        )
+        return data
+
+    @defer.inlineCallbacks
+    def validate_threepid_session(self, client_secret, sid):
+        """Validates a threepid session with only the client secret and session ID
+        Tries validating against any configured account_threepid_delegates as well as locally.
+
+        Args:
+            client_secret (str): A secret provided by the client
+
+            sid (str): The ID of the session
+
+        Returns:
+            Dict[str, str|int] if validation was successful, otherwise None
+        """
+        # XXX: We shouldn't need to keep wrapping and unwrapping this value
+        threepid_creds = {"client_secret": client_secret, "sid": sid}
+
+        # We don't actually know which medium this 3PID is. Thus we first assume it's email,
+        # and if validation fails we try msisdn
+        validation_session = None
+
+        # Try to validate as email
+        if self.hs.config.threepid_behaviour_email == ThreepidBehaviour.REMOTE:
+            # Ask our delegated email identity server
+            validation_session = yield self.threepid_from_creds(
+                self.hs.config.account_threepid_delegate_email, threepid_creds
+            )
+        elif self.hs.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            # Get a validated session matching these details
+            validation_session = yield self.store.get_threepid_validation_session(
+                "email", client_secret, sid=sid, validated=True
+            )
+
+        if validation_session:
+            return validation_session
+
+        # Try to validate as msisdn
+        if self.hs.config.account_threepid_delegate_msisdn:
+            # Ask our delegated msisdn identity server
+            validation_session = yield self.threepid_from_creds(
+                self.hs.config.account_threepid_delegate_msisdn, threepid_creds
+            )
+
+        return validation_session
+
+    @defer.inlineCallbacks
+    def proxy_msisdn_submit_token(self, id_server, client_secret, sid, token):
+        """Proxy a POST submitToken request to an identity server for verification purposes
+
+        Args:
+            id_server (str): The identity server URL to contact
+
+            client_secret (str): Secret provided by the client
+
+            sid (str): The ID of the session
+
+            token (str): The verification token
+
+        Raises:
+            SynapseError: If we failed to contact the identity server
+
+        Returns:
+            Deferred[dict]: The response dict from the identity server
+        """
+        body = {"client_secret": client_secret, "sid": sid, "token": token}
+
+        try:
+            return (
+                yield self.http_client.post_json_get_json(
+                    id_server + "/_matrix/identity/api/v1/validate/msisdn/submitToken",
+                    body,
+                )
+            )
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
+        except HttpResponseException as e:
+            logger.warning("Error contacting msisdn account_threepid_delegate: %s", e)
+            raise SynapseError(400, "Error contacting the identity server")
+
+    @defer.inlineCallbacks
+    def lookup_3pid(self, id_server, medium, address, id_access_token=None):
+        """Looks up a 3pid in the passed identity server.
+
+        Args:
+            id_server (str): The server name (including port, if required)
+                of the identity server to use.
+            medium (str): The type of the third party identifier (e.g. "email").
+            address (str): The third party identifier (e.g. "foo@example.com").
+            id_access_token (str|None): The access token to authenticate to the identity
+                server with
+
+        Returns:
+            str|None: the matrix ID of the 3pid, or None if it is not recognized.
+        """
+        if id_access_token is not None:
+            try:
+                results = yield self._lookup_3pid_v2(
+                    id_server, id_access_token, medium, address
+                )
+                return results
+
+            except Exception as e:
+                # Catch HttpResponseExcept for a non-200 response code
+                # Check if this identity server does not know about v2 lookups
+                if isinstance(e, HttpResponseException) and e.code == 404:
+                    # This is an old identity server that does not yet support v2 lookups
+                    logger.warning(
+                        "Attempted v2 lookup on v1 identity server %s. Falling "
+                        "back to v1",
+                        id_server,
+                    )
+                else:
+                    logger.warning("Error when looking up hashing details: %s", e)
+                    return None
+
+        return (yield self._lookup_3pid_v1(id_server, medium, address))
+
+    @defer.inlineCallbacks
+    def _lookup_3pid_v1(self, id_server, medium, address):
+        """Looks up a 3pid in the passed identity server using v1 lookup.
+
+        Args:
+            id_server (str): The server name (including port, if required)
+                of the identity server to use.
+            medium (str): The type of the third party identifier (e.g. "email").
+            address (str): The third party identifier (e.g. "foo@example.com").
+
+        Returns:
+            str: the matrix ID of the 3pid, or None if it is not recognized.
+        """
+        try:
+            data = yield self.blacklisting_http_client.get_json(
+                "%s%s/_matrix/identity/api/v1/lookup" % (id_server_scheme, id_server),
+                {"medium": medium, "address": address},
+            )
+
+            if "mxid" in data:
+                if "signatures" not in data:
+                    raise AuthError(401, "No signatures on 3pid binding")
+                yield self._verify_any_signature(data, id_server)
+                return data["mxid"]
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
+        except IOError as e:
+            logger.warning("Error from v1 identity server lookup: %s" % (e,))
+
+        return None
+
+    @defer.inlineCallbacks
+    def _lookup_3pid_v2(self, id_server, id_access_token, medium, address):
+        """Looks up a 3pid in the passed identity server using v2 lookup.
+
+        Args:
+            id_server (str): The server name (including port, if required)
+                of the identity server to use.
+            id_access_token (str): The access token to authenticate to the identity server with
+            medium (str): The type of the third party identifier (e.g. "email").
+            address (str): The third party identifier (e.g. "foo@example.com").
+
+        Returns:
+            Deferred[str|None]: the matrix ID of the 3pid, or None if it is not recognised.
+        """
+        # Check what hashing details are supported by this identity server
+        try:
+            hash_details = yield self.blacklisting_http_client.get_json(
+                "%s%s/_matrix/identity/v2/hash_details" % (id_server_scheme, id_server),
+                {"access_token": id_access_token},
+            )
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
+
+        if not isinstance(hash_details, dict):
+            logger.warning(
+                "Got non-dict object when checking hash details of %s%s: %s",
+                id_server_scheme,
+                id_server,
+                hash_details,
+            )
+            raise SynapseError(
+                400,
+                "Non-dict object from %s%s during v2 hash_details request: %s"
+                % (id_server_scheme, id_server, hash_details),
+            )
+
+        # Extract information from hash_details
+        supported_lookup_algorithms = hash_details.get("algorithms")
+        lookup_pepper = hash_details.get("lookup_pepper")
+        if (
+            not supported_lookup_algorithms
+            or not isinstance(supported_lookup_algorithms, list)
+            or not lookup_pepper
+            or not isinstance(lookup_pepper, str)
+        ):
+            raise SynapseError(
+                400,
+                "Invalid hash details received from identity server %s%s: %s"
+                % (id_server_scheme, id_server, hash_details),
+            )
+
+        # Check if any of the supported lookup algorithms are present
+        if LookupAlgorithm.SHA256 in supported_lookup_algorithms:
+            # Perform a hashed lookup
+            lookup_algorithm = LookupAlgorithm.SHA256
+
+            # Hash address, medium and the pepper with sha256
+            to_hash = "%s %s %s" % (address, medium, lookup_pepper)
+            lookup_value = sha256_and_url_safe_base64(to_hash)
+
+        elif LookupAlgorithm.NONE in supported_lookup_algorithms:
+            # Perform a non-hashed lookup
+            lookup_algorithm = LookupAlgorithm.NONE
+
+            # Combine together plaintext address and medium
+            lookup_value = "%s %s" % (address, medium)
+
+        else:
+            logger.warning(
+                "None of the provided lookup algorithms of %s are supported: %s",
+                id_server,
+                supported_lookup_algorithms,
+            )
+            raise SynapseError(
+                400,
+                "Provided identity server does not support any v2 lookup "
+                "algorithms that this homeserver supports.",
+            )
+
+        # Authenticate with identity server given the access token from the client
+        headers = {"Authorization": create_id_access_token_header(id_access_token)}
+
+        try:
+            lookup_results = yield self.blacklisting_http_client.post_json_get_json(
+                "%s%s/_matrix/identity/v2/lookup" % (id_server_scheme, id_server),
+                {
+                    "addresses": [lookup_value],
+                    "algorithm": lookup_algorithm,
+                    "pepper": lookup_pepper,
+                },
+                headers=headers,
+            )
+        except TimeoutError:
+            raise SynapseError(500, "Timed out contacting identity server")
+        except Exception as e:
+            logger.warning("Error when performing a v2 3pid lookup: %s", e)
+            raise SynapseError(
+                500, "Unknown error occurred during identity server lookup"
+            )
+
+        # Check for a mapping from what we looked up to an MXID
+        if "mappings" not in lookup_results or not isinstance(
+            lookup_results["mappings"], dict
+        ):
+            logger.warning("No results from 3pid lookup")
+            return None
+
+        # Return the MXID if it's available, or None otherwise
+        mxid = lookup_results["mappings"].get(lookup_value)
+        return mxid
+
+    @defer.inlineCallbacks
+    def _verify_any_signature(self, data, server_hostname):
+        if server_hostname not in data["signatures"]:
+            raise AuthError(401, "No signature from server %s" % (server_hostname,))
+        for key_name, signature in data["signatures"][server_hostname].items():
+            try:
+                key_data = yield self.blacklisting_http_client.get_json(
+                    "%s%s/_matrix/identity/api/v1/pubkey/%s"
+                    % (id_server_scheme, server_hostname, key_name)
+                )
+            except TimeoutError:
+                raise SynapseError(500, "Timed out contacting identity server")
+            if "public_key" not in key_data:
+                raise AuthError(
+                    401, "No public key named %s from %s" % (key_name, server_hostname)
+                )
+            verify_signed_json(
+                data,
+                server_hostname,
+                decode_verify_key_bytes(
+                    key_name, decode_base64(key_data["public_key"])
+                ),
+            )
+            return
+
+    @defer.inlineCallbacks
+    def ask_id_server_for_third_party_invite(
+        self,
+        requester,
+        id_server,
+        medium,
+        address,
+        room_id,
+        inviter_user_id,
+        room_alias,
+        room_avatar_url,
+        room_join_rules,
+        room_name,
+        inviter_display_name,
+        inviter_avatar_url,
+        id_access_token=None,
+    ):
+        """
+        Asks an identity server for a third party invite.
+
+        Args:
+            requester (Requester)
+            id_server (str): hostname + optional port for the identity server.
+            medium (str): The literal string "email".
+            address (str): The third party address being invited.
+            room_id (str): The ID of the room to which the user is invited.
+            inviter_user_id (str): The user ID of the inviter.
+            room_alias (str): An alias for the room, for cosmetic notifications.
+            room_avatar_url (str): The URL of the room's avatar, for cosmetic
+                notifications.
+            room_join_rules (str): The join rules of the email (e.g. "public").
+            room_name (str): The m.room.name of the room.
+            inviter_display_name (str): The current display name of the
+                inviter.
+            inviter_avatar_url (str): The URL of the inviter's avatar.
+            id_access_token (str|None): The access token to authenticate to the identity
+                server with
+
+        Returns:
+            A deferred tuple containing:
+                token (str): The token which must be signed to prove authenticity.
+                public_keys ([{"public_key": str, "key_validity_url": str}]):
+                    public_key is a base64-encoded ed25519 public key.
+                fallback_public_key: One element from public_keys.
+                display_name (str): A user-friendly name to represent the invited
+                    user.
+        """
+        invite_config = {
+            "medium": medium,
+            "address": address,
+            "room_id": room_id,
+            "room_alias": room_alias,
+            "room_avatar_url": room_avatar_url,
+            "room_join_rules": room_join_rules,
+            "room_name": room_name,
+            "sender": inviter_user_id,
+            "sender_display_name": inviter_display_name,
+            "sender_avatar_url": inviter_avatar_url,
+        }
+
+        # Add the identity service access token to the JSON body and use the v2
+        # Identity Service endpoints if id_access_token is present
+        data = None
+        base_url = "%s%s/_matrix/identity" % (id_server_scheme, id_server)
+
+        if id_access_token:
+            key_validity_url = "%s%s/_matrix/identity/v2/pubkey/isvalid" % (
+                id_server_scheme,
+                id_server,
+            )
+
+            # Attempt a v2 lookup
+            url = base_url + "/v2/store-invite"
+            try:
+                data = yield self.blacklisting_http_client.post_json_get_json(
+                    url,
+                    invite_config,
+                    {"Authorization": create_id_access_token_header(id_access_token)},
+                )
+            except TimeoutError:
+                raise SynapseError(500, "Timed out contacting identity server")
+            except HttpResponseException as e:
+                if e.code != 404:
+                    logger.info("Failed to POST %s with JSON: %s", url, e)
+                    raise e
+
+        if data is None:
+            key_validity_url = "%s%s/_matrix/identity/api/v1/pubkey/isvalid" % (
+                id_server_scheme,
+                id_server,
+            )
+            url = base_url + "/api/v1/store-invite"
+
+            try:
+                data = yield self.blacklisting_http_client.post_json_get_json(
+                    url, invite_config
+                )
+            except TimeoutError:
+                raise SynapseError(500, "Timed out contacting identity server")
+            except HttpResponseException as e:
+                logger.warning(
+                    "Error trying to call /store-invite on %s%s: %s",
+                    id_server_scheme,
+                    id_server,
+                    e,
+                )
+
+            if data is None:
+                # Some identity servers may only support application/x-www-form-urlencoded
+                # types. This is especially true with old instances of Sydent, see
+                # https://github.com/matrix-org/sydent/pull/170
+                try:
+                    data = yield self.blacklisting_http_client.post_urlencoded_get_json(
+                        url, invite_config
+                    )
+                except HttpResponseException as e:
+                    logger.warning(
+                        "Error calling /store-invite on %s%s with fallback "
+                        "encoding: %s",
+                        id_server_scheme,
+                        id_server,
+                        e,
+                    )
+                    raise e
+
+        # TODO: Check for success
+        token = data["token"]
+        public_keys = data.get("public_keys", [])
+        if "public_key" in data:
+            fallback_public_key = {
+                "public_key": data["public_key"],
+                "key_validity_url": key_validity_url,
+            }
+        else:
+            fallback_public_key = public_keys[0]
+
+        if not public_keys:
+            public_keys.append(fallback_public_key)
+        display_name = data["display_name"]
+        return token, public_keys, fallback_public_key, display_name
+
+
+def create_id_access_token_header(id_access_token):
+    """Create an Authorization header for passing to SimpleHttpClient as the header value
+    of an HTTP request.
+
+    Args:
+        id_access_token (str): An identity server access token.
+
+    Returns:
+        list[str]: The ascii-encoded bearer token encased in a list.
+    """
+    # Prefix with Bearer
+    bearer_token = "Bearer %s" % id_access_token
+
+    # Encode headers to standard ascii
+    bearer_token.encode("ascii")
+
+    # Return as a list as that's how SimpleHttpClient takes header values
+    return [bearer_token]
+
+
+class LookupAlgorithm:
+    """
+    Supported hashing algorithms when performing a 3PID lookup.
+
+    SHA256 - Hashing an (address, medium, pepper) combo with sha256, then url-safe base64
+        encoding
+    NONE - Not performing any hashing. Simply sending an (address, medium) combo in plaintext
+    """
+
+    SHA256 = "sha256"
+    NONE = "none"
diff --git a/synapse/handlers/initial_sync.py b/synapse/handlers/initial_sync.py
index 595f75400b..81dce96f4b 100644
--- a/synapse/handlers/initial_sync.py
+++ b/synapse/handlers/initial_sync.py
@@ -43,6 +43,8 @@ class InitialSyncHandler(BaseHandler):
         self.validator = EventValidator()
         self.snapshot_cache = SnapshotCache()
         self._event_serializer = hs.get_event_client_serializer()
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
 
     def snapshot_all_rooms(
         self,
@@ -126,8 +128,8 @@ class InitialSyncHandler(BaseHandler):
 
         tags_by_room = yield self.store.get_tags_for_user(user_id)
 
-        account_data, account_data_by_room = (
-            yield self.store.get_account_data_for_user(user_id)
+        account_data, account_data_by_room = yield self.store.get_account_data_for_user(
+            user_id
         )
 
         public_room_ids = yield self.store.get_public_room_ids()
@@ -169,7 +171,7 @@ class InitialSyncHandler(BaseHandler):
                 elif event.membership == Membership.LEAVE:
                     room_end_token = "s%d" % (event.stream_ordering,)
                     deferred_room_state = run_in_background(
-                        self.store.get_state_for_events, [event.event_id]
+                        self.state_store.get_state_for_events, [event.event_id]
                     )
                     deferred_room_state.addCallback(
                         lambda states: states[event.event_id]
@@ -189,7 +191,9 @@ class InitialSyncHandler(BaseHandler):
                     )
                 ).addErrback(unwrapFirstError)
 
-                messages = yield filter_events_for_client(self.store, user_id, messages)
+                messages = yield filter_events_for_client(
+                    self.storage, user_id, messages
+                )
 
                 start_token = now_token.copy_and_replace("room_key", token)
                 end_token = now_token.copy_and_replace("room_key", room_end_token)
@@ -307,7 +311,7 @@ class InitialSyncHandler(BaseHandler):
     def _room_initial_sync_parted(
         self, user_id, room_id, pagin_config, membership, member_event_id, is_peeking
     ):
-        room_state = yield self.store.get_state_for_events([member_event_id])
+        room_state = yield self.state_store.get_state_for_events([member_event_id])
 
         room_state = room_state[member_event_id]
 
@@ -322,7 +326,7 @@ class InitialSyncHandler(BaseHandler):
         )
 
         messages = yield filter_events_for_client(
-            self.store, user_id, messages, is_peeking=is_peeking
+            self.storage, user_id, messages, is_peeking=is_peeking
         )
 
         start_token = StreamToken.START.copy_and_replace("room_key", token)
@@ -414,7 +418,7 @@ class InitialSyncHandler(BaseHandler):
         )
 
         messages = yield filter_events_for_client(
-            self.store, user_id, messages, is_peeking=is_peeking
+            self.storage, user_id, messages, is_peeking=is_peeking
         )
 
         start_token = now_token.copy_and_replace("room_key", token)
@@ -450,7 +454,6 @@ class InitialSyncHandler(BaseHandler):
             # else it will throw.
             member_event = yield self.auth.check_user_was_in_room(room_id, user_id)
             return member_event.membership, member_event.event_id
-            return
         except AuthError:
             visibility = yield self.state_handler.get_current_state(
                 room_id, EventTypes.RoomHistoryVisibility, ""
@@ -460,7 +463,6 @@ class InitialSyncHandler(BaseHandler):
                 and visibility.content["history_visibility"] == "world_readable"
             ):
                 return Membership.JOIN, None
-                return
             raise AuthError(
                 403, "Guest access not allowed", errcode=Codes.GUEST_ACCESS_FORBIDDEN
             )
diff --git a/synapse/handlers/message.py b/synapse/handlers/message.py
index 111f7c7e2f..d682dc2b7a 100644
--- a/synapse/handlers/message.py
+++ b/synapse/handlers/message.py
@@ -59,6 +59,8 @@ class MessageHandler(object):
         self.clock = hs.get_clock()
         self.state = hs.get_state_handler()
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
         self._event_serializer = hs.get_event_client_serializer()
 
     @defer.inlineCallbacks
@@ -74,15 +76,16 @@ class MessageHandler(object):
         Raises:
             SynapseError if something went wrong.
         """
-        membership, membership_event_id = yield self.auth.check_in_room_or_world_readable(
-            room_id, user_id
-        )
+        (
+            membership,
+            membership_event_id,
+        ) = yield self.auth.check_in_room_or_world_readable(room_id, user_id)
 
         if membership == Membership.JOIN:
             data = yield self.state.get_current_state(room_id, event_type, state_key)
         elif membership == Membership.LEAVE:
             key = (event_type, state_key)
-            room_state = yield self.store.get_state_for_events(
+            room_state = yield self.state_store.get_state_for_events(
                 [membership_event_id], StateFilter.from_types([key])
             )
             data = room_state[membership_event_id].get(key)
@@ -135,12 +138,12 @@ class MessageHandler(object):
                 raise NotFoundError("Can't find event for token %s" % (at_token,))
 
             visible_events = yield filter_events_for_client(
-                self.store, user_id, last_events
+                self.storage, user_id, last_events
             )
 
             event = last_events[0]
             if visible_events:
-                room_state = yield self.store.get_state_for_events(
+                room_state = yield self.state_store.get_state_for_events(
                     [event.event_id], state_filter=state_filter
                 )
                 room_state = room_state[event.event_id]
@@ -151,9 +154,10 @@ class MessageHandler(object):
                     % (user_id, room_id, at_token),
                 )
         else:
-            membership, membership_event_id = (
-                yield self.auth.check_in_room_or_world_readable(room_id, user_id)
-            )
+            (
+                membership,
+                membership_event_id,
+            ) = yield self.auth.check_in_room_or_world_readable(room_id, user_id)
 
             if membership == Membership.JOIN:
                 state_ids = yield self.store.get_filtered_current_state_ids(
@@ -161,7 +165,7 @@ class MessageHandler(object):
                 )
                 room_state = yield self.store.get_events(state_ids.values())
             elif membership == Membership.LEAVE:
-                room_state = yield self.store.get_state_for_events(
+                room_state = yield self.state_store.get_state_for_events(
                     [membership_event_id], state_filter=state_filter
                 )
                 room_state = room_state[membership_event_id]
@@ -222,11 +226,19 @@ class MessageHandler(object):
         }
 
 
+# The duration (in ms) after which rooms should be removed
+# `_rooms_to_exclude_from_dummy_event_insertion` (with the effect that we will try
+# to generate a dummy event for them once more)
+#
+_DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY = 7 * 24 * 60 * 60 * 1000
+
+
 class EventCreationHandler(object):
     def __init__(self, hs):
         self.hs = hs
         self.auth = hs.get_auth()
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
         self.state = hs.get_state_handler()
         self.clock = hs.get_clock()
         self.validator = EventValidator()
@@ -258,6 +270,13 @@ class EventCreationHandler(object):
             self.config.block_events_without_consent_error
         )
 
+        # Rooms which should be excluded from dummy insertion. (For instance,
+        # those without local users who can send events into the room).
+        #
+        # map from room id to time-of-last-attempt.
+        #
+        self._rooms_to_exclude_from_dummy_event_insertion = {}  # type: dict[str, int]
+
         # we need to construct a ConsentURIBuilder here, as it checks that the necessary
         # config options, but *only* if we have a configuration for which we are
         # going to need it.
@@ -673,7 +692,7 @@ class EventCreationHandler(object):
         try:
             yield self.auth.check_from_context(room_version, event, context)
         except AuthError as err:
-            logger.warn("Denying new event %r because %s", event, err)
+            logger.warning("Denying new event %r because %s", event, err)
             raise err
 
         # Ensure that we can round trip before trying to persist in db
@@ -729,7 +748,27 @@ class EventCreationHandler(object):
         assert not self.config.worker_app
 
         if ratelimit:
-            yield self.base_handler.ratelimit(requester)
+            # We check if this is a room admin redacting an event so that we
+            # can apply different ratelimiting. We do this by simply checking
+            # it's not a self-redaction (to avoid having to look up whether the
+            # user is actually admin or not).
+            is_admin_redaction = False
+            if event.type == EventTypes.Redaction:
+                original_event = yield self.store.get_event(
+                    event.redacts,
+                    check_redacted=False,
+                    get_prev_content=False,
+                    allow_rejected=False,
+                    allow_none=True,
+                )
+
+                is_admin_redaction = (
+                    original_event and event.sender != original_event.sender
+                )
+
+            yield self.base_handler.ratelimit(
+                requester, is_admin_redaction=is_admin_redaction
+            )
 
         yield self.base_handler.maybe_kick_guest_users(event, context)
 
@@ -834,7 +873,7 @@ class EventCreationHandler(object):
             if prev_state_ids:
                 raise AuthError(403, "Changing the room create event is forbidden")
 
-        (event_stream_id, max_stream_id) = yield self.store.persist_event(
+        event_stream_id, max_stream_id = yield self.storage.persistence.persist_event(
             event, context=context
         )
 
@@ -868,9 +907,11 @@ class EventCreationHandler(object):
         """Background task to send dummy events into rooms that have a large
         number of extremities
         """
-
+        self._expire_rooms_to_exclude_from_dummy_event_insertion()
         room_ids = yield self.store.get_rooms_with_many_extremities(
-            min_count=10, limit=5
+            min_count=10,
+            limit=5,
+            room_id_filter=self._rooms_to_exclude_from_dummy_event_insertion.keys(),
         )
 
         for room_id in room_ids:
@@ -884,32 +925,61 @@ class EventCreationHandler(object):
             members = yield self.state.get_current_users_in_room(
                 room_id, latest_event_ids=latest_event_ids
             )
+            dummy_event_sent = False
+            for user_id in members:
+                if not self.hs.is_mine_id(user_id):
+                    continue
+                requester = create_requester(user_id)
+                try:
+                    event, context = yield self.create_event(
+                        requester,
+                        {
+                            "type": "org.matrix.dummy_event",
+                            "content": {},
+                            "room_id": room_id,
+                            "sender": user_id,
+                        },
+                        prev_events_and_hashes=prev_events_and_hashes,
+                    )
 
-            user_id = None
-            for member in members:
-                if self.hs.is_mine_id(member):
-                    user_id = member
+                    event.internal_metadata.proactively_send = False
+
+                    yield self.send_nonmember_event(
+                        requester, event, context, ratelimit=False
+                    )
+                    dummy_event_sent = True
                     break
+                except ConsentNotGivenError:
+                    logger.info(
+                        "Failed to send dummy event into room %s for user %s due to "
+                        "lack of consent. Will try another user" % (room_id, user_id)
+                    )
+                except AuthError:
+                    logger.info(
+                        "Failed to send dummy event into room %s for user %s due to "
+                        "lack of power. Will try another user" % (room_id, user_id)
+                    )
 
-            if not user_id:
-                # We don't have a joined user.
-                # TODO: We should do something here to stop the room from
-                # appearing next time.
-                continue
+            if not dummy_event_sent:
+                # Did not find a valid user in the room, so remove from future attempts
+                # Exclusion is time limited, so the room will be rechecked in the future
+                # dependent on _DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY
+                logger.info(
+                    "Failed to send dummy event into room %s. Will exclude it from "
+                    "future attempts until cache expires" % (room_id,)
+                )
+                now = self.clock.time_msec()
+                self._rooms_to_exclude_from_dummy_event_insertion[room_id] = now
 
-            requester = create_requester(user_id)
-
-            event, context = yield self.create_event(
-                requester,
-                {
-                    "type": "org.matrix.dummy_event",
-                    "content": {},
-                    "room_id": room_id,
-                    "sender": user_id,
-                },
-                prev_events_and_hashes=prev_events_and_hashes,
+    def _expire_rooms_to_exclude_from_dummy_event_insertion(self):
+        expire_before = self.clock.time_msec() - _DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY
+        to_expire = set()
+        for room_id, time in self._rooms_to_exclude_from_dummy_event_insertion.items():
+            if time < expire_before:
+                to_expire.add(room_id)
+        for room_id in to_expire:
+            logger.debug(
+                "Expiring room id %s from dummy event insertion exclusion cache",
+                room_id,
             )
-
-            event.internal_metadata.proactively_send = False
-
-            yield self.send_nonmember_event(requester, event, context, ratelimit=False)
+            del self._rooms_to_exclude_from_dummy_event_insertion[room_id]
diff --git a/synapse/handlers/pagination.py b/synapse/handlers/pagination.py
index 5744f4579d..97f15a1c32 100644
--- a/synapse/handlers/pagination.py
+++ b/synapse/handlers/pagination.py
@@ -69,6 +69,8 @@ class PaginationHandler(object):
         self.hs = hs
         self.auth = hs.get_auth()
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
         self.clock = hs.get_clock()
         self._server_name = hs.hostname
 
@@ -210,9 +212,10 @@ class PaginationHandler(object):
         source_config = pagin_config.get_source_config("room")
 
         with (yield self.pagination_lock.read(room_id)):
-            membership, member_event_id = yield self.auth.check_in_room_or_world_readable(
-                room_id, user_id
-            )
+            (
+                membership,
+                member_event_id,
+            ) = yield self.auth.check_in_room_or_world_readable(room_id, user_id)
 
             if source_config.direction == "b":
                 # if we're going backwards, we might need to backfill. This
@@ -255,7 +258,7 @@ class PaginationHandler(object):
                 events = event_filter.filter(events)
 
             events = yield filter_events_for_client(
-                self.store, user_id, events, is_peeking=(member_event_id is None)
+                self.storage, user_id, events, is_peeking=(member_event_id is None)
             )
 
         if not events:
@@ -274,7 +277,7 @@ class PaginationHandler(object):
                 (EventTypes.Member, event.sender) for event in events
             )
 
-            state_ids = yield self.store.get_state_ids_for_event(
+            state_ids = yield self.state_store.get_state_ids_for_event(
                 events[0].event_id, state_filter=state_filter
             )
 
@@ -295,10 +298,8 @@ class PaginationHandler(object):
         }
 
         if state:
-            chunk["state"] = (
-                yield self._event_serializer.serialize_events(
-                    state, time_now, as_client_event=as_client_event
-                )
+            chunk["state"] = yield self._event_serializer.serialize_events(
+                state, time_now, as_client_event=as_client_event
             )
 
         return chunk
diff --git a/synapse/handlers/presence.py b/synapse/handlers/presence.py
index 8377a0ddc2..eda15bc623 100644
--- a/synapse/handlers/presence.py
+++ b/synapse/handlers/presence.py
@@ -24,6 +24,7 @@ The methods that define policy are:
 
 import logging
 from contextlib import contextmanager
+from typing import Dict, Set
 
 from six import iteritems, itervalues
 
@@ -179,8 +180,9 @@ class PresenceHandler(object):
         # we assume that all the sync requests on that process have stopped.
         # Stored as a dict from process_id to set of user_id, and a dict of
         # process_id to millisecond timestamp last updated.
-        self.external_process_to_current_syncs = {}
-        self.external_process_last_updated_ms = {}
+        self.external_process_to_current_syncs = {}  # type: Dict[int, Set[str]]
+        self.external_process_last_updated_ms = {}  # type: Dict[int, int]
+
         self.external_sync_linearizer = Linearizer(name="external_sync_linearizer")
 
         # Start a LoopingCall in 30s that fires every 5s.
@@ -255,7 +257,7 @@ class PresenceHandler(object):
         self.unpersisted_users_changes = set()
 
         if unpersisted:
-            logger.info("Persisting %d upersisted presence updates", len(unpersisted))
+            logger.info("Persisting %d unpersisted presence updates", len(unpersisted))
             yield self.store.update_presence(
                 [self.user_to_current_state[user_id] for user_id in unpersisted]
             )
@@ -349,10 +351,13 @@ class PresenceHandler(object):
             if now - last_update > EXTERNAL_PROCESS_EXPIRY
         ]
         for process_id in expired_process_ids:
+            # For each expired process drop tracking info and check the users
+            # that were syncing on that process to see if they need to be timed
+            # out.
             users_to_check.update(
-                self.external_process_last_updated_ms.pop(process_id, ())
+                self.external_process_to_current_syncs.pop(process_id, ())
             )
-            self.external_process_last_update.pop(process_id)
+            self.external_process_last_updated_ms.pop(process_id)
 
         states = [
             self.user_to_current_state.get(user_id, UserPresenceState.default(user_id))
@@ -803,17 +808,25 @@ class PresenceHandler(object):
         # Loop round handling deltas until we're up to date
         while True:
             with Measure(self.clock, "presence_delta"):
-                deltas = yield self.store.get_current_state_deltas(self._event_pos)
-                if not deltas:
+                room_max_stream_ordering = self.store.get_room_max_stream_ordering()
+                if self._event_pos == room_max_stream_ordering:
                     return
 
+                logger.debug(
+                    "Processing presence stats %s->%s",
+                    self._event_pos,
+                    room_max_stream_ordering,
+                )
+                max_pos, deltas = yield self.store.get_current_state_deltas(
+                    self._event_pos, room_max_stream_ordering
+                )
                 yield self._handle_state_delta(deltas)
 
-                self._event_pos = deltas[-1]["stream_id"]
+                self._event_pos = max_pos
 
                 # Expose current event processing position to prometheus
                 synapse.metrics.event_processing_positions.labels("presence").set(
-                    self._event_pos
+                    max_pos
                 )
 
     @defer.inlineCallbacks
diff --git a/synapse/handlers/profile.py b/synapse/handlers/profile.py
index 8690f69d45..22e0a04da4 100644
--- a/synapse/handlers/profile.py
+++ b/synapse/handlers/profile.py
@@ -275,7 +275,7 @@ class BaseProfileHandler(BaseHandler):
                     ratelimit=False,  # Try to hide that these events aren't atomic.
                 )
             except Exception as e:
-                logger.warn(
+                logger.warning(
                     "Failed to update join event for room %s - %s", room_id, str(e)
                 )
 
diff --git a/synapse/handlers/read_marker.py b/synapse/handlers/read_marker.py
index 3e4d8c93a4..e3b528d271 100644
--- a/synapse/handlers/read_marker.py
+++ b/synapse/handlers/read_marker.py
@@ -15,8 +15,6 @@
 
 import logging
 
-from twisted.internet import defer
-
 from synapse.util.async_helpers import Linearizer
 
 from ._base import BaseHandler
@@ -32,8 +30,7 @@ class ReadMarkerHandler(BaseHandler):
         self.read_marker_linearizer = Linearizer(name="read_marker")
         self.notifier = hs.get_notifier()
 
-    @defer.inlineCallbacks
-    def received_client_read_marker(self, room_id, user_id, event_id):
+    async def received_client_read_marker(self, room_id, user_id, event_id):
         """Updates the read marker for a given user in a given room if the event ID given
         is ahead in the stream relative to the current read marker.
 
@@ -41,8 +38,8 @@ class ReadMarkerHandler(BaseHandler):
         the read marker has changed.
         """
 
-        with (yield self.read_marker_linearizer.queue((room_id, user_id))):
-            existing_read_marker = yield self.store.get_account_data_for_room_and_type(
+        with await self.read_marker_linearizer.queue((room_id, user_id)):
+            existing_read_marker = await self.store.get_account_data_for_room_and_type(
                 user_id, room_id, "m.fully_read"
             )
 
@@ -50,13 +47,13 @@ class ReadMarkerHandler(BaseHandler):
 
             if existing_read_marker:
                 # Only update if the new marker is ahead in the stream
-                should_update = yield self.store.is_event_after(
+                should_update = await self.store.is_event_after(
                     event_id, existing_read_marker["event_id"]
                 )
 
             if should_update:
                 content = {"event_id": event_id}
-                max_id = yield self.store.add_account_data_to_room(
+                max_id = await self.store.add_account_data_to_room(
                     user_id, room_id, "m.fully_read", content
                 )
                 self.notifier.on_new_event("account_data_key", max_id, users=[user_id])
diff --git a/synapse/handlers/receipts.py b/synapse/handlers/receipts.py
index 6854c751a6..9283c039e3 100644
--- a/synapse/handlers/receipts.py
+++ b/synapse/handlers/receipts.py
@@ -18,6 +18,7 @@ from twisted.internet import defer
 
 from synapse.handlers._base import BaseHandler
 from synapse.types import ReadReceipt, get_domain_from_id
+from synapse.util.async_helpers import maybe_awaitable
 
 logger = logging.getLogger(__name__)
 
@@ -36,8 +37,7 @@ class ReceiptsHandler(BaseHandler):
         self.clock = self.hs.get_clock()
         self.state = hs.get_state_handler()
 
-    @defer.inlineCallbacks
-    def _received_remote_receipt(self, origin, content):
+    async def _received_remote_receipt(self, origin, content):
         """Called when we receive an EDU of type m.receipt from a remote HS.
         """
         receipts = []
@@ -62,17 +62,16 @@ class ReceiptsHandler(BaseHandler):
                         )
                     )
 
-        yield self._handle_new_receipts(receipts)
+        await self._handle_new_receipts(receipts)
 
-    @defer.inlineCallbacks
-    def _handle_new_receipts(self, receipts):
+    async def _handle_new_receipts(self, receipts):
         """Takes a list of receipts, stores them and informs the notifier.
         """
         min_batch_id = None
         max_batch_id = None
 
         for receipt in receipts:
-            res = yield self.store.insert_receipt(
+            res = await self.store.insert_receipt(
                 receipt.room_id,
                 receipt.receipt_type,
                 receipt.user_id,
@@ -99,14 +98,15 @@ class ReceiptsHandler(BaseHandler):
 
         self.notifier.on_new_event("receipt_key", max_batch_id, rooms=affected_room_ids)
         # Note that the min here shouldn't be relied upon to be accurate.
-        yield self.hs.get_pusherpool().on_new_receipts(
-            min_batch_id, max_batch_id, affected_room_ids
+        await maybe_awaitable(
+            self.hs.get_pusherpool().on_new_receipts(
+                min_batch_id, max_batch_id, affected_room_ids
+            )
         )
 
         return True
 
-    @defer.inlineCallbacks
-    def received_client_receipt(self, room_id, receipt_type, user_id, event_id):
+    async def received_client_receipt(self, room_id, receipt_type, user_id, event_id):
         """Called when a client tells us a local user has read up to the given
         event_id in the room.
         """
@@ -118,24 +118,11 @@ class ReceiptsHandler(BaseHandler):
             data={"ts": int(self.clock.time_msec())},
         )
 
-        is_new = yield self._handle_new_receipts([receipt])
+        is_new = await self._handle_new_receipts([receipt])
         if not is_new:
             return
 
-        yield self.federation.send_read_receipt(receipt)
-
-    @defer.inlineCallbacks
-    def get_receipts_for_room(self, room_id, to_key):
-        """Gets all receipts for a room, upto the given key.
-        """
-        result = yield self.store.get_linearized_receipts_for_room(
-            room_id, to_key=to_key
-        )
-
-        if not result:
-            return []
-
-        return result
+        await self.federation.send_read_receipt(receipt)
 
 
 class ReceiptEventSource(object):
diff --git a/synapse/handlers/register.py b/synapse/handlers/register.py
index be0425a33b..cff6b0d375 100644
--- a/synapse/handlers/register.py
+++ b/synapse/handlers/register.py
@@ -24,13 +24,11 @@ from synapse.api.errors import (
     AuthError,
     Codes,
     ConsentNotGivenError,
-    InvalidCaptchaError,
     LimitExceededError,
     RegistrationError,
     SynapseError,
 )
 from synapse.config.server import is_threepid_reserved
-from synapse.http.client import CaptchaServerHttpClient
 from synapse.http.servlet import assert_params_in_dict
 from synapse.replication.http.login import RegisterDeviceReplicationServlet
 from synapse.replication.http.register import (
@@ -39,7 +37,6 @@ from synapse.replication.http.register import (
 )
 from synapse.types import RoomAlias, RoomID, UserID, create_requester
 from synapse.util.async_helpers import Linearizer
-from synapse.util.threepids import check_3pid_allowed
 
 from ._base import BaseHandler
 
@@ -59,7 +56,6 @@ class RegistrationHandler(BaseHandler):
         self._auth_handler = hs.get_auth_handler()
         self.profile_handler = hs.get_profile_handler()
         self.user_directory_handler = hs.get_user_directory_handler()
-        self.captcha_client = CaptchaServerHttpClient(hs)
         self.identity_handler = self.hs.get_handlers().identity_handler
         self.ratelimiter = hs.get_registration_ratelimiter()
 
@@ -221,10 +217,9 @@ class RegistrationHandler(BaseHandler):
 
         else:
             # autogen a sequential user ID
-            attempts = 0
             user = None
             while not user:
-                localpart = yield self._generate_user_id(attempts > 0)
+                localpart = yield self._generate_user_id()
                 user = UserID(localpart, self.hs.hostname)
                 user_id = user.to_string()
                 yield self.check_user_id_not_appservice_exclusive(user_id)
@@ -242,7 +237,6 @@ class RegistrationHandler(BaseHandler):
                     # if user id is taken, just generate another
                     user = None
                     user_id = None
-                    attempts += 1
 
         if not self.hs.config.user_consent_at_registration:
             yield self._auto_join_rooms(user_id)
@@ -279,16 +273,12 @@ class RegistrationHandler(BaseHandler):
         fake_requester = create_requester(user_id)
 
         # try to create the room if we're the first real user on the server. Note
-        # that an auto-generated support user is not a real user and will never be
+        # that an auto-generated support or bot user is not a real user and will never be
         # the user to create the room
         should_auto_create_rooms = False
-        is_support = yield self.store.is_support_user(user_id)
-        # There is an edge case where the first user is the support user, then
-        # the room is never created, though this seems unlikely and
-        # recoverable from given the support user being involved in the first
-        # place.
-        if self.hs.config.autocreate_auto_join_rooms and not is_support:
-            count = yield self.store.count_all_users()
+        is_real_user = yield self.store.is_real_user(user_id)
+        if self.hs.config.autocreate_auto_join_rooms and is_real_user:
+            count = yield self.store.count_real_users()
             should_auto_create_rooms = count == 1
         for r in self.hs.config.auto_join_rooms:
             logger.info("Auto-joining %s to %s", user_id, r)
@@ -362,70 +352,6 @@ class RegistrationHandler(BaseHandler):
         )
         return user_id
 
-    @defer.inlineCallbacks
-    def check_recaptcha(self, ip, private_key, challenge, response):
-        """
-        Checks a recaptcha is correct.
-
-        Used only by c/s api v1
-        """
-
-        captcha_response = yield self._validate_captcha(
-            ip, private_key, challenge, response
-        )
-        if not captcha_response["valid"]:
-            logger.info(
-                "Invalid captcha entered from %s. Error: %s",
-                ip,
-                captcha_response["error_url"],
-            )
-            raise InvalidCaptchaError(error_url=captcha_response["error_url"])
-        else:
-            logger.info("Valid captcha entered from %s", ip)
-
-    @defer.inlineCallbacks
-    def register_email(self, threepidCreds):
-        """
-        Registers emails with an identity server.
-
-        Used only by c/s api v1
-        """
-
-        for c in threepidCreds:
-            logger.info(
-                "validating threepidcred sid %s on id server %s",
-                c["sid"],
-                c["idServer"],
-            )
-            try:
-                threepid = yield self.identity_handler.threepid_from_creds(c)
-            except Exception:
-                logger.exception("Couldn't validate 3pid")
-                raise RegistrationError(400, "Couldn't validate 3pid")
-
-            if not threepid:
-                raise RegistrationError(400, "Couldn't validate 3pid")
-            logger.info(
-                "got threepid with medium '%s' and address '%s'",
-                threepid["medium"],
-                threepid["address"],
-            )
-
-            if not check_3pid_allowed(self.hs, threepid["medium"], threepid["address"]):
-                raise RegistrationError(403, "Third party identifier is not allowed")
-
-    @defer.inlineCallbacks
-    def bind_emails(self, user_id, threepidCreds):
-        """Links emails with a user ID and informs an identity server.
-
-        Used only by c/s api v1
-        """
-
-        # Now we have a matrix ID, bind it to the threepids we were given
-        for c in threepidCreds:
-            # XXX: This should be a deferred list, shouldn't it?
-            yield self.identity_handler.bind_threepid(c, user_id)
-
     def check_user_id_not_appservice_exclusive(self, user_id, allowed_appservice=None):
         # don't allow people to register the server notices mxid
         if self._server_notices_mxid is not None:
@@ -451,10 +377,10 @@ class RegistrationHandler(BaseHandler):
                 )
 
     @defer.inlineCallbacks
-    def _generate_user_id(self, reseed=False):
-        if reseed or self._next_generated_user_id is None:
+    def _generate_user_id(self):
+        if self._next_generated_user_id is None:
             with (yield self._generate_user_id_linearizer.queue(())):
-                if reseed or self._next_generated_user_id is None:
+                if self._next_generated_user_id is None:
                     self._next_generated_user_id = (
                         yield self.store.find_next_generated_user_id_localpart()
                     )
@@ -463,52 +389,15 @@ class RegistrationHandler(BaseHandler):
         self._next_generated_user_id += 1
         return str(id)
 
-    @defer.inlineCallbacks
-    def _validate_captcha(self, ip_addr, private_key, challenge, response):
-        """Validates the captcha provided.
-
-        Used only by c/s api v1
-
-        Returns:
-            dict: Containing 'valid'(bool) and 'error_url'(str) if invalid.
-
-        """
-        response = yield self._submit_captcha(ip_addr, private_key, challenge, response)
-        # parse Google's response. Lovely format..
-        lines = response.split("\n")
-        json = {
-            "valid": lines[0] == "true",
-            "error_url": "http://www.recaptcha.net/recaptcha/api/challenge?"
-            + "error=%s" % lines[1],
-        }
-        return json
-
-    @defer.inlineCallbacks
-    def _submit_captcha(self, ip_addr, private_key, challenge, response):
-        """
-        Used only by c/s api v1
-        """
-        data = yield self.captcha_client.post_urlencoded_get_raw(
-            "http://www.recaptcha.net:80/recaptcha/api/verify",
-            args={
-                "privatekey": private_key,
-                "remoteip": ip_addr,
-                "challenge": challenge,
-                "response": response,
-            },
-        )
-        return data
-
     @defer.inlineCallbacks
     def _join_user_to_room(self, requester, room_identifier):
-        room_id = None
         room_member_handler = self.hs.get_room_member_handler()
         if RoomID.is_valid(room_identifier):
             room_id = room_identifier
         elif RoomAlias.is_valid(room_identifier):
             room_alias = RoomAlias.from_string(room_identifier)
-            room_id, remote_room_hosts = (
-                yield room_member_handler.lookup_room_alias(room_alias)
+            room_id, remote_room_hosts = yield room_member_handler.lookup_room_alias(
+                room_alias
             )
             room_id = room_id.to_string()
         else:
@@ -648,9 +537,7 @@ class RegistrationHandler(BaseHandler):
         return (device_id, access_token)
 
     @defer.inlineCallbacks
-    def post_registration_actions(
-        self, user_id, auth_result, access_token, bind_email, bind_msisdn
-    ):
+    def post_registration_actions(self, user_id, auth_result, access_token):
         """A user has completed registration
 
         Args:
@@ -659,18 +546,10 @@ class RegistrationHandler(BaseHandler):
                 registered user.
             access_token (str|None): The access token of the newly logged in
                 device, or None if `inhibit_login` enabled.
-            bind_email (bool): Whether to bind the email with the identity
-                server.
-            bind_msisdn (bool): Whether to bind the msisdn with the identity
-                server.
         """
         if self.hs.config.worker_app:
             yield self._post_registration_client(
-                user_id=user_id,
-                auth_result=auth_result,
-                access_token=access_token,
-                bind_email=bind_email,
-                bind_msisdn=bind_msisdn,
+                user_id=user_id, auth_result=auth_result, access_token=access_token
             )
             return
 
@@ -683,13 +562,11 @@ class RegistrationHandler(BaseHandler):
             ):
                 yield self.store.upsert_monthly_active_user(user_id)
 
-            yield self._register_email_threepid(
-                user_id, threepid, access_token, bind_email
-            )
+            yield self._register_email_threepid(user_id, threepid, access_token)
 
         if auth_result and LoginType.MSISDN in auth_result:
             threepid = auth_result[LoginType.MSISDN]
-            yield self._register_msisdn_threepid(user_id, threepid, bind_msisdn)
+            yield self._register_msisdn_threepid(user_id, threepid)
 
         if auth_result and LoginType.TERMS in auth_result:
             yield self._on_user_consented(user_id, self.hs.config.user_consent_version)
@@ -708,14 +585,12 @@ class RegistrationHandler(BaseHandler):
         yield self.post_consent_actions(user_id)
 
     @defer.inlineCallbacks
-    def _register_email_threepid(self, user_id, threepid, token, bind_email):
+    def _register_email_threepid(self, user_id, threepid, token):
         """Add an email address as a 3pid identifier
 
         Also adds an email pusher for the email address, if configured in the
         HS config
 
-        Also optionally binds emails to the given user_id on the identity server
-
         Must be called on master.
 
         Args:
@@ -723,8 +598,6 @@ class RegistrationHandler(BaseHandler):
             threepid (object): m.login.email.identity auth response
             token (str|None): access_token for the user, or None if not logged
                 in.
-            bind_email (bool): true if the client requested the email to be
-                bound at the identity server
         Returns:
             defer.Deferred:
         """
@@ -766,29 +639,15 @@ class RegistrationHandler(BaseHandler):
                 data={},
             )
 
-        if bind_email:
-            logger.info("bind_email specified: binding")
-            logger.debug("Binding emails %s to %s" % (threepid, user_id))
-            yield self.identity_handler.bind_threepid(
-                threepid["threepid_creds"], user_id
-            )
-        else:
-            logger.info("bind_email not specified: not binding email")
-
     @defer.inlineCallbacks
-    def _register_msisdn_threepid(self, user_id, threepid, bind_msisdn):
+    def _register_msisdn_threepid(self, user_id, threepid):
         """Add a phone number as a 3pid identifier
 
-        Also optionally binds msisdn to the given user_id on the identity server
-
         Must be called on master.
 
         Args:
             user_id (str): id of user
             threepid (object): m.login.msisdn auth response
-            token (str): access_token for the user
-            bind_email (bool): true if the client requested the email to be
-                bound at the identity server
         Returns:
             defer.Deferred:
         """
@@ -804,12 +663,3 @@ class RegistrationHandler(BaseHandler):
         yield self._auth_handler.add_threepid(
             user_id, threepid["medium"], threepid["address"], threepid["validated_at"]
         )
-
-        if bind_msisdn:
-            logger.info("bind_msisdn specified: binding")
-            logger.debug("Binding msisdn %s to %s", threepid, user_id)
-            yield self.identity_handler.bind_threepid(
-                threepid["threepid_creds"], user_id
-            )
-        else:
-            logger.info("bind_msisdn not specified: not binding msisdn")
diff --git a/synapse/handlers/room.py b/synapse/handlers/room.py
index 6e47fe7867..e92b2eafd5 100644
--- a/synapse/handlers/room.py
+++ b/synapse/handlers/room.py
@@ -28,6 +28,7 @@ from twisted.internet import defer
 from synapse.api.constants import EventTypes, JoinRules, RoomCreationPreset
 from synapse.api.errors import AuthError, Codes, NotFoundError, StoreError, SynapseError
 from synapse.api.room_versions import KNOWN_ROOM_VERSIONS
+from synapse.http.endpoint import parse_and_validate_server_name
 from synapse.storage.state import StateFilter
 from synapse.types import RoomAlias, RoomID, RoomStreamToken, StreamToken, UserID
 from synapse.util import stringutils
@@ -128,6 +129,7 @@ class RoomCreationHandler(BaseHandler):
             old_room_id,
             new_version,  # args for _upgrade_room
         )
+
         return ret
 
     @defer.inlineCallbacks
@@ -146,21 +148,22 @@ class RoomCreationHandler(BaseHandler):
 
         # we create and auth the tombstone event before properly creating the new
         # room, to check our user has perms in the old room.
-        tombstone_event, tombstone_context = (
-            yield self.event_creation_handler.create_event(
-                requester,
-                {
-                    "type": EventTypes.Tombstone,
-                    "state_key": "",
-                    "room_id": old_room_id,
-                    "sender": user_id,
-                    "content": {
-                        "body": "This room has been replaced",
-                        "replacement_room": new_room_id,
-                    },
+        (
+            tombstone_event,
+            tombstone_context,
+        ) = yield self.event_creation_handler.create_event(
+            requester,
+            {
+                "type": EventTypes.Tombstone,
+                "state_key": "",
+                "room_id": old_room_id,
+                "sender": user_id,
+                "content": {
+                    "body": "This room has been replaced",
+                    "replacement_room": new_room_id,
                 },
-                token_id=requester.access_token_id,
-            )
+            },
+            token_id=requester.access_token_id,
         )
         old_room_version = yield self.store.get_room_version(old_room_id)
         yield self.auth.check_from_context(
@@ -187,7 +190,12 @@ class RoomCreationHandler(BaseHandler):
             requester, old_room_id, new_room_id, old_room_state
         )
 
-        # and finally, shut down the PLs in the old room, and update them in the new
+        # Copy over user push rules, tags and migrate room directory state
+        yield self.room_member_handler.transfer_room_state_on_room_upgrade(
+            old_room_id, new_room_id
+        )
+
+        # finally, shut down the PLs in the old room, and update them in the new
         # room.
         yield self._update_upgraded_room_pls(
             requester, old_room_id, new_room_id, old_room_state
@@ -554,7 +562,8 @@ class RoomCreationHandler(BaseHandler):
         invite_list = config.get("invite", [])
         for i in invite_list:
             try:
-                UserID.from_string(i)
+                uid = UserID.from_string(i)
+                parse_and_validate_server_name(uid.domain)
             except Exception:
                 raise SynapseError(400, "Invalid user_id: %s" % (i,))
 
@@ -579,8 +588,8 @@ class RoomCreationHandler(BaseHandler):
 
         room_id = yield self._generate_room_id(creator_id=user_id, is_public=is_public)
 
+        directory_handler = self.hs.get_handlers().directory_handler
         if room_alias:
-            directory_handler = self.hs.get_handlers().directory_handler
             yield directory_handler.create_association(
                 requester=requester,
                 room_id=room_id,
@@ -665,6 +674,7 @@ class RoomCreationHandler(BaseHandler):
 
         for invite_3pid in invite_3pid_list:
             id_server = invite_3pid["id_server"]
+            id_access_token = invite_3pid.get("id_access_token")  # optional
             address = invite_3pid["address"]
             medium = invite_3pid["medium"]
             yield self.hs.get_room_member_handler().do_3pid_invite(
@@ -675,6 +685,7 @@ class RoomCreationHandler(BaseHandler):
                 id_server,
                 requester,
                 txn_id=None,
+                id_access_token=id_access_token,
             )
 
         result = {"room_id": room_id}
@@ -818,6 +829,8 @@ class RoomContextHandler(object):
     def __init__(self, hs):
         self.hs = hs
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
 
     @defer.inlineCallbacks
     def get_event_context(self, user, room_id, event_id, limit, event_filter):
@@ -844,7 +857,7 @@ class RoomContextHandler(object):
 
         def filter_evts(events):
             return filter_events_for_client(
-                self.store, user.to_string(), events, is_peeking=is_peeking
+                self.storage, user.to_string(), events, is_peeking=is_peeking
             )
 
         event = yield self.store.get_event(
@@ -852,7 +865,6 @@ class RoomContextHandler(object):
         )
         if not event:
             return None
-            return
 
         filtered = yield (filter_evts([event]))
         if not filtered:
@@ -887,7 +899,7 @@ class RoomContextHandler(object):
         # first? Shouldn't we be consistent with /sync?
         # https://github.com/matrix-org/matrix-doc/issues/687
 
-        state = yield self.store.get_state_for_events(
+        state = yield self.state_store.get_state_for_events(
             [last_event_id], state_filter=state_filter
         )
         results["state"] = list(state[last_event_id].values())
@@ -919,7 +931,7 @@ class RoomEventSource(object):
 
         from_token = RoomStreamToken.parse(from_key)
         if from_token.topological:
-            logger.warn("Stream has topological part!!!! %r", from_key)
+            logger.warning("Stream has topological part!!!! %r", from_key)
             from_key = "s%s" % (from_token.stream,)
 
         app_service = self.store.get_app_service_by_user_id(user.to_string())
diff --git a/synapse/handlers/room_list.py b/synapse/handlers/room_list.py
index a7e55f00e5..c615206df1 100644
--- a/synapse/handlers/room_list.py
+++ b/synapse/handlers/room_list.py
@@ -16,8 +16,7 @@
 import logging
 from collections import namedtuple
 
-from six import PY3, iteritems
-from six.moves import range
+from six import iteritems
 
 import msgpack
 from unpaddedbase64 import decode_base64, encode_base64
@@ -27,7 +26,6 @@ from twisted.internet import defer
 from synapse.api.constants import EventTypes, JoinRules
 from synapse.api.errors import Codes, HttpResponseException
 from synapse.types import ThirdPartyInstanceID
-from synapse.util.async_helpers import concurrently_execute
 from synapse.util.caches.descriptors import cachedInlineCallbacks
 from synapse.util.caches.response_cache import ResponseCache
 
@@ -37,7 +35,6 @@ logger = logging.getLogger(__name__)
 
 REMOTE_ROOM_LIST_POLL_INTERVAL = 60 * 1000
 
-
 # This is used to indicate we should only return rooms published to the main list.
 EMPTY_THIRD_PARTY_ID = ThirdPartyInstanceID(None, None)
 
@@ -72,6 +69,8 @@ class RoomListHandler(BaseHandler):
                 This can be (None, None) to indicate the main list, or a particular
                 appservice and network id to use an appservice specific one.
                 Setting to None returns all public rooms across all lists.
+            from_federation (bool): true iff the request comes from the federation
+                API
         """
         if not self.enable_room_list_search:
             return defer.succeed({"chunk": [], "total_room_count_estimate": 0})
@@ -89,16 +88,8 @@ class RoomListHandler(BaseHandler):
             # appservice specific lists.
             logger.info("Bypassing cache as search request.")
 
-            # XXX: Quick hack to stop room directory queries taking too long.
-            # Timeout request after 60s. Probably want a more fundamental
-            # solution at some point
-            timeout = self.clock.time() + 60
             return self._get_public_room_list(
-                limit,
-                since_token,
-                search_filter,
-                network_tuple=network_tuple,
-                timeout=timeout,
+                limit, since_token, search_filter, network_tuple=network_tuple
             )
 
         key = (limit, since_token, network_tuple)
@@ -119,7 +110,6 @@ class RoomListHandler(BaseHandler):
         search_filter=None,
         network_tuple=EMPTY_THIRD_PARTY_ID,
         from_federation=False,
-        timeout=None,
     ):
         """Generate a public room list.
         Args:
@@ -132,240 +122,116 @@ class RoomListHandler(BaseHandler):
                 Setting to None returns all public rooms across all lists.
             from_federation (bool): Whether this request originated from a
                 federating server or a client. Used for room filtering.
-            timeout (int|None): Amount of seconds to wait for a response before
-                timing out.
         """
-        if since_token and since_token != "END":
-            since_token = RoomListNextBatch.from_token(since_token)
-        else:
-            since_token = None
 
-        rooms_to_order_value = {}
-        rooms_to_num_joined = {}
+        # Pagination tokens work by storing the room ID sent in the last batch,
+        # plus the direction (forwards or backwards). Next batch tokens always
+        # go forwards, prev batch tokens always go backwards.
 
-        newly_visible = []
-        newly_unpublished = []
         if since_token:
-            stream_token = since_token.stream_ordering
-            current_public_id = yield self.store.get_current_public_room_stream_id()
-            public_room_stream_id = since_token.public_room_stream_id
-            newly_visible, newly_unpublished = yield self.store.get_public_room_changes(
-                public_room_stream_id, current_public_id, network_tuple=network_tuple
-            )
-        else:
-            stream_token = yield self.store.get_room_max_stream_ordering()
-            public_room_stream_id = yield self.store.get_current_public_room_stream_id()
+            batch_token = RoomListNextBatch.from_token(since_token)
 
-        room_ids = yield self.store.get_public_room_ids_at_stream_id(
-            public_room_stream_id, network_tuple=network_tuple
+            bounds = (batch_token.last_joined_members, batch_token.last_room_id)
+            forwards = batch_token.direction_is_forward
+        else:
+            batch_token = None
+            bounds = None
+
+            forwards = True
+
+        # we request one more than wanted to see if there are more pages to come
+        probing_limit = limit + 1 if limit is not None else None
+
+        results = yield self.store.get_largest_public_rooms(
+            network_tuple,
+            search_filter,
+            probing_limit,
+            bounds=bounds,
+            forwards=forwards,
+            ignore_non_federatable=from_federation,
         )
 
-        # We want to return rooms in a particular order: the number of joined
-        # users. We then arbitrarily use the room_id as a tie breaker.
+        def build_room_entry(room):
+            entry = {
+                "room_id": room["room_id"],
+                "name": room["name"],
+                "topic": room["topic"],
+                "canonical_alias": room["canonical_alias"],
+                "num_joined_members": room["joined_members"],
+                "avatar_url": room["avatar"],
+                "world_readable": room["history_visibility"] == "world_readable",
+                "guest_can_join": room["guest_access"] == "can_join",
+            }
 
-        @defer.inlineCallbacks
-        def get_order_for_room(room_id):
-            # Most of the rooms won't have changed between the since token and
-            # now (especially if the since token is "now"). So, we can ask what
-            # the current users are in a room (that will hit a cache) and then
-            # check if the room has changed since the since token. (We have to
-            # do it in that order to avoid races).
-            # If things have changed then fall back to getting the current state
-            # at the since token.
-            joined_users = yield self.store.get_users_in_room(room_id)
-            if self.store.has_room_changed_since(room_id, stream_token):
-                latest_event_ids = yield self.store.get_forward_extremeties_for_room(
-                    room_id, stream_token
-                )
+            # Filter out Nones – rather omit the field altogether
+            return {k: v for k, v in entry.items() if v is not None}
 
-                if not latest_event_ids:
-                    return
+        results = [build_room_entry(r) for r in results]
 
-                joined_users = yield self.state_handler.get_current_users_in_room(
-                    room_id, latest_event_ids
-                )
+        response = {}
+        num_results = len(results)
+        if limit is not None:
+            more_to_come = num_results == probing_limit
 
-            num_joined_users = len(joined_users)
-            rooms_to_num_joined[room_id] = num_joined_users
+            # Depending on direction we trim either the front or back.
+            if forwards:
+                results = results[:limit]
+            else:
+                results = results[-limit:]
+        else:
+            more_to_come = False
 
-            if num_joined_users == 0:
-                return
+        if num_results > 0:
+            final_entry = results[-1]
+            initial_entry = results[0]
 
-            # We want larger rooms to be first, hence negating num_joined_users
-            rooms_to_order_value[room_id] = (-num_joined_users, room_id)
+            if forwards:
+                if batch_token:
+                    # If there was a token given then we assume that there
+                    # must be previous results.
+                    response["prev_batch"] = RoomListNextBatch(
+                        last_joined_members=initial_entry["num_joined_members"],
+                        last_room_id=initial_entry["room_id"],
+                        direction_is_forward=False,
+                    ).to_token()
 
-        logger.info(
-            "Getting ordering for %i rooms since %s", len(room_ids), stream_token
+                if more_to_come:
+                    response["next_batch"] = RoomListNextBatch(
+                        last_joined_members=final_entry["num_joined_members"],
+                        last_room_id=final_entry["room_id"],
+                        direction_is_forward=True,
+                    ).to_token()
+            else:
+                if batch_token:
+                    response["next_batch"] = RoomListNextBatch(
+                        last_joined_members=final_entry["num_joined_members"],
+                        last_room_id=final_entry["room_id"],
+                        direction_is_forward=True,
+                    ).to_token()
+
+                if more_to_come:
+                    response["prev_batch"] = RoomListNextBatch(
+                        last_joined_members=initial_entry["num_joined_members"],
+                        last_room_id=initial_entry["room_id"],
+                        direction_is_forward=False,
+                    ).to_token()
+
+        for room in results:
+            # populate search result entries with additional fields, namely
+            # 'aliases'
+            room_id = room["room_id"]
+
+            aliases = yield self.store.get_aliases_for_room(room_id)
+            if aliases:
+                room["aliases"] = aliases
+
+        response["chunk"] = results
+
+        response["total_room_count_estimate"] = yield self.store.count_public_rooms(
+            network_tuple, ignore_non_federatable=from_federation
         )
-        yield concurrently_execute(get_order_for_room, room_ids, 10)
 
-        sorted_entries = sorted(rooms_to_order_value.items(), key=lambda e: e[1])
-        sorted_rooms = [room_id for room_id, _ in sorted_entries]
-
-        # `sorted_rooms` should now be a list of all public room ids that is
-        # stable across pagination. Therefore, we can use indices into this
-        # list as our pagination tokens.
-
-        # Filter out rooms that we don't want to return
-        rooms_to_scan = [
-            r
-            for r in sorted_rooms
-            if r not in newly_unpublished and rooms_to_num_joined[r] > 0
-        ]
-
-        total_room_count = len(rooms_to_scan)
-
-        if since_token:
-            # Filter out rooms we've already returned previously
-            # `since_token.current_limit` is the index of the last room we
-            # sent down, so we exclude it and everything before/after it.
-            if since_token.direction_is_forward:
-                rooms_to_scan = rooms_to_scan[since_token.current_limit + 1 :]
-            else:
-                rooms_to_scan = rooms_to_scan[: since_token.current_limit]
-                rooms_to_scan.reverse()
-
-        logger.info("After sorting and filtering, %i rooms remain", len(rooms_to_scan))
-
-        # _append_room_entry_to_chunk will append to chunk but will stop if
-        # len(chunk) > limit
-        #
-        # Normally we will generate enough results on the first iteration here,
-        #  but if there is a search filter, _append_room_entry_to_chunk may
-        # filter some results out, in which case we loop again.
-        #
-        # We don't want to scan over the entire range either as that
-        # would potentially waste a lot of work.
-        #
-        # XXX if there is no limit, we may end up DoSing the server with
-        # calls to get_current_state_ids for every single room on the
-        # server. Surely we should cap this somehow?
-        #
-        if limit:
-            step = limit + 1
-        else:
-            # step cannot be zero
-            step = len(rooms_to_scan) if len(rooms_to_scan) != 0 else 1
-
-        chunk = []
-        for i in range(0, len(rooms_to_scan), step):
-            if timeout and self.clock.time() > timeout:
-                raise Exception("Timed out searching room directory")
-
-            batch = rooms_to_scan[i : i + step]
-            logger.info("Processing %i rooms for result", len(batch))
-            yield concurrently_execute(
-                lambda r: self._append_room_entry_to_chunk(
-                    r,
-                    rooms_to_num_joined[r],
-                    chunk,
-                    limit,
-                    search_filter,
-                    from_federation=from_federation,
-                ),
-                batch,
-                5,
-            )
-            logger.info("Now %i rooms in result", len(chunk))
-            if len(chunk) >= limit + 1:
-                break
-
-        chunk.sort(key=lambda e: (-e["num_joined_members"], e["room_id"]))
-
-        # Work out the new limit of the batch for pagination, or None if we
-        # know there are no more results that would be returned.
-        # i.e., [since_token.current_limit..new_limit] is the batch of rooms
-        # we've returned (or the reverse if we paginated backwards)
-        # We tried to pull out limit + 1 rooms above, so if we have <= limit
-        # then we know there are no more results to return
-        new_limit = None
-        if chunk and (not limit or len(chunk) > limit):
-
-            if not since_token or since_token.direction_is_forward:
-                if limit:
-                    chunk = chunk[:limit]
-                last_room_id = chunk[-1]["room_id"]
-            else:
-                if limit:
-                    chunk = chunk[-limit:]
-                last_room_id = chunk[0]["room_id"]
-
-            new_limit = sorted_rooms.index(last_room_id)
-
-        results = {"chunk": chunk, "total_room_count_estimate": total_room_count}
-
-        if since_token:
-            results["new_rooms"] = bool(newly_visible)
-
-        if not since_token or since_token.direction_is_forward:
-            if new_limit is not None:
-                results["next_batch"] = RoomListNextBatch(
-                    stream_ordering=stream_token,
-                    public_room_stream_id=public_room_stream_id,
-                    current_limit=new_limit,
-                    direction_is_forward=True,
-                ).to_token()
-
-            if since_token:
-                results["prev_batch"] = since_token.copy_and_replace(
-                    direction_is_forward=False,
-                    current_limit=since_token.current_limit + 1,
-                ).to_token()
-        else:
-            if new_limit is not None:
-                results["prev_batch"] = RoomListNextBatch(
-                    stream_ordering=stream_token,
-                    public_room_stream_id=public_room_stream_id,
-                    current_limit=new_limit,
-                    direction_is_forward=False,
-                ).to_token()
-
-            if since_token:
-                results["next_batch"] = since_token.copy_and_replace(
-                    direction_is_forward=True,
-                    current_limit=since_token.current_limit - 1,
-                ).to_token()
-
-        return results
-
-    @defer.inlineCallbacks
-    def _append_room_entry_to_chunk(
-        self,
-        room_id,
-        num_joined_users,
-        chunk,
-        limit,
-        search_filter,
-        from_federation=False,
-    ):
-        """Generate the entry for a room in the public room list and append it
-        to the `chunk` if it matches the search filter
-
-        Args:
-            room_id (str): The ID of the room.
-            num_joined_users (int): The number of joined users in the room.
-            chunk (list)
-            limit (int|None): Maximum amount of rooms to display. Function will
-                return if length of chunk is greater than limit + 1.
-            search_filter (dict|None)
-            from_federation (bool): Whether this request originated from a
-                federating server or a client. Used for room filtering.
-        """
-        if limit and len(chunk) > limit + 1:
-            # We've already got enough, so lets just drop it.
-            return
-
-        result = yield self.generate_room_entry(room_id, num_joined_users)
-        if not result:
-            return
-
-        if from_federation and not result.get("m.federate", True):
-            # This is a room that other servers cannot join. Do not show them
-            # this room.
-            return
-
-        if _matches_room_entry(result, search_filter):
-            chunk.append(result)
+        return response
 
     @cachedInlineCallbacks(num_args=1, cache_context=True)
     def generate_room_entry(
@@ -580,18 +446,15 @@ class RoomListNextBatch(
     namedtuple(
         "RoomListNextBatch",
         (
-            "stream_ordering",  # stream_ordering of the first public room list
-            "public_room_stream_id",  # public room stream id for first public room list
-            "current_limit",  # The number of previous rooms returned
+            "last_joined_members",  # The count to get rooms after/before
+            "last_room_id",  # The room_id to get rooms after/before
             "direction_is_forward",  # Bool if this is a next_batch, false if prev_batch
         ),
     )
 ):
-
     KEY_DICT = {
-        "stream_ordering": "s",
-        "public_room_stream_id": "p",
-        "current_limit": "n",
+        "last_joined_members": "m",
+        "last_room_id": "r",
         "direction_is_forward": "d",
     }
 
@@ -599,13 +462,7 @@ class RoomListNextBatch(
 
     @classmethod
     def from_token(cls, token):
-        if PY3:
-            # The argument raw=False is only available on new versions of
-            # msgpack, and only really needed on Python 3. Gate it behind
-            # a PY3 check to avoid causing issues on Debian-packaged versions.
-            decoded = msgpack.loads(decode_base64(token), raw=False)
-        else:
-            decoded = msgpack.loads(decode_base64(token))
+        decoded = msgpack.loads(decode_base64(token), raw=False)
         return RoomListNextBatch(
             **{cls.REVERSE_KEY_DICT[key]: val for key, val in decoded.items()}
         )
diff --git a/synapse/handlers/room_member.py b/synapse/handlers/room_member.py
index f03a2bd540..06d09c2947 100644
--- a/synapse/handlers/room_member.py
+++ b/synapse/handlers/room_member.py
@@ -20,15 +20,11 @@ import logging
 
 from six.moves import http_client
 
-from signedjson.key import decode_verify_key_bytes
-from signedjson.sign import verify_signed_json
-from unpaddedbase64 import decode_base64
-
 from twisted.internet import defer
 
 from synapse import types
 from synapse.api.constants import EventTypes, Membership
-from synapse.api.errors import AuthError, Codes, HttpResponseException, SynapseError
+from synapse.api.errors import AuthError, Codes, SynapseError
 from synapse.types import RoomID, UserID
 from synapse.util.async_helpers import Linearizer
 from synapse.util.distributor import user_joined_room, user_left_room
@@ -37,8 +33,6 @@ from ._base import BaseHandler
 
 logger = logging.getLogger(__name__)
 
-id_server_scheme = "https://"
-
 
 class RoomMemberHandler(object):
     # TODO(paul): This handler currently contains a messy conflation of
@@ -59,10 +53,10 @@ class RoomMemberHandler(object):
         self.auth = hs.get_auth()
         self.state_handler = hs.get_state_handler()
         self.config = hs.config
-        self.simple_http_client = hs.get_simple_http_client()
 
         self.federation_handler = hs.get_handlers().federation_handler
         self.directory_handler = hs.get_handlers().directory_handler
+        self.identity_handler = hs.get_handlers().identity_handler
         self.registration_handler = hs.get_registration_handler()
         self.profile_handler = hs.get_profile_handler()
         self.event_creation_handler = hs.get_event_creation_handler()
@@ -100,7 +94,7 @@ class RoomMemberHandler(object):
         raise NotImplementedError()
 
     @abc.abstractmethod
-    def _remote_reject_invite(self, remote_room_hosts, room_id, target):
+    def _remote_reject_invite(self, requester, remote_room_hosts, room_id, target):
         """Attempt to reject an invite for a room this server is not in. If we
         fail to do so we locally mark the invite as rejected.
 
@@ -210,22 +204,6 @@ class RoomMemberHandler(object):
                 newly_joined = prev_member_event.membership != Membership.JOIN
             if newly_joined:
                 yield self._user_joined_room(target, room_id)
-
-            # Copy over direct message status and room tags if this is a join
-            # on an upgraded room
-
-            # Check if this is an upgraded room
-            predecessor = yield self.store.get_room_predecessor(room_id)
-
-            if predecessor:
-                # It is an upgraded room. Copy over old tags
-                self.copy_room_tags_and_direct_to_room(
-                    predecessor["room_id"], room_id, user_id
-                )
-                # Move over old push rules
-                self.store.move_push_rules_from_room_to_room_for_user(
-                    predecessor["room_id"], room_id, user_id
-                )
         elif event.membership == Membership.LEAVE:
             if prev_member_event_id:
                 prev_member_event = yield self.store.get_event(prev_member_event_id)
@@ -469,10 +447,11 @@ class RoomMemberHandler(object):
                 if requester.is_guest:
                     content["kind"] = "guest"
 
-                ret = yield self._remote_join(
+                remote_join_response = yield self._remote_join(
                     requester, remote_room_hosts, room_id, target, content
                 )
-                return ret
+
+                return remote_join_response
 
         elif effective_membership_state == Membership.LEAVE:
             if not is_host_in_room:
@@ -510,9 +489,75 @@ class RoomMemberHandler(object):
         return res
 
     @defer.inlineCallbacks
-    def send_membership_event(
-        self, requester, event, context, remote_room_hosts=None, ratelimit=True
-    ):
+    def transfer_room_state_on_room_upgrade(self, old_room_id, room_id):
+        """Upon our server becoming aware of an upgraded room, either by upgrading a room
+        ourselves or joining one, we can transfer over information from the previous room.
+
+        Copies user state (tags/push rules) for every local user that was in the old room, as
+        well as migrating the room directory state.
+
+        Args:
+            old_room_id (str): The ID of the old room
+
+            room_id (str): The ID of the new room
+
+        Returns:
+            Deferred
+        """
+        # Find all local users that were in the old room and copy over each user's state
+        users = yield self.store.get_users_in_room(old_room_id)
+        yield self.copy_user_state_on_room_upgrade(old_room_id, room_id, users)
+
+        # Add new room to the room directory if the old room was there
+        # Remove old room from the room directory
+        old_room = yield self.store.get_room(old_room_id)
+        if old_room and old_room["is_public"]:
+            yield self.store.set_room_is_public(old_room_id, False)
+            yield self.store.set_room_is_public(room_id, True)
+
+    @defer.inlineCallbacks
+    def copy_user_state_on_room_upgrade(self, old_room_id, new_room_id, user_ids):
+        """Copy user-specific information when they join a new room when that new room is the
+        result of a room upgrade
+
+        Args:
+            old_room_id (str): The ID of upgraded room
+            new_room_id (str): The ID of the new room
+            user_ids (Iterable[str]): User IDs to copy state for
+
+        Returns:
+            Deferred
+        """
+
+        logger.debug(
+            "Copying over room tags and push rules from %s to %s for users %s",
+            old_room_id,
+            new_room_id,
+            user_ids,
+        )
+
+        for user_id in user_ids:
+            try:
+                # It is an upgraded room. Copy over old tags
+                yield self.copy_room_tags_and_direct_to_room(
+                    old_room_id, new_room_id, user_id
+                )
+                # Copy over push rules
+                yield self.store.copy_push_rules_from_room_to_room_for_user(
+                    old_room_id, new_room_id, user_id
+                )
+            except Exception:
+                logger.exception(
+                    "Error copying tags and/or push rules from rooms %s to %s for user %s. "
+                    "Skipping...",
+                    old_room_id,
+                    new_room_id,
+                    user_id,
+                )
+                continue
+
+    @defer.inlineCallbacks
+    def send_membership_event(self, requester, event, context, ratelimit=True):
         """
         Change the membership status of a user in a room.
 
@@ -522,16 +567,10 @@ class RoomMemberHandler(object):
                 act as the sender, will be skipped.
             event (SynapseEvent): The membership event.
             context: The context of the event.
-            is_guest (bool): Whether the sender is a guest.
-            room_hosts ([str]): Homeservers which are likely to already be in
-                the room, and could be danced with in order to join this
-                homeserver for the first time.
             ratelimit (bool): Whether to rate limit this request.
         Raises:
             SynapseError if there was a problem changing the membership.
         """
-        remote_room_hosts = remote_room_hosts or []
-
         target_user = UserID.from_string(event.state_key)
         room_id = event.room_id
 
@@ -634,7 +673,7 @@ class RoomMemberHandler(object):
             servers.remove(room_alias.domain)
         servers.insert(0, room_alias.domain)
 
-        return (RoomID.from_string(room_id), servers)
+        return RoomID.from_string(room_id), servers
 
     @defer.inlineCallbacks
     def _get_inviter(self, user_id, room_id):
@@ -646,7 +685,15 @@ class RoomMemberHandler(object):
 
     @defer.inlineCallbacks
     def do_3pid_invite(
-        self, room_id, inviter, medium, address, id_server, requester, txn_id
+        self,
+        room_id,
+        inviter,
+        medium,
+        address,
+        id_server,
+        requester,
+        txn_id,
+        id_access_token=None,
     ):
         if self.config.block_non_admin_invites:
             is_requester_admin = yield self.auth.is_server_admin(requester.user)
@@ -669,7 +716,14 @@ class RoomMemberHandler(object):
                 Codes.FORBIDDEN,
             )
 
-        invitee = yield self._lookup_3pid(id_server, medium, address)
+        if not self._enable_lookup:
+            raise SynapseError(
+                403, "Looking up third-party identifiers is denied from this server"
+            )
+
+        invitee = yield self.identity_handler.lookup_3pid(
+            id_server, medium, address, id_access_token
+        )
 
         if invitee:
             yield self.update_membership(
@@ -677,67 +731,27 @@ class RoomMemberHandler(object):
             )
         else:
             yield self._make_and_store_3pid_invite(
-                requester, id_server, medium, address, room_id, inviter, txn_id=txn_id
+                requester,
+                id_server,
+                medium,
+                address,
+                room_id,
+                inviter,
+                txn_id=txn_id,
+                id_access_token=id_access_token,
             )
 
-    @defer.inlineCallbacks
-    def _lookup_3pid(self, id_server, medium, address):
-        """Looks up a 3pid in the passed identity server.
-
-        Args:
-            id_server (str): The server name (including port, if required)
-                of the identity server to use.
-            medium (str): The type of the third party identifier (e.g. "email").
-            address (str): The third party identifier (e.g. "foo@example.com").
-
-        Returns:
-            str: the matrix ID of the 3pid, or None if it is not recognized.
-        """
-        if not self._enable_lookup:
-            raise SynapseError(
-                403, "Looking up third-party identifiers is denied from this server"
-            )
-        try:
-            data = yield self.simple_http_client.get_json(
-                "%s%s/_matrix/identity/api/v1/lookup" % (id_server_scheme, id_server),
-                {"medium": medium, "address": address},
-            )
-
-            if "mxid" in data:
-                if "signatures" not in data:
-                    raise AuthError(401, "No signatures on 3pid binding")
-                yield self._verify_any_signature(data, id_server)
-                return data["mxid"]
-
-        except IOError as e:
-            logger.warn("Error from identity server lookup: %s" % (e,))
-            return None
-
-    @defer.inlineCallbacks
-    def _verify_any_signature(self, data, server_hostname):
-        if server_hostname not in data["signatures"]:
-            raise AuthError(401, "No signature from server %s" % (server_hostname,))
-        for key_name, signature in data["signatures"][server_hostname].items():
-            key_data = yield self.simple_http_client.get_json(
-                "%s%s/_matrix/identity/api/v1/pubkey/%s"
-                % (id_server_scheme, server_hostname, key_name)
-            )
-            if "public_key" not in key_data:
-                raise AuthError(
-                    401, "No public key named %s from %s" % (key_name, server_hostname)
-                )
-            verify_signed_json(
-                data,
-                server_hostname,
-                decode_verify_key_bytes(
-                    key_name, decode_base64(key_data["public_key"])
-                ),
-            )
-            return
-
     @defer.inlineCallbacks
     def _make_and_store_3pid_invite(
-        self, requester, id_server, medium, address, room_id, user, txn_id
+        self,
+        requester,
+        id_server,
+        medium,
+        address,
+        room_id,
+        user,
+        txn_id,
+        id_access_token=None,
     ):
         room_state = yield self.state_handler.get_current_state(room_id)
 
@@ -772,21 +786,25 @@ class RoomMemberHandler(object):
         if room_avatar_event:
             room_avatar_url = room_avatar_event.content.get("url", "")
 
-        token, public_keys, fallback_public_key, display_name = (
-            yield self._ask_id_server_for_third_party_invite(
-                requester=requester,
-                id_server=id_server,
-                medium=medium,
-                address=address,
-                room_id=room_id,
-                inviter_user_id=user.to_string(),
-                room_alias=canonical_room_alias,
-                room_avatar_url=room_avatar_url,
-                room_join_rules=room_join_rules,
-                room_name=room_name,
-                inviter_display_name=inviter_display_name,
-                inviter_avatar_url=inviter_avatar_url,
-            )
+        (
+            token,
+            public_keys,
+            fallback_public_key,
+            display_name,
+        ) = yield self.identity_handler.ask_id_server_for_third_party_invite(
+            requester=requester,
+            id_server=id_server,
+            medium=medium,
+            address=address,
+            room_id=room_id,
+            inviter_user_id=user.to_string(),
+            room_alias=canonical_room_alias,
+            room_avatar_url=room_avatar_url,
+            room_join_rules=room_join_rules,
+            room_name=room_name,
+            inviter_display_name=inviter_display_name,
+            inviter_avatar_url=inviter_avatar_url,
+            id_access_token=id_access_token,
         )
 
         yield self.event_creation_handler.create_and_send_nonmember_event(
@@ -808,103 +826,6 @@ class RoomMemberHandler(object):
             txn_id=txn_id,
         )
 
-    @defer.inlineCallbacks
-    def _ask_id_server_for_third_party_invite(
-        self,
-        requester,
-        id_server,
-        medium,
-        address,
-        room_id,
-        inviter_user_id,
-        room_alias,
-        room_avatar_url,
-        room_join_rules,
-        room_name,
-        inviter_display_name,
-        inviter_avatar_url,
-    ):
-        """
-        Asks an identity server for a third party invite.
-
-        Args:
-            requester (Requester)
-            id_server (str): hostname + optional port for the identity server.
-            medium (str): The literal string "email".
-            address (str): The third party address being invited.
-            room_id (str): The ID of the room to which the user is invited.
-            inviter_user_id (str): The user ID of the inviter.
-            room_alias (str): An alias for the room, for cosmetic notifications.
-            room_avatar_url (str): The URL of the room's avatar, for cosmetic
-                notifications.
-            room_join_rules (str): The join rules of the email (e.g. "public").
-            room_name (str): The m.room.name of the room.
-            inviter_display_name (str): The current display name of the
-                inviter.
-            inviter_avatar_url (str): The URL of the inviter's avatar.
-
-        Returns:
-            A deferred tuple containing:
-                token (str): The token which must be signed to prove authenticity.
-                public_keys ([{"public_key": str, "key_validity_url": str}]):
-                    public_key is a base64-encoded ed25519 public key.
-                fallback_public_key: One element from public_keys.
-                display_name (str): A user-friendly name to represent the invited
-                    user.
-        """
-
-        is_url = "%s%s/_matrix/identity/api/v1/store-invite" % (
-            id_server_scheme,
-            id_server,
-        )
-
-        invite_config = {
-            "medium": medium,
-            "address": address,
-            "room_id": room_id,
-            "room_alias": room_alias,
-            "room_avatar_url": room_avatar_url,
-            "room_join_rules": room_join_rules,
-            "room_name": room_name,
-            "sender": inviter_user_id,
-            "sender_display_name": inviter_display_name,
-            "sender_avatar_url": inviter_avatar_url,
-        }
-
-        try:
-            data = yield self.simple_http_client.post_json_get_json(
-                is_url, invite_config
-            )
-        except HttpResponseException as e:
-            # Some identity servers may only support application/x-www-form-urlencoded
-            # types. This is especially true with old instances of Sydent, see
-            # https://github.com/matrix-org/sydent/pull/170
-            logger.info(
-                "Failed to POST %s with JSON, falling back to urlencoded form: %s",
-                is_url,
-                e,
-            )
-            data = yield self.simple_http_client.post_urlencoded_get_json(
-                is_url, invite_config
-            )
-
-        # TODO: Check for success
-        token = data["token"]
-        public_keys = data.get("public_keys", [])
-        if "public_key" in data:
-            fallback_public_key = {
-                "public_key": data["public_key"],
-                "key_validity_url": "%s%s/_matrix/identity/api/v1/pubkey/isvalid"
-                % (id_server_scheme, id_server),
-            }
-        else:
-            fallback_public_key = public_keys[0]
-
-        if not public_keys:
-            public_keys.append(fallback_public_key)
-        display_name = data["display_name"]
-        return token, public_keys, fallback_public_key, display_name
-
     @defer.inlineCallbacks
     def _is_host_in_room(self, current_state_ids):
         # Have we just created the room, and is this about to be the very
@@ -962,9 +883,7 @@ class RoomMemberMasterHandler(RoomMemberHandler):
         )
 
         if complexity:
-            if complexity["v1"] > max_complexity:
-                return True
-            return False
+            return complexity["v1"] > max_complexity
         return None
 
     @defer.inlineCallbacks
@@ -980,10 +899,7 @@ class RoomMemberMasterHandler(RoomMemberHandler):
         max_complexity = self.hs.config.limit_remote_rooms.complexity
         complexity = yield self.store.get_room_complexity(room_id)
 
-        if complexity["v1"] > max_complexity:
-            return True
-
-        return False
+        return complexity["v1"] > max_complexity
 
     @defer.inlineCallbacks
     def _remote_join(self, requester, remote_room_hosts, room_id, user, content):
@@ -1062,7 +978,7 @@ class RoomMemberMasterHandler(RoomMemberHandler):
             # The 'except' clause is very broad, but we need to
             # capture everything from DNS failures upwards
             #
-            logger.warn("Failed to reject invite: %s", e)
+            logger.warning("Failed to reject invite: %s", e)
 
             yield self.store.locally_reject_invite(target.to_string(), room_id)
             return {}
diff --git a/synapse/handlers/saml_handler.py b/synapse/handlers/saml_handler.py
index a1ce6929cf..cc9e6b9bd0 100644
--- a/synapse/handlers/saml_handler.py
+++ b/synapse/handlers/saml_handler.py
@@ -21,6 +21,8 @@ from saml2.client import Saml2Client
 from synapse.api.errors import SynapseError
 from synapse.http.servlet import parse_string
 from synapse.rest.client.v1.login import SSOAuthHandler
+from synapse.types import UserID, map_username_to_mxid_localpart
+from synapse.util.async_helpers import Linearizer
 
 logger = logging.getLogger(__name__)
 
@@ -29,12 +31,26 @@ class SamlHandler:
     def __init__(self, hs):
         self._saml_client = Saml2Client(hs.config.saml2_sp_config)
         self._sso_auth_handler = SSOAuthHandler(hs)
+        self._registration_handler = hs.get_registration_handler()
+
+        self._clock = hs.get_clock()
+        self._datastore = hs.get_datastore()
+        self._hostname = hs.hostname
+        self._saml2_session_lifetime = hs.config.saml2_session_lifetime
+        self._mxid_source_attribute = hs.config.saml2_mxid_source_attribute
+        self._grandfathered_mxid_source_attribute = (
+            hs.config.saml2_grandfathered_mxid_source_attribute
+        )
+        self._mxid_mapper = hs.config.saml2_mxid_mapper
+
+        # identifier for the external_ids table
+        self._auth_provider_id = "saml"
 
         # a map from saml session id to Saml2SessionData object
         self._outstanding_requests_dict = {}
 
-        self._clock = hs.get_clock()
-        self._saml2_session_lifetime = hs.config.saml2_session_lifetime
+        # a lock on the mappings
+        self._mapping_lock = Linearizer(name="saml_mapping", clock=self._clock)
 
     def handle_redirect_request(self, client_redirect_url):
         """Handle an incoming request to /login/sso/redirect
@@ -60,7 +76,7 @@ class SamlHandler:
         # this shouldn't happen!
         raise Exception("prepare_for_authenticate didn't return a Location header")
 
-    def handle_saml_response(self, request):
+    async def handle_saml_response(self, request):
         """Handle an incoming request to /_matrix/saml2/authn_response
 
         Args:
@@ -77,6 +93,10 @@ class SamlHandler:
         # the dict.
         self.expire_sessions()
 
+        user_id = await self._map_saml_response_to_user(resp_bytes)
+        self._sso_auth_handler.complete_sso_login(user_id, request, relay_state)
+
+    async def _map_saml_response_to_user(self, resp_bytes):
         try:
             saml2_auth = self._saml_client.parse_authn_request_response(
                 resp_bytes,
@@ -91,18 +111,88 @@ class SamlHandler:
             logger.warning("SAML2 response was not signed")
             raise SynapseError(400, "SAML2 response was not signed")
 
-        if "uid" not in saml2_auth.ava:
+        logger.info("SAML2 response: %s", saml2_auth.origxml)
+        logger.info("SAML2 mapped attributes: %s", saml2_auth.ava)
+
+        try:
+            remote_user_id = saml2_auth.ava["uid"][0]
+        except KeyError:
             logger.warning("SAML2 response lacks a 'uid' attestation")
             raise SynapseError(400, "uid not in SAML2 response")
 
+        try:
+            mxid_source = saml2_auth.ava[self._mxid_source_attribute][0]
+        except KeyError:
+            logger.warning(
+                "SAML2 response lacks a '%s' attestation", self._mxid_source_attribute
+            )
+            raise SynapseError(
+                400, "%s not in SAML2 response" % (self._mxid_source_attribute,)
+            )
+
         self._outstanding_requests_dict.pop(saml2_auth.in_response_to, None)
 
-        username = saml2_auth.ava["uid"][0]
         displayName = saml2_auth.ava.get("displayName", [None])[0]
 
-        return self._sso_auth_handler.on_successful_auth(
-            username, request, relay_state, user_display_name=displayName
-        )
+        with (await self._mapping_lock.queue(self._auth_provider_id)):
+            # first of all, check if we already have a mapping for this user
+            logger.info(
+                "Looking for existing mapping for user %s:%s",
+                self._auth_provider_id,
+                remote_user_id,
+            )
+            registered_user_id = await self._datastore.get_user_by_external_id(
+                self._auth_provider_id, remote_user_id
+            )
+            if registered_user_id is not None:
+                logger.info("Found existing mapping %s", registered_user_id)
+                return registered_user_id
+
+            # backwards-compatibility hack: see if there is an existing user with a
+            # suitable mapping from the uid
+            if (
+                self._grandfathered_mxid_source_attribute
+                and self._grandfathered_mxid_source_attribute in saml2_auth.ava
+            ):
+                attrval = saml2_auth.ava[self._grandfathered_mxid_source_attribute][0]
+                user_id = UserID(
+                    map_username_to_mxid_localpart(attrval), self._hostname
+                ).to_string()
+                logger.info(
+                    "Looking for existing account based on mapped %s %s",
+                    self._grandfathered_mxid_source_attribute,
+                    user_id,
+                )
+
+                users = await self._datastore.get_users_by_id_case_insensitive(user_id)
+                if users:
+                    registered_user_id = list(users.keys())[0]
+                    logger.info("Grandfathering mapping to %s", registered_user_id)
+                    await self._datastore.record_user_external_id(
+                        self._auth_provider_id, remote_user_id, registered_user_id
+                    )
+                    return registered_user_id
+
+            # figure out a new mxid for this user
+            base_mxid_localpart = self._mxid_mapper(mxid_source)
+
+            suffix = 0
+            while True:
+                localpart = base_mxid_localpart + (str(suffix) if suffix else "")
+                if not await self._datastore.get_users_by_id_case_insensitive(
+                    UserID(localpart, self._hostname).to_string()
+                ):
+                    break
+                suffix += 1
+            logger.info("Allocating mxid for new user with localpart %s", localpart)
+
+            registered_user_id = await self._registration_handler.register_user(
+                localpart=localpart, default_display_name=displayName
+            )
+            await self._datastore.record_user_external_id(
+                self._auth_provider_id, remote_user_id, registered_user_id
+            )
+            return registered_user_id
 
     def expire_sessions(self):
         expire_before = self._clock.time_msec() - self._saml2_session_lifetime
diff --git a/synapse/handlers/search.py b/synapse/handlers/search.py
index cd5e90bacb..56ed262a1f 100644
--- a/synapse/handlers/search.py
+++ b/synapse/handlers/search.py
@@ -35,6 +35,8 @@ class SearchHandler(BaseHandler):
     def __init__(self, hs):
         super(SearchHandler, self).__init__(hs)
         self._event_serializer = hs.get_event_client_serializer()
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
 
     @defer.inlineCallbacks
     def get_old_rooms_from_upgraded_room(self, room_id):
@@ -221,7 +223,7 @@ class SearchHandler(BaseHandler):
             filtered_events = search_filter.filter([r["event"] for r in results])
 
             events = yield filter_events_for_client(
-                self.store, user.to_string(), filtered_events
+                self.storage, user.to_string(), filtered_events
             )
 
             events.sort(key=lambda e: -rank_map[e.event_id])
@@ -271,7 +273,7 @@ class SearchHandler(BaseHandler):
                 filtered_events = search_filter.filter([r["event"] for r in results])
 
                 events = yield filter_events_for_client(
-                    self.store, user.to_string(), filtered_events
+                    self.storage, user.to_string(), filtered_events
                 )
 
                 room_events.extend(events)
@@ -340,11 +342,11 @@ class SearchHandler(BaseHandler):
                 )
 
                 res["events_before"] = yield filter_events_for_client(
-                    self.store, user.to_string(), res["events_before"]
+                    self.storage, user.to_string(), res["events_before"]
                 )
 
                 res["events_after"] = yield filter_events_for_client(
-                    self.store, user.to_string(), res["events_after"]
+                    self.storage, user.to_string(), res["events_after"]
                 )
 
                 res["start"] = now_token.copy_and_replace(
@@ -372,7 +374,7 @@ class SearchHandler(BaseHandler):
                         [(EventTypes.Member, sender) for sender in senders]
                     )
 
-                    state = yield self.store.get_state_for_event(
+                    state = yield self.state_store.get_state_for_event(
                         last_event_id, state_filter
                     )
 
@@ -394,15 +396,11 @@ class SearchHandler(BaseHandler):
         time_now = self.clock.time_msec()
 
         for context in contexts.values():
-            context["events_before"] = (
-                yield self._event_serializer.serialize_events(
-                    context["events_before"], time_now
-                )
+            context["events_before"] = yield self._event_serializer.serialize_events(
+                context["events_before"], time_now
             )
-            context["events_after"] = (
-                yield self._event_serializer.serialize_events(
-                    context["events_after"], time_now
-                )
+            context["events_after"] = yield self._event_serializer.serialize_events(
+                context["events_after"], time_now
             )
 
         state_results = {}
diff --git a/synapse/handlers/stats.py b/synapse/handlers/stats.py
index 4449da6669..7f7d56390e 100644
--- a/synapse/handlers/stats.py
+++ b/synapse/handlers/stats.py
@@ -14,15 +14,14 @@
 # limitations under the License.
 
 import logging
+from collections import Counter
 
 from twisted.internet import defer
 
-from synapse.api.constants import EventTypes, JoinRules, Membership
+from synapse.api.constants import EventTypes, Membership
 from synapse.handlers.state_deltas import StateDeltasHandler
 from synapse.metrics import event_processing_positions
 from synapse.metrics.background_process_metrics import run_as_background_process
-from synapse.types import UserID
-from synapse.util.metrics import Measure
 
 logger = logging.getLogger(__name__)
 
@@ -46,6 +45,8 @@ class StatsHandler(StateDeltasHandler):
         self.is_mine_id = hs.is_mine_id
         self.stats_bucket_size = hs.config.stats_bucket_size
 
+        self.stats_enabled = hs.config.stats_enabled
+
         # The current position in the current_state_delta stream
         self.pos = None
 
@@ -62,11 +63,10 @@ class StatsHandler(StateDeltasHandler):
     def notify_new_event(self):
         """Called when there may be more deltas to process
         """
-        if not self.hs.config.stats_enabled:
+        if not self.stats_enabled or self._is_processing:
             return
 
-        if self._is_processing:
-            return
+        self._is_processing = True
 
         @defer.inlineCallbacks
         def process():
@@ -75,39 +75,83 @@ class StatsHandler(StateDeltasHandler):
             finally:
                 self._is_processing = False
 
-        self._is_processing = True
         run_as_background_process("stats.notify_new_event", process)
 
     @defer.inlineCallbacks
     def _unsafe_process(self):
         # If self.pos is None then means we haven't fetched it from DB
         if self.pos is None:
-            self.pos = yield self.store.get_stats_stream_pos()
-
-        # If still None then the initial background update hasn't happened yet
-        if self.pos is None:
-            return None
+            self.pos = yield self.store.get_stats_positions()
 
         # Loop round handling deltas until we're up to date
+
         while True:
-            with Measure(self.clock, "stats_delta"):
-                deltas = yield self.store.get_current_state_deltas(self.pos)
-                if not deltas:
-                    return
+            # Be sure to read the max stream_ordering *before* checking if there are any outstanding
+            # deltas, since there is otherwise a chance that we could miss updates which arrive
+            # after we check the deltas.
+            room_max_stream_ordering = self.store.get_room_max_stream_ordering()
+            if self.pos == room_max_stream_ordering:
+                break
 
-                logger.info("Handling %d state deltas", len(deltas))
-                yield self._handle_deltas(deltas)
+            logger.debug(
+                "Processing room stats %s->%s", self.pos, room_max_stream_ordering
+            )
+            max_pos, deltas = yield self.store.get_current_state_deltas(
+                self.pos, room_max_stream_ordering
+            )
 
-                self.pos = deltas[-1]["stream_id"]
-                yield self.store.update_stats_stream_pos(self.pos)
+            if deltas:
+                logger.debug("Handling %d state deltas", len(deltas))
+                room_deltas, user_deltas = yield self._handle_deltas(deltas)
+            else:
+                room_deltas = {}
+                user_deltas = {}
 
-                event_processing_positions.labels("stats").set(self.pos)
+            # Then count deltas for total_events and total_event_bytes.
+            (
+                room_count,
+                user_count,
+            ) = yield self.store.get_changes_room_total_events_and_bytes(
+                self.pos, max_pos
+            )
+
+            for room_id, fields in room_count.items():
+                room_deltas.setdefault(room_id, {}).update(fields)
+
+            for user_id, fields in user_count.items():
+                user_deltas.setdefault(user_id, {}).update(fields)
+
+            logger.debug("room_deltas: %s", room_deltas)
+            logger.debug("user_deltas: %s", user_deltas)
+
+            # Always call this so that we update the stats position.
+            yield self.store.bulk_update_stats_delta(
+                self.clock.time_msec(),
+                updates={"room": room_deltas, "user": user_deltas},
+                stream_id=max_pos,
+            )
+
+            logger.debug("Handled room stats to %s -> %s", self.pos, max_pos)
+
+            event_processing_positions.labels("stats").set(max_pos)
+
+            self.pos = max_pos
 
     @defer.inlineCallbacks
     def _handle_deltas(self, deltas):
+        """Called with the state deltas to process
+
+        Returns:
+            Deferred[tuple[dict[str, Counter], dict[str, counter]]]
+            Resovles to two dicts, the room deltas and the user deltas,
+            mapping from room/user ID to changes in the various fields.
         """
-        Called with the state deltas to process
-        """
+
+        room_to_stats_deltas = {}
+        user_to_stats_deltas = {}
+
+        room_to_state_updates = {}
+
         for delta in deltas:
             typ = delta["type"]
             state_key = delta["state_key"]
@@ -115,11 +159,10 @@ class StatsHandler(StateDeltasHandler):
             event_id = delta["event_id"]
             stream_id = delta["stream_id"]
             prev_event_id = delta["prev_event_id"]
-            stream_pos = delta["stream_id"]
 
-            logger.debug("Handling: %r %r, %s", typ, state_key, event_id)
+            logger.debug("Handling: %r, %r %r, %s", room_id, typ, state_key, event_id)
 
-            token = yield self.store.get_earliest_token_for_room_stats(room_id)
+            token = yield self.store.get_earliest_token_for_stats("room", room_id)
 
             # If the earliest token to begin from is larger than our current
             # stream ID, skip processing this delta.
@@ -131,203 +174,133 @@ class StatsHandler(StateDeltasHandler):
                 continue
 
             if event_id is None and prev_event_id is None:
-                # Errr...
+                logger.error(
+                    "event ID is None and so is the previous event ID. stream_id: %s",
+                    stream_id,
+                )
                 continue
 
             event_content = {}
 
+            sender = None
             if event_id is not None:
                 event = yield self.store.get_event(event_id, allow_none=True)
                 if event:
                     event_content = event.content or {}
+                    sender = event.sender
 
-            # We use stream_pos here rather than fetch by event_id as event_id
-            # may be None
-            now = yield self.store.get_received_ts_by_stream_pos(stream_pos)
+            # All the values in this dict are deltas (RELATIVE changes)
+            room_stats_delta = room_to_stats_deltas.setdefault(room_id, Counter())
 
-            # quantise time to the nearest bucket
-            now = (now // 1000 // self.stats_bucket_size) * self.stats_bucket_size
+            room_state = room_to_state_updates.setdefault(room_id, {})
+
+            if prev_event_id is None:
+                # this state event doesn't overwrite another,
+                # so it is a new effective/current state event
+                room_stats_delta["current_state_events"] += 1
 
             if typ == EventTypes.Member:
                 # we could use _get_key_change here but it's a bit inefficient
                 # given we're not testing for a specific result; might as well
                 # just grab the prev_membership and membership strings and
                 # compare them.
-                prev_event_content = {}
+                # We take None rather than leave as a previous membership
+                # in the absence of a previous event because we do not want to
+                # reduce the leave count when a new-to-the-room user joins.
+                prev_membership = None
                 if prev_event_id is not None:
                     prev_event = yield self.store.get_event(
                         prev_event_id, allow_none=True
                     )
                     if prev_event:
                         prev_event_content = prev_event.content
+                        prev_membership = prev_event_content.get(
+                            "membership", Membership.LEAVE
+                        )
 
                 membership = event_content.get("membership", Membership.LEAVE)
-                prev_membership = prev_event_content.get("membership", Membership.LEAVE)
 
-                if prev_membership == membership:
-                    continue
-
-                if prev_membership == Membership.JOIN:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "joined_members", -1
-                    )
+                if prev_membership is None:
+                    logger.debug("No previous membership for this user.")
+                elif membership == prev_membership:
+                    pass  # noop
+                elif prev_membership == Membership.JOIN:
+                    room_stats_delta["joined_members"] -= 1
                 elif prev_membership == Membership.INVITE:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "invited_members", -1
-                    )
+                    room_stats_delta["invited_members"] -= 1
                 elif prev_membership == Membership.LEAVE:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "left_members", -1
-                    )
+                    room_stats_delta["left_members"] -= 1
                 elif prev_membership == Membership.BAN:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "banned_members", -1
-                    )
+                    room_stats_delta["banned_members"] -= 1
                 else:
-                    err = "%s is not a valid prev_membership" % (repr(prev_membership),)
-                    logger.error(err)
-                    raise ValueError(err)
+                    raise ValueError(
+                        "%r is not a valid prev_membership" % (prev_membership,)
+                    )
 
+                if membership == prev_membership:
+                    pass  # noop
                 if membership == Membership.JOIN:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "joined_members", +1
-                    )
+                    room_stats_delta["joined_members"] += 1
                 elif membership == Membership.INVITE:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "invited_members", +1
-                    )
+                    room_stats_delta["invited_members"] += 1
+
+                    if sender and self.is_mine_id(sender):
+                        user_to_stats_deltas.setdefault(sender, Counter())[
+                            "invites_sent"
+                        ] += 1
+
                 elif membership == Membership.LEAVE:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "left_members", +1
-                    )
+                    room_stats_delta["left_members"] += 1
                 elif membership == Membership.BAN:
-                    yield self.store.update_stats_delta(
-                        now, "room", room_id, "banned_members", +1
-                    )
+                    room_stats_delta["banned_members"] += 1
                 else:
-                    err = "%s is not a valid membership" % (repr(membership),)
-                    logger.error(err)
-                    raise ValueError(err)
+                    raise ValueError("%r is not a valid membership" % (membership,))
 
                 user_id = state_key
                 if self.is_mine_id(user_id):
-                    # update user_stats as it's one of our users
-                    public = yield self._is_public_room(room_id)
+                    # this accounts for transitions like leave → ban and so on.
+                    has_changed_joinedness = (prev_membership == Membership.JOIN) != (
+                        membership == Membership.JOIN
+                    )
 
-                    if membership == Membership.LEAVE:
-                        yield self.store.update_stats_delta(
-                            now,
-                            "user",
-                            user_id,
-                            "public_rooms" if public else "private_rooms",
-                            -1,
-                        )
-                    elif membership == Membership.JOIN:
-                        yield self.store.update_stats_delta(
-                            now,
-                            "user",
-                            user_id,
-                            "public_rooms" if public else "private_rooms",
-                            +1,
-                        )
+                    if has_changed_joinedness:
+                        delta = +1 if membership == Membership.JOIN else -1
+
+                        user_to_stats_deltas.setdefault(user_id, Counter())[
+                            "joined_rooms"
+                        ] += delta
+
+                        room_stats_delta["local_users_in_room"] += delta
 
             elif typ == EventTypes.Create:
-                # Newly created room. Add it with all blank portions.
-                yield self.store.update_room_state(
-                    room_id,
-                    {
-                        "join_rules": None,
-                        "history_visibility": None,
-                        "encryption": None,
-                        "name": None,
-                        "topic": None,
-                        "avatar": None,
-                        "canonical_alias": None,
-                    },
+                room_state["is_federatable"] = (
+                    event_content.get("m.federate", True) is True
                 )
-
+                if sender and self.is_mine_id(sender):
+                    user_to_stats_deltas.setdefault(sender, Counter())[
+                        "rooms_created"
+                    ] += 1
             elif typ == EventTypes.JoinRules:
-                yield self.store.update_room_state(
-                    room_id, {"join_rules": event_content.get("join_rule")}
-                )
-
-                is_public = yield self._get_key_change(
-                    prev_event_id, event_id, "join_rule", JoinRules.PUBLIC
-                )
-                if is_public is not None:
-                    yield self.update_public_room_stats(now, room_id, is_public)
-
+                room_state["join_rules"] = event_content.get("join_rule")
             elif typ == EventTypes.RoomHistoryVisibility:
-                yield self.store.update_room_state(
-                    room_id,
-                    {"history_visibility": event_content.get("history_visibility")},
+                room_state["history_visibility"] = event_content.get(
+                    "history_visibility"
                 )
-
-                is_public = yield self._get_key_change(
-                    prev_event_id, event_id, "history_visibility", "world_readable"
-                )
-                if is_public is not None:
-                    yield self.update_public_room_stats(now, room_id, is_public)
-
             elif typ == EventTypes.Encryption:
-                yield self.store.update_room_state(
-                    room_id, {"encryption": event_content.get("algorithm")}
-                )
+                room_state["encryption"] = event_content.get("algorithm")
             elif typ == EventTypes.Name:
-                yield self.store.update_room_state(
-                    room_id, {"name": event_content.get("name")}
-                )
+                room_state["name"] = event_content.get("name")
             elif typ == EventTypes.Topic:
-                yield self.store.update_room_state(
-                    room_id, {"topic": event_content.get("topic")}
-                )
+                room_state["topic"] = event_content.get("topic")
             elif typ == EventTypes.RoomAvatar:
-                yield self.store.update_room_state(
-                    room_id, {"avatar": event_content.get("url")}
-                )
+                room_state["avatar"] = event_content.get("url")
             elif typ == EventTypes.CanonicalAlias:
-                yield self.store.update_room_state(
-                    room_id, {"canonical_alias": event_content.get("alias")}
-                )
+                room_state["canonical_alias"] = event_content.get("alias")
+            elif typ == EventTypes.GuestAccess:
+                room_state["guest_access"] = event_content.get("guest_access")
 
-    @defer.inlineCallbacks
-    def update_public_room_stats(self, ts, room_id, is_public):
-        """
-        Increment/decrement a user's number of public rooms when a room they are
-        in changes to/from public visibility.
+        for room_id, state in room_to_state_updates.items():
+            logger.info("Updating room_stats_state for %s: %s", room_id, state)
+            yield self.store.update_room_state(room_id, state)
 
-        Args:
-            ts (int): Timestamp in seconds
-            room_id (str)
-            is_public (bool)
-        """
-        # For now, blindly iterate over all local users in the room so that
-        # we can handle the whole problem of copying buckets over as needed
-        user_ids = yield self.store.get_users_in_room(room_id)
-
-        for user_id in user_ids:
-            if self.hs.is_mine(UserID.from_string(user_id)):
-                yield self.store.update_stats_delta(
-                    ts, "user", user_id, "public_rooms", +1 if is_public else -1
-                )
-                yield self.store.update_stats_delta(
-                    ts, "user", user_id, "private_rooms", -1 if is_public else +1
-                )
-
-    @defer.inlineCallbacks
-    def _is_public_room(self, room_id):
-        join_rules = yield self.state.get_current_state(room_id, EventTypes.JoinRules)
-        history_visibility = yield self.state.get_current_state(
-            room_id, EventTypes.RoomHistoryVisibility
-        )
-
-        if (join_rules and join_rules.content.get("join_rule") == JoinRules.PUBLIC) or (
-            (
-                history_visibility
-                and history_visibility.content.get("history_visibility")
-                == "world_readable"
-            )
-        ):
-            return True
-        else:
-            return False
+        return room_to_stats_deltas, user_to_stats_deltas
diff --git a/synapse/handlers/sync.py b/synapse/handlers/sync.py
index d582f8e494..b536d410e5 100644
--- a/synapse/handlers/sync.py
+++ b/synapse/handlers/sync.py
@@ -1,6 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2015, 2016 OpenMarket Ltd
-# Copyright 2018 New Vector Ltd
+# Copyright 2018, 2019 New Vector Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -230,6 +230,8 @@ class SyncHandler(object):
         self.response_cache = ResponseCache(hs, "sync")
         self.state = hs.get_state_handler()
         self.auth = hs.get_auth()
+        self.storage = hs.get_storage()
+        self.state_store = self.storage.state
 
         # ExpiringCache((User, Device)) -> LruCache(state_key => event_id)
         self.lazy_loaded_members_cache = ExpiringCache(
@@ -417,7 +419,7 @@ class SyncHandler(object):
                     current_state_ids = frozenset(itervalues(current_state_ids))
 
                 recents = yield filter_events_for_client(
-                    self.store,
+                    self.storage,
                     sync_config.user.to_string(),
                     recents,
                     always_include_ids=current_state_ids,
@@ -470,7 +472,7 @@ class SyncHandler(object):
                     current_state_ids = frozenset(itervalues(current_state_ids))
 
                 loaded_recents = yield filter_events_for_client(
-                    self.store,
+                    self.storage,
                     sync_config.user.to_string(),
                     loaded_recents,
                     always_include_ids=current_state_ids,
@@ -509,7 +511,7 @@ class SyncHandler(object):
         Returns:
             A Deferred map from ((type, state_key)->Event)
         """
-        state_ids = yield self.store.get_state_ids_for_event(
+        state_ids = yield self.state_store.get_state_ids_for_event(
             event.event_id, state_filter=state_filter
         )
         if event.is_state():
@@ -578,10 +580,9 @@ class SyncHandler(object):
 
         if not last_events:
             return None
-            return
 
         last_event = last_events[-1]
-        state_ids = yield self.store.get_state_ids_for_event(
+        state_ids = yield self.state_store.get_state_ids_for_event(
             last_event.event_id,
             state_filter=StateFilter.from_types(
                 [(EventTypes.Name, ""), (EventTypes.CanonicalAlias, "")]
@@ -758,11 +759,11 @@ class SyncHandler(object):
 
             if full_state:
                 if batch:
-                    current_state_ids = yield self.store.get_state_ids_for_event(
+                    current_state_ids = yield self.state_store.get_state_ids_for_event(
                         batch.events[-1].event_id, state_filter=state_filter
                     )
 
-                    state_ids = yield self.store.get_state_ids_for_event(
+                    state_ids = yield self.state_store.get_state_ids_for_event(
                         batch.events[0].event_id, state_filter=state_filter
                     )
 
@@ -782,7 +783,7 @@ class SyncHandler(object):
                 )
             elif batch.limited:
                 if batch:
-                    state_at_timeline_start = yield self.store.get_state_ids_for_event(
+                    state_at_timeline_start = yield self.state_store.get_state_ids_for_event(
                         batch.events[0].event_id, state_filter=state_filter
                     )
                 else:
@@ -811,7 +812,7 @@ class SyncHandler(object):
                 )
 
                 if batch:
-                    current_state_ids = yield self.store.get_state_ids_for_event(
+                    current_state_ids = yield self.state_store.get_state_ids_for_event(
                         batch.events[-1].event_id, state_filter=state_filter
                     )
                 else:
@@ -842,7 +843,7 @@ class SyncHandler(object):
                         # So we fish out all the member events corresponding to the
                         # timeline here, and then dedupe any redundant ones below.
 
-                        state_ids = yield self.store.get_state_ids_for_event(
+                        state_ids = yield self.state_store.get_state_ids_for_event(
                             batch.events[0].event_id,
                             # we only want members!
                             state_filter=StateFilter.from_types(
@@ -1125,6 +1126,11 @@ class SyncHandler(object):
             # weren't in the previous sync *or* they left and rejoined.
             users_that_have_changed.update(newly_joined_or_invited_users)
 
+            user_signatures_changed = yield self.store.get_users_whose_signatures_changed(
+                user_id, since_token.device_list_key
+            )
+            users_that_have_changed.update(user_signatures_changed)
+
             # Now find users that we no longer track
             for room_id in newly_left_rooms:
                 left_users = yield self.state.get_current_users_in_room(room_id)
@@ -1200,10 +1206,11 @@ class SyncHandler(object):
         since_token = sync_result_builder.since_token
 
         if since_token and not sync_result_builder.full_state:
-            account_data, account_data_by_room = (
-                yield self.store.get_updated_account_data_for_user(
-                    user_id, since_token.account_data_key
-                )
+            (
+                account_data,
+                account_data_by_room,
+            ) = yield self.store.get_updated_account_data_for_user(
+                user_id, since_token.account_data_key
             )
 
             push_rules_changed = yield self.store.have_push_rules_changed_for_user(
@@ -1215,9 +1222,10 @@ class SyncHandler(object):
                     sync_config.user
                 )
         else:
-            account_data, account_data_by_room = (
-                yield self.store.get_account_data_for_user(sync_config.user.to_string())
-            )
+            (
+                account_data,
+                account_data_by_room,
+            ) = yield self.store.get_account_data_for_user(sync_config.user.to_string())
 
             account_data["m.push_rules"] = yield self.push_rules_for_user(
                 sync_config.user
diff --git a/synapse/handlers/ui_auth/__init__.py b/synapse/handlers/ui_auth/__init__.py
new file mode 100644
index 0000000000..824f37f8f8
--- /dev/null
+++ b/synapse/handlers/ui_auth/__init__.py
@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""This module implements user-interactive auth verification.
+
+TODO: move more stuff out of AuthHandler in here.
+
+"""
+
+from synapse.handlers.ui_auth.checkers import INTERACTIVE_AUTH_CHECKERS  # noqa: F401
diff --git a/synapse/handlers/ui_auth/checkers.py b/synapse/handlers/ui_auth/checkers.py
new file mode 100644
index 0000000000..8363d887a9
--- /dev/null
+++ b/synapse/handlers/ui_auth/checkers.py
@@ -0,0 +1,247 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import logging
+
+from canonicaljson import json
+
+from twisted.internet import defer
+from twisted.web.client import PartialDownloadError
+
+from synapse.api.constants import LoginType
+from synapse.api.errors import Codes, LoginError, SynapseError
+from synapse.config.emailconfig import ThreepidBehaviour
+
+logger = logging.getLogger(__name__)
+
+
+class UserInteractiveAuthChecker:
+    """Abstract base class for an interactive auth checker"""
+
+    def __init__(self, hs):
+        pass
+
+    def is_enabled(self):
+        """Check if the configuration of the homeserver allows this checker to work
+
+        Returns:
+            bool: True if this login type is enabled.
+        """
+
+    def check_auth(self, authdict, clientip):
+        """Given the authentication dict from the client, attempt to check this step
+
+        Args:
+            authdict (dict): authentication dictionary from the client
+            clientip (str): The IP address of the client.
+
+        Raises:
+            SynapseError if authentication failed
+
+        Returns:
+            Deferred: the result of authentication (to pass back to the client?)
+        """
+        raise NotImplementedError()
+
+
+class DummyAuthChecker(UserInteractiveAuthChecker):
+    AUTH_TYPE = LoginType.DUMMY
+
+    def is_enabled(self):
+        return True
+
+    def check_auth(self, authdict, clientip):
+        return defer.succeed(True)
+
+
+class TermsAuthChecker(UserInteractiveAuthChecker):
+    AUTH_TYPE = LoginType.TERMS
+
+    def is_enabled(self):
+        return True
+
+    def check_auth(self, authdict, clientip):
+        return defer.succeed(True)
+
+
+class RecaptchaAuthChecker(UserInteractiveAuthChecker):
+    AUTH_TYPE = LoginType.RECAPTCHA
+
+    def __init__(self, hs):
+        super().__init__(hs)
+        self._enabled = bool(hs.config.recaptcha_private_key)
+        self._http_client = hs.get_proxied_http_client()
+        self._url = hs.config.recaptcha_siteverify_api
+        self._secret = hs.config.recaptcha_private_key
+
+    def is_enabled(self):
+        return self._enabled
+
+    @defer.inlineCallbacks
+    def check_auth(self, authdict, clientip):
+        try:
+            user_response = authdict["response"]
+        except KeyError:
+            # Client tried to provide captcha but didn't give the parameter:
+            # bad request.
+            raise LoginError(
+                400, "Captcha response is required", errcode=Codes.CAPTCHA_NEEDED
+            )
+
+        logger.info(
+            "Submitting recaptcha response %s with remoteip %s", user_response, clientip
+        )
+
+        # TODO: get this from the homeserver rather than creating a new one for
+        # each request
+        try:
+            resp_body = yield self._http_client.post_urlencoded_get_json(
+                self._url,
+                args={
+                    "secret": self._secret,
+                    "response": user_response,
+                    "remoteip": clientip,
+                },
+            )
+        except PartialDownloadError as pde:
+            # Twisted is silly
+            data = pde.response
+            resp_body = json.loads(data)
+
+        if "success" in resp_body:
+            # Note that we do NOT check the hostname here: we explicitly
+            # intend the CAPTCHA to be presented by whatever client the
+            # user is using, we just care that they have completed a CAPTCHA.
+            logger.info(
+                "%s reCAPTCHA from hostname %s",
+                "Successful" if resp_body["success"] else "Failed",
+                resp_body.get("hostname"),
+            )
+            if resp_body["success"]:
+                return True
+        raise LoginError(401, "", errcode=Codes.UNAUTHORIZED)
+
+
+class _BaseThreepidAuthChecker:
+    def __init__(self, hs):
+        self.hs = hs
+        self.store = hs.get_datastore()
+
+    @defer.inlineCallbacks
+    def _check_threepid(self, medium, authdict):
+        if "threepid_creds" not in authdict:
+            raise LoginError(400, "Missing threepid_creds", Codes.MISSING_PARAM)
+
+        threepid_creds = authdict["threepid_creds"]
+
+        identity_handler = self.hs.get_handlers().identity_handler
+
+        logger.info("Getting validated threepid. threepidcreds: %r", (threepid_creds,))
+
+        # msisdns are currently always ThreepidBehaviour.REMOTE
+        if medium == "msisdn":
+            if not self.hs.config.account_threepid_delegate_msisdn:
+                raise SynapseError(
+                    400, "Phone number verification is not enabled on this homeserver"
+                )
+            threepid = yield identity_handler.threepid_from_creds(
+                self.hs.config.account_threepid_delegate_msisdn, threepid_creds
+            )
+        elif medium == "email":
+            if self.hs.config.threepid_behaviour_email == ThreepidBehaviour.REMOTE:
+                assert self.hs.config.account_threepid_delegate_email
+                threepid = yield identity_handler.threepid_from_creds(
+                    self.hs.config.account_threepid_delegate_email, threepid_creds
+                )
+            elif self.hs.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+                threepid = None
+                row = yield self.store.get_threepid_validation_session(
+                    medium,
+                    threepid_creds["client_secret"],
+                    sid=threepid_creds["sid"],
+                    validated=True,
+                )
+
+                if row:
+                    threepid = {
+                        "medium": row["medium"],
+                        "address": row["address"],
+                        "validated_at": row["validated_at"],
+                    }
+
+                    # Valid threepid returned, delete from the db
+                    yield self.store.delete_threepid_session(threepid_creds["sid"])
+            else:
+                raise SynapseError(
+                    400, "Email address verification is not enabled on this homeserver"
+                )
+        else:
+            # this can't happen!
+            raise AssertionError("Unrecognized threepid medium: %s" % (medium,))
+
+        if not threepid:
+            raise LoginError(401, "", errcode=Codes.UNAUTHORIZED)
+
+        if threepid["medium"] != medium:
+            raise LoginError(
+                401,
+                "Expecting threepid of type '%s', got '%s'"
+                % (medium, threepid["medium"]),
+                errcode=Codes.UNAUTHORIZED,
+            )
+
+        threepid["threepid_creds"] = authdict["threepid_creds"]
+
+        return threepid
+
+
+class EmailIdentityAuthChecker(UserInteractiveAuthChecker, _BaseThreepidAuthChecker):
+    AUTH_TYPE = LoginType.EMAIL_IDENTITY
+
+    def __init__(self, hs):
+        UserInteractiveAuthChecker.__init__(self, hs)
+        _BaseThreepidAuthChecker.__init__(self, hs)
+
+    def is_enabled(self):
+        return self.hs.config.threepid_behaviour_email in (
+            ThreepidBehaviour.REMOTE,
+            ThreepidBehaviour.LOCAL,
+        )
+
+    def check_auth(self, authdict, clientip):
+        return self._check_threepid("email", authdict)
+
+
+class MsisdnAuthChecker(UserInteractiveAuthChecker, _BaseThreepidAuthChecker):
+    AUTH_TYPE = LoginType.MSISDN
+
+    def __init__(self, hs):
+        UserInteractiveAuthChecker.__init__(self, hs)
+        _BaseThreepidAuthChecker.__init__(self, hs)
+
+    def is_enabled(self):
+        return bool(self.hs.config.account_threepid_delegate_msisdn)
+
+    def check_auth(self, authdict, clientip):
+        return self._check_threepid("msisdn", authdict)
+
+
+INTERACTIVE_AUTH_CHECKERS = [
+    DummyAuthChecker,
+    TermsAuthChecker,
+    RecaptchaAuthChecker,
+    EmailIdentityAuthChecker,
+    MsisdnAuthChecker,
+]
+"""A list of UserInteractiveAuthChecker classes"""
diff --git a/synapse/handlers/user_directory.py b/synapse/handlers/user_directory.py
index e53669e40d..624f05ab5b 100644
--- a/synapse/handlers/user_directory.py
+++ b/synapse/handlers/user_directory.py
@@ -138,21 +138,28 @@ class UserDirectoryHandler(StateDeltasHandler):
         # Loop round handling deltas until we're up to date
         while True:
             with Measure(self.clock, "user_dir_delta"):
-                deltas = yield self.store.get_current_state_deltas(self.pos)
-                if not deltas:
+                room_max_stream_ordering = self.store.get_room_max_stream_ordering()
+                if self.pos == room_max_stream_ordering:
                     return
 
+                logger.debug(
+                    "Processing user stats %s->%s", self.pos, room_max_stream_ordering
+                )
+                max_pos, deltas = yield self.store.get_current_state_deltas(
+                    self.pos, room_max_stream_ordering
+                )
+
                 logger.info("Handling %d state deltas", len(deltas))
                 yield self._handle_deltas(deltas)
 
-                self.pos = deltas[-1]["stream_id"]
+                self.pos = max_pos
 
                 # Expose current event processing position to prometheus
                 synapse.metrics.event_processing_positions.labels("user_dir").set(
-                    self.pos
+                    max_pos
                 )
 
-                yield self.store.update_user_directory_stream_pos(self.pos)
+                yield self.store.update_user_directory_stream_pos(max_pos)
 
     @defer.inlineCallbacks
     def _handle_deltas(self, deltas):
diff --git a/synapse/http/__init__.py b/synapse/http/__init__.py
index 3acf772cd1..3880ce0d94 100644
--- a/synapse/http/__init__.py
+++ b/synapse/http/__init__.py
@@ -42,11 +42,13 @@ def cancelled_to_request_timed_out_error(value, timeout):
 
 
 ACCESS_TOKEN_RE = re.compile(r"(\?.*access(_|%5[Ff])token=)[^&]*(.*)$")
+CLIENT_SECRET_RE = re.compile(r"(\?.*client(_|%5[Ff])secret=)[^&]*(.*)$")
 
 
 def redact_uri(uri):
-    """Strips access tokens from the uri replaces with """
-    return ACCESS_TOKEN_RE.sub(r"\1\3", uri)
+    """Strips sensitive information from the uri replaces with """
+    uri = ACCESS_TOKEN_RE.sub(r"\1\3", uri)
+    return CLIENT_SECRET_RE.sub(r"\1\3", uri)
 
 
 class QuieterFileBodyProducer(FileBodyProducer):
diff --git a/synapse/http/client.py b/synapse/http/client.py
index 0ac20ebefc..d4c285445e 100644
--- a/synapse/http/client.py
+++ b/synapse/http/client.py
@@ -35,7 +35,7 @@ from twisted.internet.interfaces import (
 )
 from twisted.python.failure import Failure
 from twisted.web._newclient import ResponseDone
-from twisted.web.client import Agent, HTTPConnectionPool, PartialDownloadError, readBody
+from twisted.web.client import Agent, HTTPConnectionPool, readBody
 from twisted.web.http import PotentialDataLoss
 from twisted.web.http_headers import Headers
 
@@ -45,7 +45,9 @@ from synapse.http import (
     cancelled_to_request_timed_out_error,
     redact_uri,
 )
+from synapse.http.proxyagent import ProxyAgent
 from synapse.logging.context import make_deferred_yieldable
+from synapse.logging.opentracing import set_tag, start_active_span, tags
 from synapse.util.async_helpers import timeout_deferred
 from synapse.util.caches import CACHE_SIZE_FACTOR
 
@@ -182,7 +184,15 @@ class SimpleHttpClient(object):
     using HTTP in Matrix
     """
 
-    def __init__(self, hs, treq_args={}, ip_whitelist=None, ip_blacklist=None):
+    def __init__(
+        self,
+        hs,
+        treq_args={},
+        ip_whitelist=None,
+        ip_blacklist=None,
+        http_proxy=None,
+        https_proxy=None,
+    ):
         """
         Args:
             hs (synapse.server.HomeServer)
@@ -191,6 +201,8 @@ class SimpleHttpClient(object):
                 we may not request.
             ip_whitelist (netaddr.IPSet): The whitelisted IP addresses, that we can
                request if it were otherwise caught in a blacklist.
+            http_proxy (bytes): proxy server to use for http connections. host[:port]
+            https_proxy (bytes): proxy server to use for https connections. host[:port]
         """
         self.hs = hs
 
@@ -235,11 +247,13 @@ class SimpleHttpClient(object):
         # The default context factory in Twisted 14.0.0 (which we require) is
         # BrowserLikePolicyForHTTPS which will do regular cert validation
         # 'like a browser'
-        self.agent = Agent(
+        self.agent = ProxyAgent(
             self.reactor,
             connectTimeout=15,
             contextFactory=self.hs.get_http_client_context_factory(),
             pool=pool,
+            http_proxy=http_proxy,
+            https_proxy=https_proxy,
         )
 
         if self._ip_blacklist:
@@ -269,42 +283,56 @@ class SimpleHttpClient(object):
         # log request but strip `access_token` (AS requests for example include this)
         logger.info("Sending request %s %s", method, redact_uri(uri))
 
-        try:
-            body_producer = None
-            if data is not None:
-                body_producer = QuieterFileBodyProducer(BytesIO(data))
+        with start_active_span(
+            "outgoing-client-request",
+            tags={
+                tags.SPAN_KIND: tags.SPAN_KIND_RPC_CLIENT,
+                tags.HTTP_METHOD: method,
+                tags.HTTP_URL: uri,
+            },
+            finish_on_close=True,
+        ):
+            try:
+                body_producer = None
+                if data is not None:
+                    body_producer = QuieterFileBodyProducer(BytesIO(data))
 
-            request_deferred = treq.request(
-                method,
-                uri,
-                agent=self.agent,
-                data=body_producer,
-                headers=headers,
-                **self._extra_treq_args
-            )
-            request_deferred = timeout_deferred(
-                request_deferred,
-                60,
-                self.hs.get_reactor(),
-                cancelled_to_request_timed_out_error,
-            )
-            response = yield make_deferred_yieldable(request_deferred)
+                request_deferred = treq.request(
+                    method,
+                    uri,
+                    agent=self.agent,
+                    data=body_producer,
+                    headers=headers,
+                    **self._extra_treq_args
+                )
+                request_deferred = timeout_deferred(
+                    request_deferred,
+                    60,
+                    self.hs.get_reactor(),
+                    cancelled_to_request_timed_out_error,
+                )
+                response = yield make_deferred_yieldable(request_deferred)
 
-            incoming_responses_counter.labels(method, response.code).inc()
-            logger.info(
-                "Received response to %s %s: %s", method, redact_uri(uri), response.code
-            )
-            return response
-        except Exception as e:
-            incoming_responses_counter.labels(method, "ERR").inc()
-            logger.info(
-                "Error sending request to  %s %s: %s %s",
-                method,
-                redact_uri(uri),
-                type(e).__name__,
-                e.args[0],
-            )
-            raise
+                incoming_responses_counter.labels(method, response.code).inc()
+                logger.info(
+                    "Received response to %s %s: %s",
+                    method,
+                    redact_uri(uri),
+                    response.code,
+                )
+                return response
+            except Exception as e:
+                incoming_responses_counter.labels(method, "ERR").inc()
+                logger.info(
+                    "Error sending request to  %s %s: %s %s",
+                    method,
+                    redact_uri(uri),
+                    type(e).__name__,
+                    e.args[0],
+                )
+                set_tag(tags.ERROR, True)
+                set_tag("error_reason", e.args[0])
+                raise
 
     @defer.inlineCallbacks
     def post_urlencoded_get_json(self, uri, args={}, headers=None):
@@ -312,7 +340,7 @@ class SimpleHttpClient(object):
         Args:
             uri (str):
             args (dict[str, str|List[str]]): query params
-            headers (dict[str, List[str]]|None): If not None, a map from
+            headers (dict[str|bytes, List[str|bytes]]|None): If not None, a map from
                header name to a list of values for that header
 
         Returns:
@@ -356,7 +384,7 @@ class SimpleHttpClient(object):
         Args:
             uri (str):
             post_json (object):
-            headers (dict[str, List[str]]|None): If not None, a map from
+            headers (dict[str|bytes, List[str|bytes]]|None): If not None, a map from
                header name to a list of values for that header
 
         Returns:
@@ -399,7 +427,7 @@ class SimpleHttpClient(object):
                 None.
                 **Note**: The value of each key is assumed to be an iterable
                 and *not* a string.
-            headers (dict[str, List[str]]|None): If not None, a map from
+            headers (dict[str|bytes, List[str|bytes]]|None): If not None, a map from
                header name to a list of values for that header
         Returns:
             Deferred: Succeeds when we get *any* 2xx HTTP response, with the
@@ -423,7 +451,7 @@ class SimpleHttpClient(object):
                 None.
                 **Note**: The value of each key is assumed to be an iterable
                 and *not* a string.
-            headers (dict[str, List[str]]|None): If not None, a map from
+            headers (dict[str|bytes, List[str|bytes]]|None): If not None, a map from
                header name to a list of values for that header
         Returns:
             Deferred: Succeeds when we get *any* 2xx HTTP response, with the
@@ -467,7 +495,7 @@ class SimpleHttpClient(object):
                 None.
                 **Note**: The value of each key is assumed to be an iterable
                 and *not* a string.
-            headers (dict[str, List[str]]|None): If not None, a map from
+            headers (dict[str|bytes, List[str|bytes]]|None): If not None, a map from
                header name to a list of values for that header
         Returns:
             Deferred: Succeeds when we get *any* 2xx HTTP response, with the
@@ -501,7 +529,7 @@ class SimpleHttpClient(object):
         Args:
             url (str): The URL to GET
             output_stream (file): File to write the response body to.
-            headers (dict[str, List[str]]|None): If not None, a map from
+            headers (dict[str|bytes, List[str|bytes]]|None): If not None, a map from
                header name to a list of values for that header
         Returns:
             A (int,dict,string,int) tuple of the file length, dict of the response
@@ -520,7 +548,7 @@ class SimpleHttpClient(object):
             b"Content-Length" in resp_headers
             and int(resp_headers[b"Content-Length"][0]) > max_size
         ):
-            logger.warn("Requested URL is too large > %r bytes" % (self.max_size,))
+            logger.warning("Requested URL is too large > %r bytes" % (self.max_size,))
             raise SynapseError(
                 502,
                 "Requested file is too large > %r bytes" % (self.max_size,),
@@ -528,7 +556,7 @@ class SimpleHttpClient(object):
             )
 
         if response.code > 299:
-            logger.warn("Got %d when downloading %s" % (response.code, url))
+            logger.warning("Got %d when downloading %s" % (response.code, url))
             raise SynapseError(502, "Got error %d" % (response.code,), Codes.UNKNOWN)
 
         # TODO: if our Content-Type is HTML or something, just read the first
@@ -599,38 +627,6 @@ def _readBodyToFile(response, stream, max_size):
     return d
 
 
-class CaptchaServerHttpClient(SimpleHttpClient):
-    """
-    Separate HTTP client for talking to google's captcha servers
-    Only slightly special because accepts partial download responses
-
-    used only by c/s api v1
-    """
-
-    @defer.inlineCallbacks
-    def post_urlencoded_get_raw(self, url, args={}):
-        query_bytes = urllib.parse.urlencode(encode_urlencode_args(args), True)
-
-        response = yield self.request(
-            "POST",
-            url,
-            data=query_bytes,
-            headers=Headers(
-                {
-                    b"Content-Type": [b"application/x-www-form-urlencoded"],
-                    b"User-Agent": [self.user_agent],
-                }
-            ),
-        )
-
-        try:
-            body = yield make_deferred_yieldable(readBody(response))
-            return body
-        except PartialDownloadError as e:
-            # twisted dislikes google's response, no content length.
-            return e.response
-
-
 def encode_urlencode_args(args):
     return {k: encode_urlencode_arg(v) for k, v in args.items()}
 
diff --git a/synapse/http/connectproxyclient.py b/synapse/http/connectproxyclient.py
new file mode 100644
index 0000000000..be7b2ceb8e
--- /dev/null
+++ b/synapse/http/connectproxyclient.py
@@ -0,0 +1,195 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+
+from zope.interface import implementer
+
+from twisted.internet import defer, protocol
+from twisted.internet.error import ConnectError
+from twisted.internet.interfaces import IStreamClientEndpoint
+from twisted.internet.protocol import connectionDone
+from twisted.web import http
+
+logger = logging.getLogger(__name__)
+
+
+class ProxyConnectError(ConnectError):
+    pass
+
+
+@implementer(IStreamClientEndpoint)
+class HTTPConnectProxyEndpoint(object):
+    """An Endpoint implementation which will send a CONNECT request to an http proxy
+
+    Wraps an existing HostnameEndpoint for the proxy.
+
+    When we get the connect() request from the connection pool (via the TLS wrapper),
+    we'll first connect to the proxy endpoint with a ProtocolFactory which will make the
+    CONNECT request. Once that completes, we invoke the protocolFactory which was passed
+    in.
+
+    Args:
+        reactor: the Twisted reactor to use for the connection
+        proxy_endpoint (IStreamClientEndpoint): the endpoint to use to connect to the
+            proxy
+        host (bytes): hostname that we want to CONNECT to
+        port (int): port that we want to connect to
+    """
+
+    def __init__(self, reactor, proxy_endpoint, host, port):
+        self._reactor = reactor
+        self._proxy_endpoint = proxy_endpoint
+        self._host = host
+        self._port = port
+
+    def __repr__(self):
+        return "" % (self._proxy_endpoint,)
+
+    def connect(self, protocolFactory):
+        f = HTTPProxiedClientFactory(self._host, self._port, protocolFactory)
+        d = self._proxy_endpoint.connect(f)
+        # once the tcp socket connects successfully, we need to wait for the
+        # CONNECT to complete.
+        d.addCallback(lambda conn: f.on_connection)
+        return d
+
+
+class HTTPProxiedClientFactory(protocol.ClientFactory):
+    """ClientFactory wrapper that triggers an HTTP proxy CONNECT on connect.
+
+    Once the CONNECT completes, invokes the original ClientFactory to build the
+    HTTP Protocol object and run the rest of the connection.
+
+    Args:
+        dst_host (bytes): hostname that we want to CONNECT to
+        dst_port (int): port that we want to connect to
+        wrapped_factory (protocol.ClientFactory): The original Factory
+    """
+
+    def __init__(self, dst_host, dst_port, wrapped_factory):
+        self.dst_host = dst_host
+        self.dst_port = dst_port
+        self.wrapped_factory = wrapped_factory
+        self.on_connection = defer.Deferred()
+
+    def startedConnecting(self, connector):
+        return self.wrapped_factory.startedConnecting(connector)
+
+    def buildProtocol(self, addr):
+        wrapped_protocol = self.wrapped_factory.buildProtocol(addr)
+
+        return HTTPConnectProtocol(
+            self.dst_host, self.dst_port, wrapped_protocol, self.on_connection
+        )
+
+    def clientConnectionFailed(self, connector, reason):
+        logger.debug("Connection to proxy failed: %s", reason)
+        if not self.on_connection.called:
+            self.on_connection.errback(reason)
+        return self.wrapped_factory.clientConnectionFailed(connector, reason)
+
+    def clientConnectionLost(self, connector, reason):
+        logger.debug("Connection to proxy lost: %s", reason)
+        if not self.on_connection.called:
+            self.on_connection.errback(reason)
+        return self.wrapped_factory.clientConnectionLost(connector, reason)
+
+
+class HTTPConnectProtocol(protocol.Protocol):
+    """Protocol that wraps an existing Protocol to do a CONNECT handshake at connect
+
+    Args:
+        host (bytes): The original HTTP(s) hostname or IPv4 or IPv6 address literal
+            to put in the CONNECT request
+
+        port (int): The original HTTP(s) port to put in the CONNECT request
+
+        wrapped_protocol (interfaces.IProtocol): the original protocol (probably
+            HTTPChannel or TLSMemoryBIOProtocol, but could be anything really)
+
+        connected_deferred (Deferred): a Deferred which will be callbacked with
+            wrapped_protocol when the CONNECT completes
+    """
+
+    def __init__(self, host, port, wrapped_protocol, connected_deferred):
+        self.host = host
+        self.port = port
+        self.wrapped_protocol = wrapped_protocol
+        self.connected_deferred = connected_deferred
+        self.http_setup_client = HTTPConnectSetupClient(self.host, self.port)
+        self.http_setup_client.on_connected.addCallback(self.proxyConnected)
+
+    def connectionMade(self):
+        self.http_setup_client.makeConnection(self.transport)
+
+    def connectionLost(self, reason=connectionDone):
+        if self.wrapped_protocol.connected:
+            self.wrapped_protocol.connectionLost(reason)
+
+        self.http_setup_client.connectionLost(reason)
+
+        if not self.connected_deferred.called:
+            self.connected_deferred.errback(reason)
+
+    def proxyConnected(self, _):
+        self.wrapped_protocol.makeConnection(self.transport)
+
+        self.connected_deferred.callback(self.wrapped_protocol)
+
+        # Get any pending data from the http buf and forward it to the original protocol
+        buf = self.http_setup_client.clearLineBuffer()
+        if buf:
+            self.wrapped_protocol.dataReceived(buf)
+
+    def dataReceived(self, data):
+        # if we've set up the HTTP protocol, we can send the data there
+        if self.wrapped_protocol.connected:
+            return self.wrapped_protocol.dataReceived(data)
+
+        # otherwise, we must still be setting up the connection: send the data to the
+        # setup client
+        return self.http_setup_client.dataReceived(data)
+
+
+class HTTPConnectSetupClient(http.HTTPClient):
+    """HTTPClient protocol to send a CONNECT message for proxies and read the response.
+
+    Args:
+        host (bytes): The hostname to send in the CONNECT message
+        port (int): The port to send in the CONNECT message
+    """
+
+    def __init__(self, host, port):
+        self.host = host
+        self.port = port
+        self.on_connected = defer.Deferred()
+
+    def connectionMade(self):
+        logger.debug("Connected to proxy, sending CONNECT")
+        self.sendCommand(b"CONNECT", b"%s:%d" % (self.host, self.port))
+        self.endHeaders()
+
+    def handleStatus(self, version, status, message):
+        logger.debug("Got Status: %s %s %s", status, message, version)
+        if status != b"200":
+            raise ProxyConnectError("Unexpected status on CONNECT: %s" % status)
+
+    def handleEndHeaders(self):
+        logger.debug("End Headers")
+        self.on_connected.callback(None)
+
+    def handleResponse(self, body):
+        pass
diff --git a/synapse/http/federation/matrix_federation_agent.py b/synapse/http/federation/matrix_federation_agent.py
index feae7de5be..647d26dc56 100644
--- a/synapse/http/federation/matrix_federation_agent.py
+++ b/synapse/http/federation/matrix_federation_agent.py
@@ -217,7 +217,7 @@ class MatrixHostnameEndpoint(object):
             self._tls_options = None
         else:
             self._tls_options = tls_client_options_factory.get_options(
-                self._parsed_uri.host.decode("ascii")
+                self._parsed_uri.host
             )
 
         self._srv_resolver = srv_resolver
diff --git a/synapse/http/federation/srv_resolver.py b/synapse/http/federation/srv_resolver.py
index 3fe4ffb9e5..021b233a7d 100644
--- a/synapse/http/federation/srv_resolver.py
+++ b/synapse/http/federation/srv_resolver.py
@@ -148,7 +148,7 @@ class SrvResolver(object):
             # Try something in the cache, else rereaise
             cache_entry = self._cache.get(service_name, None)
             if cache_entry:
-                logger.warn(
+                logger.warning(
                     "Failed to resolve %r, falling back to cache. %r", service_name, e
                 )
                 return list(cache_entry)
diff --git a/synapse/http/matrixfederationclient.py b/synapse/http/matrixfederationclient.py
index 4326e98a28..691380abda 100644
--- a/synapse/http/matrixfederationclient.py
+++ b/synapse/http/matrixfederationclient.py
@@ -149,7 +149,7 @@ def _handle_json_response(reactor, timeout_sec, request, response):
 
         body = yield make_deferred_yieldable(d)
     except Exception as e:
-        logger.warn(
+        logger.warning(
             "{%s} [%s] Error reading response: %s",
             request.txn_id,
             request.destination,
@@ -345,7 +345,6 @@ class MatrixFederationHttpClient(object):
         else:
             query_bytes = b""
 
-        # Retreive current span
         scope = start_active_span(
             "outgoing-federation-request",
             tags={
@@ -458,7 +457,7 @@ class MatrixFederationHttpClient(object):
                         except Exception as e:
                             # Eh, we're already going to raise an exception so lets
                             # ignore if this fails.
-                            logger.warn(
+                            logger.warning(
                                 "{%s} [%s] Failed to get error response: %s %s: %s",
                                 request.txn_id,
                                 request.destination,
@@ -479,7 +478,7 @@ class MatrixFederationHttpClient(object):
 
                     break
                 except RequestSendFailed as e:
-                    logger.warn(
+                    logger.warning(
                         "{%s} [%s] Request failed: %s %s: %s",
                         request.txn_id,
                         request.destination,
@@ -514,7 +513,7 @@ class MatrixFederationHttpClient(object):
                         raise
 
                 except Exception as e:
-                    logger.warn(
+                    logger.warning(
                         "{%s} [%s] Request failed: %s %s: %s",
                         request.txn_id,
                         request.destination,
@@ -890,7 +889,7 @@ class MatrixFederationHttpClient(object):
             d.addTimeout(self.default_timeout, self.reactor)
             length = yield make_deferred_yieldable(d)
         except Exception as e:
-            logger.warn(
+            logger.warning(
                 "{%s} [%s] Error reading response: %s",
                 request.txn_id,
                 request.destination,
diff --git a/synapse/http/proxyagent.py b/synapse/http/proxyagent.py
new file mode 100644
index 0000000000..332da02a8d
--- /dev/null
+++ b/synapse/http/proxyagent.py
@@ -0,0 +1,195 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import logging
+import re
+
+from zope.interface import implementer
+
+from twisted.internet import defer
+from twisted.internet.endpoints import HostnameEndpoint, wrapClientTLS
+from twisted.python.failure import Failure
+from twisted.web.client import URI, BrowserLikePolicyForHTTPS, _AgentBase
+from twisted.web.error import SchemeNotSupported
+from twisted.web.iweb import IAgent
+
+from synapse.http.connectproxyclient import HTTPConnectProxyEndpoint
+
+logger = logging.getLogger(__name__)
+
+_VALID_URI = re.compile(br"\A[\x21-\x7e]+\Z")
+
+
+@implementer(IAgent)
+class ProxyAgent(_AgentBase):
+    """An Agent implementation which will use an HTTP proxy if one was requested
+
+    Args:
+        reactor: twisted reactor to place outgoing
+            connections.
+
+        contextFactory (IPolicyForHTTPS): A factory for TLS contexts, to control the
+            verification parameters of OpenSSL.  The default is to use a
+            `BrowserLikePolicyForHTTPS`, so unless you have special
+            requirements you can leave this as-is.
+
+        connectTimeout (float): The amount of time that this Agent will wait
+            for the peer to accept a connection.
+
+        bindAddress (bytes): The local address for client sockets to bind to.
+
+        pool (HTTPConnectionPool|None): connection pool to be used. If None, a
+            non-persistent pool instance will be created.
+    """
+
+    def __init__(
+        self,
+        reactor,
+        contextFactory=BrowserLikePolicyForHTTPS(),
+        connectTimeout=None,
+        bindAddress=None,
+        pool=None,
+        http_proxy=None,
+        https_proxy=None,
+    ):
+        _AgentBase.__init__(self, reactor, pool)
+
+        self._endpoint_kwargs = {}
+        if connectTimeout is not None:
+            self._endpoint_kwargs["timeout"] = connectTimeout
+        if bindAddress is not None:
+            self._endpoint_kwargs["bindAddress"] = bindAddress
+
+        self.http_proxy_endpoint = _http_proxy_endpoint(
+            http_proxy, reactor, **self._endpoint_kwargs
+        )
+
+        self.https_proxy_endpoint = _http_proxy_endpoint(
+            https_proxy, reactor, **self._endpoint_kwargs
+        )
+
+        self._policy_for_https = contextFactory
+        self._reactor = reactor
+
+    def request(self, method, uri, headers=None, bodyProducer=None):
+        """
+        Issue a request to the server indicated by the given uri.
+
+        Supports `http` and `https` schemes.
+
+        An existing connection from the connection pool may be used or a new one may be
+        created.
+
+        See also: twisted.web.iweb.IAgent.request
+
+        Args:
+            method (bytes): The request method to use, such as `GET`, `POST`, etc
+
+            uri (bytes): The location of the resource to request.
+
+            headers (Headers|None): Extra headers to send with the request
+
+            bodyProducer (IBodyProducer|None): An object which can generate bytes to
+                make up the body of this request (for example, the properly encoded
+                contents of a file for a file upload). Or, None if the request is to
+                have no body.
+
+        Returns:
+            Deferred[IResponse]: completes when the header of the response has
+                 been received (regardless of the response status code).
+        """
+        uri = uri.strip()
+        if not _VALID_URI.match(uri):
+            raise ValueError("Invalid URI {!r}".format(uri))
+
+        parsed_uri = URI.fromBytes(uri)
+        pool_key = (parsed_uri.scheme, parsed_uri.host, parsed_uri.port)
+        request_path = parsed_uri.originForm
+
+        if parsed_uri.scheme == b"http" and self.http_proxy_endpoint:
+            # Cache *all* connections under the same key, since we are only
+            # connecting to a single destination, the proxy:
+            pool_key = ("http-proxy", self.http_proxy_endpoint)
+            endpoint = self.http_proxy_endpoint
+            request_path = uri
+        elif parsed_uri.scheme == b"https" and self.https_proxy_endpoint:
+            endpoint = HTTPConnectProxyEndpoint(
+                self._reactor,
+                self.https_proxy_endpoint,
+                parsed_uri.host,
+                parsed_uri.port,
+            )
+        else:
+            # not using a proxy
+            endpoint = HostnameEndpoint(
+                self._reactor, parsed_uri.host, parsed_uri.port, **self._endpoint_kwargs
+            )
+
+        logger.debug("Requesting %s via %s", uri, endpoint)
+
+        if parsed_uri.scheme == b"https":
+            tls_connection_creator = self._policy_for_https.creatorForNetloc(
+                parsed_uri.host, parsed_uri.port
+            )
+            endpoint = wrapClientTLS(tls_connection_creator, endpoint)
+        elif parsed_uri.scheme == b"http":
+            pass
+        else:
+            return defer.fail(
+                Failure(
+                    SchemeNotSupported("Unsupported scheme: %r" % (parsed_uri.scheme,))
+                )
+            )
+
+        return self._requestWithEndpoint(
+            pool_key, endpoint, method, parsed_uri, headers, bodyProducer, request_path
+        )
+
+
+def _http_proxy_endpoint(proxy, reactor, **kwargs):
+    """Parses an http proxy setting and returns an endpoint for the proxy
+
+    Args:
+        proxy (bytes|None):  the proxy setting
+        reactor: reactor to be used to connect to the proxy
+        kwargs: other args to be passed to HostnameEndpoint
+
+    Returns:
+        interfaces.IStreamClientEndpoint|None: endpoint to use to connect to the proxy,
+            or None
+    """
+    if proxy is None:
+        return None
+
+    # currently we only support hostname:port. Some apps also support
+    # protocol://[:port], which allows a way of requiring a TLS connection to the
+    # proxy.
+
+    host, port = parse_host_port(proxy, default_port=1080)
+    return HostnameEndpoint(reactor, host, port, **kwargs)
+
+
+def parse_host_port(hostport, default_port=None):
+    # could have sworn we had one of these somewhere else...
+    if b":" in hostport:
+        host, port = hostport.rsplit(b":", 1)
+        try:
+            port = int(port)
+            return host, port
+        except ValueError:
+            # the thing after the : wasn't a valid port; presumably this is an
+            # IPv6 address.
+            pass
+
+    return hostport, default_port
diff --git a/synapse/http/request_metrics.py b/synapse/http/request_metrics.py
index 46af27c8f6..58f9cc61c8 100644
--- a/synapse/http/request_metrics.py
+++ b/synapse/http/request_metrics.py
@@ -170,7 +170,7 @@ class RequestMetrics(object):
             tag = context.tag
 
             if context != self.start_context:
-                logger.warn(
+                logger.warning(
                     "Context have unexpectedly changed %r, %r",
                     context,
                     self.start_context,
diff --git a/synapse/http/server.py b/synapse/http/server.py
index e6f351ba3b..943d12c907 100644
--- a/synapse/http/server.py
+++ b/synapse/http/server.py
@@ -40,6 +40,7 @@ from synapse.api.errors import (
     UnrecognizedRequestError,
 )
 from synapse.logging.context import preserve_fn
+from synapse.logging.opentracing import trace_servlet
 from synapse.util.caches import intern_dict
 
 logger = logging.getLogger(__name__)
@@ -257,7 +258,9 @@ class JsonResource(HttpServer, resource.Resource):
         self.path_regexs = {}
         self.hs = hs
 
-    def register_paths(self, method, path_patterns, callback, servlet_classname):
+    def register_paths(
+        self, method, path_patterns, callback, servlet_classname, trace=True
+    ):
         """
         Registers a request handler against a regular expression. Later request URLs are
         checked against these regular expressions in order to identify an appropriate
@@ -273,8 +276,16 @@ class JsonResource(HttpServer, resource.Resource):
 
             servlet_classname (str): The name of the handler to be used in prometheus
                 and opentracing logs.
+
+            trace (bool): Whether we should start a span to trace the servlet.
         """
         method = method.encode("utf-8")  # method is bytes on py3
+
+        if trace:
+            # We don't extract the context from the servlet because we can't
+            # trust the sender
+            callback = trace_servlet(servlet_classname)(callback)
+
         for path_pattern in path_patterns:
             logger.debug("Registering for %s %s", method, path_pattern.pattern)
             self.path_regexs.setdefault(method, []).append(
@@ -377,7 +388,7 @@ class DirectServeResource(resource.Resource):
         if not callback:
             return super().render(request)
 
-        resp = callback(request)
+        resp = trace_servlet(self.__class__.__name__)(callback)(request)
 
         # If it's a coroutine, turn it into a Deferred
         if isinstance(resp, types.CoroutineType):
@@ -443,7 +454,7 @@ def respond_with_json(
     # the Deferred fires, but since the flag is RIGHT THERE it seems like
     # a waste.
     if request._disconnected:
-        logger.warn(
+        logger.warning(
             "Not sending response to request %s, already disconnected.", request
         )
         return
diff --git a/synapse/http/servlet.py b/synapse/http/servlet.py
index c186b31f59..e9a5e46ced 100644
--- a/synapse/http/servlet.py
+++ b/synapse/http/servlet.py
@@ -20,7 +20,6 @@ import logging
 from canonicaljson import json
 
 from synapse.api.errors import Codes, SynapseError
-from synapse.logging.opentracing import trace_servlet
 
 logger = logging.getLogger(__name__)
 
@@ -220,13 +219,13 @@ def parse_json_value_from_request(request, allow_empty_body=False):
     try:
         content_unicode = content_bytes.decode("utf8")
     except UnicodeDecodeError:
-        logger.warn("Unable to decode UTF-8")
+        logger.warning("Unable to decode UTF-8")
         raise SynapseError(400, "Content not JSON.", errcode=Codes.NOT_JSON)
 
     try:
         content = json.loads(content_unicode)
     except Exception as e:
-        logger.warn("Unable to parse JSON: %s", e)
+        logger.warning("Unable to parse JSON: %s", e)
         raise SynapseError(400, "Content not JSON.", errcode=Codes.NOT_JSON)
 
     return content
@@ -298,10 +297,7 @@ class RestServlet(object):
                     servlet_classname = self.__class__.__name__
                     method_handler = getattr(self, "on_%s" % (method,))
                     http_server.register_paths(
-                        method,
-                        patterns,
-                        trace_servlet(servlet_classname)(method_handler),
-                        servlet_classname,
+                        method, patterns, method_handler, servlet_classname
                     )
 
         else:
diff --git a/synapse/http/site.py b/synapse/http/site.py
index df5274c177..ff8184a3d0 100644
--- a/synapse/http/site.py
+++ b/synapse/http/site.py
@@ -199,7 +199,7 @@ class SynapseRequest(Request):
         # It's useful to log it here so that we can get an idea of when
         # the client disconnects.
         with PreserveLoggingContext(self.logcontext):
-            logger.warn(
+            logger.warning(
                 "Error processing request %r: %s %s", self, reason.type, reason.value
             )
 
@@ -305,7 +305,7 @@ class SynapseRequest(Request):
         try:
             self.request_metrics.stop(self.finish_time, self.code, self.sentLength)
         except Exception as e:
-            logger.warn("Failed to stop metrics: %r", e)
+            logger.warning("Failed to stop metrics: %r", e)
 
 
 class XForwardedForRequest(SynapseRequest):
diff --git a/synapse/logging/_structured.py b/synapse/logging/_structured.py
index 0367d6dfc4..334ddaf39a 100644
--- a/synapse/logging/_structured.py
+++ b/synapse/logging/_structured.py
@@ -18,6 +18,7 @@ import os.path
 import sys
 import typing
 import warnings
+from typing import List
 
 import attr
 from constantly import NamedConstant, Names, ValueConstant, Values
@@ -33,7 +34,6 @@ from twisted.logger import (
     LogLevelFilterPredicate,
     LogPublisher,
     eventAsText,
-    globalLogBeginner,
     jsonFileLogObserver,
 )
 
@@ -134,7 +134,7 @@ class PythonStdlibToTwistedLogger(logging.Handler):
         )
 
 
-def SynapseFileLogObserver(outFile: typing.io.TextIO) -> FileLogObserver:
+def SynapseFileLogObserver(outFile: typing.IO[str]) -> FileLogObserver:
     """
     A log observer that formats events like the traditional log formatter and
     sends them to `outFile`.
@@ -185,7 +185,7 @@ DEFAULT_LOGGERS = {"synapse": {"level": "INFO"}}
 
 
 def parse_drain_configs(
-    drains: dict
+    drains: dict,
 ) -> typing.Generator[DrainConfiguration, None, None]:
     """
     Parse the drain configurations.
@@ -265,7 +265,7 @@ def setup_structured_logging(
     hs,
     config,
     log_config: dict,
-    logBeginner: LogBeginner = globalLogBeginner,
+    logBeginner: LogBeginner,
     redirect_stdlib_logging: bool = True,
 ) -> LogPublisher:
     """
@@ -286,7 +286,7 @@ def setup_structured_logging(
     if "drains" not in log_config:
         raise ConfigError("The logging configuration requires a list of drains.")
 
-    observers = []
+    observers = []  # type: List[ILogObserver]
 
     for observer in parse_drain_configs(log_config["drains"]):
         # Pipe drains
diff --git a/synapse/logging/_terse_json.py b/synapse/logging/_terse_json.py
index 7f1e8f23fe..0ebbde06f2 100644
--- a/synapse/logging/_terse_json.py
+++ b/synapse/logging/_terse_json.py
@@ -21,10 +21,11 @@ import sys
 from collections import deque
 from ipaddress import IPv4Address, IPv6Address, ip_address
 from math import floor
-from typing.io import TextIO
+from typing import IO
 
 import attr
 from simplejson import dumps
+from zope.interface import implementer
 
 from twisted.application.internet import ClientService
 from twisted.internet.endpoints import (
@@ -33,7 +34,7 @@ from twisted.internet.endpoints import (
     TCP6ClientEndpoint,
 )
 from twisted.internet.protocol import Factory, Protocol
-from twisted.logger import FileLogObserver, Logger
+from twisted.logger import FileLogObserver, ILogObserver, Logger
 from twisted.python.failure import Failure
 
 
@@ -129,7 +130,7 @@ def flatten_event(event: dict, metadata: dict, include_time: bool = False):
     return new_event
 
 
-def TerseJSONToConsoleLogObserver(outFile: TextIO, metadata: dict) -> FileLogObserver:
+def TerseJSONToConsoleLogObserver(outFile: IO[str], metadata: dict) -> FileLogObserver:
     """
     A log observer that formats events to a flattened JSON representation.
 
@@ -146,6 +147,7 @@ def TerseJSONToConsoleLogObserver(outFile: TextIO, metadata: dict) -> FileLogObs
 
 
 @attr.s
+@implementer(ILogObserver)
 class TerseJSONToTCPLogObserver(object):
     """
     An IObserver that writes JSON logs to a TCP target.
diff --git a/synapse/logging/context.py b/synapse/logging/context.py
index 63379bfb93..2c1fb9ddac 100644
--- a/synapse/logging/context.py
+++ b/synapse/logging/context.py
@@ -1,4 +1,5 @@
 # Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -42,13 +43,17 @@ try:
     # exception.
     resource.getrusage(RUSAGE_THREAD)
 
+    is_thread_resource_usage_supported = True
+
     def get_thread_resource_usage():
         return resource.getrusage(RUSAGE_THREAD)
 
 
 except Exception:
     # If the system doesn't support resource.getrusage(RUSAGE_THREAD) then we
-    # won't track resource usage by returning None.
+    # won't track resource usage.
+    is_thread_resource_usage_supported = False
+
     def get_thread_resource_usage():
         return None
 
@@ -289,7 +294,7 @@ class LoggingContext(object):
         """Enters this logging context into thread local storage"""
         old_context = self.set_current_context(self)
         if self.previous_context != old_context:
-            logger.warn(
+            logger.warning(
                 "Expected previous context %r, found %r",
                 self.previous_context,
                 old_context,
@@ -359,7 +364,11 @@ class LoggingContext(object):
 
         # When we stop, let's record the cpu used since we started
         if not self.usage_start:
-            logger.warning("Called stop on logcontext %s without calling start", self)
+            # Log a warning on platforms that support thread usage tracking
+            if is_thread_resource_usage_supported:
+                logger.warning(
+                    "Called stop on logcontext %s without calling start", self
+                )
             return
 
         utime_delta, stime_delta = self._get_cputime()
diff --git a/synapse/logging/opentracing.py b/synapse/logging/opentracing.py
index dd296027a1..0638cec429 100644
--- a/synapse/logging/opentracing.py
+++ b/synapse/logging/opentracing.py
@@ -85,14 +85,14 @@ the function becomes the operation name for the span.
        return something_usual_and_useful
 
 
-Operation names can be explicitly set for functions by using
-``trace_using_operation_name``
+Operation names can be explicitly set for a function by passing the
+operation name to ``trace``
 
 .. code-block:: python
 
-   from synapse.logging.opentracing import trace_using_operation_name
+   from synapse.logging.opentracing import trace
 
-   @trace_using_operation_name("A *much* better operation name")
+   @trace(opname="a_better_operation_name")
    def interesting_badly_named_function(*args, **kwargs):
        # Does all kinds of cool and expected things
        return something_usual_and_useful
@@ -169,7 +169,9 @@ import contextlib
 import inspect
 import logging
 import re
+import types
 from functools import wraps
+from typing import Dict
 
 from canonicaljson import json
 
@@ -223,8 +225,8 @@ try:
     from jaeger_client import Config as JaegerConfig
     from synapse.logging.scopecontextmanager import LogContextScopeManager
 except ImportError:
-    JaegerConfig = None
-    LogContextScopeManager = None
+    JaegerConfig = None  # type: ignore
+    LogContextScopeManager = None  # type: ignore
 
 
 logger = logging.getLogger(__name__)
@@ -239,8 +241,7 @@ _homeserver_whitelist = None
 
 
 def only_if_tracing(func):
-    """Executes the function only if we're tracing. Otherwise return.
-    Assumes the function wrapped may return None"""
+    """Executes the function only if we're tracing. Otherwise returns None."""
 
     @wraps(func)
     def _only_if_tracing_inner(*args, **kwargs):
@@ -252,6 +253,41 @@ def only_if_tracing(func):
     return _only_if_tracing_inner
 
 
+def ensure_active_span(message, ret=None):
+    """Executes the operation only if opentracing is enabled and there is an active span.
+    If there is no active span it logs message at the error level.
+
+    Args:
+        message (str): Message which fills in "There was no active span when trying to %s"
+            in the error log if there is no active span and opentracing is enabled.
+        ret (object): return value if opentracing is None or there is no active span.
+
+    Returns (object): The result of the func or ret if opentracing is disabled or there
+        was no active span.
+    """
+
+    def ensure_active_span_inner_1(func):
+        @wraps(func)
+        def ensure_active_span_inner_2(*args, **kwargs):
+            if not opentracing:
+                return ret
+
+            if not opentracing.tracer.active_span:
+                logger.error(
+                    "There was no active span when trying to %s."
+                    " Did you forget to start one or did a context slip?",
+                    message,
+                )
+
+                return ret
+
+            return func(*args, **kwargs)
+
+        return ensure_active_span_inner_2
+
+    return ensure_active_span_inner_1
+
+
 @contextlib.contextmanager
 def _noop_context_manager(*args, **kwargs):
     """Does exactly what it says on the tin"""
@@ -319,7 +355,7 @@ def whitelisted_homeserver(destination):
     Args:
         destination (str)
         """
-    _homeserver_whitelist
+
     if _homeserver_whitelist:
         return _homeserver_whitelist.match(destination)
     return False
@@ -349,26 +385,24 @@ def start_active_span(
     if opentracing is None:
         return _noop_context_manager()
 
-    else:
-        # We need to enter the scope here for the logcontext to become active
-        return opentracing.tracer.start_active_span(
-            operation_name,
-            child_of=child_of,
-            references=references,
-            tags=tags,
-            start_time=start_time,
-            ignore_active_span=ignore_active_span,
-            finish_on_close=finish_on_close,
-        )
+    return opentracing.tracer.start_active_span(
+        operation_name,
+        child_of=child_of,
+        references=references,
+        tags=tags,
+        start_time=start_time,
+        ignore_active_span=ignore_active_span,
+        finish_on_close=finish_on_close,
+    )
 
 
 def start_active_span_follows_from(operation_name, contexts):
     if opentracing is None:
         return _noop_context_manager()
-    else:
-        references = [opentracing.follows_from(context) for context in contexts]
-        scope = start_active_span(operation_name, references=references)
-        return scope
+
+    references = [opentracing.follows_from(context) for context in contexts]
+    scope = start_active_span(operation_name, references=references)
+    return scope
 
 
 def start_active_span_from_request(
@@ -465,19 +499,19 @@ def start_active_span_from_edu(
 # Opentracing setters for tags, logs, etc
 
 
-@only_if_tracing
+@ensure_active_span("set a tag")
 def set_tag(key, value):
     """Sets a tag on the active span"""
     opentracing.tracer.active_span.set_tag(key, value)
 
 
-@only_if_tracing
+@ensure_active_span("log")
 def log_kv(key_values, timestamp=None):
     """Log to the active span"""
     opentracing.tracer.active_span.log_kv(key_values, timestamp)
 
 
-@only_if_tracing
+@ensure_active_span("set the traces operation name")
 def set_operation_name(operation_name):
     """Sets the operation name of the active span"""
     opentracing.tracer.active_span.set_operation_name(operation_name)
@@ -486,13 +520,18 @@ def set_operation_name(operation_name):
 # Injection and extraction
 
 
-@only_if_tracing
+@ensure_active_span("inject the span into a header")
 def inject_active_span_twisted_headers(headers, destination, check_destination=True):
     """
     Injects a span context into twisted headers in-place
 
     Args:
         headers (twisted.web.http_headers.Headers)
+        destination (str): address of entity receiving the span context. If check_destination
+            is true the context will only be injected if the destination matches the
+            opentracing whitelist
+        check_destination (bool): If false, destination will be ignored and the context
+            will always be injected.
         span (opentracing.Span)
 
     Returns:
@@ -510,14 +549,14 @@ def inject_active_span_twisted_headers(headers, destination, check_destination=T
         return
 
     span = opentracing.tracer.active_span
-    carrier = {}
+    carrier = {}  # type: Dict[str, str]
     opentracing.tracer.inject(span, opentracing.Format.HTTP_HEADERS, carrier)
 
     for key, value in carrier.items():
         headers.addRawHeaders(key, value)
 
 
-@only_if_tracing
+@ensure_active_span("inject the span into a byte dict")
 def inject_active_span_byte_dict(headers, destination, check_destination=True):
     """
     Injects a span context into a dict where the headers are encoded as byte
@@ -525,6 +564,11 @@ def inject_active_span_byte_dict(headers, destination, check_destination=True):
 
     Args:
         headers (dict)
+        destination (str): address of entity receiving the span context. If check_destination
+            is true the context will only be injected if the destination matches the
+            opentracing whitelist
+        check_destination (bool): If false, destination will be ignored and the context
+            will always be injected.
         span (opentracing.Span)
 
     Returns:
@@ -537,28 +581,30 @@ def inject_active_span_byte_dict(headers, destination, check_destination=True):
         here:
         https://github.com/jaegertracing/jaeger-client-python/blob/master/jaeger_client/constants.py
     """
-    if not whitelisted_homeserver(destination):
+    if check_destination and not whitelisted_homeserver(destination):
         return
 
     span = opentracing.tracer.active_span
 
-    carrier = {}
+    carrier = {}  # type: Dict[str, str]
     opentracing.tracer.inject(span, opentracing.Format.HTTP_HEADERS, carrier)
 
     for key, value in carrier.items():
         headers[key.encode()] = [value.encode()]
 
 
-@only_if_tracing
+@ensure_active_span("inject the span into a text map")
 def inject_active_span_text_map(carrier, destination, check_destination=True):
     """
     Injects a span context into a dict
 
     Args:
         carrier (dict)
-        destination (str): the name of the remote server. The span context
-        will only be injected if the destination matches the homeserver_whitelist
-        or destination is None.
+        destination (str): address of entity receiving the span context. If check_destination
+            is true the context will only be injected if the destination matches the
+            opentracing whitelist
+        check_destination (bool): If false, destination will be ignored and the context
+            will always be injected.
 
     Returns:
         In-place modification of carrier
@@ -579,6 +625,7 @@ def inject_active_span_text_map(carrier, destination, check_destination=True):
     )
 
 
+@ensure_active_span("get the active span context as a dict", ret={})
 def get_active_span_text_map(destination=None):
     """
     Gets a span context as a dict. This can be used instead of manually
@@ -591,10 +638,10 @@ def get_active_span_text_map(destination=None):
         dict: the active span's context if opentracing is enabled, otherwise empty.
     """
 
-    if not opentracing or (destination and not whitelisted_homeserver(destination)):
+    if destination and not whitelisted_homeserver(destination):
         return {}
 
-    carrier = {}
+    carrier = {}  # type: Dict[str, str]
     opentracing.tracer.inject(
         opentracing.tracer.active_span, opentracing.Format.TEXT_MAP, carrier
     )
@@ -602,12 +649,13 @@ def get_active_span_text_map(destination=None):
     return carrier
 
 
+@ensure_active_span("get the span context as a string.", ret={})
 def active_span_context_as_string():
     """
     Returns:
         The active span context encoded as a string.
     """
-    carrier = {}
+    carrier = {}  # type: Dict[str, str]
     if opentracing:
         opentracing.tracer.inject(
             opentracing.tracer.active_span, opentracing.Format.TEXT_MAP, carrier
@@ -641,70 +689,30 @@ def extract_text_map(carrier):
 # Tracing decorators
 
 
-def trace(func):
+def trace(func=None, opname=None):
     """
     Decorator to trace a function.
-    Sets the operation name to that of the function's.
+    Sets the operation name to that of the function's or that given
+    as operation_name. See the module's doc string for usage
+    examples.
     """
-    if opentracing is None:
-        return func
 
-    @wraps(func)
-    def _trace_inner(self, *args, **kwargs):
-        if opentracing is None:
-            return func(self, *args, **kwargs)
-
-        scope = start_active_span(func.__name__)
-        scope.__enter__()
-
-        try:
-            result = func(self, *args, **kwargs)
-            if isinstance(result, defer.Deferred):
-
-                def call_back(result):
-                    scope.__exit__(None, None, None)
-                    return result
-
-                def err_back(result):
-                    scope.span.set_tag(tags.ERROR, True)
-                    scope.__exit__(None, None, None)
-                    return result
-
-                result.addCallbacks(call_back, err_back)
-
-            else:
-                scope.__exit__(None, None, None)
-
-            return result
-
-        except Exception as e:
-            scope.__exit__(type(e), None, e.__traceback__)
-            raise
-
-    return _trace_inner
-
-
-def trace_using_operation_name(operation_name):
-    """Decorator to trace a function. Explicitely sets the operation_name."""
-
-    def trace(func):
-        """
-        Decorator to trace a function.
-        Sets the operation name to that of the function's.
-        """
+    def decorator(func):
         if opentracing is None:
             return func
 
-        @wraps(func)
-        def _trace_inner(self, *args, **kwargs):
-            if opentracing is None:
-                return func(self, *args, **kwargs)
+        _opname = opname if opname else func.__name__
 
-            scope = start_active_span(operation_name)
+        @wraps(func)
+        def _trace_inner(*args, **kwargs):
+            if opentracing is None:
+                return func(*args, **kwargs)
+
+            scope = start_active_span(_opname)
             scope.__enter__()
 
             try:
-                result = func(self, *args, **kwargs)
+                result = func(*args, **kwargs)
                 if isinstance(result, defer.Deferred):
 
                     def call_back(result):
@@ -717,6 +725,7 @@ def trace_using_operation_name(operation_name):
                         return result
 
                     result.addCallbacks(call_back, err_back)
+
                 else:
                     scope.__exit__(None, None, None)
 
@@ -728,7 +737,10 @@ def trace_using_operation_name(operation_name):
 
         return _trace_inner
 
-    return trace
+    if func:
+        return decorator(func)
+    else:
+        return decorator
 
 
 def tag_args(func):
@@ -740,13 +752,13 @@ def tag_args(func):
         return func
 
     @wraps(func)
-    def _tag_args_inner(self, *args, **kwargs):
+    def _tag_args_inner(*args, **kwargs):
         argspec = inspect.getargspec(func)
         for i, arg in enumerate(argspec.args[1:]):
             set_tag("ARG_" + arg, args[i])
         set_tag("args", args[len(argspec.args) :])
         set_tag("kwargs", kwargs)
-        return func(self, *args, **kwargs)
+        return func(*args, **kwargs)
 
     return _tag_args_inner
 
@@ -767,8 +779,7 @@ def trace_servlet(servlet_name, extract_context=False):
             return func
 
         @wraps(func)
-        @defer.inlineCallbacks
-        def _trace_servlet_inner(request, *args, **kwargs):
+        async def _trace_servlet_inner(request, *args, **kwargs):
             request_tags = {
                 "request_id": request.get_request_id(),
                 tags.SPAN_KIND: tags.SPAN_KIND_RPC_SERVER,
@@ -785,8 +796,14 @@ def trace_servlet(servlet_name, extract_context=False):
                 scope = start_active_span(servlet_name, tags=request_tags)
 
             with scope:
-                result = yield defer.maybeDeferred(func, request, *args, **kwargs)
-                return result
+                result = func(request, *args, **kwargs)
+
+                if not isinstance(result, (types.CoroutineType, defer.Deferred)):
+                    # Some servlets aren't async and just return results
+                    # directly, so we handle that here.
+                    return result
+
+                return await result
 
         return _trace_servlet_inner
 
diff --git a/synapse/logging/utils.py b/synapse/logging/utils.py
index 7df0fa6087..6073fc2725 100644
--- a/synapse/logging/utils.py
+++ b/synapse/logging/utils.py
@@ -119,7 +119,11 @@ def trace_function(f):
         logger = logging.getLogger(name)
         level = logging.DEBUG
 
-        s = inspect.currentframe().f_back
+        frame = inspect.currentframe()
+        if frame is None:
+            raise Exception("Can't get current frame!")
+
+        s = frame.f_back
 
         to_print = [
             "\t%s:%s %s. Args: args=%s, kwargs=%s"
@@ -144,7 +148,7 @@ def trace_function(f):
             pathname=pathname,
             lineno=lineno,
             msg=msg,
-            args=None,
+            args=tuple(),
             exc_info=None,
         )
 
@@ -157,7 +161,12 @@ def trace_function(f):
 
 
 def get_previous_frames():
-    s = inspect.currentframe().f_back.f_back
+
+    frame = inspect.currentframe()
+    if frame is None:
+        raise Exception("Can't get current frame!")
+
+    s = frame.f_back.f_back
     to_return = []
     while s:
         if s.f_globals["__name__"].startswith("synapse"):
@@ -174,7 +183,10 @@ def get_previous_frames():
 
 
 def get_previous_frame(ignore=[]):
-    s = inspect.currentframe().f_back.f_back
+    frame = inspect.currentframe()
+    if frame is None:
+        raise Exception("Can't get current frame!")
+    s = frame.f_back.f_back
 
     while s:
         if s.f_globals["__name__"].startswith("synapse"):
diff --git a/synapse/metrics/__init__.py b/synapse/metrics/__init__.py
index 488280b4a6..0b45e1f52a 100644
--- a/synapse/metrics/__init__.py
+++ b/synapse/metrics/__init__.py
@@ -20,6 +20,7 @@ import os
 import platform
 import threading
 import time
+from typing import Dict, Union
 
 import six
 
@@ -29,20 +30,20 @@ from prometheus_client.core import REGISTRY, GaugeMetricFamily, HistogramMetricF
 
 from twisted.internet import reactor
 
+import synapse
 from synapse.metrics._exposition import (
     MetricsResource,
     generate_latest,
     start_http_server,
 )
+from synapse.util.versionstring import get_version_string
 
 logger = logging.getLogger(__name__)
 
 METRICS_PREFIX = "/_synapse/metrics"
 
 running_on_pypy = platform.python_implementation() == "PyPy"
-all_metrics = []
-all_collectors = []
-all_gauges = {}
+all_gauges = {}  # type: Dict[str, Union[LaterGauge, InFlightGauge, BucketCollector]]
 
 HAVE_PROC_SELF_STAT = os.path.exists("/proc/self/stat")
 
@@ -124,7 +125,7 @@ class InFlightGauge(object):
         )
 
         # Counts number of in flight blocks for a given set of label values
-        self._registrations = {}
+        self._registrations = {}  # type: Dict
 
         # Protects access to _registrations
         self._lock = threading.Lock()
@@ -225,7 +226,7 @@ class BucketCollector(object):
         # Fetch the data -- this must be synchronous!
         data = self.data_collector()
 
-        buckets = {}
+        buckets = {}  # type: Dict[float, int]
 
         res = []
         for x in data.keys():
@@ -385,6 +386,16 @@ event_processing_last_ts = Gauge("synapse_event_processing_last_ts", "", ["name"
 # finished being processed.
 event_processing_lag = Gauge("synapse_event_processing_lag", "", ["name"])
 
+# Build info of the running server.
+build_info = Gauge(
+    "synapse_build_info", "Build information", ["pythonversion", "version", "osversion"]
+)
+build_info.labels(
+    " ".join([platform.python_implementation(), platform.python_version()]),
+    get_version_string(synapse),
+    " ".join([platform.system(), platform.release()]),
+).set(1)
+
 last_ticked = time.time()
 
 
diff --git a/synapse/metrics/_exposition.py b/synapse/metrics/_exposition.py
index 1933ecd3e3..a248103191 100644
--- a/synapse/metrics/_exposition.py
+++ b/synapse/metrics/_exposition.py
@@ -36,7 +36,9 @@ from twisted.web.resource import Resource
 try:
     from prometheus_client.samples import Sample
 except ImportError:
-    Sample = namedtuple("Sample", ["name", "labels", "value", "timestamp", "exemplar"])
+    Sample = namedtuple(  # type: ignore[no-redef] # noqa
+        "Sample", ["name", "labels", "value", "timestamp", "exemplar"]
+    )
 
 
 CONTENT_TYPE_LATEST = str("text/plain; version=0.0.4; charset=utf-8")
diff --git a/synapse/metrics/background_process_metrics.py b/synapse/metrics/background_process_metrics.py
index edd6b42db3..c53d2a0d40 100644
--- a/synapse/metrics/background_process_metrics.py
+++ b/synapse/metrics/background_process_metrics.py
@@ -15,6 +15,8 @@
 
 import logging
 import threading
+from asyncio import iscoroutine
+from functools import wraps
 
 import six
 
@@ -173,7 +175,7 @@ def run_as_background_process(desc, func, *args, **kwargs):
 
     Args:
         desc (str): a description for this background process type
-        func: a function, which may return a Deferred
+        func: a function, which may return a Deferred or a coroutine
         args: positional args for func
         kwargs: keyword args for func
 
@@ -197,7 +199,17 @@ def run_as_background_process(desc, func, *args, **kwargs):
                 _background_processes.setdefault(desc, set()).add(proc)
 
             try:
-                yield func(*args, **kwargs)
+                result = func(*args, **kwargs)
+
+                # We probably don't have an ensureDeferred in our call stack to handle
+                # coroutine results, so we need to ensureDeferred here.
+                #
+                # But we need this check because ensureDeferred doesn't like being
+                # called on immediate values (as opposed to Deferreds or coroutines).
+                if iscoroutine(result):
+                    result = defer.ensureDeferred(result)
+
+                return (yield result)
             except Exception:
                 logger.exception("Background process '%s' threw an exception", desc)
             finally:
@@ -208,3 +220,20 @@ def run_as_background_process(desc, func, *args, **kwargs):
 
     with PreserveLoggingContext():
         return run()
+
+
+def wrap_as_background_process(desc):
+    """Decorator that wraps a function that gets called as a background
+    process.
+
+    Equivalent of calling the function with `run_as_background_process`
+    """
+
+    def wrap_as_background_process_inner(func):
+        @wraps(func)
+        def wrap_as_background_process_inner_2(*args, **kwargs):
+            return run_as_background_process(desc, func, *args, **kwargs)
+
+        return wrap_as_background_process_inner_2
+
+    return wrap_as_background_process_inner
diff --git a/synapse/notifier.py b/synapse/notifier.py
index 4e091314e6..af161a81d7 100644
--- a/synapse/notifier.py
+++ b/synapse/notifier.py
@@ -159,6 +159,7 @@ class Notifier(object):
         self.room_to_user_streams = {}
 
         self.hs = hs
+        self.storage = hs.get_storage()
         self.event_sources = hs.get_event_sources()
         self.store = hs.get_datastore()
         self.pending_new_room_events = []
@@ -425,7 +426,10 @@ class Notifier(object):
 
                 if name == "room":
                     new_events = yield filter_events_for_client(
-                        self.store, user.to_string(), new_events, is_peeking=is_peeking
+                        self.storage,
+                        user.to_string(),
+                        new_events,
+                        is_peeking=is_peeking,
                     )
                 elif name == "presence":
                     now = self.clock.time_msec()
diff --git a/synapse/push/bulk_push_rule_evaluator.py b/synapse/push/bulk_push_rule_evaluator.py
index 22491f3700..1ba7bcd4d8 100644
--- a/synapse/push/bulk_push_rule_evaluator.py
+++ b/synapse/push/bulk_push_rule_evaluator.py
@@ -79,7 +79,7 @@ class BulkPushRuleEvaluator(object):
             dict of user_id -> push_rules
         """
         room_id = event.room_id
-        rules_for_room = self._get_rules_for_room(room_id)
+        rules_for_room = yield self._get_rules_for_room(room_id)
 
         rules_by_user = yield rules_for_room.get_rules(event, context)
 
@@ -149,9 +149,10 @@ class BulkPushRuleEvaluator(object):
 
         room_members = yield self.store.get_joined_users_from_context(event, context)
 
-        (power_levels, sender_power_level) = (
-            yield self._get_power_levels_and_sender_level(event, context)
-        )
+        (
+            power_levels,
+            sender_power_level,
+        ) = yield self._get_power_levels_and_sender_level(event, context)
 
         evaluator = PushRuleEvaluatorForEvent(
             event, len(room_members), sender_power_level, power_levels
diff --git a/synapse/push/emailpusher.py b/synapse/push/emailpusher.py
index 42e5b0c0a5..8c818a86bf 100644
--- a/synapse/push/emailpusher.py
+++ b/synapse/push/emailpusher.py
@@ -234,14 +234,12 @@ class EmailPusher(object):
             return
 
         self.last_stream_ordering = last_stream_ordering
-        pusher_still_exists = (
-            yield self.store.update_pusher_last_stream_ordering_and_success(
-                self.app_id,
-                self.email,
-                self.user_id,
-                last_stream_ordering,
-                self.clock.time_msec(),
-            )
+        pusher_still_exists = yield self.store.update_pusher_last_stream_ordering_and_success(
+            self.app_id,
+            self.email,
+            self.user_id,
+            last_stream_ordering,
+            self.clock.time_msec(),
         )
         if not pusher_still_exists:
             # The pusher has been deleted while we were processing, so
diff --git a/synapse/push/httppusher.py b/synapse/push/httppusher.py
index bd5d53af91..e994037be6 100644
--- a/synapse/push/httppusher.py
+++ b/synapse/push/httppusher.py
@@ -22,6 +22,7 @@ from prometheus_client import Counter
 from twisted.internet import defer
 from twisted.internet.error import AlreadyCalled, AlreadyCancelled
 
+from synapse.logging import opentracing
 from synapse.metrics.background_process_metrics import run_as_background_process
 from synapse.push import PusherConfigException
 
@@ -63,6 +64,7 @@ class HttpPusher(object):
     def __init__(self, hs, pusherdict):
         self.hs = hs
         self.store = self.hs.get_datastore()
+        self.storage = self.hs.get_storage()
         self.clock = self.hs.get_clock()
         self.state_handler = self.hs.get_state_handler()
         self.user_id = pusherdict["user_name"]
@@ -101,7 +103,7 @@ class HttpPusher(object):
         if "url" not in self.data:
             raise PusherConfigException("'url' required in data for HTTP pusher")
         self.url = self.data["url"]
-        self.http_client = hs.get_simple_http_client()
+        self.http_client = hs.get_proxied_http_client()
         self.data_minus_url = {}
         self.data_minus_url.update(self.data)
         del self.data_minus_url["url"]
@@ -194,19 +196,27 @@ class HttpPusher(object):
         )
 
         for push_action in unprocessed:
-            processed = yield self._process_one(push_action)
+            with opentracing.start_active_span(
+                "http-push",
+                tags={
+                    "authenticated_entity": self.user_id,
+                    "event_id": push_action["event_id"],
+                    "app_id": self.app_id,
+                    "app_display_name": self.app_display_name,
+                },
+            ):
+                processed = yield self._process_one(push_action)
+
             if processed:
                 http_push_processed_counter.inc()
                 self.backoff_delay = HttpPusher.INITIAL_BACKOFF_SEC
                 self.last_stream_ordering = push_action["stream_ordering"]
-                pusher_still_exists = (
-                    yield self.store.update_pusher_last_stream_ordering_and_success(
-                        self.app_id,
-                        self.pushkey,
-                        self.user_id,
-                        self.last_stream_ordering,
-                        self.clock.time_msec(),
-                    )
+                pusher_still_exists = yield self.store.update_pusher_last_stream_ordering_and_success(
+                    self.app_id,
+                    self.pushkey,
+                    self.user_id,
+                    self.last_stream_ordering,
+                    self.clock.time_msec(),
                 )
                 if not pusher_still_exists:
                     # The pusher has been deleted while we were processing, so
@@ -235,7 +245,7 @@ class HttpPusher(object):
                     # we really only give up so that if the URL gets
                     # fixed, we don't suddenly deliver a load
                     # of old notifications.
-                    logger.warn(
+                    logger.warning(
                         "Giving up on a notification to user %s, " "pushkey %s",
                         self.user_id,
                         self.pushkey,
@@ -288,7 +298,7 @@ class HttpPusher(object):
                 if pk != self.pushkey:
                     # for sanity, we only remove the pushkey if it
                     # was the one we actually sent...
-                    logger.warn(
+                    logger.warning(
                         ("Ignoring rejected pushkey %s because we" " didn't send it"),
                         pk,
                     )
@@ -318,7 +328,7 @@ class HttpPusher(object):
             return d
 
         ctx = yield push_tools.get_context_for_event(
-            self.store, self.state_handler, event, self.user_id
+            self.storage, self.state_handler, event, self.user_id
         )
 
         d = {
diff --git a/synapse/push/mailer.py b/synapse/push/mailer.py
index 4245ce26f3..1d15a06a58 100644
--- a/synapse/push/mailer.py
+++ b/synapse/push/mailer.py
@@ -119,6 +119,7 @@ class Mailer(object):
         self.store = self.hs.get_datastore()
         self.macaroon_gen = self.hs.get_macaroon_generator()
         self.state_handler = self.hs.get_state_handler()
+        self.storage = hs.get_storage()
         self.app_name = app_name
 
         logger.info("Created Mailer for app_name %s" % app_name)
@@ -131,25 +132,80 @@ class Mailer(object):
             email_address (str): Email address we're sending the password
                 reset to
             token (str): Unique token generated by the server to verify
-                password reset email was received
+                the email was received
             client_secret (str): Unique token generated by the client to
                 group together multiple email sending attempts
             sid (str): The generated session ID
         """
-        if email.utils.parseaddr(email_address)[1] == "":
-            raise RuntimeError("Invalid 'to' email address")
-
+        params = {"token": token, "client_secret": client_secret, "sid": sid}
         link = (
             self.hs.config.public_baseurl
-            + "_matrix/client/unstable/password_reset/email/submit_token"
-            "?token=%s&client_secret=%s&sid=%s" % (token, client_secret, sid)
+            + "_matrix/client/unstable/password_reset/email/submit_token?%s"
+            % urllib.parse.urlencode(params)
         )
 
         template_vars = {"link": link}
 
         yield self.send_email(
             email_address,
-            "[%s] Password Reset Email" % self.hs.config.server_name,
+            "[%s] Password Reset" % self.hs.config.server_name,
+            template_vars,
+        )
+
+    @defer.inlineCallbacks
+    def send_registration_mail(self, email_address, token, client_secret, sid):
+        """Send an email with a registration confirmation link to a user
+
+        Args:
+            email_address (str): Email address we're sending the registration
+                link to
+            token (str): Unique token generated by the server to verify
+                the email was received
+            client_secret (str): Unique token generated by the client to
+                group together multiple email sending attempts
+            sid (str): The generated session ID
+        """
+        params = {"token": token, "client_secret": client_secret, "sid": sid}
+        link = (
+            self.hs.config.public_baseurl
+            + "_matrix/client/unstable/registration/email/submit_token?%s"
+            % urllib.parse.urlencode(params)
+        )
+
+        template_vars = {"link": link}
+
+        yield self.send_email(
+            email_address,
+            "[%s] Register your Email Address" % self.hs.config.server_name,
+            template_vars,
+        )
+
+    @defer.inlineCallbacks
+    def send_add_threepid_mail(self, email_address, token, client_secret, sid):
+        """Send an email with a validation link to a user for adding a 3pid to their account
+
+        Args:
+            email_address (str): Email address we're sending the validation link to
+
+            token (str): Unique token generated by the server to verify the email was received
+
+            client_secret (str): Unique token generated by the client to group together
+                multiple email sending attempts
+
+            sid (str): The generated session ID
+        """
+        params = {"token": token, "client_secret": client_secret, "sid": sid}
+        link = (
+            self.hs.config.public_baseurl
+            + "_matrix/client/unstable/add_threepid/email/submit_token?%s"
+            % urllib.parse.urlencode(params)
+        )
+
+        template_vars = {"link": link}
+
+        yield self.send_email(
+            email_address,
+            "[%s] Validate Your Email" % self.hs.config.server_name,
             template_vars,
         )
 
@@ -256,7 +312,7 @@ class Mailer(object):
         multipart_msg.attach(text_part)
         multipart_msg.attach(html_part)
 
-        logger.info("Sending email notification to %s" % email_address)
+        logger.info("Sending email to %s" % email_address)
 
         yield make_deferred_yieldable(
             self.sendmail(
@@ -334,7 +390,7 @@ class Mailer(object):
         }
 
         the_events = yield filter_events_for_client(
-            self.store, user_id, results["events_before"]
+            self.storage, user_id, results["events_before"]
         )
         the_events.append(notif_event)
 
@@ -605,25 +661,50 @@ def format_ts_filter(value, format):
     return time.strftime(format, time.localtime(value / 1000))
 
 
-def load_jinja2_templates(config, template_html_name, template_text_name):
-    """Load the jinja2 email templates from disk
+def load_jinja2_templates(
+    template_dir,
+    template_filenames,
+    apply_format_ts_filter=False,
+    apply_mxc_to_http_filter=False,
+    public_baseurl=None,
+):
+    """Loads and returns one or more jinja2 templates and applies optional filters
+
+    Args:
+        template_dir (str): The directory where templates are stored
+        template_filenames (list[str]): A list of template filenames
+        apply_format_ts_filter (bool): Whether to apply a template filter that formats
+            timestamps
+        apply_mxc_to_http_filter (bool): Whether to apply a template filter that converts
+            mxc urls to http urls
+        public_baseurl (str|None): The public baseurl of the server. Required for
+            apply_mxc_to_http_filter to be enabled
 
     Returns:
-        (template_html, template_text)
+        A list of jinja2 templates corresponding to the given list of filenames,
+        with order preserved
     """
-    logger.info("loading email templates from '%s'", config.email_template_dir)
-    loader = jinja2.FileSystemLoader(config.email_template_dir)
+    logger.info(
+        "loading email templates %s from '%s'", template_filenames, template_dir
+    )
+    loader = jinja2.FileSystemLoader(template_dir)
     env = jinja2.Environment(loader=loader)
-    env.filters["format_ts"] = format_ts_filter
-    env.filters["mxc_to_http"] = _create_mxc_to_http_filter(config)
 
-    template_html = env.get_template(template_html_name)
-    template_text = env.get_template(template_text_name)
+    if apply_format_ts_filter:
+        env.filters["format_ts"] = format_ts_filter
 
-    return template_html, template_text
+    if apply_mxc_to_http_filter and public_baseurl:
+        env.filters["mxc_to_http"] = _create_mxc_to_http_filter(public_baseurl)
+
+    templates = []
+    for template_filename in template_filenames:
+        template = env.get_template(template_filename)
+        templates.append(template)
+
+    return templates
 
 
-def _create_mxc_to_http_filter(config):
+def _create_mxc_to_http_filter(public_baseurl):
     def mxc_to_http_filter(value, width, height, resize_method="crop"):
         if value[0:6] != "mxc://":
             return ""
@@ -636,7 +717,7 @@ def _create_mxc_to_http_filter(config):
 
         params = {"width": width, "height": height, "method": resize_method}
         return "%s_matrix/media/v1/thumbnail/%s?%s%s" % (
-            config.public_baseurl,
+            public_baseurl,
             serverAndMediaId,
             urllib.parse.urlencode(params),
             fragment or "",
diff --git a/synapse/push/push_rule_evaluator.py b/synapse/push/push_rule_evaluator.py
index 5ed9147de4..b1587183a8 100644
--- a/synapse/push/push_rule_evaluator.py
+++ b/synapse/push/push_rule_evaluator.py
@@ -117,7 +117,7 @@ class PushRuleEvaluatorForEvent(object):
                 pattern = UserID.from_string(user_id).localpart
 
         if not pattern:
-            logger.warn("event_match condition with no pattern")
+            logger.warning("event_match condition with no pattern")
             return False
 
         # XXX: optimisation: cache our pattern regexps
@@ -173,7 +173,7 @@ def _glob_matches(glob, value, word_boundary=False):
             regex_cache[(glob, word_boundary)] = r
         return r.search(value)
     except re.error:
-        logger.warn("Failed to parse glob to regex: %r", glob)
+        logger.warning("Failed to parse glob to regex: %r", glob)
         return False
 
 
diff --git a/synapse/push/push_tools.py b/synapse/push/push_tools.py
index a54051a726..de5c101a58 100644
--- a/synapse/push/push_tools.py
+++ b/synapse/push/push_tools.py
@@ -16,6 +16,7 @@
 from twisted.internet import defer
 
 from synapse.push.presentable_names import calculate_room_name, name_from_member_event
+from synapse.storage import Storage
 
 
 @defer.inlineCallbacks
@@ -43,22 +44,22 @@ def get_badge_count(store, user_id):
 
 
 @defer.inlineCallbacks
-def get_context_for_event(store, state_handler, ev, user_id):
+def get_context_for_event(storage: Storage, state_handler, ev, user_id):
     ctx = {}
 
-    room_state_ids = yield store.get_state_ids_for_event(ev.event_id)
+    room_state_ids = yield storage.state.get_state_ids_for_event(ev.event_id)
 
     # we no longer bother setting room_alias, and make room_name the
     # human-readable name instead, be that m.room.name, an alias or
     # a list of people in the room
     name = yield calculate_room_name(
-        store, room_state_ids, user_id, fallback_to_single_member=False
+        storage.main, room_state_ids, user_id, fallback_to_single_member=False
     )
     if name:
         ctx["name"] = name
 
     sender_state_event_id = room_state_ids[("m.room.member", ev.sender)]
-    sender_state_event = yield store.get_event(sender_state_event_id)
+    sender_state_event = yield storage.main.get_event(sender_state_event_id)
     ctx["sender_display_name"] = name_from_member_event(sender_state_event)
 
     return ctx
diff --git a/synapse/push/pusher.py b/synapse/push/pusher.py
index a9c64a9c54..f277aeb131 100644
--- a/synapse/push/pusher.py
+++ b/synapse/push/pusher.py
@@ -35,6 +35,7 @@ except Exception:
 class PusherFactory(object):
     def __init__(self, hs):
         self.hs = hs
+        self.config = hs.config
 
         self.pusher_types = {"http": HttpPusher}
 
@@ -42,12 +43,16 @@ class PusherFactory(object):
         if hs.config.email_enable_notifs:
             self.mailers = {}  # app_name -> Mailer
 
-            templates = load_jinja2_templates(
-                config=hs.config,
-                template_html_name=hs.config.email_notif_template_html,
-                template_text_name=hs.config.email_notif_template_text,
+            self.notif_template_html, self.notif_template_text = load_jinja2_templates(
+                self.config.email_template_dir,
+                [
+                    self.config.email_notif_template_html,
+                    self.config.email_notif_template_text,
+                ],
+                apply_format_ts_filter=True,
+                apply_mxc_to_http_filter=True,
+                public_baseurl=self.config.public_baseurl,
             )
-            self.notif_template_html, self.notif_template_text = templates
 
             self.pusher_types["email"] = self._create_email_pusher
 
@@ -78,6 +83,6 @@ class PusherFactory(object):
         if "data" in pusherdict and "brand" in pusherdict["data"]:
             app_name = pusherdict["data"]["brand"]
         else:
-            app_name = self.hs.config.email_app_name
+            app_name = self.config.email_app_name
 
         return app_name
diff --git a/synapse/push/pusherpool.py b/synapse/push/pusherpool.py
index 08e840fdc2..0f6992202d 100644
--- a/synapse/push/pusherpool.py
+++ b/synapse/push/pusherpool.py
@@ -103,9 +103,7 @@ class PusherPool:
         # create the pusher setting last_stream_ordering to the current maximum
         # stream ordering in event_push_actions, so it will process
         # pushes from this point onwards.
-        last_stream_ordering = (
-            yield self.store.get_latest_push_action_stream_ordering()
-        )
+        last_stream_ordering = yield self.store.get_latest_push_action_stream_ordering()
 
         yield self.store.add_pusher(
             user_id=user_id,
diff --git a/synapse/python_dependencies.py b/synapse/python_dependencies.py
index ec0ac547c1..5871feaafd 100644
--- a/synapse/python_dependencies.py
+++ b/synapse/python_dependencies.py
@@ -15,6 +15,7 @@
 # limitations under the License.
 
 import logging
+from typing import List, Set
 
 from pkg_resources import (
     DistributionNotFound,
@@ -60,7 +61,6 @@ REQUIREMENTS = [
     "bcrypt>=3.1.0",
     "pillow>=4.3.0",
     "sortedcontainers>=1.4.4",
-    "psutil>=2.0.0",
     "pymacaroons>=0.13.0",
     "msgpack>=0.5.2",
     "phonenumbers>=8.2.0",
@@ -72,6 +72,7 @@ REQUIREMENTS = [
     "netaddr>=0.7.18",
     "Jinja2>=2.9",
     "bleach>=1.4.3",
+    "typing-extensions>=3.7.4",
 ]
 
 CONDITIONAL_REQUIREMENTS = {
@@ -97,7 +98,7 @@ CONDITIONAL_REQUIREMENTS = {
     "jwt": ["pyjwt>=1.6.4"],
 }
 
-ALL_OPTIONAL_REQUIREMENTS = set()
+ALL_OPTIONAL_REQUIREMENTS = set()  # type: Set[str]
 
 for name, optional_deps in CONDITIONAL_REQUIREMENTS.items():
     # Exclude systemd as it's a system-based requirement.
@@ -143,16 +144,26 @@ def check_requirements(for_feature=None):
             deps_needed.append(dependency)
             errors.append(
                 "Needed %s, got %s==%s"
-                % (dependency, e.dist.project_name, e.dist.version)
+                % (
+                    dependency,
+                    e.dist.project_name,  # type: ignore[attr-defined] # noqa
+                    e.dist.version,  # type: ignore[attr-defined] # noqa
+                )
             )
         except DistributionNotFound:
             deps_needed.append(dependency)
-            errors.append("Needed %s but it was not installed" % (dependency,))
+            if for_feature:
+                errors.append(
+                    "Needed %s for the '%s' feature but it was not installed"
+                    % (dependency, for_feature)
+                )
+            else:
+                errors.append("Needed %s but it was not installed" % (dependency,))
 
     if not for_feature:
         # Check the optional dependencies are up to date. We allow them to not be
         # installed.
-        OPTS = sum(CONDITIONAL_REQUIREMENTS.values(), [])
+        OPTS = sum(CONDITIONAL_REQUIREMENTS.values(), [])  # type: List[str]
 
         for dependency in OPTS:
             try:
@@ -161,15 +172,19 @@ def check_requirements(for_feature=None):
                 deps_needed.append(dependency)
                 errors.append(
                     "Needed optional %s, got %s==%s"
-                    % (dependency, e.dist.project_name, e.dist.version)
+                    % (
+                        dependency,
+                        e.dist.project_name,  # type: ignore[attr-defined] # noqa
+                        e.dist.version,  # type: ignore[attr-defined] # noqa
+                    )
                 )
             except DistributionNotFound:
                 # If it's not found, we don't care
                 pass
 
     if deps_needed:
-        for e in errors:
-            logging.error(e)
+        for err in errors:
+            logging.error(err)
 
         raise DependencyException(deps_needed)
 
diff --git a/synapse/replication/http/_base.py b/synapse/replication/http/_base.py
index c4be9273f6..c8056b0c0c 100644
--- a/synapse/replication/http/_base.py
+++ b/synapse/replication/http/_base.py
@@ -22,13 +22,17 @@ from six.moves import urllib
 
 from twisted.internet import defer
 
-import synapse.logging.opentracing as opentracing
 from synapse.api.errors import (
     CodeMessageException,
     HttpResponseException,
     RequestSendFailed,
     SynapseError,
 )
+from synapse.logging.opentracing import (
+    inject_active_span_byte_dict,
+    trace,
+    trace_servlet,
+)
 from synapse.util.caches.response_cache import ResponseCache
 from synapse.util.stringutils import random_string
 
@@ -106,14 +110,14 @@ class ReplicationEndpoint(object):
         return {}
 
     @abc.abstractmethod
-    def _handle_request(self, request, **kwargs):
+    async def _handle_request(self, request, **kwargs):
         """Handle incoming request.
 
         This is called with the request object and PATH_ARGS.
 
         Returns:
-            Deferred[dict]: A JSON serialisable dict to be used as response
-            body of request.
+            tuple[int, dict]: HTTP status code and a JSON serialisable dict
+            to be used as response body of request.
         """
         pass
 
@@ -129,6 +133,7 @@ class ReplicationEndpoint(object):
 
         client = hs.get_simple_http_client()
 
+        @trace(opname="outgoing_replication_request")
         @defer.inlineCallbacks
         def send_request(**kwargs):
             data = yield cls._serialize_payload(**kwargs)
@@ -167,9 +172,7 @@ class ReplicationEndpoint(object):
                 # the master, and so whether we should clean up or not.
                 while True:
                     headers = {}
-                    opentracing.inject_active_span_byte_dict(
-                        headers, None, check_destination=False
-                    )
+                    inject_active_span_byte_dict(headers, None, check_destination=False)
                     try:
                         result = yield request_func(uri, data, headers=headers)
                         break
@@ -177,7 +180,7 @@ class ReplicationEndpoint(object):
                         if e.code != 504 or not cls.RETRY_ON_TIMEOUT:
                             raise
 
-                    logger.warn("%s request timed out", cls.NAME)
+                    logger.warning("%s request timed out", cls.NAME)
 
                     # If we timed out we probably don't need to worry about backing
                     # off too much, but lets just wait a little anyway.
@@ -210,13 +213,11 @@ class ReplicationEndpoint(object):
         args = "/".join("(?P<%s>[^/]+)" % (arg,) for arg in url_args)
         pattern = re.compile("^/_synapse/replication/%s/%s$" % (self.NAME, args))
 
+        handler = trace_servlet(self.__class__.__name__, extract_context=True)(handler)
+        # We don't let register paths trace this servlet using the default tracing
+        # options because we wish to extract the context explicitly.
         http_server.register_paths(
-            method,
-            [pattern],
-            opentracing.trace_servlet(self.__class__.__name__, extract_context=True)(
-                handler
-            ),
-            self.__class__.__name__,
+            method, [pattern], handler, self.__class__.__name__, trace=False
         )
 
     def _cached_handler(self, request, txn_id, **kwargs):
diff --git a/synapse/replication/http/federation.py b/synapse/replication/http/federation.py
index 2f16955954..9af4e7e173 100644
--- a/synapse/replication/http/federation.py
+++ b/synapse/replication/http/federation.py
@@ -82,8 +82,7 @@ class ReplicationFederationSendEventsRestServlet(ReplicationEndpoint):
 
         return payload
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request):
+    async def _handle_request(self, request):
         with Measure(self.clock, "repl_fed_send_events_parse"):
             content = parse_json_object_from_request(request)
 
@@ -101,15 +100,13 @@ class ReplicationFederationSendEventsRestServlet(ReplicationEndpoint):
                 EventType = event_type_from_format_version(format_ver)
                 event = EventType(event_dict, internal_metadata, rejected_reason)
 
-                context = yield EventContext.deserialize(
-                    self.store, event_payload["context"]
-                )
+                context = EventContext.deserialize(self.store, event_payload["context"])
 
                 event_and_contexts.append((event, context))
 
         logger.info("Got %d events from federation", len(event_and_contexts))
 
-        yield self.federation_handler.persist_events_and_notify(
+        await self.federation_handler.persist_events_and_notify(
             event_and_contexts, backfilled
         )
 
@@ -144,8 +141,7 @@ class ReplicationFederationSendEduRestServlet(ReplicationEndpoint):
     def _serialize_payload(edu_type, origin, content):
         return {"origin": origin, "content": content}
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, edu_type):
+    async def _handle_request(self, request, edu_type):
         with Measure(self.clock, "repl_fed_send_edu_parse"):
             content = parse_json_object_from_request(request)
 
@@ -154,7 +150,7 @@ class ReplicationFederationSendEduRestServlet(ReplicationEndpoint):
 
         logger.info("Got %r edu from %s", edu_type, origin)
 
-        result = yield self.registry.on_edu(edu_type, origin, edu_content)
+        result = await self.registry.on_edu(edu_type, origin, edu_content)
 
         return 200, result
 
@@ -193,8 +189,7 @@ class ReplicationGetQueryRestServlet(ReplicationEndpoint):
         """
         return {"args": args}
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, query_type):
+    async def _handle_request(self, request, query_type):
         with Measure(self.clock, "repl_fed_query_parse"):
             content = parse_json_object_from_request(request)
 
@@ -202,7 +197,7 @@ class ReplicationGetQueryRestServlet(ReplicationEndpoint):
 
         logger.info("Got %r query", query_type)
 
-        result = yield self.registry.on_query(query_type, args)
+        result = await self.registry.on_query(query_type, args)
 
         return 200, result
 
@@ -234,9 +229,8 @@ class ReplicationCleanRoomRestServlet(ReplicationEndpoint):
         """
         return {}
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, room_id):
-        yield self.store.clean_room_for_join(room_id)
+    async def _handle_request(self, request, room_id):
+        await self.store.clean_room_for_join(room_id)
 
         return 200, {}
 
diff --git a/synapse/replication/http/login.py b/synapse/replication/http/login.py
index 786f5232b2..798b9d3af5 100644
--- a/synapse/replication/http/login.py
+++ b/synapse/replication/http/login.py
@@ -15,8 +15,6 @@
 
 import logging
 
-from twisted.internet import defer
-
 from synapse.http.servlet import parse_json_object_from_request
 from synapse.replication.http._base import ReplicationEndpoint
 
@@ -52,15 +50,14 @@ class RegisterDeviceReplicationServlet(ReplicationEndpoint):
             "is_guest": is_guest,
         }
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, user_id):
+    async def _handle_request(self, request, user_id):
         content = parse_json_object_from_request(request)
 
         device_id = content["device_id"]
         initial_display_name = content["initial_display_name"]
         is_guest = content["is_guest"]
 
-        device_id, access_token = yield self.registration_handler.register_device(
+        device_id, access_token = await self.registration_handler.register_device(
             user_id, device_id, initial_display_name, is_guest
         )
 
diff --git a/synapse/replication/http/membership.py b/synapse/replication/http/membership.py
index b9ce3477ad..cc1f249740 100644
--- a/synapse/replication/http/membership.py
+++ b/synapse/replication/http/membership.py
@@ -15,8 +15,6 @@
 
 import logging
 
-from twisted.internet import defer
-
 from synapse.http.servlet import parse_json_object_from_request
 from synapse.replication.http._base import ReplicationEndpoint
 from synapse.types import Requester, UserID
@@ -65,8 +63,7 @@ class ReplicationRemoteJoinRestServlet(ReplicationEndpoint):
             "content": content,
         }
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, room_id, user_id):
+    async def _handle_request(self, request, room_id, user_id):
         content = parse_json_object_from_request(request)
 
         remote_room_hosts = content["remote_room_hosts"]
@@ -79,7 +76,7 @@ class ReplicationRemoteJoinRestServlet(ReplicationEndpoint):
 
         logger.info("remote_join: %s into room: %s", user_id, room_id)
 
-        yield self.federation_handler.do_invite_join(
+        await self.federation_handler.do_invite_join(
             remote_room_hosts, room_id, user_id, event_content
         )
 
@@ -123,8 +120,7 @@ class ReplicationRemoteRejectInviteRestServlet(ReplicationEndpoint):
             "remote_room_hosts": remote_room_hosts,
         }
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, room_id, user_id):
+    async def _handle_request(self, request, room_id, user_id):
         content = parse_json_object_from_request(request)
 
         remote_room_hosts = content["remote_room_hosts"]
@@ -137,7 +133,7 @@ class ReplicationRemoteRejectInviteRestServlet(ReplicationEndpoint):
         logger.info("remote_reject_invite: %s out of room: %s", user_id, room_id)
 
         try:
-            event = yield self.federation_handler.do_remotely_reject_invite(
+            event = await self.federation_handler.do_remotely_reject_invite(
                 remote_room_hosts, room_id, user_id
             )
             ret = event.get_pdu_json()
@@ -148,9 +144,9 @@ class ReplicationRemoteRejectInviteRestServlet(ReplicationEndpoint):
             # The 'except' clause is very broad, but we need to
             # capture everything from DNS failures upwards
             #
-            logger.warn("Failed to reject invite: %s", e)
+            logger.warning("Failed to reject invite: %s", e)
 
-            yield self.store.locally_reject_invite(user_id, room_id)
+            await self.store.locally_reject_invite(user_id, room_id)
             ret = {}
 
         return 200, ret
diff --git a/synapse/replication/http/register.py b/synapse/replication/http/register.py
index 87fe2dd9b0..915cfb9430 100644
--- a/synapse/replication/http/register.py
+++ b/synapse/replication/http/register.py
@@ -15,8 +15,6 @@
 
 import logging
 
-from twisted.internet import defer
-
 from synapse.http.servlet import parse_json_object_from_request
 from synapse.replication.http._base import ReplicationEndpoint
 
@@ -74,11 +72,10 @@ class ReplicationRegisterServlet(ReplicationEndpoint):
             "address": address,
         }
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, user_id):
+    async def _handle_request(self, request, user_id):
         content = parse_json_object_from_request(request)
 
-        yield self.registration_handler.register_with_store(
+        await self.registration_handler.register_with_store(
             user_id=user_id,
             password_hash=content["password_hash"],
             was_guest=content["was_guest"],
@@ -106,7 +103,7 @@ class ReplicationPostRegisterActionsServlet(ReplicationEndpoint):
         self.registration_handler = hs.get_registration_handler()
 
     @staticmethod
-    def _serialize_payload(user_id, auth_result, access_token, bind_email, bind_msisdn):
+    def _serialize_payload(user_id, auth_result, access_token):
         """
         Args:
             user_id (str): The user ID that consented
@@ -114,33 +111,17 @@ class ReplicationPostRegisterActionsServlet(ReplicationEndpoint):
                 registered user.
             access_token (str|None): The access token of the newly logged in
                 device, or None if `inhibit_login` enabled.
-            bind_email (bool): Whether to bind the email with the identity
-                server
-            bind_msisdn (bool): Whether to bind the msisdn with the identity
-                server
         """
-        return {
-            "auth_result": auth_result,
-            "access_token": access_token,
-            "bind_email": bind_email,
-            "bind_msisdn": bind_msisdn,
-        }
+        return {"auth_result": auth_result, "access_token": access_token}
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, user_id):
+    async def _handle_request(self, request, user_id):
         content = parse_json_object_from_request(request)
 
         auth_result = content["auth_result"]
         access_token = content["access_token"]
-        bind_email = content["bind_email"]
-        bind_msisdn = content["bind_msisdn"]
 
-        yield self.registration_handler.post_registration_actions(
-            user_id=user_id,
-            auth_result=auth_result,
-            access_token=access_token,
-            bind_email=bind_email,
-            bind_msisdn=bind_msisdn,
+        await self.registration_handler.post_registration_actions(
+            user_id=user_id, auth_result=auth_result, access_token=access_token
         )
 
         return 200, {}
diff --git a/synapse/replication/http/send_event.py b/synapse/replication/http/send_event.py
index adb9b2f7f4..9bafd60b14 100644
--- a/synapse/replication/http/send_event.py
+++ b/synapse/replication/http/send_event.py
@@ -87,8 +87,7 @@ class ReplicationSendEventRestServlet(ReplicationEndpoint):
 
         return payload
 
-    @defer.inlineCallbacks
-    def _handle_request(self, request, event_id):
+    async def _handle_request(self, request, event_id):
         with Measure(self.clock, "repl_send_event_parse"):
             content = parse_json_object_from_request(request)
 
@@ -101,7 +100,7 @@ class ReplicationSendEventRestServlet(ReplicationEndpoint):
             event = EventType(event_dict, internal_metadata, rejected_reason)
 
             requester = Requester.deserialize(self.store, content["requester"])
-            context = yield EventContext.deserialize(self.store, content["context"])
+            context = EventContext.deserialize(self.store, content["context"])
 
             ratelimit = content["ratelimit"]
             extra_users = [UserID.from_string(u) for u in content["extra_users"]]
@@ -113,7 +112,7 @@ class ReplicationSendEventRestServlet(ReplicationEndpoint):
             "Got event to send with ID: %s into room: %s", event.event_id, event.room_id
         )
 
-        yield self.event_creation_handler.persist_and_notify_client_event(
+        await self.event_creation_handler.persist_and_notify_client_event(
             requester, event, context, ratelimit=ratelimit, extra_users=extra_users
         )
 
diff --git a/synapse/replication/slave/storage/_base.py b/synapse/replication/slave/storage/_base.py
index 182cb2a1d8..456bc005a0 100644
--- a/synapse/replication/slave/storage/_base.py
+++ b/synapse/replication/slave/storage/_base.py
@@ -14,6 +14,7 @@
 # limitations under the License.
 
 import logging
+from typing import Dict
 
 import six
 
@@ -44,7 +45,14 @@ class BaseSlavedStore(SQLBaseStore):
 
         self.hs = hs
 
-    def stream_positions(self):
+    def stream_positions(self) -> Dict[str, int]:
+        """
+        Get the current positions of all the streams this store wants to subscribe to
+
+        Returns:
+            map from stream name to the most recent update we have for
+            that stream (ie, the point we want to start replicating from)
+        """
         pos = {}
         if self._cache_id_gen:
             pos["caches"] = self._cache_id_gen.get_current_token()
diff --git a/synapse/replication/slave/storage/account_data.py b/synapse/replication/slave/storage/account_data.py
index 3c44d1d48d..bc2f6a12ae 100644
--- a/synapse/replication/slave/storage/account_data.py
+++ b/synapse/replication/slave/storage/account_data.py
@@ -16,8 +16,8 @@
 
 from synapse.replication.slave.storage._base import BaseSlavedStore
 from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
-from synapse.storage.account_data import AccountDataWorkerStore
-from synapse.storage.tags import TagsWorkerStore
+from synapse.storage.data_stores.main.account_data import AccountDataWorkerStore
+from synapse.storage.data_stores.main.tags import TagsWorkerStore
 
 
 class SlavedAccountDataStore(TagsWorkerStore, AccountDataWorkerStore, BaseSlavedStore):
diff --git a/synapse/replication/slave/storage/appservice.py b/synapse/replication/slave/storage/appservice.py
index cda12ea70d..a67fbeffb7 100644
--- a/synapse/replication/slave/storage/appservice.py
+++ b/synapse/replication/slave/storage/appservice.py
@@ -14,7 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.appservice import (
+from synapse.storage.data_stores.main.appservice import (
     ApplicationServiceTransactionWorkerStore,
     ApplicationServiceWorkerStore,
 )
diff --git a/synapse/replication/slave/storage/client_ips.py b/synapse/replication/slave/storage/client_ips.py
index 14ced32333..b4f58cea19 100644
--- a/synapse/replication/slave/storage/client_ips.py
+++ b/synapse/replication/slave/storage/client_ips.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.client_ips import LAST_SEEN_GRANULARITY
+from synapse.storage.data_stores.main.client_ips import LAST_SEEN_GRANULARITY
 from synapse.util.caches import CACHE_SIZE_FACTOR
 from synapse.util.caches.descriptors import Cache
 
diff --git a/synapse/replication/slave/storage/deviceinbox.py b/synapse/replication/slave/storage/deviceinbox.py
index 284fd30d89..9fb6c5c6ff 100644
--- a/synapse/replication/slave/storage/deviceinbox.py
+++ b/synapse/replication/slave/storage/deviceinbox.py
@@ -15,7 +15,7 @@
 
 from synapse.replication.slave.storage._base import BaseSlavedStore
 from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
-from synapse.storage.deviceinbox import DeviceInboxWorkerStore
+from synapse.storage.data_stores.main.deviceinbox import DeviceInboxWorkerStore
 from synapse.util.caches.expiringcache import ExpiringCache
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 
diff --git a/synapse/replication/slave/storage/devices.py b/synapse/replication/slave/storage/devices.py
index d9300fce33..de50748c30 100644
--- a/synapse/replication/slave/storage/devices.py
+++ b/synapse/replication/slave/storage/devices.py
@@ -15,8 +15,9 @@
 
 from synapse.replication.slave.storage._base import BaseSlavedStore
 from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker
-from synapse.storage.devices import DeviceWorkerStore
-from synapse.storage.end_to_end_keys import EndToEndKeyWorkerStore
+from synapse.replication.tcp.streams._base import DeviceListsStream, UserSignatureStream
+from synapse.storage.data_stores.main.devices import DeviceWorkerStore
+from synapse.storage.data_stores.main.end_to_end_keys import EndToEndKeyWorkerStore
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 
 
@@ -33,20 +34,31 @@ class SlavedDeviceStore(EndToEndKeyWorkerStore, DeviceWorkerStore, BaseSlavedSto
         self._device_list_stream_cache = StreamChangeCache(
             "DeviceListStreamChangeCache", device_list_max
         )
+        self._user_signature_stream_cache = StreamChangeCache(
+            "UserSignatureStreamChangeCache", device_list_max
+        )
         self._device_list_federation_stream_cache = StreamChangeCache(
             "DeviceListFederationStreamChangeCache", device_list_max
         )
 
     def stream_positions(self):
         result = super(SlavedDeviceStore, self).stream_positions()
-        result["device_lists"] = self._device_list_id_gen.get_current_token()
+        # The user signature stream uses the same stream ID generator as the
+        # device list stream, so set them both to the device list ID
+        # generator's current token.
+        current_token = self._device_list_id_gen.get_current_token()
+        result[DeviceListsStream.NAME] = current_token
+        result[UserSignatureStream.NAME] = current_token
         return result
 
     def process_replication_rows(self, stream_name, token, rows):
-        if stream_name == "device_lists":
+        if stream_name == DeviceListsStream.NAME:
             self._device_list_id_gen.advance(token)
             for row in rows:
                 self._invalidate_caches_for_devices(token, row.user_id, row.destination)
+        elif stream_name == UserSignatureStream.NAME:
+            for row in rows:
+                self._user_signature_stream_cache.entity_has_changed(row.user_id, token)
         return super(SlavedDeviceStore, self).process_replication_rows(
             stream_name, token, rows
         )
diff --git a/synapse/replication/slave/storage/directory.py b/synapse/replication/slave/storage/directory.py
index 1d1d48709a..8b9717c46f 100644
--- a/synapse/replication/slave/storage/directory.py
+++ b/synapse/replication/slave/storage/directory.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.directory import DirectoryWorkerStore
+from synapse.storage.data_stores.main.directory import DirectoryWorkerStore
 
 from ._base import BaseSlavedStore
 
diff --git a/synapse/replication/slave/storage/events.py b/synapse/replication/slave/storage/events.py
index ab5937e638..d0a0eaf75b 100644
--- a/synapse/replication/slave/storage/events.py
+++ b/synapse/replication/slave/storage/events.py
@@ -20,15 +20,17 @@ from synapse.replication.tcp.streams.events import (
     EventsStreamCurrentStateRow,
     EventsStreamEventRow,
 )
-from synapse.storage.event_federation import EventFederationWorkerStore
-from synapse.storage.event_push_actions import EventPushActionsWorkerStore
-from synapse.storage.events_worker import EventsWorkerStore
-from synapse.storage.relations import RelationsWorkerStore
-from synapse.storage.roommember import RoomMemberWorkerStore
-from synapse.storage.signatures import SignatureWorkerStore
-from synapse.storage.state import StateGroupWorkerStore
-from synapse.storage.stream import StreamWorkerStore
-from synapse.storage.user_erasure_store import UserErasureWorkerStore
+from synapse.storage.data_stores.main.event_federation import EventFederationWorkerStore
+from synapse.storage.data_stores.main.event_push_actions import (
+    EventPushActionsWorkerStore,
+)
+from synapse.storage.data_stores.main.events_worker import EventsWorkerStore
+from synapse.storage.data_stores.main.relations import RelationsWorkerStore
+from synapse.storage.data_stores.main.roommember import RoomMemberWorkerStore
+from synapse.storage.data_stores.main.signatures import SignatureWorkerStore
+from synapse.storage.data_stores.main.state import StateGroupWorkerStore
+from synapse.storage.data_stores.main.stream import StreamWorkerStore
+from synapse.storage.data_stores.main.user_erasure_store import UserErasureWorkerStore
 
 from ._base import BaseSlavedStore
 from ._slaved_id_tracker import SlavedIdTracker
diff --git a/synapse/replication/slave/storage/filtering.py b/synapse/replication/slave/storage/filtering.py
index 456a14cd5c..5c84ebd125 100644
--- a/synapse/replication/slave/storage/filtering.py
+++ b/synapse/replication/slave/storage/filtering.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.filtering import FilteringStore
+from synapse.storage.data_stores.main.filtering import FilteringStore
 
 from ._base import BaseSlavedStore
 
diff --git a/synapse/replication/slave/storage/keys.py b/synapse/replication/slave/storage/keys.py
index cc6f7f009f..3def367ae9 100644
--- a/synapse/replication/slave/storage/keys.py
+++ b/synapse/replication/slave/storage/keys.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage import KeyStore
+from synapse.storage.data_stores.main.keys import KeyStore
 
 # KeyStore isn't really safe to use from a worker, but for now we do so and hope that
 # the races it creates aren't too bad.
diff --git a/synapse/replication/slave/storage/presence.py b/synapse/replication/slave/storage/presence.py
index 82d808af4c..747ced0c84 100644
--- a/synapse/replication/slave/storage/presence.py
+++ b/synapse/replication/slave/storage/presence.py
@@ -14,7 +14,7 @@
 # limitations under the License.
 
 from synapse.storage import DataStore
-from synapse.storage.presence import PresenceStore
+from synapse.storage.data_stores.main.presence import PresenceStore
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 
 from ._base import BaseSlavedStore, __func__
diff --git a/synapse/replication/slave/storage/profile.py b/synapse/replication/slave/storage/profile.py
index 46c28d4171..28c508aad3 100644
--- a/synapse/replication/slave/storage/profile.py
+++ b/synapse/replication/slave/storage/profile.py
@@ -14,7 +14,7 @@
 # limitations under the License.
 
 from synapse.replication.slave.storage._base import BaseSlavedStore
-from synapse.storage.profile import ProfileWorkerStore
+from synapse.storage.data_stores.main.profile import ProfileWorkerStore
 
 
 class SlavedProfileStore(ProfileWorkerStore, BaseSlavedStore):
diff --git a/synapse/replication/slave/storage/push_rule.py b/synapse/replication/slave/storage/push_rule.py
index af7012702e..3655f05e54 100644
--- a/synapse/replication/slave/storage/push_rule.py
+++ b/synapse/replication/slave/storage/push_rule.py
@@ -14,7 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.push_rule import PushRulesWorkerStore
+from synapse.storage.data_stores.main.push_rule import PushRulesWorkerStore
 
 from ._slaved_id_tracker import SlavedIdTracker
 from .events import SlavedEventStore
diff --git a/synapse/replication/slave/storage/pushers.py b/synapse/replication/slave/storage/pushers.py
index 8eeb267d61..b4331d0799 100644
--- a/synapse/replication/slave/storage/pushers.py
+++ b/synapse/replication/slave/storage/pushers.py
@@ -14,7 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.pusher import PusherWorkerStore
+from synapse.storage.data_stores.main.pusher import PusherWorkerStore
 
 from ._base import BaseSlavedStore
 from ._slaved_id_tracker import SlavedIdTracker
diff --git a/synapse/replication/slave/storage/receipts.py b/synapse/replication/slave/storage/receipts.py
index 91afa5a72b..43d823c601 100644
--- a/synapse/replication/slave/storage/receipts.py
+++ b/synapse/replication/slave/storage/receipts.py
@@ -14,7 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.receipts import ReceiptsWorkerStore
+from synapse.storage.data_stores.main.receipts import ReceiptsWorkerStore
 
 from ._base import BaseSlavedStore
 from ._slaved_id_tracker import SlavedIdTracker
diff --git a/synapse/replication/slave/storage/registration.py b/synapse/replication/slave/storage/registration.py
index 408d91df1c..4b8553e250 100644
--- a/synapse/replication/slave/storage/registration.py
+++ b/synapse/replication/slave/storage/registration.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.registration import RegistrationWorkerStore
+from synapse.storage.data_stores.main.registration import RegistrationWorkerStore
 
 from ._base import BaseSlavedStore
 
diff --git a/synapse/replication/slave/storage/room.py b/synapse/replication/slave/storage/room.py
index f68b3378e3..d9ad386b28 100644
--- a/synapse/replication/slave/storage/room.py
+++ b/synapse/replication/slave/storage/room.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.room import RoomWorkerStore
+from synapse.storage.data_stores.main.room import RoomWorkerStore
 
 from ._base import BaseSlavedStore
 from ._slaved_id_tracker import SlavedIdTracker
diff --git a/synapse/replication/slave/storage/transactions.py b/synapse/replication/slave/storage/transactions.py
index 3527beb3c9..ac88e6b8c3 100644
--- a/synapse/replication/slave/storage/transactions.py
+++ b/synapse/replication/slave/storage/transactions.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from synapse.storage.transactions import TransactionStore
+from synapse.storage.data_stores.main.transactions import TransactionStore
 
 from ._base import BaseSlavedStore
 
diff --git a/synapse/replication/tcp/client.py b/synapse/replication/tcp/client.py
index a44ceb00e7..fead78388c 100644
--- a/synapse/replication/tcp/client.py
+++ b/synapse/replication/tcp/client.py
@@ -16,10 +16,17 @@
 """
 
 import logging
+from typing import Dict
 
 from twisted.internet import defer
 from twisted.internet.protocol import ReconnectingClientFactory
 
+from synapse.replication.slave.storage._base import BaseSlavedStore
+from synapse.replication.tcp.protocol import (
+    AbstractReplicationClientHandler,
+    ClientReplicationStreamProtocol,
+)
+
 from .commands import (
     FederationAckCommand,
     InvalidateCacheCommand,
@@ -27,7 +34,6 @@ from .commands import (
     UserIpCommand,
     UserSyncCommand,
 )
-from .protocol import ClientReplicationStreamProtocol
 
 logger = logging.getLogger(__name__)
 
@@ -42,7 +48,7 @@ class ReplicationClientFactory(ReconnectingClientFactory):
 
     maxDelay = 30  # Try at least once every N seconds
 
-    def __init__(self, hs, client_name, handler):
+    def __init__(self, hs, client_name, handler: AbstractReplicationClientHandler):
         self.client_name = client_name
         self.handler = handler
         self.server_name = hs.config.server_name
@@ -68,13 +74,13 @@ class ReplicationClientFactory(ReconnectingClientFactory):
         ReconnectingClientFactory.clientConnectionFailed(self, connector, reason)
 
 
-class ReplicationClientHandler(object):
+class ReplicationClientHandler(AbstractReplicationClientHandler):
     """A base handler that can be passed to the ReplicationClientFactory.
 
     By default proxies incoming replication data to the SlaveStore.
     """
 
-    def __init__(self, store):
+    def __init__(self, store: BaseSlavedStore):
         self.store = store
 
         # The current connection. None if we are currently (re)connecting
@@ -138,11 +144,13 @@ class ReplicationClientHandler(object):
         if d:
             d.callback(data)
 
-    def get_streams_to_replicate(self):
+    def get_streams_to_replicate(self) -> Dict[str, int]:
         """Called when a new connection has been established and we need to
         subscribe to streams.
 
-        Returns a dictionary of stream name to token.
+        Returns:
+            map from stream name to the most recent update we have for
+            that stream (ie, the point we want to start replicating from)
         """
         args = self.store.stream_positions()
         user_account_data = args.pop("user_account_data", None)
@@ -168,7 +176,7 @@ class ReplicationClientHandler(object):
         if self.connection:
             self.connection.send_command(cmd)
         else:
-            logger.warn("Queuing command as not connected: %r", cmd.NAME)
+            logger.warning("Queuing command as not connected: %r", cmd.NAME)
             self.pending_commands.append(cmd)
 
     def send_federation_ack(self, token):
diff --git a/synapse/replication/tcp/protocol.py b/synapse/replication/tcp/protocol.py
index 5ffdf2675d..afaf002fe6 100644
--- a/synapse/replication/tcp/protocol.py
+++ b/synapse/replication/tcp/protocol.py
@@ -48,7 +48,7 @@ indicate which side is sending, these are *not* included on the wire::
     > ERROR server stopping
     * connection closed by server *
 """
-
+import abc
 import fcntl
 import logging
 import struct
@@ -65,6 +65,7 @@ from twisted.python.failure import Failure
 from synapse.logging.context import make_deferred_yieldable, run_in_background
 from synapse.metrics import LaterGauge
 from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.util import Clock
 from synapse.util.stringutils import random_string
 
 from .commands import (
@@ -249,7 +250,7 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver):
         return handler(cmd)
 
     def close(self):
-        logger.warn("[%s] Closing connection", self.id())
+        logger.warning("[%s] Closing connection", self.id())
         self.time_we_closed = self.clock.time_msec()
         self.transport.loseConnection()
         self.on_connection_closed()
@@ -558,11 +559,80 @@ class ServerReplicationStreamProtocol(BaseReplicationStreamProtocol):
         self.streamer.lost_connection(self)
 
 
+class AbstractReplicationClientHandler(metaclass=abc.ABCMeta):
+    """
+    The interface for the handler that should be passed to
+    ClientReplicationStreamProtocol
+    """
+
+    @abc.abstractmethod
+    def on_rdata(self, stream_name, token, rows):
+        """Called to handle a batch of replication data with a given stream token.
+
+        Args:
+            stream_name (str): name of the replication stream for this batch of rows
+            token (int): stream token for this batch of rows
+            rows (list): a list of Stream.ROW_TYPE objects as returned by
+                Stream.parse_row.
+
+        Returns:
+            Deferred|None
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def on_position(self, stream_name, token):
+        """Called when we get new position data."""
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def on_sync(self, data):
+        """Called when get a new SYNC command."""
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_streams_to_replicate(self):
+        """Called when a new connection has been established and we need to
+        subscribe to streams.
+
+        Returns:
+            map from stream name to the most recent update we have for
+            that stream (ie, the point we want to start replicating from)
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_currently_syncing_users(self):
+        """Get the list of currently syncing users (if any). This is called
+        when a connection has been established and we need to send the
+        currently syncing users."""
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def update_connection(self, connection):
+        """Called when a connection has been established (or lost with None).
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def finished_connecting(self):
+        """Called when we have successfully subscribed and caught up to all
+        streams we're interested in.
+        """
+        raise NotImplementedError()
+
+
 class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol):
     VALID_INBOUND_COMMANDS = VALID_SERVER_COMMANDS
     VALID_OUTBOUND_COMMANDS = VALID_CLIENT_COMMANDS
 
-    def __init__(self, client_name, server_name, clock, handler):
+    def __init__(
+        self,
+        client_name: str,
+        server_name: str,
+        clock: Clock,
+        handler: AbstractReplicationClientHandler,
+    ):
         BaseReplicationStreamProtocol.__init__(self, clock)
 
         self.client_name = client_name
diff --git a/synapse/replication/tcp/streams/__init__.py b/synapse/replication/tcp/streams/__init__.py
index 634f636dc9..5f52264e84 100644
--- a/synapse/replication/tcp/streams/__init__.py
+++ b/synapse/replication/tcp/streams/__init__.py
@@ -45,5 +45,6 @@ STREAMS_MAP = {
         _base.TagAccountDataStream,
         _base.AccountDataStream,
         _base.GroupServerStream,
+        _base.UserSignatureStream,
     )
 }
diff --git a/synapse/replication/tcp/streams/_base.py b/synapse/replication/tcp/streams/_base.py
index f03111c259..9e45429d49 100644
--- a/synapse/replication/tcp/streams/_base.py
+++ b/synapse/replication/tcp/streams/_base.py
@@ -95,6 +95,7 @@ GroupsStreamRow = namedtuple(
     "GroupsStreamRow",
     ("group_id", "user_id", "type", "content"),  # str  # str  # str  # dict
 )
+UserSignatureStreamRow = namedtuple("UserSignatureStreamRow", ("user_id"))  # str
 
 
 class Stream(object):
@@ -438,3 +439,20 @@ class GroupServerStream(Stream):
         self.update_function = store.get_all_groups_changes
 
         super(GroupServerStream, self).__init__(hs)
+
+
+class UserSignatureStream(Stream):
+    """A user has signed their own device with their user-signing key
+    """
+
+    NAME = "user_signature"
+    _LIMITED = False
+    ROW_TYPE = UserSignatureStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_device_stream_token
+        self.update_function = store.get_all_user_signature_changes_for_remotes
+
+        super(UserSignatureStream, self).__init__(hs)
diff --git a/synapse/res/templates/add_threepid.html b/synapse/res/templates/add_threepid.html
new file mode 100644
index 0000000000..cc4ab07e09
--- /dev/null
+++ b/synapse/res/templates/add_threepid.html
@@ -0,0 +1,9 @@
+
+
+    
A request to add an email address to your Matrix account has been received. If this was you, please click the link below to confirm adding this email:

+
+    {{ link }}
+
+    
If this was not you, you can safely ignore this email. Thank you.

+
+
diff --git a/synapse/res/templates/add_threepid.txt b/synapse/res/templates/add_threepid.txt
new file mode 100644
index 0000000000..a60c1ff659
--- /dev/null
+++ b/synapse/res/templates/add_threepid.txt
@@ -0,0 +1,6 @@
+A request to add an email address to your Matrix account has been received. If this was you,
+please click the link below to confirm adding this email:
+
+{{ link }}
+
+If this was not you, you can safely ignore this email. Thank you.
diff --git a/synapse/res/templates/add_threepid_failure.html b/synapse/res/templates/add_threepid_failure.html
new file mode 100644
index 0000000000..441d11c846
--- /dev/null
+++ b/synapse/res/templates/add_threepid_failure.html
@@ -0,0 +1,8 @@
+
+
+
+
The request failed for the following reason: {{ failure_reason }}.

+
+
No changes have been made to your account.

+
+
diff --git a/synapse/res/templates/add_threepid_success.html b/synapse/res/templates/add_threepid_success.html
new file mode 100644
index 0000000000..fbd6e4018f
--- /dev/null
+++ b/synapse/res/templates/add_threepid_success.html
@@ -0,0 +1,6 @@
+
+
+
+
Your email has now been validated, please return to your client. You may now close this window.

+
+
diff --git a/synapse/res/templates/password_reset.html b/synapse/res/templates/password_reset.html
index 4fa7b36734..a197bf872c 100644
--- a/synapse/res/templates/password_reset.html
+++ b/synapse/res/templates/password_reset.html
@@ -4,6 +4,6 @@
 
     {{ link }}
 
-    
If this was not you, please disregard this email and contact your server administrator. Thank you.

+    
If this was not you, do not click the link above and instead contact your server administrator. Thank you.

 
 
diff --git a/synapse/res/templates/password_reset.txt b/synapse/res/templates/password_reset.txt
index f0deff59a7..6aa6527560 100644
--- a/synapse/res/templates/password_reset.txt
+++ b/synapse/res/templates/password_reset.txt
@@ -3,5 +3,5 @@ was you, please click the link below to confirm resetting your password:
 
 {{ link }}
 
-If this was not you, please disregard this email and contact your server
-administrator. Thank you.
+If this was not you, DO NOT click the link above and instead contact your
+server administrator. Thank you.
diff --git a/synapse/res/templates/password_reset_failure.html b/synapse/res/templates/password_reset_failure.html
index 0b132cf8db..9e3c4446e3 100644
--- a/synapse/res/templates/password_reset_failure.html
+++ b/synapse/res/templates/password_reset_failure.html
@@ -1,6 +1,8 @@
 
 
 
-
{{ failure_reason }}. Your password has not been reset.

+
The request failed for the following reason: {{ failure_reason }}.

+
+
Your password has not been reset.

 
 
diff --git a/synapse/res/templates/registration.html b/synapse/res/templates/registration.html
new file mode 100644
index 0000000000..16730a527f
--- /dev/null
+++ b/synapse/res/templates/registration.html
@@ -0,0 +1,11 @@
+
+
+    
You have asked us to register this email with a new Matrix account. If this was you, please click the link below to confirm your email address:

+
+    Verify Your Email Address
+
+    
If this was not you, you can safely disregard this email.

+
+    
Thank you.

+
+
diff --git a/synapse/res/templates/registration.txt b/synapse/res/templates/registration.txt
new file mode 100644
index 0000000000..cb4f16a90c
--- /dev/null
+++ b/synapse/res/templates/registration.txt
@@ -0,0 +1,10 @@
+Hello there,
+
+You have asked us to register this email with a new Matrix account. If this
+was you, please click the link below to confirm your email address:
+
+{{ link }}
+
+If this was not you, you can safely disregard this email.
+
+Thank you.
diff --git a/synapse/res/templates/registration_failure.html b/synapse/res/templates/registration_failure.html
new file mode 100644
index 0000000000..2833d79c37
--- /dev/null
+++ b/synapse/res/templates/registration_failure.html
@@ -0,0 +1,6 @@
+
+
+
+
Validation failed for the following reason: {{ failure_reason }}.

+
+
diff --git a/synapse/res/templates/registration_success.html b/synapse/res/templates/registration_success.html
new file mode 100644
index 0000000000..fbd6e4018f
--- /dev/null
+++ b/synapse/res/templates/registration_success.html
@@ -0,0 +1,6 @@
+
+
+
+
Your email has now been validated, please return to your client. You may now close this window.

+
+
diff --git a/synapse/rest/__init__.py b/synapse/rest/__init__.py
index 1d20b96d03..4a1fc2ec2b 100644
--- a/synapse/rest/__init__.py
+++ b/synapse/rest/__init__.py
@@ -73,7 +73,7 @@ class ClientRestResource(JsonResource):
 
     @staticmethod
     def register_servlets(client_resource, hs):
-        versions.register_servlets(client_resource)
+        versions.register_servlets(hs, client_resource)
 
         # Deprecated in r0
         initial_sync.register_servlets(hs, client_resource)
diff --git a/synapse/rest/admin/__init__.py b/synapse/rest/admin/__init__.py
index b4761adaed..5c2a2eb593 100644
--- a/synapse/rest/admin/__init__.py
+++ b/synapse/rest/admin/__init__.py
@@ -23,8 +23,6 @@ import re
 from six import text_type
 from six.moves import http_client
 
-from twisted.internet import defer
-
 import synapse
 from synapse.api.constants import Membership, UserTypes
 from synapse.api.errors import Codes, NotFoundError, SynapseError
@@ -41,11 +39,12 @@ from synapse.rest.admin._base import (
     assert_user_is_admin,
     historical_admin_path_patterns,
 )
-from synapse.rest.admin.media import register_servlets_for_media_repo
+from synapse.rest.admin.media import ListMediaInRoom, register_servlets_for_media_repo
 from synapse.rest.admin.purge_room_servlet import PurgeRoomServlet
 from synapse.rest.admin.server_notice_servlet import SendServerNoticeServlet
 from synapse.rest.admin.users import UserAdminServlet
 from synapse.types import UserID, create_requester
+from synapse.util.async_helpers import maybe_awaitable
 from synapse.util.versionstring import get_version_string
 
 logger = logging.getLogger(__name__)
@@ -59,15 +58,14 @@ class UsersRestServlet(RestServlet):
         self.auth = hs.get_auth()
         self.handlers = hs.get_handlers()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, user_id):
+    async def on_GET(self, request, user_id):
         target_user = UserID.from_string(user_id)
-        yield assert_requester_is_admin(self.auth, request)
+        await assert_requester_is_admin(self.auth, request)
 
         if not self.hs.is_mine(target_user):
             raise SynapseError(400, "Can only users a local user")
 
-        ret = yield self.handlers.admin_handler.get_users()
+        ret = await self.handlers.admin_handler.get_users()
 
         return 200, ret
 
@@ -122,8 +120,7 @@ class UserRegisterServlet(RestServlet):
         self.nonces[nonce] = int(self.reactor.seconds())
         return 200, {"nonce": nonce}
 
-    @defer.inlineCallbacks
-    def on_POST(self, request):
+    async def on_POST(self, request):
         self._clear_old_nonces()
 
         if not self.hs.config.registration_shared_secret:
@@ -204,14 +201,14 @@ class UserRegisterServlet(RestServlet):
 
         register = RegisterRestServlet(self.hs)
 
-        user_id = yield register.registration_handler.register_user(
+        user_id = await register.registration_handler.register_user(
             localpart=body["username"].lower(),
             password=body["password"],
             admin=bool(admin),
             user_type=user_type,
         )
 
-        result = yield register._create_registration_details(user_id, body)
+        result = await register._create_registration_details(user_id, body)
         return 200, result
 
 
@@ -223,19 +220,18 @@ class WhoisRestServlet(RestServlet):
         self.auth = hs.get_auth()
         self.handlers = hs.get_handlers()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, user_id):
+    async def on_GET(self, request, user_id):
         target_user = UserID.from_string(user_id)
-        requester = yield self.auth.get_user_by_req(request)
+        requester = await self.auth.get_user_by_req(request)
         auth_user = requester.user
 
         if target_user != auth_user:
-            yield assert_user_is_admin(self.auth, auth_user)
+            await assert_user_is_admin(self.auth, auth_user)
 
         if not self.hs.is_mine(target_user):
             raise SynapseError(400, "Can only whois a local user")
 
-        ret = yield self.handlers.admin_handler.get_whois(target_user)
+        ret = await self.handlers.admin_handler.get_whois(target_user)
 
         return 200, ret
 
@@ -255,9 +251,8 @@ class PurgeHistoryRestServlet(RestServlet):
         self.store = hs.get_datastore()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id, event_id):
-        yield assert_requester_is_admin(self.auth, request)
+    async def on_POST(self, request, room_id, event_id):
+        await assert_requester_is_admin(self.auth, request)
 
         body = parse_json_object_from_request(request, allow_empty_body=True)
 
@@ -270,12 +265,12 @@ class PurgeHistoryRestServlet(RestServlet):
             event_id = body.get("purge_up_to_event_id")
 
         if event_id is not None:
-            event = yield self.store.get_event(event_id)
+            event = await self.store.get_event(event_id)
 
             if event.room_id != room_id:
                 raise SynapseError(400, "Event is for wrong room.")
 
-            token = yield self.store.get_topological_token_for_event(event_id)
+            token = await self.store.get_topological_token_for_event(event_id)
 
             logger.info("[purge] purging up to token %s (event_id %s)", token, event_id)
         elif "purge_up_to_ts" in body:
@@ -285,15 +280,13 @@ class PurgeHistoryRestServlet(RestServlet):
                     400, "purge_up_to_ts must be an int", errcode=Codes.BAD_JSON
                 )
 
-            stream_ordering = (yield self.store.find_first_stream_ordering_after_ts(ts))
+            stream_ordering = await self.store.find_first_stream_ordering_after_ts(ts)
 
-            r = (
-                yield self.store.get_room_event_after_stream_ordering(
-                    room_id, stream_ordering
-                )
+            r = await self.store.get_room_event_after_stream_ordering(
+                room_id, stream_ordering
             )
             if not r:
-                logger.warn(
+                logger.warning(
                     "[purge] purging events not possible: No event found "
                     "(received_ts %i => stream_ordering %i)",
                     ts,
@@ -318,7 +311,7 @@ class PurgeHistoryRestServlet(RestServlet):
                 errcode=Codes.BAD_JSON,
             )
 
-        purge_id = yield self.pagination_handler.start_purge_history(
+        purge_id = self.pagination_handler.start_purge_history(
             room_id, token, delete_local_events=delete_local_events
         )
 
@@ -339,9 +332,8 @@ class PurgeHistoryStatusRestServlet(RestServlet):
         self.pagination_handler = hs.get_pagination_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, purge_id):
-        yield assert_requester_is_admin(self.auth, request)
+    async def on_GET(self, request, purge_id):
+        await assert_requester_is_admin(self.auth, request)
 
         purge_status = self.pagination_handler.get_purge_status(purge_id)
         if purge_status is None:
@@ -357,9 +349,8 @@ class DeactivateAccountRestServlet(RestServlet):
         self._deactivate_account_handler = hs.get_deactivate_account_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, target_user_id):
-        yield assert_requester_is_admin(self.auth, request)
+    async def on_POST(self, request, target_user_id):
+        await assert_requester_is_admin(self.auth, request)
         body = parse_json_object_from_request(request, allow_empty_body=True)
         erase = body.get("erase", False)
         if not isinstance(erase, bool):
@@ -371,7 +362,7 @@ class DeactivateAccountRestServlet(RestServlet):
 
         UserID.from_string(target_user_id)
 
-        result = yield self._deactivate_account_handler.deactivate_account(
+        result = await self._deactivate_account_handler.deactivate_account(
             target_user_id, erase
         )
         if result:
@@ -405,10 +396,9 @@ class ShutdownRoomRestServlet(RestServlet):
         self.room_member_handler = hs.get_room_member_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request)
-        yield assert_user_is_admin(self.auth, requester.user)
+    async def on_POST(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request)
+        await assert_user_is_admin(self.auth, requester.user)
 
         content = parse_json_object_from_request(request)
         assert_params_in_dict(content, ["new_room_user_id"])
@@ -419,7 +409,7 @@ class ShutdownRoomRestServlet(RestServlet):
         message = content.get("message", self.DEFAULT_MESSAGE)
         room_name = content.get("room_name", "Content Violation Notification")
 
-        info = yield self._room_creation_handler.create_room(
+        info = await self._room_creation_handler.create_room(
             room_creator_requester,
             config={
                 "preset": "public_chat",
@@ -438,9 +428,9 @@ class ShutdownRoomRestServlet(RestServlet):
 
         # This will work even if the room is already blocked, but that is
         # desirable in case the first attempt at blocking the room failed below.
-        yield self.store.block_room(room_id, requester_user_id)
+        await self.store.block_room(room_id, requester_user_id)
 
-        users = yield self.state.get_current_users_in_room(room_id)
+        users = await self.state.get_current_users_in_room(room_id)
         kicked_users = []
         failed_to_kick_users = []
         for user_id in users:
@@ -451,7 +441,7 @@ class ShutdownRoomRestServlet(RestServlet):
 
             try:
                 target_requester = create_requester(user_id)
-                yield self.room_member_handler.update_membership(
+                await self.room_member_handler.update_membership(
                     requester=target_requester,
                     target=target_requester.user,
                     room_id=room_id,
@@ -461,9 +451,9 @@ class ShutdownRoomRestServlet(RestServlet):
                     require_consent=False,
                 )
 
-                yield self.room_member_handler.forget(target_requester.user, room_id)
+                await self.room_member_handler.forget(target_requester.user, room_id)
 
-                yield self.room_member_handler.update_membership(
+                await self.room_member_handler.update_membership(
                     requester=target_requester,
                     target=target_requester.user,
                     room_id=new_room_id,
@@ -480,7 +470,7 @@ class ShutdownRoomRestServlet(RestServlet):
                 )
                 failed_to_kick_users.append(user_id)
 
-        yield self.event_creation_handler.create_and_send_nonmember_event(
+        await self.event_creation_handler.create_and_send_nonmember_event(
             room_creator_requester,
             {
                 "type": "m.room.message",
@@ -491,9 +481,11 @@ class ShutdownRoomRestServlet(RestServlet):
             ratelimit=False,
         )
 
-        aliases_for_room = yield self.store.get_aliases_for_room(room_id)
+        aliases_for_room = await maybe_awaitable(
+            self.store.get_aliases_for_room(room_id)
+        )
 
-        yield self.store.update_aliases_for_room(
+        await self.store.update_aliases_for_room(
             room_id, new_room_id, requester_user_id
         )
 
@@ -532,13 +524,12 @@ class ResetPasswordRestServlet(RestServlet):
         self.auth = hs.get_auth()
         self._set_password_handler = hs.get_set_password_handler()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, target_user_id):
+    async def on_POST(self, request, target_user_id):
         """Post request to allow an administrator reset password for a user.
         This needs user to have administrator access in Synapse.
         """
-        requester = yield self.auth.get_user_by_req(request)
-        yield assert_user_is_admin(self.auth, requester.user)
+        requester = await self.auth.get_user_by_req(request)
+        await assert_user_is_admin(self.auth, requester.user)
 
         UserID.from_string(target_user_id)
 
@@ -546,7 +537,7 @@ class ResetPasswordRestServlet(RestServlet):
         assert_params_in_dict(params, ["new_password"])
         new_password = params["new_password"]
 
-        yield self._set_password_handler.set_password(
+        await self._set_password_handler.set_password(
             target_user_id, new_password, requester
         )
         return 200, {}
@@ -572,12 +563,11 @@ class GetUsersPaginatedRestServlet(RestServlet):
         self.auth = hs.get_auth()
         self.handlers = hs.get_handlers()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, target_user_id):
+    async def on_GET(self, request, target_user_id):
         """Get request to get specific number of users from Synapse.
         This needs user to have administrator access in Synapse.
         """
-        yield assert_requester_is_admin(self.auth, request)
+        await assert_requester_is_admin(self.auth, request)
 
         target_user = UserID.from_string(target_user_id)
 
@@ -590,11 +580,10 @@ class GetUsersPaginatedRestServlet(RestServlet):
 
         logger.info("limit: %s, start: %s", limit, start)
 
-        ret = yield self.handlers.admin_handler.get_users_paginate(order, start, limit)
+        ret = await self.handlers.admin_handler.get_users_paginate(order, start, limit)
         return 200, ret
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, target_user_id):
+    async def on_POST(self, request, target_user_id):
         """Post request to get specific number of users from Synapse..
         This needs user to have administrator access in Synapse.
         Example:
@@ -608,7 +597,7 @@ class GetUsersPaginatedRestServlet(RestServlet):
         Returns:
             200 OK with json object {list[dict[str, Any]], count} or empty object.
         """
-        yield assert_requester_is_admin(self.auth, request)
+        await assert_requester_is_admin(self.auth, request)
         UserID.from_string(target_user_id)
 
         order = "name"  # order by name in user table
@@ -618,7 +607,7 @@ class GetUsersPaginatedRestServlet(RestServlet):
         start = params["start"]
         logger.info("limit: %s, start: %s", limit, start)
 
-        ret = yield self.handlers.admin_handler.get_users_paginate(order, start, limit)
+        ret = await self.handlers.admin_handler.get_users_paginate(order, start, limit)
         return 200, ret
 
 
@@ -641,13 +630,12 @@ class SearchUsersRestServlet(RestServlet):
         self.auth = hs.get_auth()
         self.handlers = hs.get_handlers()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, target_user_id):
+    async def on_GET(self, request, target_user_id):
         """Get request to search user table for specific users according to
         search term.
         This needs user to have a administrator access in Synapse.
         """
-        yield assert_requester_is_admin(self.auth, request)
+        await assert_requester_is_admin(self.auth, request)
 
         target_user = UserID.from_string(target_user_id)
 
@@ -661,7 +649,7 @@ class SearchUsersRestServlet(RestServlet):
         term = parse_string(request, "term", required=True)
         logger.info("term: %s ", term)
 
-        ret = yield self.handlers.admin_handler.search_users(term)
+        ret = await self.handlers.admin_handler.search_users(term)
         return 200, ret
 
 
@@ -676,15 +664,14 @@ class DeleteGroupAdminRestServlet(RestServlet):
         self.is_mine_id = hs.is_mine_id
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, group_id):
-        requester = yield self.auth.get_user_by_req(request)
-        yield assert_user_is_admin(self.auth, requester.user)
+    async def on_POST(self, request, group_id):
+        requester = await self.auth.get_user_by_req(request)
+        await assert_user_is_admin(self.auth, requester.user)
 
         if not self.is_mine_id(group_id):
             raise SynapseError(400, "Can only delete local groups")
 
-        yield self.group_server.delete_group(group_id, requester.user.to_string())
+        await self.group_server.delete_group(group_id, requester.user.to_string())
         return 200, {}
 
 
@@ -700,16 +687,15 @@ class AccountValidityRenewServlet(RestServlet):
         self.account_activity_handler = hs.get_account_validity_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request):
-        yield assert_requester_is_admin(self.auth, request)
+    async def on_POST(self, request):
+        await assert_requester_is_admin(self.auth, request)
 
         body = parse_json_object_from_request(request)
 
         if "user_id" not in body:
             raise SynapseError(400, "Missing property 'user_id' in the request body")
 
-        expiration_ts = yield self.account_activity_handler.renew_account_for_user(
+        expiration_ts = await self.account_activity_handler.renew_account_for_user(
             body["user_id"],
             body.get("expiration_ts"),
             not body.get("enable_renewal_emails", True),
@@ -761,9 +747,12 @@ def register_servlets_for_client_rest_resource(hs, http_server):
     DeleteGroupAdminRestServlet(hs).register(http_server)
     AccountValidityRenewServlet(hs).register(http_server)
 
-    # Load the media repo ones if we're using them.
+    # Load the media repo ones if we're using them. Otherwise load the servlets which
+    # don't need a media repo (typically readonly admin APIs).
     if hs.config.can_load_media_repo:
         register_servlets_for_media_repo(hs, http_server)
+    else:
+        ListMediaInRoom(hs).register(http_server)
 
     # don't add more things here: new servlets should only be exposed on
     # /_synapse/admin so should not go here. Instead register them in AdminRestResource.
diff --git a/synapse/rest/admin/_base.py b/synapse/rest/admin/_base.py
index 5a9b08d3ef..afd0647205 100644
--- a/synapse/rest/admin/_base.py
+++ b/synapse/rest/admin/_base.py
@@ -15,8 +15,6 @@
 
 import re
 
-from twisted.internet import defer
-
 from synapse.api.errors import AuthError
 
 
@@ -42,8 +40,7 @@ def historical_admin_path_patterns(path_regex):
     )
 
 
-@defer.inlineCallbacks
-def assert_requester_is_admin(auth, request):
+async def assert_requester_is_admin(auth, request):
     """Verify that the requester is an admin user
 
     WARNING: MAKE SURE YOU YIELD ON THE RESULT!
@@ -58,12 +55,11 @@ def assert_requester_is_admin(auth, request):
     Raises:
         AuthError if the requester is not an admin
     """
-    requester = yield auth.get_user_by_req(request)
-    yield assert_user_is_admin(auth, requester.user)
+    requester = await auth.get_user_by_req(request)
+    await assert_user_is_admin(auth, requester.user)
 
 
-@defer.inlineCallbacks
-def assert_user_is_admin(auth, user_id):
+async def assert_user_is_admin(auth, user_id):
     """Verify that the given user is an admin user
 
     WARNING: MAKE SURE YOU YIELD ON THE RESULT!
@@ -79,6 +75,6 @@ def assert_user_is_admin(auth, user_id):
         AuthError if the user is not an admin
     """
 
-    is_admin = yield auth.is_server_admin(user_id)
+    is_admin = await auth.is_server_admin(user_id)
     if not is_admin:
         raise AuthError(403, "You are not a server admin")
diff --git a/synapse/rest/admin/media.py b/synapse/rest/admin/media.py
index f3f63f0be7..fa833e54cf 100644
--- a/synapse/rest/admin/media.py
+++ b/synapse/rest/admin/media.py
@@ -16,8 +16,6 @@
 
 import logging
 
-from twisted.internet import defer
-
 from synapse.api.errors import AuthError
 from synapse.http.servlet import RestServlet, parse_integer
 from synapse.rest.admin._base import (
@@ -40,12 +38,11 @@ class QuarantineMediaInRoom(RestServlet):
         self.store = hs.get_datastore()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request)
-        yield assert_user_is_admin(self.auth, requester.user)
+    async def on_POST(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request)
+        await assert_user_is_admin(self.auth, requester.user)
 
-        num_quarantined = yield self.store.quarantine_media_ids_in_room(
+        num_quarantined = await self.store.quarantine_media_ids_in_room(
             room_id, requester.user.to_string()
         )
 
@@ -60,15 +57,15 @@ class ListMediaInRoom(RestServlet):
 
     def __init__(self, hs):
         self.store = hs.get_datastore()
+        self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request)
-        is_admin = yield self.auth.is_server_admin(requester.user)
+    async def on_GET(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request)
+        is_admin = await self.auth.is_server_admin(requester.user)
         if not is_admin:
             raise AuthError(403, "You are not a server admin")
 
-        local_mxcs, remote_mxcs = yield self.store.get_media_mxcs_in_room(room_id)
+        local_mxcs, remote_mxcs = await self.store.get_media_mxcs_in_room(room_id)
 
         return 200, {"local": local_mxcs, "remote": remote_mxcs}
 
@@ -80,14 +77,13 @@ class PurgeMediaCacheRestServlet(RestServlet):
         self.media_repository = hs.get_media_repository()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request):
-        yield assert_requester_is_admin(self.auth, request)
+    async def on_POST(self, request):
+        await assert_requester_is_admin(self.auth, request)
 
         before_ts = parse_integer(request, "before_ts", required=True)
         logger.info("before_ts: %r", before_ts)
 
-        ret = yield self.media_repository.delete_old_remote_media(before_ts)
+        ret = await self.media_repository.delete_old_remote_media(before_ts)
 
         return 200, ret
 
diff --git a/synapse/rest/admin/server_notice_servlet.py b/synapse/rest/admin/server_notice_servlet.py
index ae2cbe2e0a..6e9a874121 100644
--- a/synapse/rest/admin/server_notice_servlet.py
+++ b/synapse/rest/admin/server_notice_servlet.py
@@ -14,8 +14,6 @@
 # limitations under the License.
 import re
 
-from twisted.internet import defer
-
 from synapse.api.constants import EventTypes
 from synapse.api.errors import SynapseError
 from synapse.http.servlet import (
@@ -69,9 +67,8 @@ class SendServerNoticeServlet(RestServlet):
             self.__class__.__name__,
         )
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, txn_id=None):
-        yield assert_requester_is_admin(self.auth, request)
+    async def on_POST(self, request, txn_id=None):
+        await assert_requester_is_admin(self.auth, request)
         body = parse_json_object_from_request(request)
         assert_params_in_dict(body, ("user_id", "content"))
         event_type = body.get("type", EventTypes.Message)
@@ -85,7 +82,7 @@ class SendServerNoticeServlet(RestServlet):
         if not self.hs.is_mine_id(user_id):
             raise SynapseError(400, "Server notices can only be sent to local users")
 
-        event = yield self.snm.send_notice(
+        event = await self.snm.send_notice(
             user_id=body["user_id"],
             type=event_type,
             state_key=state_key,
diff --git a/synapse/rest/admin/users.py b/synapse/rest/admin/users.py
index 9720a3bab0..d5d124a0dc 100644
--- a/synapse/rest/admin/users.py
+++ b/synapse/rest/admin/users.py
@@ -14,8 +14,6 @@
 # limitations under the License.
 import re
 
-from twisted.internet import defer
-
 from synapse.api.errors import SynapseError
 from synapse.http.servlet import (
     RestServlet,
@@ -59,24 +57,22 @@ class UserAdminServlet(RestServlet):
         self.auth = hs.get_auth()
         self.handlers = hs.get_handlers()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, user_id):
-        yield assert_requester_is_admin(self.auth, request)
+    async def on_GET(self, request, user_id):
+        await assert_requester_is_admin(self.auth, request)
 
         target_user = UserID.from_string(user_id)
 
         if not self.hs.is_mine(target_user):
             raise SynapseError(400, "Only local users can be admins of this homeserver")
 
-        is_admin = yield self.handlers.admin_handler.get_user_server_admin(target_user)
+        is_admin = await self.handlers.admin_handler.get_user_server_admin(target_user)
         is_admin = bool(is_admin)
 
         return 200, {"admin": is_admin}
 
-    @defer.inlineCallbacks
-    def on_PUT(self, request, user_id):
-        requester = yield self.auth.get_user_by_req(request)
-        yield assert_user_is_admin(self.auth, requester.user)
+    async def on_PUT(self, request, user_id):
+        requester = await self.auth.get_user_by_req(request)
+        await assert_user_is_admin(self.auth, requester.user)
         auth_user = requester.user
 
         target_user = UserID.from_string(user_id)
@@ -93,7 +89,7 @@ class UserAdminServlet(RestServlet):
         if target_user == auth_user and not set_admin_to:
             raise SynapseError(400, "You may not demote yourself.")
 
-        yield self.handlers.admin_handler.set_user_server_admin(
+        await self.handlers.admin_handler.set_user_server_admin(
             target_user, set_admin_to
         )
 
diff --git a/synapse/rest/client/v1/login.py b/synapse/rest/client/v1/login.py
index 25a1b67092..24a0ce74f2 100644
--- a/synapse/rest/client/v1/login.py
+++ b/synapse/rest/client/v1/login.py
@@ -29,6 +29,7 @@ from synapse.http.servlet import (
     parse_json_object_from_request,
     parse_string,
 )
+from synapse.http.site import SynapseRequest
 from synapse.rest.client.v2_alpha._base import client_patterns
 from synapse.rest.well_known import WellKnownBuilder
 from synapse.types import UserID, map_username_to_mxid_localpart
@@ -202,10 +203,11 @@ class LoginRestServlet(RestServlet):
                 address = address.lower()
 
             # Check for login providers that support 3pid login types
-            canonical_user_id, callback_3pid = (
-                yield self.auth_handler.check_password_provider_3pid(
-                    medium, address, login_submission["password"]
-                )
+            (
+                canonical_user_id,
+                callback_3pid,
+            ) = yield self.auth_handler.check_password_provider_3pid(
+                medium, address, login_submission["password"]
             )
             if canonical_user_id:
                 # Authentication through password provider and 3pid succeeded
@@ -220,7 +222,7 @@ class LoginRestServlet(RestServlet):
                 medium, address
             )
             if not user_id:
-                logger.warn(
+                logger.warning(
                     "unknown 3pid identifier medium %s, address %r", medium, address
                 )
                 raise LoginError(403, "", errcode=Codes.FORBIDDEN)
@@ -279,8 +281,8 @@ class LoginRestServlet(RestServlet):
     def do_token_login(self, login_submission):
         token = login_submission["token"]
         auth_handler = self.auth_handler
-        user_id = (
-            yield auth_handler.validate_short_term_login_token_and_get_user_id(token)
+        user_id = yield auth_handler.validate_short_term_login_token_and_get_user_id(
+            token
         )
 
         result = yield self._register_device_with_callback(user_id, login_submission)
@@ -376,9 +378,10 @@ class CasTicketServlet(RestServlet):
         super(CasTicketServlet, self).__init__()
         self.cas_server_url = hs.config.cas_server_url
         self.cas_service_url = hs.config.cas_service_url
+        self.cas_displayname_attribute = hs.config.cas_displayname_attribute
         self.cas_required_attributes = hs.config.cas_required_attributes
         self._sso_auth_handler = SSOAuthHandler(hs)
-        self._http_client = hs.get_simple_http_client()
+        self._http_client = hs.get_proxied_http_client()
 
     @defer.inlineCallbacks
     def on_GET(self, request):
@@ -399,6 +402,7 @@ class CasTicketServlet(RestServlet):
 
     def handle_cas_response(self, request, cas_response_body, client_redirect_url):
         user, attributes = self.parse_cas_response(cas_response_body)
+        displayname = attributes.pop(self.cas_displayname_attribute, None)
 
         for required_attribute, required_value in self.cas_required_attributes.items():
             # If required attribute was not in CAS Response - Forbidden
@@ -413,7 +417,7 @@ class CasTicketServlet(RestServlet):
                     raise LoginError(401, "Unauthorized", errcode=Codes.UNAUTHORIZED)
 
         return self._sso_auth_handler.on_successful_auth(
-            user, request, client_redirect_url
+            user, request, client_redirect_url, displayname
         )
 
     def parse_cas_response(self, cas_response_body):
@@ -507,6 +511,19 @@ class SSOAuthHandler(object):
                 localpart=localpart, default_display_name=user_display_name
             )
 
+        self.complete_sso_login(registered_user_id, request, client_redirect_url)
+
+    def complete_sso_login(
+        self, registered_user_id: str, request: SynapseRequest, client_redirect_url: str
+    ):
+        """Having figured out a mxid for this user, complete the HTTP request
+
+        Args:
+            registered_user_id:
+            request:
+            client_redirect_url:
+        """
+
         login_token = self._macaroon_gen.generate_short_term_login_token(
             registered_user_id
         )
diff --git a/synapse/rest/client/v1/room.py b/synapse/rest/client/v1/room.py
index f244e8f469..86bbcc0eea 100644
--- a/synapse/rest/client/v1/room.py
+++ b/synapse/rest/client/v1/room.py
@@ -21,8 +21,6 @@ from six.moves.urllib import parse as urlparse
 
 from canonicaljson import json
 
-from twisted.internet import defer
-
 from synapse.api.constants import EventTypes, Membership
 from synapse.api.errors import (
     AuthError,
@@ -39,6 +37,7 @@ from synapse.http.servlet import (
     parse_json_object_from_request,
     parse_string,
 )
+from synapse.logging.opentracing import set_tag
 from synapse.rest.client.transactions import HttpTransactionCache
 from synapse.rest.client.v2_alpha._base import client_patterns
 from synapse.storage.state import StateFilter
@@ -81,13 +80,13 @@ class RoomCreateRestServlet(TransactionRestServlet):
         )
 
     def on_PUT(self, request, txn_id):
+        set_tag("txn_id", txn_id)
         return self.txns.fetch_or_execute_request(request, self.on_POST, request)
 
-    @defer.inlineCallbacks
-    def on_POST(self, request):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_POST(self, request):
+        requester = await self.auth.get_user_by_req(request)
 
-        info = yield self._room_creation_handler.create_room(
+        info = await self._room_creation_handler.create_room(
             requester, self.get_room_config(request)
         )
 
@@ -152,15 +151,14 @@ class RoomStateEventRestServlet(TransactionRestServlet):
     def on_PUT_no_state_key(self, request, room_id, event_type):
         return self.on_PUT(request, room_id, event_type, "")
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id, event_type, state_key):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_GET(self, request, room_id, event_type, state_key):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
         format = parse_string(
             request, "format", default="content", allowed_values=["content", "event"]
         )
 
         msg_handler = self.message_handler
-        data = yield msg_handler.get_room_data(
+        data = await msg_handler.get_room_data(
             user_id=requester.user.to_string(),
             room_id=room_id,
             event_type=event_type,
@@ -177,9 +175,11 @@ class RoomStateEventRestServlet(TransactionRestServlet):
         elif format == "content":
             return 200, data.get_dict()["content"]
 
-    @defer.inlineCallbacks
-    def on_PUT(self, request, room_id, event_type, state_key, txn_id=None):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_PUT(self, request, room_id, event_type, state_key, txn_id=None):
+        requester = await self.auth.get_user_by_req(request)
+
+        if txn_id:
+            set_tag("txn_id", txn_id)
 
         content = parse_json_object_from_request(request)
 
@@ -195,7 +195,7 @@ class RoomStateEventRestServlet(TransactionRestServlet):
 
         if event_type == EventTypes.Member:
             membership = content.get("membership", None)
-            event = yield self.room_member_handler.update_membership(
+            event = await self.room_member_handler.update_membership(
                 requester,
                 target=UserID.from_string(state_key),
                 room_id=room_id,
@@ -203,12 +203,13 @@ class RoomStateEventRestServlet(TransactionRestServlet):
                 content=content,
             )
         else:
-            event = yield self.event_creation_handler.create_and_send_nonmember_event(
+            event = await self.event_creation_handler.create_and_send_nonmember_event(
                 requester, event_dict, txn_id=txn_id
             )
 
         ret = {}
         if event:
+            set_tag("event_id", event.event_id)
             ret = {"event_id": event.event_id}
         return 200, ret
 
@@ -225,9 +226,8 @@ class RoomSendEventRestServlet(TransactionRestServlet):
         PATTERNS = "/rooms/(?P[^/]*)/send/(?P[^/]*)"
         register_txn_path(self, PATTERNS, http_server, with_get=True)
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id, event_type, txn_id=None):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_POST(self, request, room_id, event_type, txn_id=None):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
         content = parse_json_object_from_request(request)
 
         event_dict = {
@@ -240,16 +240,19 @@ class RoomSendEventRestServlet(TransactionRestServlet):
         if b"ts" in request.args and requester.app_service:
             event_dict["origin_server_ts"] = parse_integer(request, "ts", 0)
 
-        event = yield self.event_creation_handler.create_and_send_nonmember_event(
+        event = await self.event_creation_handler.create_and_send_nonmember_event(
             requester, event_dict, txn_id=txn_id
         )
 
+        set_tag("event_id", event.event_id)
         return 200, {"event_id": event.event_id}
 
     def on_GET(self, request, room_id, event_type, txn_id):
         return 200, "Not implemented"
 
     def on_PUT(self, request, room_id, event_type, txn_id):
+        set_tag("txn_id", txn_id)
+
         return self.txns.fetch_or_execute_request(
             request, self.on_POST, request, room_id, event_type, txn_id
         )
@@ -267,9 +270,8 @@ class JoinRoomAliasServlet(TransactionRestServlet):
         PATTERNS = "/join/(?P[^/]*)"
         register_txn_path(self, PATTERNS, http_server)
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_identifier, txn_id=None):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_POST(self, request, room_identifier, txn_id=None):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
 
         try:
             content = parse_json_object_from_request(request)
@@ -289,14 +291,14 @@ class JoinRoomAliasServlet(TransactionRestServlet):
         elif RoomAlias.is_valid(room_identifier):
             handler = self.room_member_handler
             room_alias = RoomAlias.from_string(room_identifier)
-            room_id, remote_room_hosts = yield handler.lookup_room_alias(room_alias)
+            room_id, remote_room_hosts = await handler.lookup_room_alias(room_alias)
             room_id = room_id.to_string()
         else:
             raise SynapseError(
                 400, "%s was not legal room ID or room alias" % (room_identifier,)
             )
 
-        yield self.room_member_handler.update_membership(
+        await self.room_member_handler.update_membership(
             requester=requester,
             target=requester.user,
             room_id=room_id,
@@ -310,6 +312,8 @@ class JoinRoomAliasServlet(TransactionRestServlet):
         return 200, {"room_id": room_id}
 
     def on_PUT(self, request, room_identifier, txn_id):
+        set_tag("txn_id", txn_id)
+
         return self.txns.fetch_or_execute_request(
             request, self.on_POST, request, room_identifier, txn_id
         )
@@ -324,12 +328,11 @@ class PublicRoomListRestServlet(TransactionRestServlet):
         self.hs = hs
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request):
+    async def on_GET(self, request):
         server = parse_string(request, "server", default=None)
 
         try:
-            yield self.auth.get_user_by_req(request, allow_guest=True)
+            await self.auth.get_user_by_req(request, allow_guest=True)
         except InvalidClientCredentialsError as e:
             # Option to allow servers to require auth when accessing
             # /publicRooms via CS API. This is especially helpful in private
@@ -350,21 +353,24 @@ class PublicRoomListRestServlet(TransactionRestServlet):
         limit = parse_integer(request, "limit", 0)
         since_token = parse_string(request, "since", None)
 
+        if limit == 0:
+            # zero is a special value which corresponds to no limit.
+            limit = None
+
         handler = self.hs.get_room_list_handler()
         if server:
-            data = yield handler.get_remote_public_room_list(
+            data = await handler.get_remote_public_room_list(
                 server, limit=limit, since_token=since_token
             )
         else:
-            data = yield handler.get_local_public_room_list(
+            data = await handler.get_local_public_room_list(
                 limit=limit, since_token=since_token
             )
 
         return 200, data
 
-    @defer.inlineCallbacks
-    def on_POST(self, request):
-        yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_POST(self, request):
+        await self.auth.get_user_by_req(request, allow_guest=True)
 
         server = parse_string(request, "server", default=None)
         content = parse_json_object_from_request(request)
@@ -387,9 +393,13 @@ class PublicRoomListRestServlet(TransactionRestServlet):
         else:
             network_tuple = ThirdPartyInstanceID.from_string(third_party_instance_id)
 
+        if limit == 0:
+            # zero is a special value which corresponds to no limit.
+            limit = None
+
         handler = self.hs.get_room_list_handler()
         if server:
-            data = yield handler.get_remote_public_room_list(
+            data = await handler.get_remote_public_room_list(
                 server,
                 limit=limit,
                 since_token=since_token,
@@ -398,7 +408,7 @@ class PublicRoomListRestServlet(TransactionRestServlet):
                 third_party_instance_id=third_party_instance_id,
             )
         else:
-            data = yield handler.get_local_public_room_list(
+            data = await handler.get_local_public_room_list(
                 limit=limit,
                 since_token=since_token,
                 search_filter=search_filter,
@@ -417,10 +427,9 @@ class RoomMemberListRestServlet(RestServlet):
         self.message_handler = hs.get_message_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id):
+    async def on_GET(self, request, room_id):
         # TODO support Pagination stream API (limit/tokens)
-        requester = yield self.auth.get_user_by_req(request)
+        requester = await self.auth.get_user_by_req(request)
         handler = self.message_handler
 
         # request the state as of a given event, as identified by a stream token,
@@ -440,7 +449,7 @@ class RoomMemberListRestServlet(RestServlet):
         membership = parse_string(request, "membership")
         not_membership = parse_string(request, "not_membership")
 
-        events = yield handler.get_state_events(
+        events = await handler.get_state_events(
             room_id=room_id,
             user_id=requester.user.to_string(),
             at_token=at_token,
@@ -469,11 +478,10 @@ class JoinedRoomMemberListRestServlet(RestServlet):
         self.message_handler = hs.get_message_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_GET(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request)
 
-        users_with_profile = yield self.message_handler.get_joined_members(
+        users_with_profile = await self.message_handler.get_joined_members(
             requester, room_id
         )
 
@@ -489,9 +497,8 @@ class RoomMessageListRestServlet(RestServlet):
         self.pagination_handler = hs.get_pagination_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_GET(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
         pagination_config = PaginationConfig.from_request(request, default_limit=10)
         as_client_event = b"raw" not in request.args
         filter_bytes = parse_string(request, b"filter", encoding=None)
@@ -502,7 +509,7 @@ class RoomMessageListRestServlet(RestServlet):
                 as_client_event = False
         else:
             event_filter = None
-        msgs = yield self.pagination_handler.get_messages(
+        msgs = await self.pagination_handler.get_messages(
             room_id=room_id,
             requester=requester,
             pagin_config=pagination_config,
@@ -522,11 +529,10 @@ class RoomStateRestServlet(RestServlet):
         self.message_handler = hs.get_message_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_GET(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
         # Get all the current state for this room
-        events = yield self.message_handler.get_state_events(
+        events = await self.message_handler.get_state_events(
             room_id=room_id,
             user_id=requester.user.to_string(),
             is_guest=requester.is_guest,
@@ -543,11 +549,10 @@ class RoomInitialSyncRestServlet(RestServlet):
         self.initial_sync_handler = hs.get_initial_sync_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_GET(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
         pagination_config = PaginationConfig.from_request(request)
-        content = yield self.initial_sync_handler.room_initial_sync(
+        content = await self.initial_sync_handler.room_initial_sync(
             room_id=room_id, requester=requester, pagin_config=pagination_config
         )
         return 200, content
@@ -565,11 +570,10 @@ class RoomEventServlet(RestServlet):
         self._event_serializer = hs.get_event_client_serializer()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id, event_id):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_GET(self, request, room_id, event_id):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
         try:
-            event = yield self.event_handler.get_event(
+            event = await self.event_handler.get_event(
                 requester.user, room_id, event_id
             )
         except AuthError:
@@ -580,7 +584,7 @@ class RoomEventServlet(RestServlet):
 
         time_now = self.clock.time_msec()
         if event:
-            event = yield self._event_serializer.serialize_event(event, time_now)
+            event = await self._event_serializer.serialize_event(event, time_now)
             return 200, event
 
         return SynapseError(404, "Event not found.", errcode=Codes.NOT_FOUND)
@@ -598,9 +602,8 @@ class RoomEventContextServlet(RestServlet):
         self._event_serializer = hs.get_event_client_serializer()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request, room_id, event_id):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_GET(self, request, room_id, event_id):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
 
         limit = parse_integer(request, "limit", default=10)
 
@@ -612,7 +615,7 @@ class RoomEventContextServlet(RestServlet):
         else:
             event_filter = None
 
-        results = yield self.room_context_handler.get_event_context(
+        results = await self.room_context_handler.get_event_context(
             requester.user, room_id, event_id, limit, event_filter
         )
 
@@ -620,16 +623,16 @@ class RoomEventContextServlet(RestServlet):
             raise SynapseError(404, "Event not found.", errcode=Codes.NOT_FOUND)
 
         time_now = self.clock.time_msec()
-        results["events_before"] = yield self._event_serializer.serialize_events(
+        results["events_before"] = await self._event_serializer.serialize_events(
             results["events_before"], time_now
         )
-        results["event"] = yield self._event_serializer.serialize_event(
+        results["event"] = await self._event_serializer.serialize_event(
             results["event"], time_now
         )
-        results["events_after"] = yield self._event_serializer.serialize_events(
+        results["events_after"] = await self._event_serializer.serialize_events(
             results["events_after"], time_now
         )
-        results["state"] = yield self._event_serializer.serialize_events(
+        results["state"] = await self._event_serializer.serialize_events(
             results["state"], time_now
         )
 
@@ -646,15 +649,16 @@ class RoomForgetRestServlet(TransactionRestServlet):
         PATTERNS = "/rooms/(?P[^/]*)/forget"
         register_txn_path(self, PATTERNS, http_server)
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id, txn_id=None):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=False)
+    async def on_POST(self, request, room_id, txn_id=None):
+        requester = await self.auth.get_user_by_req(request, allow_guest=False)
 
-        yield self.room_member_handler.forget(user=requester.user, room_id=room_id)
+        await self.room_member_handler.forget(user=requester.user, room_id=room_id)
 
         return 200, {}
 
     def on_PUT(self, request, room_id, txn_id):
+        set_tag("txn_id", txn_id)
+
         return self.txns.fetch_or_execute_request(
             request, self.on_POST, request, room_id, txn_id
         )
@@ -675,9 +679,8 @@ class RoomMembershipRestServlet(TransactionRestServlet):
         )
         register_txn_path(self, PATTERNS, http_server)
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id, membership_action, txn_id=None):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_POST(self, request, room_id, membership_action, txn_id=None):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
 
         if requester.is_guest and membership_action not in {
             Membership.JOIN,
@@ -693,7 +696,7 @@ class RoomMembershipRestServlet(TransactionRestServlet):
             content = {}
 
         if membership_action == "invite" and self._has_3pid_invite_keys(content):
-            yield self.room_member_handler.do_3pid_invite(
+            await self.room_member_handler.do_3pid_invite(
                 room_id,
                 requester.user,
                 content["medium"],
@@ -701,9 +704,9 @@ class RoomMembershipRestServlet(TransactionRestServlet):
                 content["id_server"],
                 requester,
                 txn_id,
+                content.get("id_access_token"),
             )
             return 200, {}
-            return
 
         target = requester.user
         if membership_action in ["invite", "ban", "unban", "kick"]:
@@ -714,7 +717,7 @@ class RoomMembershipRestServlet(TransactionRestServlet):
         if "reason" in content and membership_action in ["kick", "ban"]:
             event_content = {"reason": content["reason"]}
 
-        yield self.room_member_handler.update_membership(
+        await self.room_member_handler.update_membership(
             requester=requester,
             target=target,
             room_id=room_id,
@@ -738,6 +741,8 @@ class RoomMembershipRestServlet(TransactionRestServlet):
         return True
 
     def on_PUT(self, request, room_id, membership_action, txn_id):
+        set_tag("txn_id", txn_id)
+
         return self.txns.fetch_or_execute_request(
             request, self.on_POST, request, room_id, membership_action, txn_id
         )
@@ -754,12 +759,11 @@ class RoomRedactEventRestServlet(TransactionRestServlet):
         PATTERNS = "/rooms/(?P[^/]*)/redact/(?P[^/]*)"
         register_txn_path(self, PATTERNS, http_server)
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id, event_id, txn_id=None):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_POST(self, request, room_id, event_id, txn_id=None):
+        requester = await self.auth.get_user_by_req(request)
         content = parse_json_object_from_request(request)
 
-        event = yield self.event_creation_handler.create_and_send_nonmember_event(
+        event = await self.event_creation_handler.create_and_send_nonmember_event(
             requester,
             {
                 "type": EventTypes.Redaction,
@@ -771,9 +775,12 @@ class RoomRedactEventRestServlet(TransactionRestServlet):
             txn_id=txn_id,
         )
 
+        set_tag("event_id", event.event_id)
         return 200, {"event_id": event.event_id}
 
     def on_PUT(self, request, room_id, event_id, txn_id):
+        set_tag("txn_id", txn_id)
+
         return self.txns.fetch_or_execute_request(
             request, self.on_POST, request, room_id, event_id, txn_id
         )
@@ -790,29 +797,28 @@ class RoomTypingRestServlet(RestServlet):
         self.typing_handler = hs.get_typing_handler()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_PUT(self, request, room_id, user_id):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_PUT(self, request, room_id, user_id):
+        requester = await self.auth.get_user_by_req(request)
 
         room_id = urlparse.unquote(room_id)
         target_user = UserID.from_string(urlparse.unquote(user_id))
 
         content = parse_json_object_from_request(request)
 
-        yield self.presence_handler.bump_presence_active_time(requester.user)
+        await self.presence_handler.bump_presence_active_time(requester.user)
 
         # Limit timeout to stop people from setting silly typing timeouts.
         timeout = min(content.get("timeout", 30000), 120000)
 
         if content["typing"]:
-            yield self.typing_handler.started_typing(
+            await self.typing_handler.started_typing(
                 target_user=target_user,
                 auth_user=requester.user,
                 room_id=room_id,
                 timeout=timeout,
             )
         else:
-            yield self.typing_handler.stopped_typing(
+            await self.typing_handler.stopped_typing(
                 target_user=target_user, auth_user=requester.user, room_id=room_id
             )
 
@@ -827,14 +833,13 @@ class SearchRestServlet(RestServlet):
         self.handlers = hs.get_handlers()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_POST(self, request):
+        requester = await self.auth.get_user_by_req(request)
 
         content = parse_json_object_from_request(request)
 
         batch = parse_string(request, "next_batch")
-        results = yield self.handlers.search_handler.search(
+        results = await self.handlers.search_handler.search(
             requester.user, content, batch
         )
 
@@ -849,11 +854,10 @@ class JoinedRoomsRestServlet(RestServlet):
         self.store = hs.get_datastore()
         self.auth = hs.get_auth()
 
-    @defer.inlineCallbacks
-    def on_GET(self, request):
-        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+    async def on_GET(self, request):
+        requester = await self.auth.get_user_by_req(request, allow_guest=True)
 
-        room_ids = yield self.store.get_rooms_for_user(requester.user.to_string())
+        room_ids = await self.store.get_rooms_for_user(requester.user.to_string())
         return 200, {"joined_rooms": list(room_ids)}
 
 
diff --git a/synapse/rest/client/v2_alpha/_base.py b/synapse/rest/client/v2_alpha/_base.py
index e3d59ac3ac..8250ae0ae1 100644
--- a/synapse/rest/client/v2_alpha/_base.py
+++ b/synapse/rest/client/v2_alpha/_base.py
@@ -37,6 +37,7 @@ def client_patterns(path_regex, releases=(0,), unstable=True, v1=False):
         SRE_Pattern
     """
     patterns = []
+
     if unstable:
         unstable_prefix = CLIENT_API_PREFIX + "/unstable"
         patterns.append(re.compile("^" + unstable_prefix + path_regex))
@@ -46,6 +47,7 @@ def client_patterns(path_regex, releases=(0,), unstable=True, v1=False):
     for release in releases:
         new_prefix = CLIENT_API_PREFIX + "/r%d" % (release,)
         patterns.append(re.compile("^" + new_prefix + path_regex))
+
     return patterns
 
 
diff --git a/synapse/rest/client/v2_alpha/account.py b/synapse/rest/client/v2_alpha/account.py
index 0620a4d0cf..f26eae794c 100644
--- a/synapse/rest/client/v2_alpha/account.py
+++ b/synapse/rest/client/v2_alpha/account.py
@@ -18,12 +18,11 @@ import logging
 
 from six.moves import http_client
 
-import jinja2
-
 from twisted.internet import defer
 
 from synapse.api.constants import LoginType
 from synapse.api.errors import Codes, SynapseError, ThreepidValidationError
+from synapse.config.emailconfig import ThreepidBehaviour
 from synapse.http.server import finish_request
 from synapse.http.servlet import (
     RestServlet,
@@ -31,8 +30,8 @@ from synapse.http.servlet import (
     parse_json_object_from_request,
     parse_string,
 )
+from synapse.push.mailer import Mailer, load_jinja2_templates
 from synapse.util.msisdn import phone_number_to_msisdn
-from synapse.util.stringutils import random_string
 from synapse.util.threepids import check_3pid_allowed
 
 from ._base import client_patterns, interactive_auth_handler
@@ -50,26 +49,29 @@ class EmailPasswordRequestTokenRestServlet(RestServlet):
         self.config = hs.config
         self.identity_handler = hs.get_handlers().identity_handler
 
-        if self.config.email_password_reset_behaviour == "local":
-            from synapse.push.mailer import Mailer, load_jinja2_templates
-
-            templates = load_jinja2_templates(
-                config=hs.config,
-                template_html_name=hs.config.email_password_reset_template_html,
-                template_text_name=hs.config.email_password_reset_template_text,
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            template_html, template_text = load_jinja2_templates(
+                self.config.email_template_dir,
+                [
+                    self.config.email_password_reset_template_html,
+                    self.config.email_password_reset_template_text,
+                ],
+                apply_format_ts_filter=True,
+                apply_mxc_to_http_filter=True,
+                public_baseurl=self.config.public_baseurl,
             )
             self.mailer = Mailer(
                 hs=self.hs,
                 app_name=self.config.email_app_name,
-                template_html=templates[0],
-                template_text=templates[1],
+                template_html=template_html,
+                template_text=template_text,
             )
 
     @defer.inlineCallbacks
     def on_POST(self, request):
-        if self.config.email_password_reset_behaviour == "off":
-            if self.config.password_resets_were_disabled_due_to_email_config:
-                logger.warn(
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.OFF:
+            if self.config.local_threepid_handling_disabled_due_to_email_config:
+                logger.warning(
                     "User password resets have been disabled due to lack of email config"
                 )
             raise SynapseError(
@@ -93,25 +95,32 @@ class EmailPasswordRequestTokenRestServlet(RestServlet):
                 Codes.THREEPID_DENIED,
             )
 
-        existingUid = yield self.hs.get_datastore().get_user_id_by_threepid(
+        existing_user_id = yield self.hs.get_datastore().get_user_id_by_threepid(
             "email", email
         )
 
-        if existingUid is None:
+        if existing_user_id is None:
             raise SynapseError(400, "Email not found", Codes.THREEPID_NOT_FOUND)
 
-        if self.config.email_password_reset_behaviour == "remote":
-            if "id_server" not in body:
-                raise SynapseError(400, "Missing 'id_server' param in body")
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.REMOTE:
+            assert self.hs.config.account_threepid_delegate_email
 
-            # Have the identity server handle the password reset flow
+            # Have the configured identity server handle the request
             ret = yield self.identity_handler.requestEmailToken(
-                body["id_server"], email, client_secret, send_attempt, next_link
+                self.hs.config.account_threepid_delegate_email,
+                email,
+                client_secret,
+                send_attempt,
+                next_link,
             )
         else:
             # Send password reset emails from Synapse
-            sid = yield self.send_password_reset(
-                email, client_secret, send_attempt, next_link
+            sid = yield self.identity_handler.send_threepid_validation(
+                email,
+                client_secret,
+                send_attempt,
+                self.mailer.send_password_reset_mail,
+                next_link,
             )
 
             # Wrap the session id in a JSON object
@@ -119,116 +128,12 @@ class EmailPasswordRequestTokenRestServlet(RestServlet):
 
         return 200, ret
 
-    @defer.inlineCallbacks
-    def send_password_reset(self, email, client_secret, send_attempt, next_link=None):
-        """Send a password reset email
-
-        Args:
-            email (str): The user's email address
-            client_secret (str): The provided client secret
-            send_attempt (int): Which send attempt this is
-
-        Returns:
-            The new session_id upon success
-
-        Raises:
-            SynapseError is an error occurred when sending the email
-        """
-        # Check that this email/client_secret/send_attempt combo is new or
-        # greater than what we've seen previously
-        session = yield self.datastore.get_threepid_validation_session(
-            "email", client_secret, address=email, validated=False
-        )
-
-        # Check to see if a session already exists and that it is not yet
-        # marked as validated
-        if session and session.get("validated_at") is None:
-            session_id = session["session_id"]
-            last_send_attempt = session["last_send_attempt"]
-
-            # Check that the send_attempt is higher than previous attempts
-            if send_attempt <= last_send_attempt:
-                # If not, just return a success without sending an email
-                return session_id
-        else:
-            # An non-validated session does not exist yet.
-            # Generate a session id
-            session_id = random_string(16)
-
-        # Generate a new validation token
-        token = random_string(32)
-
-        # Send the mail with the link containing the token, client_secret
-        # and session_id
-        try:
-            yield self.mailer.send_password_reset_mail(
-                email, token, client_secret, session_id
-            )
-        except Exception:
-            logger.exception("Error sending a password reset email to %s", email)
-            raise SynapseError(
-                500, "An error was encountered when sending the password reset email"
-            )
-
-        token_expires = (
-            self.hs.clock.time_msec() + self.config.email_validation_token_lifetime
-        )
-
-        yield self.datastore.start_or_continue_validation_session(
-            "email",
-            email,
-            session_id,
-            client_secret,
-            send_attempt,
-            next_link,
-            token,
-            token_expires,
-        )
-
-        return session_id
-
-
-class MsisdnPasswordRequestTokenRestServlet(RestServlet):
-    PATTERNS = client_patterns("/account/password/msisdn/requestToken$")
-
-    def __init__(self, hs):
-        super(MsisdnPasswordRequestTokenRestServlet, self).__init__()
-        self.hs = hs
-        self.datastore = self.hs.get_datastore()
-        self.identity_handler = hs.get_handlers().identity_handler
-
-    @defer.inlineCallbacks
-    def on_POST(self, request):
-        body = parse_json_object_from_request(request)
-
-        assert_params_in_dict(
-            body,
-            ["id_server", "client_secret", "country", "phone_number", "send_attempt"],
-        )
-
-        msisdn = phone_number_to_msisdn(body["country"], body["phone_number"])
-
-        if not check_3pid_allowed(self.hs, "msisdn", msisdn):
-            raise SynapseError(
-                403,
-                "Account phone numbers are not authorized on this server",
-                Codes.THREEPID_DENIED,
-            )
-
-        existingUid = yield self.datastore.get_user_id_by_threepid("msisdn", msisdn)
-
-        if existingUid is None:
-            raise SynapseError(400, "MSISDN not found", Codes.THREEPID_NOT_FOUND)
-
-        ret = yield self.identity_handler.requestMsisdnToken(**body)
-        return 200, ret
-
 
 class PasswordResetSubmitTokenServlet(RestServlet):
     """Handles 3PID validation token submission"""
 
     PATTERNS = client_patterns(
-        "/password_reset/(?P[^/]*)/submit_token/*$", releases=(), unstable=True
+        "/password_reset/(?P[^/]*)/submit_token$", releases=(), unstable=True
     )
 
     def __init__(self, hs):
@@ -241,38 +146,44 @@ class PasswordResetSubmitTokenServlet(RestServlet):
         self.auth = hs.get_auth()
         self.config = hs.config
         self.clock = hs.get_clock()
-        self.datastore = hs.get_datastore()
+        self.store = hs.get_datastore()
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            (self.failure_email_template,) = load_jinja2_templates(
+                self.config.email_template_dir,
+                [self.config.email_password_reset_template_failure_html],
+            )
 
     @defer.inlineCallbacks
     def on_GET(self, request, medium):
+        # We currently only handle threepid token submissions for email
         if medium != "email":
             raise SynapseError(
                 400, "This medium is currently not supported for password resets"
             )
-        if self.config.email_password_reset_behaviour == "off":
-            if self.config.password_resets_were_disabled_due_to_email_config:
-                logger.warn(
-                    "User password resets have been disabled due to lack of email config"
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.OFF:
+            if self.config.local_threepid_handling_disabled_due_to_email_config:
+                logger.warning(
+                    "Password reset emails have been disabled due to lack of an email config"
                 )
             raise SynapseError(
-                400, "Email-based password resets have been disabled on this server"
+                400, "Email-based password resets are disabled on this server"
             )
 
-        sid = parse_string(request, "sid")
-        client_secret = parse_string(request, "client_secret")
-        token = parse_string(request, "token")
+        sid = parse_string(request, "sid", required=True)
+        client_secret = parse_string(request, "client_secret", required=True)
+        token = parse_string(request, "token", required=True)
 
-        # Attempt to validate a 3PID sesssion
+        # Attempt to validate a 3PID session
         try:
             # Mark the session as valid
-            next_link = yield self.datastore.validate_threepid_session(
+            next_link = yield self.store.validate_threepid_session(
                 sid, client_secret, token, self.clock.time_msec()
             )
 
             # Perform a 302 redirect if next_link is set
             if next_link:
                 if next_link.startswith("file:///"):
-                    logger.warn(
+                    logger.warning(
                         "Not redirecting to next_link as it is a local file: address"
                     )
                 else:
@@ -282,55 +193,17 @@ class PasswordResetSubmitTokenServlet(RestServlet):
                     return None
 
             # Otherwise show the success template
-            html = self.config.email_password_reset_template_success_html_content
+            html = self.config.email_password_reset_template_success_html
             request.setResponseCode(200)
         except ThreepidValidationError as e:
-            # Show a failure page with a reason
-            html = self.load_jinja2_template(
-                self.config.email_template_dir,
-                self.config.email_password_reset_template_failure_html,
-                template_vars={"failure_reason": e.msg},
-            )
             request.setResponseCode(e.code)
 
+            # Show a failure page with a reason
+            template_vars = {"failure_reason": e.msg}
+            html = self.failure_email_template.render(**template_vars)
+
         request.write(html.encode("utf-8"))
         finish_request(request)
-        return None
-
-    def load_jinja2_template(self, template_dir, template_filename, template_vars):
-        """Loads a jinja2 template with variables to insert
-
-        Args:
-            template_dir (str): The directory where templates are stored
-            template_filename (str): The name of the template in the template_dir
-            template_vars (Dict): Dictionary of keys in the template
-                alongside their values to insert
-
-        Returns:
-            str containing the contents of the rendered template
-        """
-        loader = jinja2.FileSystemLoader(template_dir)
-        env = jinja2.Environment(loader=loader)
-
-        template = env.get_template(template_filename)
-        return template.render(**template_vars)
-
-    @defer.inlineCallbacks
-    def on_POST(self, request, medium):
-        if medium != "email":
-            raise SynapseError(
-                400, "This medium is currently not supported for password resets"
-            )
-
-        body = parse_json_object_from_request(request)
-        assert_params_in_dict(body, ["sid", "client_secret", "token"])
-
-        valid, _ = yield self.datastore.validate_threepid_validation_token(
-            body["sid"], body["client_secret"], body["token"], self.clock.time_msec()
-        )
-        response_code = 200 if valid else 400
-
-        return response_code, {"success": valid}
 
 
 class PasswordRestServlet(RestServlet):
@@ -368,10 +241,7 @@ class PasswordRestServlet(RestServlet):
         else:
             requester = None
             result, params, _ = yield self.auth_handler.check_auth(
-                [[LoginType.EMAIL_IDENTITY], [LoginType.MSISDN]],
-                body,
-                self.hs.get_ip_from_request(request),
-                password_servlet=True,
+                [[LoginType.EMAIL_IDENTITY]], body, self.hs.get_ip_from_request(request)
             )
 
             if LoginType.EMAIL_IDENTITY in result:
@@ -454,33 +324,84 @@ class EmailThreepidRequestTokenRestServlet(RestServlet):
     PATTERNS = client_patterns("/account/3pid/email/requestToken$")
 
     def __init__(self, hs):
-        self.hs = hs
         super(EmailThreepidRequestTokenRestServlet, self).__init__()
+        self.hs = hs
+        self.config = hs.config
         self.identity_handler = hs.get_handlers().identity_handler
-        self.datastore = self.hs.get_datastore()
+        self.store = self.hs.get_datastore()
+
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            template_html, template_text = load_jinja2_templates(
+                self.config.email_template_dir,
+                [
+                    self.config.email_add_threepid_template_html,
+                    self.config.email_add_threepid_template_text,
+                ],
+                public_baseurl=self.config.public_baseurl,
+            )
+            self.mailer = Mailer(
+                hs=self.hs,
+                app_name=self.config.email_app_name,
+                template_html=template_html,
+                template_text=template_text,
+            )
 
     @defer.inlineCallbacks
     def on_POST(self, request):
-        body = parse_json_object_from_request(request)
-        assert_params_in_dict(
-            body, ["id_server", "client_secret", "email", "send_attempt"]
-        )
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.OFF:
+            if self.config.local_threepid_handling_disabled_due_to_email_config:
+                logger.warning(
+                    "Adding emails have been disabled due to lack of an email config"
+                )
+            raise SynapseError(
+                400, "Adding an email to your account is disabled on this server"
+            )
 
-        if not check_3pid_allowed(self.hs, "email", body["email"]):
+        body = parse_json_object_from_request(request)
+        assert_params_in_dict(body, ["client_secret", "email", "send_attempt"])
+        client_secret = body["client_secret"]
+        email = body["email"]
+        send_attempt = body["send_attempt"]
+        next_link = body.get("next_link")  # Optional param
+
+        if not check_3pid_allowed(self.hs, "email", email):
             raise SynapseError(
                 403,
                 "Your email domain is not authorized on this server",
                 Codes.THREEPID_DENIED,
             )
 
-        existingUid = yield self.datastore.get_user_id_by_threepid(
+        existing_user_id = yield self.store.get_user_id_by_threepid(
             "email", body["email"]
         )
 
-        if existingUid is not None:
+        if existing_user_id is not None:
             raise SynapseError(400, "Email is already in use", Codes.THREEPID_IN_USE)
 
-        ret = yield self.identity_handler.requestEmailToken(**body)
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.REMOTE:
+            assert self.hs.config.account_threepid_delegate_email
+
+            # Have the configured identity server handle the request
+            ret = yield self.identity_handler.requestEmailToken(
+                self.hs.config.account_threepid_delegate_email,
+                email,
+                client_secret,
+                send_attempt,
+                next_link,
+            )
+        else:
+            # Send threepid validation emails from Synapse
+            sid = yield self.identity_handler.send_threepid_validation(
+                email,
+                client_secret,
+                send_attempt,
+                self.mailer.send_add_threepid_mail,
+                next_link,
+            )
+
+            # Wrap the session id in a JSON object
+            ret = {"sid": sid}
+
         return 200, ret
 
 
@@ -490,18 +411,22 @@ class MsisdnThreepidRequestTokenRestServlet(RestServlet):
     def __init__(self, hs):
         self.hs = hs
         super(MsisdnThreepidRequestTokenRestServlet, self).__init__()
+        self.store = self.hs.get_datastore()
         self.identity_handler = hs.get_handlers().identity_handler
-        self.datastore = self.hs.get_datastore()
 
     @defer.inlineCallbacks
     def on_POST(self, request):
         body = parse_json_object_from_request(request)
         assert_params_in_dict(
-            body,
-            ["id_server", "client_secret", "country", "phone_number", "send_attempt"],
+            body, ["client_secret", "country", "phone_number", "send_attempt"]
         )
+        client_secret = body["client_secret"]
+        country = body["country"]
+        phone_number = body["phone_number"]
+        send_attempt = body["send_attempt"]
+        next_link = body.get("next_link")  # Optional param
 
-        msisdn = phone_number_to_msisdn(body["country"], body["phone_number"])
+        msisdn = phone_number_to_msisdn(country, phone_number)
 
         if not check_3pid_allowed(self.hs, "msisdn", msisdn):
             raise SynapseError(
@@ -510,15 +435,151 @@ class MsisdnThreepidRequestTokenRestServlet(RestServlet):
                 Codes.THREEPID_DENIED,
             )
 
-        existingUid = yield self.datastore.get_user_id_by_threepid("msisdn", msisdn)
+        existing_user_id = yield self.store.get_user_id_by_threepid("msisdn", msisdn)
 
-        if existingUid is not None:
+        if existing_user_id is not None:
             raise SynapseError(400, "MSISDN is already in use", Codes.THREEPID_IN_USE)
 
-        ret = yield self.identity_handler.requestMsisdnToken(**body)
+        if not self.hs.config.account_threepid_delegate_msisdn:
+            logger.warning(
+                "No upstream msisdn account_threepid_delegate configured on the server to "
+                "handle this request"
+            )
+            raise SynapseError(
+                400,
+                "Adding phone numbers to user account is not supported by this homeserver",
+            )
+
+        ret = yield self.identity_handler.requestMsisdnToken(
+            self.hs.config.account_threepid_delegate_msisdn,
+            country,
+            phone_number,
+            client_secret,
+            send_attempt,
+            next_link,
+        )
+
         return 200, ret
 
 
+class AddThreepidEmailSubmitTokenServlet(RestServlet):
+    """Handles 3PID validation token submission for adding an email to a user's account"""
+
+    PATTERNS = client_patterns(
+        "/add_threepid/email/submit_token$", releases=(), unstable=True
+    )
+
+    def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer): server
+        """
+        super().__init__()
+        self.config = hs.config
+        self.clock = hs.get_clock()
+        self.store = hs.get_datastore()
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            (self.failure_email_template,) = load_jinja2_templates(
+                self.config.email_template_dir,
+                [self.config.email_add_threepid_template_failure_html],
+            )
+
+    @defer.inlineCallbacks
+    def on_GET(self, request):
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.OFF:
+            if self.config.local_threepid_handling_disabled_due_to_email_config:
+                logger.warning(
+                    "Adding emails have been disabled due to lack of an email config"
+                )
+            raise SynapseError(
+                400, "Adding an email to your account is disabled on this server"
+            )
+        elif self.config.threepid_behaviour_email == ThreepidBehaviour.REMOTE:
+            raise SynapseError(
+                400,
+                "This homeserver is not validating threepids. Use an identity server "
+                "instead.",
+            )
+
+        sid = parse_string(request, "sid", required=True)
+        client_secret = parse_string(request, "client_secret", required=True)
+        token = parse_string(request, "token", required=True)
+
+        # Attempt to validate a 3PID session
+        try:
+            # Mark the session as valid
+            next_link = yield self.store.validate_threepid_session(
+                sid, client_secret, token, self.clock.time_msec()
+            )
+
+            # Perform a 302 redirect if next_link is set
+            if next_link:
+                if next_link.startswith("file:///"):
+                    logger.warning(
+                        "Not redirecting to next_link as it is a local file: address"
+                    )
+                else:
+                    request.setResponseCode(302)
+                    request.setHeader("Location", next_link)
+                    finish_request(request)
+                    return None
+
+            # Otherwise show the success template
+            html = self.config.email_add_threepid_template_success_html_content
+            request.setResponseCode(200)
+        except ThreepidValidationError as e:
+            request.setResponseCode(e.code)
+
+            # Show a failure page with a reason
+            template_vars = {"failure_reason": e.msg}
+            html = self.failure_email_template.render(**template_vars)
+
+        request.write(html.encode("utf-8"))
+        finish_request(request)
+
+
+class AddThreepidMsisdnSubmitTokenServlet(RestServlet):
+    """Handles 3PID validation token submission for adding a phone number to a user's
+    account
+    """
+
+    PATTERNS = client_patterns(
+        "/add_threepid/msisdn/submit_token$", releases=(), unstable=True
+    )
+
+    def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer): server
+        """
+        super().__init__()
+        self.config = hs.config
+        self.clock = hs.get_clock()
+        self.store = hs.get_datastore()
+        self.identity_handler = hs.get_handlers().identity_handler
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        if not self.config.account_threepid_delegate_msisdn:
+            raise SynapseError(
+                400,
+                "This homeserver is not validating phone numbers. Use an identity server "
+                "instead.",
+            )
+
+        body = parse_json_object_from_request(request)
+        assert_params_in_dict(body, ["client_secret", "sid", "token"])
+
+        # Proxy submit_token request to msisdn threepid delegate
+        response = yield self.identity_handler.proxy_msisdn_submit_token(
+            self.config.account_threepid_delegate_msisdn,
+            body["client_secret"],
+            body["sid"],
+            body["token"],
+        )
+        return 200, response
+
+
 class ThreepidRestServlet(RestServlet):
     PATTERNS = client_patterns("/account/3pid$")
 
@@ -540,37 +601,135 @@ class ThreepidRestServlet(RestServlet):
 
     @defer.inlineCallbacks
     def on_POST(self, request):
+        requester = yield self.auth.get_user_by_req(request)
+        user_id = requester.user.to_string()
         body = parse_json_object_from_request(request)
 
-        threePidCreds = body.get("threePidCreds")
-        threePidCreds = body.get("three_pid_creds", threePidCreds)
-        if threePidCreds is None:
-            raise SynapseError(400, "Missing param", Codes.MISSING_PARAM)
+        threepid_creds = body.get("threePidCreds") or body.get("three_pid_creds")
+        if threepid_creds is None:
+            raise SynapseError(
+                400, "Missing param three_pid_creds", Codes.MISSING_PARAM
+            )
+        assert_params_in_dict(threepid_creds, ["client_secret", "sid"])
+
+        client_secret = threepid_creds["client_secret"]
+        sid = threepid_creds["sid"]
+
+        validation_session = yield self.identity_handler.validate_threepid_session(
+            client_secret, sid
+        )
+        if validation_session:
+            yield self.auth_handler.add_threepid(
+                user_id,
+                validation_session["medium"],
+                validation_session["address"],
+                validation_session["validated_at"],
+            )
+            return 200, {}
+
+        raise SynapseError(
+            400, "No validated 3pid session found", Codes.THREEPID_AUTH_FAILED
+        )
+
+
+class ThreepidAddRestServlet(RestServlet):
+    PATTERNS = client_patterns("/account/3pid/add$", releases=(), unstable=True)
+
+    def __init__(self, hs):
+        super(ThreepidAddRestServlet, self).__init__()
+        self.hs = hs
+        self.identity_handler = hs.get_handlers().identity_handler
+        self.auth = hs.get_auth()
+        self.auth_handler = hs.get_auth_handler()
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        requester = yield self.auth.get_user_by_req(request)
+        user_id = requester.user.to_string()
+        body = parse_json_object_from_request(request)
+
+        assert_params_in_dict(body, ["client_secret", "sid"])
+        client_secret = body["client_secret"]
+        sid = body["sid"]
+
+        validation_session = yield self.identity_handler.validate_threepid_session(
+            client_secret, sid
+        )
+        if validation_session:
+            yield self.auth_handler.add_threepid(
+                user_id,
+                validation_session["medium"],
+                validation_session["address"],
+                validation_session["validated_at"],
+            )
+            return 200, {}
+
+        raise SynapseError(
+            400, "No validated 3pid session found", Codes.THREEPID_AUTH_FAILED
+        )
+
+
+class ThreepidBindRestServlet(RestServlet):
+    PATTERNS = client_patterns("/account/3pid/bind$", releases=(), unstable=True)
+
+    def __init__(self, hs):
+        super(ThreepidBindRestServlet, self).__init__()
+        self.hs = hs
+        self.identity_handler = hs.get_handlers().identity_handler
+        self.auth = hs.get_auth()
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        body = parse_json_object_from_request(request)
+
+        assert_params_in_dict(body, ["id_server", "sid", "client_secret"])
+        id_server = body["id_server"]
+        sid = body["sid"]
+        client_secret = body["client_secret"]
+        id_access_token = body.get("id_access_token")  # optional
 
         requester = yield self.auth.get_user_by_req(request)
         user_id = requester.user.to_string()
 
-        threepid = yield self.identity_handler.threepid_from_creds(threePidCreds)
-
-        if not threepid:
-            raise SynapseError(400, "Failed to auth 3pid", Codes.THREEPID_AUTH_FAILED)
-
-        for reqd in ["medium", "address", "validated_at"]:
-            if reqd not in threepid:
-                logger.warn("Couldn't add 3pid: invalid response from ID server")
-                raise SynapseError(500, "Invalid response from ID Server")
-
-        yield self.auth_handler.add_threepid(
-            user_id, threepid["medium"], threepid["address"], threepid["validated_at"]
+        yield self.identity_handler.bind_threepid(
+            client_secret, sid, user_id, id_server, id_access_token
         )
 
-        if "bind" in body and body["bind"]:
-            logger.debug("Binding threepid %s to %s", threepid, user_id)
-            yield self.identity_handler.bind_threepid(threePidCreds, user_id)
-
         return 200, {}
 
 
+class ThreepidUnbindRestServlet(RestServlet):
+    PATTERNS = client_patterns("/account/3pid/unbind$", releases=(), unstable=True)
+
+    def __init__(self, hs):
+        super(ThreepidUnbindRestServlet, self).__init__()
+        self.hs = hs
+        self.identity_handler = hs.get_handlers().identity_handler
+        self.auth = hs.get_auth()
+        self.datastore = self.hs.get_datastore()
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        """Unbind the given 3pid from a specific identity server, or identity servers that are
+        known to have this 3pid bound
+        """
+        requester = yield self.auth.get_user_by_req(request)
+        body = parse_json_object_from_request(request)
+        assert_params_in_dict(body, ["medium", "address"])
+
+        medium = body.get("medium")
+        address = body.get("address")
+        id_server = body.get("id_server")
+
+        # Attempt to unbind the threepid from an identity server. If id_server is None, try to
+        # unbind from all identity servers this threepid has been added to in the past
+        result = yield self.identity_handler.try_unbind_threepid(
+            requester.user.to_string(),
+            {"address": address, "medium": medium, "id_server": id_server},
+        )
+        return 200, {"id_server_unbind_result": "success" if result else "no-support"}
+
+
 class ThreepidDeleteRestServlet(RestServlet):
     PATTERNS = client_patterns("/account/3pid/delete$")
 
@@ -622,12 +781,16 @@ class WhoamiRestServlet(RestServlet):
 
 def register_servlets(hs, http_server):
     EmailPasswordRequestTokenRestServlet(hs).register(http_server)
-    MsisdnPasswordRequestTokenRestServlet(hs).register(http_server)
     PasswordResetSubmitTokenServlet(hs).register(http_server)
     PasswordRestServlet(hs).register(http_server)
     DeactivateAccountRestServlet(hs).register(http_server)
     EmailThreepidRequestTokenRestServlet(hs).register(http_server)
     MsisdnThreepidRequestTokenRestServlet(hs).register(http_server)
+    AddThreepidEmailSubmitTokenServlet(hs).register(http_server)
+    AddThreepidMsisdnSubmitTokenServlet(hs).register(http_server)
     ThreepidRestServlet(hs).register(http_server)
+    ThreepidAddRestServlet(hs).register(http_server)
+    ThreepidBindRestServlet(hs).register(http_server)
+    ThreepidUnbindRestServlet(hs).register(http_server)
     ThreepidDeleteRestServlet(hs).register(http_server)
     WhoamiRestServlet(hs).register(http_server)
diff --git a/synapse/rest/client/v2_alpha/filter.py b/synapse/rest/client/v2_alpha/filter.py
index c6ddf24c8d..17a8bc7366 100644
--- a/synapse/rest/client/v2_alpha/filter.py
+++ b/synapse/rest/client/v2_alpha/filter.py
@@ -17,7 +17,7 @@ import logging
 
 from twisted.internet import defer
 
-from synapse.api.errors import AuthError, Codes, StoreError, SynapseError
+from synapse.api.errors import AuthError, NotFoundError, StoreError, SynapseError
 from synapse.http.servlet import RestServlet, parse_json_object_from_request
 from synapse.types import UserID
 
@@ -52,13 +52,15 @@ class GetFilterRestServlet(RestServlet):
             raise SynapseError(400, "Invalid filter_id")
 
         try:
-            filter = yield self.filtering.get_user_filter(
+            filter_collection = yield self.filtering.get_user_filter(
                 user_localpart=target_user.localpart, filter_id=filter_id
             )
+        except StoreError as e:
+            if e.code != 404:
+                raise
+            raise NotFoundError("No such filter")
 
-            return 200, filter.get_filter_json()
-        except (KeyError, StoreError):
-            raise SynapseError(400, "No such filter", errcode=Codes.NOT_FOUND)
+        return 200, filter_collection.get_filter_json()
 
 
 class CreateFilterRestServlet(RestServlet):
diff --git a/synapse/rest/client/v2_alpha/keys.py b/synapse/rest/client/v2_alpha/keys.py
index 64b6898eb8..341567ae21 100644
--- a/synapse/rest/client/v2_alpha/keys.py
+++ b/synapse/rest/client/v2_alpha/keys.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2015, 2016 OpenMarket Ltd
+# Copyright 2019 New Vector Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -24,10 +25,10 @@ from synapse.http.servlet import (
     parse_json_object_from_request,
     parse_string,
 )
-from synapse.logging.opentracing import log_kv, set_tag, trace_using_operation_name
+from synapse.logging.opentracing import log_kv, set_tag, trace
 from synapse.types import StreamToken
 
-from ._base import client_patterns
+from ._base import client_patterns, interactive_auth_handler
 
 logger = logging.getLogger(__name__)
 
@@ -69,7 +70,7 @@ class KeyUploadServlet(RestServlet):
         self.auth = hs.get_auth()
         self.e2e_keys_handler = hs.get_e2e_keys_handler()
 
-    @trace_using_operation_name("upload_keys")
+    @trace(opname="upload_keys")
     @defer.inlineCallbacks
     def on_POST(self, request, device_id):
         requester = yield self.auth.get_user_by_req(request, allow_guest=True)
@@ -155,10 +156,11 @@ class KeyQueryServlet(RestServlet):
 
     @defer.inlineCallbacks
     def on_POST(self, request):
-        yield self.auth.get_user_by_req(request, allow_guest=True)
+        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+        user_id = requester.user.to_string()
         timeout = parse_integer(request, "timeout", 10 * 1000)
         body = parse_json_object_from_request(request)
-        result = yield self.e2e_keys_handler.query_devices(body, timeout)
+        result = yield self.e2e_keys_handler.query_devices(body, timeout, user_id)
         return 200, result
 
 
@@ -238,8 +240,97 @@ class OneTimeKeyServlet(RestServlet):
         return 200, result
 
 
+class SigningKeyUploadServlet(RestServlet):
+    """
+    POST /keys/device_signing/upload HTTP/1.1
+    Content-Type: application/json
+
+    {
+    }
+    """
+
+    PATTERNS = client_patterns("/keys/device_signing/upload$", releases=())
+
+    def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer): server
+        """
+        super(SigningKeyUploadServlet, self).__init__()
+        self.hs = hs
+        self.auth = hs.get_auth()
+        self.e2e_keys_handler = hs.get_e2e_keys_handler()
+        self.auth_handler = hs.get_auth_handler()
+
+    @interactive_auth_handler
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        requester = yield self.auth.get_user_by_req(request)
+        user_id = requester.user.to_string()
+        body = parse_json_object_from_request(request)
+
+        yield self.auth_handler.validate_user_via_ui_auth(
+            requester, body, self.hs.get_ip_from_request(request)
+        )
+
+        result = yield self.e2e_keys_handler.upload_signing_keys_for_user(user_id, body)
+        return 200, result
+
+
+class SignaturesUploadServlet(RestServlet):
+    """
+    POST /keys/signatures/upload HTTP/1.1
+    Content-Type: application/json
+
+    {
+      "@alice:example.com": {
+        "": {
+          "user_id": "",
+          "device_id": "",
+          "algorithms": [
+            "m.olm.curve25519-aes-sha256",
+            "m.megolm.v1.aes-sha"
+          ],
+          "keys": {
+            ":": "",
+          },
+          "signatures": {
+            "": {
+              ":": ">"
+            }
+          }
+        }
+      }
+    }
+    """
+
+    PATTERNS = client_patterns("/keys/signatures/upload$")
+
+    def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer): server
+        """
+        super(SignaturesUploadServlet, self).__init__()
+        self.auth = hs.get_auth()
+        self.e2e_keys_handler = hs.get_e2e_keys_handler()
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        requester = yield self.auth.get_user_by_req(request, allow_guest=True)
+        user_id = requester.user.to_string()
+        body = parse_json_object_from_request(request)
+
+        result = yield self.e2e_keys_handler.upload_signatures_for_device_keys(
+            user_id, body
+        )
+        return 200, result
+
+
 def register_servlets(hs, http_server):
     KeyUploadServlet(hs).register(http_server)
     KeyQueryServlet(hs).register(http_server)
     KeyChangesServlet(hs).register(http_server)
     OneTimeKeyServlet(hs).register(http_server)
+    SigningKeyUploadServlet(hs).register(http_server)
+    SignaturesUploadServlet(hs).register(http_server)
diff --git a/synapse/rest/client/v2_alpha/read_marker.py b/synapse/rest/client/v2_alpha/read_marker.py
index b3bf8567e1..67cbc37312 100644
--- a/synapse/rest/client/v2_alpha/read_marker.py
+++ b/synapse/rest/client/v2_alpha/read_marker.py
@@ -15,8 +15,6 @@
 
 import logging
 
-from twisted.internet import defer
-
 from synapse.http.servlet import RestServlet, parse_json_object_from_request
 
 from ._base import client_patterns
@@ -34,17 +32,16 @@ class ReadMarkerRestServlet(RestServlet):
         self.read_marker_handler = hs.get_read_marker_handler()
         self.presence_handler = hs.get_presence_handler()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_POST(self, request, room_id):
+        requester = await self.auth.get_user_by_req(request)
 
-        yield self.presence_handler.bump_presence_active_time(requester.user)
+        await self.presence_handler.bump_presence_active_time(requester.user)
 
         body = parse_json_object_from_request(request)
 
         read_event_id = body.get("m.read", None)
         if read_event_id:
-            yield self.receipts_handler.received_client_receipt(
+            await self.receipts_handler.received_client_receipt(
                 room_id,
                 "m.read",
                 user_id=requester.user.to_string(),
@@ -53,7 +50,7 @@ class ReadMarkerRestServlet(RestServlet):
 
         read_marker_event_id = body.get("m.fully_read", None)
         if read_marker_event_id:
-            yield self.read_marker_handler.received_client_read_marker(
+            await self.read_marker_handler.received_client_read_marker(
                 room_id,
                 user_id=requester.user.to_string(),
                 event_id=read_marker_event_id,
diff --git a/synapse/rest/client/v2_alpha/receipts.py b/synapse/rest/client/v2_alpha/receipts.py
index 0dab03d227..92555bd4a9 100644
--- a/synapse/rest/client/v2_alpha/receipts.py
+++ b/synapse/rest/client/v2_alpha/receipts.py
@@ -15,8 +15,6 @@
 
 import logging
 
-from twisted.internet import defer
-
 from synapse.api.errors import SynapseError
 from synapse.http.servlet import RestServlet
 
@@ -39,16 +37,15 @@ class ReceiptRestServlet(RestServlet):
         self.receipts_handler = hs.get_receipts_handler()
         self.presence_handler = hs.get_presence_handler()
 
-    @defer.inlineCallbacks
-    def on_POST(self, request, room_id, receipt_type, event_id):
-        requester = yield self.auth.get_user_by_req(request)
+    async def on_POST(self, request, room_id, receipt_type, event_id):
+        requester = await self.auth.get_user_by_req(request)
 
         if receipt_type != "m.read":
             raise SynapseError(400, "Receipt type must be 'm.read'")
 
-        yield self.presence_handler.bump_presence_active_time(requester.user)
+        await self.presence_handler.bump_presence_active_time(requester.user)
 
-        yield self.receipts_handler.received_client_receipt(
+        await self.receipts_handler.received_client_receipt(
             room_id, receipt_type, user_id=requester.user.to_string(), event_id=event_id
         )
 
diff --git a/synapse/rest/client/v2_alpha/register.py b/synapse/rest/client/v2_alpha/register.py
index 65f9fce2ff..91db923814 100644
--- a/synapse/rest/client/v2_alpha/register.py
+++ b/synapse/rest/client/v2_alpha/register.py
@@ -16,6 +16,7 @@
 
 import hmac
 import logging
+from typing import List, Union
 
 from six import string_types
 
@@ -28,16 +29,25 @@ from synapse.api.errors import (
     Codes,
     LimitExceededError,
     SynapseError,
+    ThreepidValidationError,
     UnrecognizedRequestError,
 )
+from synapse.config import ConfigError
+from synapse.config.captcha import CaptchaConfig
+from synapse.config.consent_config import ConsentConfig
+from synapse.config.emailconfig import ThreepidBehaviour
 from synapse.config.ratelimiting import FederationRateLimitConfig
+from synapse.config.registration import RegistrationConfig
 from synapse.config.server import is_threepid_reserved
+from synapse.handlers.auth import AuthHandler
+from synapse.http.server import finish_request
 from synapse.http.servlet import (
     RestServlet,
     assert_params_in_dict,
     parse_json_object_from_request,
     parse_string,
 )
+from synapse.push.mailer import load_jinja2_templates
 from synapse.util.msisdn import phone_number_to_msisdn
 from synapse.util.ratelimitutils import FederationRateLimiter
 from synapse.util.threepids import check_3pid_allowed
@@ -70,30 +80,86 @@ class EmailRegisterRequestTokenRestServlet(RestServlet):
         super(EmailRegisterRequestTokenRestServlet, self).__init__()
         self.hs = hs
         self.identity_handler = hs.get_handlers().identity_handler
+        self.config = hs.config
+
+        if self.hs.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            from synapse.push.mailer import Mailer, load_jinja2_templates
+
+            template_html, template_text = load_jinja2_templates(
+                self.config.email_template_dir,
+                [
+                    self.config.email_registration_template_html,
+                    self.config.email_registration_template_text,
+                ],
+                apply_format_ts_filter=True,
+                apply_mxc_to_http_filter=True,
+                public_baseurl=self.config.public_baseurl,
+            )
+            self.mailer = Mailer(
+                hs=self.hs,
+                app_name=self.config.email_app_name,
+                template_html=template_html,
+                template_text=template_text,
+            )
 
     @defer.inlineCallbacks
     def on_POST(self, request):
+        if self.hs.config.threepid_behaviour_email == ThreepidBehaviour.OFF:
+            if self.hs.config.local_threepid_handling_disabled_due_to_email_config:
+                logger.warning(
+                    "Email registration has been disabled due to lack of email config"
+                )
+            raise SynapseError(
+                400, "Email-based registration has been disabled on this server"
+            )
         body = parse_json_object_from_request(request)
 
-        assert_params_in_dict(
-            body, ["id_server", "client_secret", "email", "send_attempt"]
-        )
+        assert_params_in_dict(body, ["client_secret", "email", "send_attempt"])
 
-        if not check_3pid_allowed(self.hs, "email", body["email"]):
+        # Extract params from body
+        client_secret = body["client_secret"]
+        email = body["email"]
+        send_attempt = body["send_attempt"]
+        next_link = body.get("next_link")  # Optional param
+
+        if not check_3pid_allowed(self.hs, "email", email):
             raise SynapseError(
                 403,
                 "Your email domain is not authorized to register on this server",
                 Codes.THREEPID_DENIED,
             )
 
-        existingUid = yield self.hs.get_datastore().get_user_id_by_threepid(
+        existing_user_id = yield self.hs.get_datastore().get_user_id_by_threepid(
             "email", body["email"]
         )
 
-        if existingUid is not None:
+        if existing_user_id is not None:
             raise SynapseError(400, "Email is already in use", Codes.THREEPID_IN_USE)
 
-        ret = yield self.identity_handler.requestEmailToken(**body)
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.REMOTE:
+            assert self.hs.config.account_threepid_delegate_email
+
+            # Have the configured identity server handle the request
+            ret = yield self.identity_handler.requestEmailToken(
+                self.hs.config.account_threepid_delegate_email,
+                email,
+                client_secret,
+                send_attempt,
+                next_link,
+            )
+        else:
+            # Send registration emails from Synapse
+            sid = yield self.identity_handler.send_threepid_validation(
+                email,
+                client_secret,
+                send_attempt,
+                self.mailer.send_registration_mail,
+                next_link,
+            )
+
+            # Wrap the session id in a JSON object
+            ret = {"sid": sid}
+
         return 200, ret
 
 
@@ -114,11 +180,15 @@ class MsisdnRegisterRequestTokenRestServlet(RestServlet):
         body = parse_json_object_from_request(request)
 
         assert_params_in_dict(
-            body,
-            ["id_server", "client_secret", "country", "phone_number", "send_attempt"],
+            body, ["client_secret", "country", "phone_number", "send_attempt"]
         )
+        client_secret = body["client_secret"]
+        country = body["country"]
+        phone_number = body["phone_number"]
+        send_attempt = body["send_attempt"]
+        next_link = body.get("next_link")  # Optional param
 
-        msisdn = phone_number_to_msisdn(body["country"], body["phone_number"])
+        msisdn = phone_number_to_msisdn(country, phone_number)
 
         if not check_3pid_allowed(self.hs, "msisdn", msisdn):
             raise SynapseError(
@@ -127,19 +197,120 @@ class MsisdnRegisterRequestTokenRestServlet(RestServlet):
                 Codes.THREEPID_DENIED,
             )
 
-        existingUid = yield self.hs.get_datastore().get_user_id_by_threepid(
+        existing_user_id = yield self.hs.get_datastore().get_user_id_by_threepid(
             "msisdn", msisdn
         )
 
-        if existingUid is not None:
+        if existing_user_id is not None:
             raise SynapseError(
                 400, "Phone number is already in use", Codes.THREEPID_IN_USE
             )
 
-        ret = yield self.identity_handler.requestMsisdnToken(**body)
+        if not self.hs.config.account_threepid_delegate_msisdn:
+            logger.warning(
+                "No upstream msisdn account_threepid_delegate configured on the server to "
+                "handle this request"
+            )
+            raise SynapseError(
+                400, "Registration by phone number is not supported on this homeserver"
+            )
+
+        ret = yield self.identity_handler.requestMsisdnToken(
+            self.hs.config.account_threepid_delegate_msisdn,
+            country,
+            phone_number,
+            client_secret,
+            send_attempt,
+            next_link,
+        )
+
         return 200, ret
 
 
+class RegistrationSubmitTokenServlet(RestServlet):
+    """Handles registration 3PID validation token submission"""
+
+    PATTERNS = client_patterns(
+        "/registration/(?P[^/]*)/submit_token$", releases=(), unstable=True
+    )
+
+    def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer): server
+        """
+        super(RegistrationSubmitTokenServlet, self).__init__()
+        self.hs = hs
+        self.auth = hs.get_auth()
+        self.config = hs.config
+        self.clock = hs.get_clock()
+        self.store = hs.get_datastore()
+
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            (self.failure_email_template,) = load_jinja2_templates(
+                self.config.email_template_dir,
+                [self.config.email_registration_template_failure_html],
+            )
+
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.LOCAL:
+            (self.failure_email_template,) = load_jinja2_templates(
+                self.config.email_template_dir,
+                [self.config.email_registration_template_failure_html],
+            )
+
+    @defer.inlineCallbacks
+    def on_GET(self, request, medium):
+        if medium != "email":
+            raise SynapseError(
+                400, "This medium is currently not supported for registration"
+            )
+        if self.config.threepid_behaviour_email == ThreepidBehaviour.OFF:
+            if self.config.local_threepid_handling_disabled_due_to_email_config:
+                logger.warning(
+                    "User registration via email has been disabled due to lack of email config"
+                )
+            raise SynapseError(
+                400, "Email-based registration is disabled on this server"
+            )
+
+        sid = parse_string(request, "sid", required=True)
+        client_secret = parse_string(request, "client_secret", required=True)
+        token = parse_string(request, "token", required=True)
+
+        # Attempt to validate a 3PID session
+        try:
+            # Mark the session as valid
+            next_link = yield self.store.validate_threepid_session(
+                sid, client_secret, token, self.clock.time_msec()
+            )
+
+            # Perform a 302 redirect if next_link is set
+            if next_link:
+                if next_link.startswith("file:///"):
+                    logger.warning(
+                        "Not redirecting to next_link as it is a local file: address"
+                    )
+                else:
+                    request.setResponseCode(302)
+                    request.setHeader("Location", next_link)
+                    finish_request(request)
+                    return None
+
+            # Otherwise show the success template
+            html = self.config.email_registration_template_success_html_content
+
+            request.setResponseCode(200)
+        except ThreepidValidationError as e:
+            request.setResponseCode(e.code)
+
+            # Show a failure page with a reason
+            template_vars = {"failure_reason": e.msg}
+            html = self.failure_email_template.render(**template_vars)
+
+        request.write(html.encode("utf-8"))
+        finish_request(request)
+
+
 class UsernameAvailabilityRestServlet(RestServlet):
     PATTERNS = client_patterns("/register/available")
 
@@ -169,6 +340,11 @@ class UsernameAvailabilityRestServlet(RestServlet):
 
     @defer.inlineCallbacks
     def on_GET(self, request):
+        if not self.hs.config.enable_registration:
+            raise SynapseError(
+                403, "Registration has been disabled", errcode=Codes.FORBIDDEN
+            )
+
         ip = self.hs.get_ip_from_request(request)
         with self.ratelimiter.ratelimit(ip) as wait_deferred:
             yield wait_deferred
@@ -201,6 +377,10 @@ class RegisterRestServlet(RestServlet):
         self.ratelimiter = hs.get_registration_ratelimiter()
         self.clock = hs.get_clock()
 
+        self._registration_flows = _calculate_registration_flows(
+            hs.config, self.auth_handler
+        )
+
     @interactive_auth_handler
     @defer.inlineCallbacks
     def on_POST(self, request):
@@ -230,7 +410,6 @@ class RegisterRestServlet(RestServlet):
         if kind == b"guest":
             ret = yield self._do_guest_registration(body, address=client_addr)
             return ret
-            return
         elif kind != b"user":
             raise UnrecognizedRequestError(
                 "Do not understand membership kind: %s" % (kind,)
@@ -280,7 +459,6 @@ class RegisterRestServlet(RestServlet):
                     desired_username, access_token, body
                 )
             return 200, result  # we throw for non 200 responses
-            return
 
         # for regular registration, downcase the provided username before
         # attempting to register it. This should mean
@@ -302,7 +480,7 @@ class RegisterRestServlet(RestServlet):
             # a password to work around a client bug where it sent
             # the 'initial_device_display_name' param alone, wiping out
             # the original registration params
-            logger.warn("Ignoring initial_device_display_name without password")
+            logger.warning("Ignoring initial_device_display_name without password")
             del body["initial_device_display_name"]
 
         session_id = self.auth_handler.get_session_id(body)
@@ -323,69 +501,8 @@ class RegisterRestServlet(RestServlet):
                 assigned_user_id=registered_user_id,
             )
 
-        # FIXME: need a better error than "no auth flow found" for scenarios
-        # where we required 3PID for registration but the user didn't give one
-        require_email = "email" in self.hs.config.registrations_require_3pid
-        require_msisdn = "msisdn" in self.hs.config.registrations_require_3pid
-
-        show_msisdn = True
-        if self.hs.config.disable_msisdn_registration:
-            show_msisdn = False
-            require_msisdn = False
-
-        flows = []
-        if self.hs.config.enable_registration_captcha:
-            # only support 3PIDless registration if no 3PIDs are required
-            if not require_email and not require_msisdn:
-                # Also add a dummy flow here, otherwise if a client completes
-                # recaptcha first we'll assume they were going for this flow
-                # and complete the request, when they could have been trying to
-                # complete one of the flows with email/msisdn auth.
-                flows.extend([[LoginType.RECAPTCHA, LoginType.DUMMY]])
-            # only support the email-only flow if we don't require MSISDN 3PIDs
-            if not require_msisdn:
-                flows.extend([[LoginType.RECAPTCHA, LoginType.EMAIL_IDENTITY]])
-
-            if show_msisdn:
-                # only support the MSISDN-only flow if we don't require email 3PIDs
-                if not require_email:
-                    flows.extend([[LoginType.RECAPTCHA, LoginType.MSISDN]])
-                # always let users provide both MSISDN & email
-                flows.extend(
-                    [[LoginType.RECAPTCHA, LoginType.MSISDN, LoginType.EMAIL_IDENTITY]]
-                )
-        else:
-            # only support 3PIDless registration if no 3PIDs are required
-            if not require_email and not require_msisdn:
-                flows.extend([[LoginType.DUMMY]])
-            # only support the email-only flow if we don't require MSISDN 3PIDs
-            if not require_msisdn:
-                flows.extend([[LoginType.EMAIL_IDENTITY]])
-
-            if show_msisdn:
-                # only support the MSISDN-only flow if we don't require email 3PIDs
-                if not require_email or require_msisdn:
-                    flows.extend([[LoginType.MSISDN]])
-                # always let users provide both MSISDN & email
-                flows.extend([[LoginType.MSISDN, LoginType.EMAIL_IDENTITY]])
-
-        # Append m.login.terms to all flows if we're requiring consent
-        if self.hs.config.user_consent_at_registration:
-            new_flows = []
-            for flow in flows:
-                inserted = False
-                # m.login.terms should go near the end but before msisdn or email auth
-                for i, stage in enumerate(flow):
-                    if stage == LoginType.EMAIL_IDENTITY or stage == LoginType.MSISDN:
-                        flow.insert(i, LoginType.TERMS)
-                        inserted = True
-                        break
-                if not inserted:
-                    flow.append(LoginType.TERMS)
-            flows.extend(new_flows)
-
         auth_result, params, session_id = yield self.auth_handler.check_auth(
-            flows, body, self.hs.get_ip_from_request(request)
+            self._registration_flows, body, self.hs.get_ip_from_request(request)
         )
 
         # Check that we're not trying to register a denied 3pid.
@@ -440,11 +557,11 @@ class RegisterRestServlet(RestServlet):
                         medium = auth_result[login_type]["medium"]
                         address = auth_result[login_type]["address"]
 
-                        existingUid = yield self.store.get_user_id_by_threepid(
+                        existing_user_id = yield self.store.get_user_id_by_threepid(
                             medium, address
                         )
 
-                        if existingUid is not None:
+                        if existing_user_id is not None:
                             raise SynapseError(
                                 400,
                                 "%s is already in use" % medium,
@@ -483,8 +600,6 @@ class RegisterRestServlet(RestServlet):
                 user_id=registered_user_id,
                 auth_result=auth_result,
                 access_token=return_dict.get("access_token"),
-                bind_email=params.get("bind_email"),
-                bind_msisdn=params.get("bind_msisdn"),
             )
 
         return 200, return_dict
@@ -550,8 +665,86 @@ class RegisterRestServlet(RestServlet):
         )
 
 
+def _calculate_registration_flows(
+    # technically `config` has to provide *all* of these interfaces, not just one
+    config: Union[RegistrationConfig, ConsentConfig, CaptchaConfig],
+    auth_handler: AuthHandler,
+) -> List[List[str]]:
+    """Get a suitable flows list for registration
+
+    Args:
+        config: server configuration
+        auth_handler: authorization handler
+
+    Returns: a list of supported flows
+    """
+    # FIXME: need a better error than "no auth flow found" for scenarios
+    # where we required 3PID for registration but the user didn't give one
+    require_email = "email" in config.registrations_require_3pid
+    require_msisdn = "msisdn" in config.registrations_require_3pid
+
+    show_msisdn = True
+    show_email = True
+
+    if config.disable_msisdn_registration:
+        show_msisdn = False
+        require_msisdn = False
+
+    enabled_auth_types = auth_handler.get_enabled_auth_types()
+    if LoginType.EMAIL_IDENTITY not in enabled_auth_types:
+        show_email = False
+        if require_email:
+            raise ConfigError(
+                "Configuration requires email address at registration, but email "
+                "validation is not configured"
+            )
+
+    if LoginType.MSISDN not in enabled_auth_types:
+        show_msisdn = False
+        if require_msisdn:
+            raise ConfigError(
+                "Configuration requires msisdn at registration, but msisdn "
+                "validation is not configured"
+            )
+
+    flows = []
+
+    # only support 3PIDless registration if no 3PIDs are required
+    if not require_email and not require_msisdn:
+        # Add a dummy step here, otherwise if a client completes
+        # recaptcha first we'll assume they were going for this flow
+        # and complete the request, when they could have been trying to
+        # complete one of the flows with email/msisdn auth.
+        flows.append([LoginType.DUMMY])
+
+    # only support the email-only flow if we don't require MSISDN 3PIDs
+    if show_email and not require_msisdn:
+        flows.append([LoginType.EMAIL_IDENTITY])
+
+    # only support the MSISDN-only flow if we don't require email 3PIDs
+    if show_msisdn and not require_email:
+        flows.append([LoginType.MSISDN])
+
+    if show_email and show_msisdn:
+        # always let users provide both MSISDN & email
+        flows.append([LoginType.MSISDN, LoginType.EMAIL_IDENTITY])
+
+    # Prepend m.login.terms to all flows if we're requiring consent
+    if config.user_consent_at_registration:
+        for flow in flows:
+            flow.insert(0, LoginType.TERMS)
+
+    # Prepend recaptcha to all flows if we're requiring captcha
+    if config.enable_registration_captcha:
+        for flow in flows:
+            flow.insert(0, LoginType.RECAPTCHA)
+
+    return flows
+
+
 def register_servlets(hs, http_server):
     EmailRegisterRequestTokenRestServlet(hs).register(http_server)
     MsisdnRegisterRequestTokenRestServlet(hs).register(http_server)
     UsernameAvailabilityRestServlet(hs).register(http_server)
+    RegistrationSubmitTokenServlet(hs).register(http_server)
     RegisterRestServlet(hs).register(http_server)
diff --git a/synapse/rest/client/v2_alpha/room_keys.py b/synapse/rest/client/v2_alpha/room_keys.py
index c39a1453c8..d83ac8e3c5 100644
--- a/synapse/rest/client/v2_alpha/room_keys.py
+++ b/synapse/rest/client/v2_alpha/room_keys.py
@@ -375,7 +375,7 @@ class RoomKeysVersionServlet(RestServlet):
                     "ed25519:something": "hijklmnop"
                 }
             },
-            "version": "42"
+            "version": "12345"
         }
 
         HTTP/1.1 200 OK
diff --git a/synapse/rest/client/v2_alpha/sendtodevice.py b/synapse/rest/client/v2_alpha/sendtodevice.py
index 2613648d82..d90e52ed1a 100644
--- a/synapse/rest/client/v2_alpha/sendtodevice.py
+++ b/synapse/rest/client/v2_alpha/sendtodevice.py
@@ -19,6 +19,7 @@ from twisted.internet import defer
 
 from synapse.http import servlet
 from synapse.http.servlet import parse_json_object_from_request
+from synapse.logging.opentracing import set_tag, trace
 from synapse.rest.client.transactions import HttpTransactionCache
 
 from ._base import client_patterns
@@ -42,7 +43,10 @@ class SendToDeviceRestServlet(servlet.RestServlet):
         self.txns = HttpTransactionCache(hs)
         self.device_message_handler = hs.get_device_message_handler()
 
+    @trace(opname="sendToDevice")
     def on_PUT(self, request, message_type, txn_id):
+        set_tag("message_type", message_type)
+        set_tag("txn_id", txn_id)
         return self.txns.fetch_or_execute_request(
             request, self._put, request, message_type, txn_id
         )
diff --git a/synapse/rest/client/v2_alpha/sync.py b/synapse/rest/client/v2_alpha/sync.py
index c98c5a3802..ccd8b17b23 100644
--- a/synapse/rest/client/v2_alpha/sync.py
+++ b/synapse/rest/client/v2_alpha/sync.py
@@ -21,7 +21,7 @@ from canonicaljson import json
 from twisted.internet import defer
 
 from synapse.api.constants import PresenceState
-from synapse.api.errors import SynapseError
+from synapse.api.errors import Codes, StoreError, SynapseError
 from synapse.api.filtering import DEFAULT_FILTER_COLLECTION, FilterCollection
 from synapse.events.utils import (
     format_event_for_client_v2_without_room_id,
@@ -112,32 +112,44 @@ class SyncRestServlet(RestServlet):
         full_state = parse_boolean(request, "full_state", default=False)
 
         logger.debug(
-            "/sync: user=%r, timeout=%r, since=%r,"
-            " set_presence=%r, filter_id=%r, device_id=%r"
-            % (user, timeout, since, set_presence, filter_id, device_id)
+            "/sync: user=%r, timeout=%r, since=%r, "
+            "set_presence=%r, filter_id=%r, device_id=%r",
+            user,
+            timeout,
+            since,
+            set_presence,
+            filter_id,
+            device_id,
         )
 
         request_key = (user, timeout, since, filter_id, full_state, device_id)
 
-        if filter_id:
-            if filter_id.startswith("{"):
-                try:
-                    filter_object = json.loads(filter_id)
-                    set_timeline_upper_limit(
-                        filter_object, self.hs.config.filter_timeline_limit
-                    )
-                except Exception:
-                    raise SynapseError(400, "Invalid filter JSON")
-                self.filtering.check_valid_filter(filter_object)
-                filter = FilterCollection(filter_object)
-            else:
-                filter = yield self.filtering.get_user_filter(user.localpart, filter_id)
+        if filter_id is None:
+            filter_collection = DEFAULT_FILTER_COLLECTION
+        elif filter_id.startswith("{"):
+            try:
+                filter_object = json.loads(filter_id)
+                set_timeline_upper_limit(
+                    filter_object, self.hs.config.filter_timeline_limit
+                )
+            except Exception:
+                raise SynapseError(400, "Invalid filter JSON")
+            self.filtering.check_valid_filter(filter_object)
+            filter_collection = FilterCollection(filter_object)
         else:
-            filter = DEFAULT_FILTER_COLLECTION
+            try:
+                filter_collection = yield self.filtering.get_user_filter(
+                    user.localpart, filter_id
+                )
+            except StoreError as err:
+                if err.code != 404:
+                    raise
+                # fix up the description and errcode to be more useful
+                raise SynapseError(400, "No such filter", errcode=Codes.INVALID_PARAM)
 
         sync_config = SyncConfig(
             user=user,
-            filter_collection=filter,
+            filter_collection=filter_collection,
             is_guest=requester.is_guest,
             request_key=request_key,
             device_id=device_id,
@@ -171,7 +183,7 @@ class SyncRestServlet(RestServlet):
 
         time_now = self.clock.time_msec()
         response_content = yield self.encode_response(
-            time_now, sync_result, requester.access_token_id, filter
+            time_now, sync_result, requester.access_token_id, filter_collection
         )
 
         return 200, response_content
@@ -382,7 +394,7 @@ class SyncRestServlet(RestServlet):
             # We've had bug reports that events were coming down under the
             # wrong room.
             if event.room_id != room.room_id:
-                logger.warn(
+                logger.warning(
                     "Event %r is under room %r instead of %r",
                     event.event_id,
                     room.room_id,
diff --git a/synapse/rest/client/versions.py b/synapse/rest/client/versions.py
index 0e09191632..bb30ce3f34 100644
--- a/synapse/rest/client/versions.py
+++ b/synapse/rest/client/versions.py
@@ -24,6 +24,10 @@ logger = logging.getLogger(__name__)
 class VersionsRestServlet(RestServlet):
     PATTERNS = [re.compile("^/_matrix/client/versions$")]
 
+    def __init__(self, hs):
+        super(VersionsRestServlet, self).__init__()
+        self.config = hs.config
+
     def on_GET(self, request):
         return (
             200,
@@ -44,10 +48,30 @@ class VersionsRestServlet(RestServlet):
                     "r0.5.0",
                 ],
                 # as per MSC1497:
-                "unstable_features": {"m.lazy_load_members": True},
+                "unstable_features": {
+                    "m.lazy_load_members": True,
+                    # as per MSC2190, as amended by MSC2264
+                    # to be removed in r0.6.0
+                    "m.id_access_token": True,
+                    # Advertise to clients that they need not include an `id_server`
+                    # parameter during registration or password reset, as Synapse now decides
+                    # itself which identity server to use (or none at all).
+                    #
+                    # This is also used by a client when they wish to bind a 3PID to their
+                    # account, but not bind it to an identity server, the endpoint for which
+                    # also requires `id_server`. If the homeserver is handling 3PID
+                    # verification itself, there is no need to ask the user for `id_server` to
+                    # be supplied.
+                    "m.require_identity_server": False,
+                    # as per MSC2290
+                    "m.separate_add_and_bind": True,
+                    # Implements support for label-based filtering as described in
+                    # MSC2326.
+                    "org.matrix.label_based_filtering": True,
+                },
             },
         )
 
 
-def register_servlets(http_server):
-    VersionsRestServlet().register(http_server)
+def register_servlets(hs, http_server):
+    VersionsRestServlet(hs).register(http_server)
diff --git a/synapse/rest/key/v2/remote_key_resource.py b/synapse/rest/key/v2/remote_key_resource.py
index 55580bc59e..e7fc3f0431 100644
--- a/synapse/rest/key/v2/remote_key_resource.py
+++ b/synapse/rest/key/v2/remote_key_resource.py
@@ -102,7 +102,7 @@ class RemoteKey(DirectServeResource):
     @wrap_json_request_handler
     async def _async_render_GET(self, request):
         if len(request.postpath) == 1:
-            server, = request.postpath
+            (server,) = request.postpath
             query = {server.decode("ascii"): {}}
         elif len(request.postpath) == 2:
             server, key_id = request.postpath
diff --git a/synapse/rest/media/v1/_base.py b/synapse/rest/media/v1/_base.py
index 5fefee4dde..65bbf00073 100644
--- a/synapse/rest/media/v1/_base.py
+++ b/synapse/rest/media/v1/_base.py
@@ -195,7 +195,7 @@ def respond_with_responder(request, responder, media_type, file_size, upload_nam
         respond_404(request)
         return
 
-    logger.debug("Responding to media request with responder %s")
+    logger.debug("Responding to media request with responder %s", responder)
     add_file_headers(request, media_type, file_size, upload_name)
     try:
         with responder:
diff --git a/synapse/rest/media/v1/media_repository.py b/synapse/rest/media/v1/media_repository.py
index d4ea09260c..bd9186fe50 100644
--- a/synapse/rest/media/v1/media_repository.py
+++ b/synapse/rest/media/v1/media_repository.py
@@ -363,7 +363,7 @@ class MediaRepository(object):
                     },
                 )
             except RequestSendFailed as e:
-                logger.warn(
+                logger.warning(
                     "Request failed fetching remote media %s/%s: %r",
                     server_name,
                     media_id,
@@ -372,7 +372,7 @@ class MediaRepository(object):
                 raise SynapseError(502, "Failed to fetch remote media")
 
             except HttpResponseException as e:
-                logger.warn(
+                logger.warning(
                     "HTTP error fetching remote media %s/%s: %s",
                     server_name,
                     media_id,
@@ -383,10 +383,12 @@ class MediaRepository(object):
                 raise SynapseError(502, "Failed to fetch remote media")
 
             except SynapseError:
-                logger.warn("Failed to fetch remote media %s/%s", server_name, media_id)
+                logger.warning(
+                    "Failed to fetch remote media %s/%s", server_name, media_id
+                )
                 raise
             except NotRetryingDestination:
-                logger.warn("Not retrying destination %r", server_name)
+                logger.warning("Not retrying destination %r", server_name)
                 raise SynapseError(502, "Failed to fetch remote media")
             except Exception:
                 logger.exception(
@@ -526,7 +528,7 @@ class MediaRepository(object):
             try:
                 file_info = FileInfo(
                     server_name=server_name,
-                    file_id=media_id,
+                    file_id=file_id,
                     thumbnail=True,
                     thumbnail_width=t_width,
                     thumbnail_height=t_height,
@@ -691,7 +693,7 @@ class MediaRepository(object):
                 try:
                     os.remove(full_path)
                 except OSError as e:
-                    logger.warn("Failed to remove file: %r", full_path)
+                    logger.warning("Failed to remove file: %r", full_path)
                     if e.errno == errno.ENOENT:
                         pass
                     else:
diff --git a/synapse/rest/media/v1/preview_url_resource.py b/synapse/rest/media/v1/preview_url_resource.py
index bd40891a7f..15c15a12f5 100644
--- a/synapse/rest/media/v1/preview_url_resource.py
+++ b/synapse/rest/media/v1/preview_url_resource.py
@@ -56,6 +56,9 @@ logger = logging.getLogger(__name__)
 _charset_match = re.compile(br"<\s*meta[^>]*charset\s*=\s*([a-z0-9-]+)", flags=re.I)
 _content_type_match = re.compile(r'.*; *charset="?(.*?)"?(;|$)', flags=re.I)
 
+OG_TAG_NAME_MAXLEN = 50
+OG_TAG_VALUE_MAXLEN = 1000
+
 
 class PreviewUrlResource(DirectServeResource):
     isLeaf = True
@@ -74,6 +77,8 @@ class PreviewUrlResource(DirectServeResource):
             treq_args={"browser_like_redirects": True},
             ip_whitelist=hs.config.url_preview_ip_range_whitelist,
             ip_blacklist=hs.config.url_preview_ip_range_blacklist,
+            http_proxy=os.getenv("http_proxy"),
+            https_proxy=os.getenv("HTTPS_PROXY"),
         )
         self.media_repo = media_repo
         self.primary_base_path = media_repo.primary_base_path
@@ -117,8 +122,10 @@ class PreviewUrlResource(DirectServeResource):
                 pattern = entry[attrib]
                 value = getattr(url_tuple, attrib)
                 logger.debug(
-                    ("Matching attrib '%s' with value '%s' against" " pattern '%s'")
-                    % (attrib, value, pattern)
+                    "Matching attrib '%s' with value '%s' against" " pattern '%s'",
+                    attrib,
+                    value,
+                    pattern,
                 )
 
                 if value is None:
@@ -134,7 +141,7 @@ class PreviewUrlResource(DirectServeResource):
                         match = False
                         continue
             if match:
-                logger.warn("URL %s blocked by url_blacklist entry %s", url, entry)
+                logger.warning("URL %s blocked by url_blacklist entry %s", url, entry)
                 raise SynapseError(
                     403, "URL blocked by url pattern blacklist entry", Codes.UNKNOWN
                 )
@@ -167,7 +174,7 @@ class PreviewUrlResource(DirectServeResource):
             ts (int):
 
         Returns:
-            Deferred[str]: json-encoded og data
+            Deferred[bytes]: json-encoded og data
         """
         # check the URL cache in the DB (which will also provide us with
         # historical previews, if we have any)
@@ -183,11 +190,10 @@ class PreviewUrlResource(DirectServeResource):
             if isinstance(og, six.text_type):
                 og = og.encode("utf8")
             return og
-            return
 
         media_info = yield self._download_url(url, user)
 
-        logger.debug("got media_info of '%s'" % media_info)
+        logger.debug("got media_info of '%s'", media_info)
 
         if _is_media(media_info["media_type"]):
             file_id = media_info["filesystem_id"]
@@ -207,7 +213,7 @@ class PreviewUrlResource(DirectServeResource):
                 og["og:image:width"] = dims["width"]
                 og["og:image:height"] = dims["height"]
             else:
-                logger.warn("Couldn't get dims for %s" % url)
+                logger.warning("Couldn't get dims for %s" % url)
 
             # define our OG response for this media
         elif _is_html(media_info["media_type"]):
@@ -255,7 +261,7 @@ class PreviewUrlResource(DirectServeResource):
                         og["og:image:width"] = dims["width"]
                         og["og:image:height"] = dims["height"]
                     else:
-                        logger.warn("Couldn't get dims for %s" % og["og:image"])
+                        logger.warning("Couldn't get dims for %s", og["og:image"])
 
                     og["og:image"] = "mxc://%s/%s" % (
                         self.server_name,
@@ -266,12 +272,24 @@ class PreviewUrlResource(DirectServeResource):
                 else:
                     del og["og:image"]
         else:
-            logger.warn("Failed to find any OG data in %s", url)
+            logger.warning("Failed to find any OG data in %s", url)
             og = {}
 
-        logger.debug("Calculated OG for %s as %s" % (url, og))
+        # filter out any stupidly long values
+        keys_to_remove = []
+        for k, v in og.items():
+            # values can be numeric as well as strings, hence the cast to str
+            if len(k) > OG_TAG_NAME_MAXLEN or len(str(v)) > OG_TAG_VALUE_MAXLEN:
+                logger.warning(
+                    "Pruning overlong tag %s from OG data", k[:OG_TAG_NAME_MAXLEN]
+                )
+                keys_to_remove.append(k)
+        for k in keys_to_remove:
+            del og[k]
 
-        jsonog = json.dumps(og).encode("utf8")
+        logger.debug("Calculated OG for %s as %s", url, og)
+
+        jsonog = json.dumps(og)
 
         # store OG in history-aware DB cache
         yield self.store.store_url_cache(
@@ -284,7 +302,7 @@ class PreviewUrlResource(DirectServeResource):
             media_info["created_ts"],
         )
 
-        return jsonog
+        return jsonog.encode("utf8")
 
     @defer.inlineCallbacks
     def _download_url(self, url, user):
@@ -298,7 +316,7 @@ class PreviewUrlResource(DirectServeResource):
 
         with self.media_storage.store_into_file(file_info) as (f, fname, finish):
             try:
-                logger.debug("Trying to get url '%s'" % url)
+                logger.debug("Trying to get url '%s'", url)
                 length, headers, uri, code = yield self.client.get_file(
                     url, output_stream=f, max_size=self.max_spider_size
                 )
@@ -318,7 +336,7 @@ class PreviewUrlResource(DirectServeResource):
                 )
             except Exception as e:
                 # FIXME: pass through 404s and other error messages nicely
-                logger.warn("Error downloading %s: %r", url, e)
+                logger.warning("Error downloading %s: %r", url, e)
 
                 raise SynapseError(
                     500,
@@ -399,7 +417,7 @@ class PreviewUrlResource(DirectServeResource):
             except OSError as e:
                 # If the path doesn't exist, meh
                 if e.errno != errno.ENOENT:
-                    logger.warn("Failed to remove media: %r: %s", media_id, e)
+                    logger.warning("Failed to remove media: %r: %s", media_id, e)
                     continue
 
             removed_media.append(media_id)
@@ -431,7 +449,7 @@ class PreviewUrlResource(DirectServeResource):
             except OSError as e:
                 # If the path doesn't exist, meh
                 if e.errno != errno.ENOENT:
-                    logger.warn("Failed to remove media: %r: %s", media_id, e)
+                    logger.warning("Failed to remove media: %r: %s", media_id, e)
                     continue
 
             try:
@@ -447,7 +465,7 @@ class PreviewUrlResource(DirectServeResource):
             except OSError as e:
                 # If the path doesn't exist, meh
                 if e.errno != errno.ENOENT:
-                    logger.warn("Failed to remove media: %r: %s", media_id, e)
+                    logger.warning("Failed to remove media: %r: %s", media_id, e)
                     continue
 
             removed_media.append(media_id)
@@ -503,6 +521,10 @@ def _calc_og(tree, media_uri):
     og = {}
     for tag in tree.xpath("//*/meta[starts-with(@property, 'og:')]"):
         if "content" in tag.attrib:
+            # if we've got more than 50 tags, someone is taking the piss
+            if len(og) >= 50:
+                logger.warning("Skipping OG for page with too many 'og:' tags")
+                return {}
             og[tag.attrib["property"]] = tag.attrib["content"]
 
     # TODO: grab article: meta tags too, e.g.:
diff --git a/synapse/rest/media/v1/thumbnail_resource.py b/synapse/rest/media/v1/thumbnail_resource.py
index 08329884ac..931ce79be8 100644
--- a/synapse/rest/media/v1/thumbnail_resource.py
+++ b/synapse/rest/media/v1/thumbnail_resource.py
@@ -182,7 +182,7 @@ class ThumbnailResource(DirectServeResource):
         if file_path:
             yield respond_with_file(request, desired_type, file_path)
         else:
-            logger.warn("Failed to generate thumbnail")
+            logger.warning("Failed to generate thumbnail")
             respond_404(request)
 
     @defer.inlineCallbacks
@@ -245,7 +245,7 @@ class ThumbnailResource(DirectServeResource):
         if file_path:
             yield respond_with_file(request, desired_type, file_path)
         else:
-            logger.warn("Failed to generate thumbnail")
+            logger.warning("Failed to generate thumbnail")
             respond_404(request)
 
     @defer.inlineCallbacks
diff --git a/synapse/rest/media/v1/thumbnailer.py b/synapse/rest/media/v1/thumbnailer.py
index c995d7e043..8cf415e29d 100644
--- a/synapse/rest/media/v1/thumbnailer.py
+++ b/synapse/rest/media/v1/thumbnailer.py
@@ -82,13 +82,21 @@ class Thumbnailer(object):
         else:
             return (max_height * self.width) // self.height, max_height
 
+    def _resize(self, width, height):
+        # 1-bit or 8-bit color palette images need converting to RGB
+        # otherwise they will be scaled using nearest neighbour which
+        # looks awful
+        if self.image.mode in ["1", "P"]:
+            self.image = self.image.convert("RGB")
+        return self.image.resize((width, height), Image.ANTIALIAS)
+
     def scale(self, width, height, output_type):
         """Rescales the image to the given dimensions.
 
         Returns:
             BytesIO: the bytes of the encoded image ready to be written to disk
         """
-        scaled = self.image.resize((width, height), Image.ANTIALIAS)
+        scaled = self._resize(width, height)
         return self._encode_image(scaled, output_type)
 
     def crop(self, width, height, output_type):
@@ -107,13 +115,13 @@ class Thumbnailer(object):
         """
         if width * self.height > height * self.width:
             scaled_height = (width * self.height) // self.width
-            scaled_image = self.image.resize((width, scaled_height), Image.ANTIALIAS)
+            scaled_image = self._resize(width, scaled_height)
             crop_top = (scaled_height - height) // 2
             crop_bottom = height + crop_top
             cropped = scaled_image.crop((0, crop_top, width, crop_bottom))
         else:
             scaled_width = (height * self.width) // self.height
-            scaled_image = self.image.resize((scaled_width, height), Image.ANTIALIAS)
+            scaled_image = self._resize(scaled_width, height)
             crop_left = (scaled_width - width) // 2
             crop_right = width + crop_left
             cropped = scaled_image.crop((crop_left, 0, crop_right, height))
diff --git a/synapse/rest/media/v1/upload_resource.py b/synapse/rest/media/v1/upload_resource.py
index 5d76bbdf68..83d005812d 100644
--- a/synapse/rest/media/v1/upload_resource.py
+++ b/synapse/rest/media/v1/upload_resource.py
@@ -17,7 +17,7 @@ import logging
 
 from twisted.web.server import NOT_DONE_YET
 
-from synapse.api.errors import SynapseError
+from synapse.api.errors import Codes, SynapseError
 from synapse.http.server import (
     DirectServeResource,
     respond_with_json,
@@ -56,7 +56,11 @@ class UploadResource(DirectServeResource):
         if content_length is None:
             raise SynapseError(msg="Request must specify a Content-Length", code=400)
         if int(content_length) > self.max_upload_size:
-            raise SynapseError(msg="Upload request body is too large", code=413)
+            raise SynapseError(
+                msg="Upload request body is too large",
+                code=413,
+                errcode=Codes.TOO_LARGE,
+            )
 
         upload_name = parse_string(request, b"filename", encoding=None)
         if upload_name:
diff --git a/synapse/server.py b/synapse/server.py
index 9e28dba2b1..90c3b072e8 100644
--- a/synapse/server.py
+++ b/synapse/server.py
@@ -23,6 +23,7 @@
 # Imports required for the default HomeServer() implementation
 import abc
 import logging
+import os
 
 from twisted.enterprise import adbapi
 from twisted.mail.smtp import sendmail
@@ -95,6 +96,7 @@ from synapse.server_notices.worker_server_notices_sender import (
     WorkerServerNoticesSender,
 )
 from synapse.state import StateHandler, StateResolutionHandler
+from synapse.storage import DataStores, Storage
 from synapse.streams.events import EventSources
 from synapse.util import Clock
 from synapse.util.distributor import Distributor
@@ -167,6 +169,7 @@ class HomeServer(object):
         "filtering",
         "http_client_context_factory",
         "simple_http_client",
+        "proxied_http_client",
         "media_repository",
         "media_repository_resource",
         "federation_transport_client",
@@ -196,6 +199,7 @@ class HomeServer(object):
         "account_validity_handler",
         "saml_handler",
         "event_client_serializer",
+        "storage",
     ]
 
     REQUIRED_ON_MASTER_STARTUP = ["user_directory_handler", "stats_handler"]
@@ -217,13 +221,15 @@ class HomeServer(object):
         self.hostname = hostname
         self._building = {}
         self._listening_services = []
+        self.start_time = None
 
         self.clock = Clock(reactor)
         self.distributor = Distributor()
         self.ratelimiter = Ratelimiter()
+        self.admin_redaction_ratelimiter = Ratelimiter()
         self.registration_ratelimiter = Ratelimiter()
 
-        self.datastore = None
+        self.datastores = None
 
         # Other kwargs are explicit dependencies
         for depname in kwargs:
@@ -232,8 +238,10 @@ class HomeServer(object):
     def setup(self):
         logger.info("Setting up.")
         with self.get_db_conn() as conn:
-            self.datastore = self.DATASTORE_CLASS(conn, self)
+            datastore = self.DATASTORE_CLASS(conn, self)
+            self.datastores = DataStores(datastore, conn, self)
             conn.commit()
+        self.start_time = int(self.get_clock().time())
         logger.info("Finished setting up.")
 
     def setup_master(self):
@@ -265,7 +273,7 @@ class HomeServer(object):
         return self.clock
 
     def get_datastore(self):
-        return self.datastore
+        return self.datastores.main
 
     def get_config(self):
         return self.config
@@ -279,6 +287,9 @@ class HomeServer(object):
     def get_registration_ratelimiter(self):
         return self.registration_ratelimiter
 
+    def get_admin_redaction_ratelimiter(self):
+        return self.admin_redaction_ratelimiter
+
     def build_federation_client(self):
         return FederationClient(self)
 
@@ -304,6 +315,13 @@ class HomeServer(object):
     def build_simple_http_client(self):
         return SimpleHttpClient(self)
 
+    def build_proxied_http_client(self):
+        return SimpleHttpClient(
+            self,
+            http_proxy=os.getenv("http_proxy"),
+            https_proxy=os.getenv("HTTPS_PROXY"),
+        )
+
     def build_room_creation_handler(self):
         return RoomCreationHandler(self)
 
@@ -533,6 +551,9 @@ class HomeServer(object):
     def build_event_client_serializer(self):
         return EventClientSerializer(self)
 
+    def build_storage(self) -> Storage:
+        return Storage(self, self.datastores)
+
     def remove_pusher(self, app_id, push_key, user_id):
         return self.get_pusherpool().remove_pusher(app_id, push_key, user_id)
 
diff --git a/synapse/server.pyi b/synapse/server.pyi
index 16f8f6b573..b5e0b57095 100644
--- a/synapse/server.pyi
+++ b/synapse/server.pyi
@@ -12,6 +12,7 @@ import synapse.handlers.message
 import synapse.handlers.room
 import synapse.handlers.room_member
 import synapse.handlers.set_password
+import synapse.http.client
 import synapse.rest.media.v1.media_repository
 import synapse.server_notices.server_notices_manager
 import synapse.server_notices.server_notices_sender
@@ -38,8 +39,16 @@ class HomeServer(object):
         pass
     def get_state_resolution_handler(self) -> synapse.state.StateResolutionHandler:
         pass
+    def get_simple_http_client(self) -> synapse.http.client.SimpleHttpClient:
+        """Fetch an HTTP client implementation which doesn't do any blacklisting
+        or support any HTTP_PROXY settings"""
+        pass
+    def get_proxied_http_client(self) -> synapse.http.client.SimpleHttpClient:
+        """Fetch an HTTP client implementation which doesn't do any blacklisting
+        but does support HTTP_PROXY settings"""
+        pass
     def get_deactivate_account_handler(
-        self
+        self,
     ) -> synapse.handlers.deactivate_account.DeactivateAccountHandler:
         pass
     def get_room_creation_handler(self) -> synapse.handlers.room.RoomCreationHandler:
@@ -47,32 +56,32 @@ class HomeServer(object):
     def get_room_member_handler(self) -> synapse.handlers.room_member.RoomMemberHandler:
         pass
     def get_event_creation_handler(
-        self
+        self,
     ) -> synapse.handlers.message.EventCreationHandler:
         pass
     def get_set_password_handler(
-        self
+        self,
     ) -> synapse.handlers.set_password.SetPasswordHandler:
         pass
     def get_federation_sender(self) -> synapse.federation.sender.FederationSender:
         pass
     def get_federation_transport_client(
-        self
+        self,
     ) -> synapse.federation.transport.client.TransportLayerClient:
         pass
     def get_media_repository_resource(
-        self
+        self,
     ) -> synapse.rest.media.v1.media_repository.MediaRepositoryResource:
         pass
     def get_media_repository(
-        self
+        self,
     ) -> synapse.rest.media.v1.media_repository.MediaRepository:
         pass
     def get_server_notices_manager(
-        self
+        self,
     ) -> synapse.server_notices.server_notices_manager.ServerNoticesManager:
         pass
     def get_server_notices_sender(
-        self
+        self,
     ) -> synapse.server_notices.server_notices_sender.ServerNoticesSender:
         pass
diff --git a/synapse/server_notices/resource_limits_server_notices.py b/synapse/server_notices/resource_limits_server_notices.py
index 81c4aff496..9fae2e0afe 100644
--- a/synapse/server_notices/resource_limits_server_notices.py
+++ b/synapse/server_notices/resource_limits_server_notices.py
@@ -20,6 +20,7 @@ from twisted.internet import defer
 
 from synapse.api.constants import (
     EventTypes,
+    LimitBlockingTypes,
     ServerNoticeLimitReached,
     ServerNoticeMsgType,
 )
@@ -70,7 +71,7 @@ class ResourceLimitsServerNotices(object):
             return
 
         if not self._server_notices_manager.is_enabled():
-            # Don't try and send server notices unles they've been enabled
+            # Don't try and send server notices unless they've been enabled
             return
 
         timestamp = yield self._store.user_last_seen_monthly_active(user_id)
@@ -79,59 +80,92 @@ class ResourceLimitsServerNotices(object):
             # In practice, not sure we can ever get here
             return
 
-        # Determine current state of room
-
         room_id = yield self._server_notices_manager.get_notice_room_for_user(user_id)
 
         if not room_id:
-            logger.warn("Failed to get server notices room")
+            logger.warning("Failed to get server notices room")
             return
 
         yield self._check_and_set_tags(user_id, room_id)
+
+        # Determine current state of room
         currently_blocked, ref_events = yield self._is_room_currently_blocked(room_id)
 
+        limit_msg = None
+        limit_type = None
         try:
-            # Normally should always pass in user_id if you have it, but in
-            # this case are checking what would happen to other users if they
-            # were to arrive.
-            try:
-                yield self._auth.check_auth_blocking()
-                is_auth_blocking = False
-            except ResourceLimitError as e:
-                is_auth_blocking = True
-                event_content = e.msg
-                event_limit_type = e.limit_type
+            # Normally should always pass in user_id to check_auth_blocking
+            # if you have it, but in this case are checking what would happen
+            # to other users if they were to arrive.
+            yield self._auth.check_auth_blocking()
+        except ResourceLimitError as e:
+            limit_msg = e.msg
+            limit_type = e.limit_type
 
-            if currently_blocked and not is_auth_blocking:
+        try:
+            if (
+                limit_type == LimitBlockingTypes.MONTHLY_ACTIVE_USER
+                and not self._config.mau_limit_alerting
+            ):
+                # We have hit the MAU limit, but MAU alerting is disabled:
+                # reset room if necessary and return
+                if currently_blocked:
+                    self._remove_limit_block_notification(user_id, ref_events)
+                return
+
+            if currently_blocked and not limit_msg:
                 # Room is notifying of a block, when it ought not to be.
-                # Remove block notification
-                content = {"pinned": ref_events}
-                yield self._server_notices_manager.send_notice(
-                    user_id, content, EventTypes.Pinned, ""
-                )
-
-            elif not currently_blocked and is_auth_blocking:
+                yield self._remove_limit_block_notification(user_id, ref_events)
+            elif not currently_blocked and limit_msg:
                 # Room is not notifying of a block, when it ought to be.
-                # Add block notification
-                content = {
-                    "body": event_content,
-                    "msgtype": ServerNoticeMsgType,
-                    "server_notice_type": ServerNoticeLimitReached,
-                    "admin_contact": self._config.admin_contact,
-                    "limit_type": event_limit_type,
-                }
-                event = yield self._server_notices_manager.send_notice(
-                    user_id, content, EventTypes.Message
+                yield self._apply_limit_block_notification(
+                    user_id, limit_msg, limit_type
                 )
-
-                content = {"pinned": [event.event_id]}
-                yield self._server_notices_manager.send_notice(
-                    user_id, content, EventTypes.Pinned, ""
-                )
-
         except SynapseError as e:
             logger.error("Error sending resource limits server notice: %s", e)
 
+    @defer.inlineCallbacks
+    def _remove_limit_block_notification(self, user_id, ref_events):
+        """Utility method to remove limit block notifications from the server
+        notices room.
+
+        Args:
+            user_id (str): user to notify
+            ref_events (list[str]): The event_ids of pinned events that are unrelated to
+            limit blocking and need to be preserved.
+        """
+        content = {"pinned": ref_events}
+        yield self._server_notices_manager.send_notice(
+            user_id, content, EventTypes.Pinned, ""
+        )
+
+    @defer.inlineCallbacks
+    def _apply_limit_block_notification(self, user_id, event_body, event_limit_type):
+        """Utility method to apply limit block notifications in the server
+        notices room.
+
+        Args:
+            user_id (str): user to notify
+            event_body(str): The human readable text that describes the block.
+            event_limit_type(str): Specifies the type of block e.g. monthly active user
+            limit has been exceeded.
+        """
+        content = {
+            "body": event_body,
+            "msgtype": ServerNoticeMsgType,
+            "server_notice_type": ServerNoticeLimitReached,
+            "admin_contact": self._config.admin_contact,
+            "limit_type": event_limit_type,
+        }
+        event = yield self._server_notices_manager.send_notice(
+            user_id, content, EventTypes.Message
+        )
+
+        content = {"pinned": [event.event_id]}
+        yield self._server_notices_manager.send_notice(
+            user_id, content, EventTypes.Pinned, ""
+        )
+
     @defer.inlineCallbacks
     def _check_and_set_tags(self, user_id, room_id):
         """
diff --git a/synapse/spam_checker_api/__init__.py b/synapse/spam_checker_api/__init__.py
new file mode 100644
index 0000000000..efcc10f808
--- /dev/null
+++ b/synapse/spam_checker_api/__init__.py
@@ -0,0 +1,51 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import logging
+
+from twisted.internet import defer
+
+from synapse.storage.state import StateFilter
+
+logger = logging.getLogger(__name__)
+
+
+class SpamCheckerApi(object):
+    """A proxy object that gets passed to spam checkers so they can get
+    access to rooms and other relevant information.
+    """
+
+    def __init__(self, hs):
+        self.hs = hs
+
+        self._store = hs.get_datastore()
+
+    @defer.inlineCallbacks
+    def get_state_events_in_room(self, room_id, types):
+        """Gets state events for the given room.
+
+        Args:
+            room_id (string): The room ID to get state events in.
+            types (tuple): The event type and state key (using None
+                to represent 'any') of the room state to acquire.
+
+        Returns:
+            twisted.internet.defer.Deferred[list(synapse.events.FrozenEvent)]:
+                The filtered state events in the room.
+        """
+        state_ids = yield self._store.get_filtered_current_state_ids(
+            room_id=room_id, state_filter=StateFilter.from_types(types)
+        )
+        state = yield self._store.get_events(state_ids.values())
+        return state.values()
diff --git a/synapse/state/__init__.py b/synapse/state/__init__.py
index a0d34f16ea..139beef8ed 100644
--- a/synapse/state/__init__.py
+++ b/synapse/state/__init__.py
@@ -16,6 +16,7 @@
 
 import logging
 from collections import namedtuple
+from typing import Iterable, Optional
 
 from six import iteritems, itervalues
 
@@ -27,13 +28,14 @@ from twisted.internet import defer
 
 from synapse.api.constants import EventTypes
 from synapse.api.room_versions import KNOWN_ROOM_VERSIONS, StateResolutionVersions
+from synapse.events import EventBase
 from synapse.events.snapshot import EventContext
 from synapse.logging.utils import log_function
 from synapse.state import v1, v2
 from synapse.util.async_helpers import Linearizer
 from synapse.util.caches import get_cache_factor_for
 from synapse.util.caches.expiringcache import ExpiringCache
-from synapse.util.metrics import Measure
+from synapse.util.metrics import Measure, measure_func
 
 logger = logging.getLogger(__name__)
 
@@ -103,6 +105,7 @@ class StateHandler(object):
     def __init__(self, hs):
         self.clock = hs.get_clock()
         self.store = hs.get_datastore()
+        self.state_store = hs.get_storage().state
         self.hs = hs
         self._state_resolution_handler = hs.get_state_resolution_handler()
 
@@ -136,7 +139,6 @@ class StateHandler(object):
             if event_id:
                 event = yield self.store.get_event(event_id, allow_none=True)
             return event
-            return
 
         state_map = yield self.store.get_events(
             list(state.values()), get_prev_content=False
@@ -192,24 +194,37 @@ class StateHandler(object):
         return joined_users
 
     @defer.inlineCallbacks
-    def get_current_hosts_in_room(self, room_id, latest_event_ids=None):
-        if not latest_event_ids:
-            latest_event_ids = yield self.store.get_latest_event_ids_in_room(room_id)
-        logger.debug("calling resolve_state_groups from get_current_hosts_in_room")
-        entry = yield self.resolve_state_groups_for_events(room_id, latest_event_ids)
+    def get_current_hosts_in_room(self, room_id):
+        event_ids = yield self.store.get_latest_event_ids_in_room(room_id)
+        return (yield self.get_hosts_in_room_at_events(room_id, event_ids))
+
+    @defer.inlineCallbacks
+    def get_hosts_in_room_at_events(self, room_id, event_ids):
+        """Get the hosts that were in a room at the given event ids
+
+        Args:
+            room_id (str):
+            event_ids (list[str]):
+
+        Returns:
+            Deferred[list[str]]: the hosts in the room at the given events
+        """
+        entry = yield self.resolve_state_groups_for_events(room_id, event_ids)
         joined_hosts = yield self.store.get_joined_hosts(room_id, entry)
         return joined_hosts
 
     @defer.inlineCallbacks
-    def compute_event_context(self, event, old_state=None):
+    def compute_event_context(
+        self, event: EventBase, old_state: Optional[Iterable[EventBase]] = None
+    ):
         """Build an EventContext structure for the event.
 
         This works out what the current state should be for the event, and
         generates a new state group if necessary.
 
         Args:
-            event (synapse.events.EventBase):
-            old_state (dict|None): The state at the event if it can't be
+            event:
+            old_state: The state at the event if it can't be
                 calculated from existing events. This is normally only specified
                 when receiving an event from federation where we don't have the
                 prev events for, e.g. when backfilling.
@@ -221,6 +236,9 @@ class StateHandler(object):
             # If this is an outlier, then we know it shouldn't have any current
             # state. Certainly store.get_current_state won't return any, and
             # persisting the event won't store the state group.
+
+            # FIXME: why do we populate current_state_ids? I thought the point was
+            # that we weren't supposed to have any state for outliers?
             if old_state:
                 prev_state_ids = {(s.type, s.state_key): s.event_id for s in old_state}
                 if event.is_state():
@@ -237,114 +255,105 @@ class StateHandler(object):
             # group for it.
             context = EventContext.with_state(
                 state_group=None,
+                state_group_before_event=None,
                 current_state_ids=current_state_ids,
                 prev_state_ids=prev_state_ids,
             )
 
             return context
 
+        #
+        # first of all, figure out the state before the event
+        #
+
         if old_state:
-            # We already have the state, so we don't need to calculate it.
-            # Let's just correctly fill out the context and create a
-            # new state group for it.
+            # if we're given the state before the event, then we use that
+            state_ids_before_event = {
+                (s.type, s.state_key): s.event_id for s in old_state
+            }
+            state_group_before_event = None
+            state_group_before_event_prev_group = None
+            deltas_to_state_group_before_event = None
 
-            prev_state_ids = {(s.type, s.state_key): s.event_id for s in old_state}
+        else:
+            # otherwise, we'll need to resolve the state across the prev_events.
+            logger.debug("calling resolve_state_groups from compute_event_context")
 
-            if event.is_state():
-                key = (event.type, event.state_key)
-                if key in prev_state_ids:
-                    replaces = prev_state_ids[key]
-                    if replaces != event.event_id:  # Paranoia check
-                        event.unsigned["replaces_state"] = replaces
-                current_state_ids = dict(prev_state_ids)
-                current_state_ids[key] = event.event_id
-            else:
-                current_state_ids = prev_state_ids
+            entry = yield self.resolve_state_groups_for_events(
+                event.room_id, event.prev_event_ids()
+            )
 
-            state_group = yield self.store.store_state_group(
+            state_ids_before_event = entry.state
+            state_group_before_event = entry.state_group
+            state_group_before_event_prev_group = entry.prev_group
+            deltas_to_state_group_before_event = entry.delta_ids
+
+        #
+        # make sure that we have a state group at that point. If it's not a state event,
+        # that will be the state group for the new event. If it *is* a state event,
+        # it might get rejected (in which case we'll need to persist it with the
+        # previous state group)
+        #
+
+        if not state_group_before_event:
+            state_group_before_event = yield self.state_store.store_state_group(
                 event.event_id,
                 event.room_id,
-                prev_group=None,
-                delta_ids=None,
-                current_state_ids=current_state_ids,
+                prev_group=state_group_before_event_prev_group,
+                delta_ids=deltas_to_state_group_before_event,
+                current_state_ids=state_ids_before_event,
             )
 
-            context = EventContext.with_state(
-                state_group=state_group,
-                current_state_ids=current_state_ids,
-                prev_state_ids=prev_state_ids,
+            # XXX: can we update the state cache entry for the new state group? or
+            # could we set a flag on resolve_state_groups_for_events to tell it to
+            # always make a state group?
+
+        #
+        # now if it's not a state event, we're done
+        #
+
+        if not event.is_state():
+            return EventContext.with_state(
+                state_group_before_event=state_group_before_event,
+                state_group=state_group_before_event,
+                current_state_ids=state_ids_before_event,
+                prev_state_ids=state_ids_before_event,
+                prev_group=state_group_before_event_prev_group,
+                delta_ids=deltas_to_state_group_before_event,
             )
 
-            return context
+        #
+        # otherwise, we'll need to create a new state group for after the event
+        #
 
-        logger.debug("calling resolve_state_groups from compute_event_context")
-
-        entry = yield self.resolve_state_groups_for_events(
-            event.room_id, event.prev_event_ids()
-        )
-
-        prev_state_ids = entry.state
-        prev_group = None
-        delta_ids = None
-
-        if event.is_state():
-            # If this is a state event then we need to create a new state
-            # group for the state after this event.
-
-            key = (event.type, event.state_key)
-            if key in prev_state_ids:
-                replaces = prev_state_ids[key]
+        key = (event.type, event.state_key)
+        if key in state_ids_before_event:
+            replaces = state_ids_before_event[key]
+            if replaces != event.event_id:
                 event.unsigned["replaces_state"] = replaces
 
-            current_state_ids = dict(prev_state_ids)
-            current_state_ids[key] = event.event_id
+        state_ids_after_event = dict(state_ids_before_event)
+        state_ids_after_event[key] = event.event_id
+        delta_ids = {key: event.event_id}
 
-            if entry.state_group:
-                # If the state at the event has a state group assigned then
-                # we can use that as the prev group
-                prev_group = entry.state_group
-                delta_ids = {key: event.event_id}
-            elif entry.prev_group:
-                # If the state at the event only has a prev group, then we can
-                # use that as a prev group too.
-                prev_group = entry.prev_group
-                delta_ids = dict(entry.delta_ids)
-                delta_ids[key] = event.event_id
+        state_group_after_event = yield self.state_store.store_state_group(
+            event.event_id,
+            event.room_id,
+            prev_group=state_group_before_event,
+            delta_ids=delta_ids,
+            current_state_ids=state_ids_after_event,
+        )
 
-            state_group = yield self.store.store_state_group(
-                event.event_id,
-                event.room_id,
-                prev_group=prev_group,
-                delta_ids=delta_ids,
-                current_state_ids=current_state_ids,
-            )
-        else:
-            current_state_ids = prev_state_ids
-            prev_group = entry.prev_group
-            delta_ids = entry.delta_ids
-
-            if entry.state_group is None:
-                entry.state_group = yield self.store.store_state_group(
-                    event.event_id,
-                    event.room_id,
-                    prev_group=entry.prev_group,
-                    delta_ids=entry.delta_ids,
-                    current_state_ids=current_state_ids,
-                )
-                entry.state_id = entry.state_group
-
-            state_group = entry.state_group
-
-        context = EventContext.with_state(
-            state_group=state_group,
-            current_state_ids=current_state_ids,
-            prev_state_ids=prev_state_ids,
-            prev_group=prev_group,
+        return EventContext.with_state(
+            state_group=state_group_after_event,
+            state_group_before_event=state_group_before_event,
+            current_state_ids=state_ids_after_event,
+            prev_state_ids=state_ids_before_event,
+            prev_group=state_group_before_event,
             delta_ids=delta_ids,
         )
 
-        return context
-
+    @measure_func()
     @defer.inlineCallbacks
     def resolve_state_groups_for_events(self, room_id, event_ids):
         """ Given a list of event_ids this method fetches the state at each
@@ -365,14 +374,16 @@ class StateHandler(object):
         # map from state group id to the state in that state group (where
         # 'state' is a map from state key to event id)
         # dict[int, dict[(str, str), str]]
-        state_groups_ids = yield self.store.get_state_groups_ids(room_id, event_ids)
+        state_groups_ids = yield self.state_store.get_state_groups_ids(
+            room_id, event_ids
+        )
 
         if len(state_groups_ids) == 0:
             return _StateCacheEntry(state={}, state_group=None)
         elif len(state_groups_ids) == 1:
             name, state_list = list(state_groups_ids.items()).pop()
 
-            prev_group, delta_ids = yield self.store.get_state_group_delta(name)
+            prev_group, delta_ids = yield self.state_store.get_state_group_delta(name)
 
             return _StateCacheEntry(
                 state=state_list,
diff --git a/synapse/static/client/login/js/login.js b/synapse/static/client/login/js/login.js
index e02663f50e..276c271bbe 100644
--- a/synapse/static/client/login/js/login.js
+++ b/synapse/static/client/login/js/login.js
@@ -62,7 +62,7 @@ var show_login = function() {
         $("#sso_flow").show();
     }
 
-    if (!matrixLogin.serverAcceptsPassword && !matrixLogin.serverAcceptsCas) {
+    if (!matrixLogin.serverAcceptsPassword && !matrixLogin.serverAcceptsCas && !matrixLogin.serverAcceptsSso) {
         $("#no_login_types").show();
     }
 };
diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py
index e7f6ea7286..0a1a8cc1e5 100644
--- a/synapse/storage/__init__.py
+++ b/synapse/storage/__init__.py
@@ -1,6 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2014-2016 OpenMarket Ltd
-# Copyright 2018 New Vector Ltd
+# Copyright 2018,2019 New Vector Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,509 +14,39 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import calendar
-import logging
-import time
+"""
+The storage layer is split up into multiple parts to allow Synapse to run
+against different configurations of databases (e.g. single or multiple
+databases). The `data_stores` are classes that talk directly to a single
+database and have associated schemas, background updates, etc. On top of those
+there are (or will be) classes that provide high level interfaces that combine
+calls to multiple `data_stores`.
 
-from twisted.internet import defer
+There are also schemas that get applied to every database, regardless of the
+data stores associated with them (e.g. the schema version tables), which are
+stored in `synapse.storage.schema`.
+"""
 
-from synapse.api.constants import PresenceState
-from synapse.storage.devices import DeviceStore
-from synapse.storage.user_erasure_store import UserErasureStore
-from synapse.util.caches.stream_change_cache import StreamChangeCache
+from synapse.storage.data_stores import DataStores
+from synapse.storage.data_stores.main import DataStore
+from synapse.storage.persist_events import EventsPersistenceStorage
+from synapse.storage.state import StateGroupStorage
 
-from .account_data import AccountDataStore
-from .appservice import ApplicationServiceStore, ApplicationServiceTransactionStore
-from .client_ips import ClientIpStore
-from .deviceinbox import DeviceInboxStore
-from .directory import DirectoryStore
-from .e2e_room_keys import EndToEndRoomKeyStore
-from .end_to_end_keys import EndToEndKeyStore
-from .engines import PostgresEngine
-from .event_federation import EventFederationStore
-from .event_push_actions import EventPushActionsStore
-from .events import EventsStore
-from .events_bg_updates import EventsBackgroundUpdatesStore
-from .filtering import FilteringStore
-from .group_server import GroupServerStore
-from .keys import KeyStore
-from .media_repository import MediaRepositoryStore
-from .monthly_active_users import MonthlyActiveUsersStore
-from .openid import OpenIdStore
-from .presence import PresenceStore, UserPresenceState
-from .profile import ProfileStore
-from .push_rule import PushRuleStore
-from .pusher import PusherStore
-from .receipts import ReceiptsStore
-from .registration import RegistrationStore
-from .rejections import RejectionsStore
-from .relations import RelationsStore
-from .room import RoomStore
-from .roommember import RoomMemberStore
-from .search import SearchStore
-from .signatures import SignatureStore
-from .state import StateStore
-from .stats import StatsStore
-from .stream import StreamStore
-from .tags import TagsStore
-from .transactions import TransactionStore
-from .user_directory import UserDirectoryStore
-from .util.id_generators import ChainedIdGenerator, IdGenerator, StreamIdGenerator
-
-logger = logging.getLogger(__name__)
+__all__ = ["DataStores", "DataStore"]
 
 
-class DataStore(
-    EventsBackgroundUpdatesStore,
-    RoomMemberStore,
-    RoomStore,
-    RegistrationStore,
-    StreamStore,
-    ProfileStore,
-    PresenceStore,
-    TransactionStore,
-    DirectoryStore,
-    KeyStore,
-    StateStore,
-    SignatureStore,
-    ApplicationServiceStore,
-    EventsStore,
-    EventFederationStore,
-    MediaRepositoryStore,
-    RejectionsStore,
-    FilteringStore,
-    PusherStore,
-    PushRuleStore,
-    ApplicationServiceTransactionStore,
-    ReceiptsStore,
-    EndToEndKeyStore,
-    EndToEndRoomKeyStore,
-    SearchStore,
-    TagsStore,
-    AccountDataStore,
-    EventPushActionsStore,
-    OpenIdStore,
-    ClientIpStore,
-    DeviceStore,
-    DeviceInboxStore,
-    UserDirectoryStore,
-    GroupServerStore,
-    UserErasureStore,
-    MonthlyActiveUsersStore,
-    StatsStore,
-    RelationsStore,
-):
-    def __init__(self, db_conn, hs):
-        self.hs = hs
-        self._clock = hs.get_clock()
-        self.database_engine = hs.database_engine
+class Storage(object):
+    """The high level interfaces for talking to various storage layers.
+    """
 
-        self._stream_id_gen = StreamIdGenerator(
-            db_conn,
-            "events",
-            "stream_ordering",
-            extra_tables=[("local_invites", "stream_id")],
-        )
-        self._backfill_id_gen = StreamIdGenerator(
-            db_conn,
-            "events",
-            "stream_ordering",
-            step=-1,
-            extra_tables=[("ex_outlier_stream", "event_stream_ordering")],
-        )
-        self._presence_id_gen = StreamIdGenerator(
-            db_conn, "presence_stream", "stream_id"
-        )
-        self._device_inbox_id_gen = StreamIdGenerator(
-            db_conn, "device_max_stream_id", "stream_id"
-        )
-        self._public_room_id_gen = StreamIdGenerator(
-            db_conn, "public_room_list_stream", "stream_id"
-        )
-        self._device_list_id_gen = StreamIdGenerator(
-            db_conn, "device_lists_stream", "stream_id"
-        )
+    def __init__(self, hs, stores: DataStores):
+        # We include the main data store here mainly so that we don't have to
+        # rewrite all the existing code to split it into high vs low level
+        # interfaces.
+        self.main = stores.main
 
-        self._access_tokens_id_gen = IdGenerator(db_conn, "access_tokens", "id")
-        self._event_reports_id_gen = IdGenerator(db_conn, "event_reports", "id")
-        self._push_rule_id_gen = IdGenerator(db_conn, "push_rules", "id")
-        self._push_rules_enable_id_gen = IdGenerator(db_conn, "push_rules_enable", "id")
-        self._push_rules_stream_id_gen = ChainedIdGenerator(
-            self._stream_id_gen, db_conn, "push_rules_stream", "stream_id"
-        )
-        self._pushers_id_gen = StreamIdGenerator(
-            db_conn, "pushers", "id", extra_tables=[("deleted_pushers", "stream_id")]
-        )
-        self._group_updates_id_gen = StreamIdGenerator(
-            db_conn, "local_group_updates", "stream_id"
-        )
-
-        if isinstance(self.database_engine, PostgresEngine):
-            self._cache_id_gen = StreamIdGenerator(
-                db_conn, "cache_invalidation_stream", "stream_id"
-            )
-        else:
-            self._cache_id_gen = None
-
-        self._presence_on_startup = self._get_active_presence(db_conn)
-
-        presence_cache_prefill, min_presence_val = self._get_cache_dict(
-            db_conn,
-            "presence_stream",
-            entity_column="user_id",
-            stream_column="stream_id",
-            max_value=self._presence_id_gen.get_current_token(),
-        )
-        self.presence_stream_cache = StreamChangeCache(
-            "PresenceStreamChangeCache",
-            min_presence_val,
-            prefilled_cache=presence_cache_prefill,
-        )
-
-        max_device_inbox_id = self._device_inbox_id_gen.get_current_token()
-        device_inbox_prefill, min_device_inbox_id = self._get_cache_dict(
-            db_conn,
-            "device_inbox",
-            entity_column="user_id",
-            stream_column="stream_id",
-            max_value=max_device_inbox_id,
-            limit=1000,
-        )
-        self._device_inbox_stream_cache = StreamChangeCache(
-            "DeviceInboxStreamChangeCache",
-            min_device_inbox_id,
-            prefilled_cache=device_inbox_prefill,
-        )
-        # The federation outbox and the local device inbox uses the same
-        # stream_id generator.
-        device_outbox_prefill, min_device_outbox_id = self._get_cache_dict(
-            db_conn,
-            "device_federation_outbox",
-            entity_column="destination",
-            stream_column="stream_id",
-            max_value=max_device_inbox_id,
-            limit=1000,
-        )
-        self._device_federation_outbox_stream_cache = StreamChangeCache(
-            "DeviceFederationOutboxStreamChangeCache",
-            min_device_outbox_id,
-            prefilled_cache=device_outbox_prefill,
-        )
-
-        device_list_max = self._device_list_id_gen.get_current_token()
-        self._device_list_stream_cache = StreamChangeCache(
-            "DeviceListStreamChangeCache", device_list_max
-        )
-        self._device_list_federation_stream_cache = StreamChangeCache(
-            "DeviceListFederationStreamChangeCache", device_list_max
-        )
-
-        events_max = self._stream_id_gen.get_current_token()
-        curr_state_delta_prefill, min_curr_state_delta_id = self._get_cache_dict(
-            db_conn,
-            "current_state_delta_stream",
-            entity_column="room_id",
-            stream_column="stream_id",
-            max_value=events_max,  # As we share the stream id with events token
-            limit=1000,
-        )
-        self._curr_state_delta_stream_cache = StreamChangeCache(
-            "_curr_state_delta_stream_cache",
-            min_curr_state_delta_id,
-            prefilled_cache=curr_state_delta_prefill,
-        )
-
-        _group_updates_prefill, min_group_updates_id = self._get_cache_dict(
-            db_conn,
-            "local_group_updates",
-            entity_column="user_id",
-            stream_column="stream_id",
-            max_value=self._group_updates_id_gen.get_current_token(),
-            limit=1000,
-        )
-        self._group_updates_stream_cache = StreamChangeCache(
-            "_group_updates_stream_cache",
-            min_group_updates_id,
-            prefilled_cache=_group_updates_prefill,
-        )
-
-        self._stream_order_on_start = self.get_room_max_stream_ordering()
-        self._min_stream_order_on_start = self.get_room_min_stream_ordering()
-
-        # Used in _generate_user_daily_visits to keep track of progress
-        self._last_user_visit_update = self._get_start_of_day()
-
-        super(DataStore, self).__init__(db_conn, hs)
-
-    def take_presence_startup_info(self):
-        active_on_startup = self._presence_on_startup
-        self._presence_on_startup = None
-        return active_on_startup
-
-    def _get_active_presence(self, db_conn):
-        """Fetch non-offline presence from the database so that we can register
-        the appropriate time outs.
-        """
-
-        sql = (
-            "SELECT user_id, state, last_active_ts, last_federation_update_ts,"
-            " last_user_sync_ts, status_msg, currently_active FROM presence_stream"
-            " WHERE state != ?"
-        )
-        sql = self.database_engine.convert_param_style(sql)
-
-        txn = db_conn.cursor()
-        txn.execute(sql, (PresenceState.OFFLINE,))
-        rows = self.cursor_to_dict(txn)
-        txn.close()
-
-        for row in rows:
-            row["currently_active"] = bool(row["currently_active"])
-
-        return [UserPresenceState(**row) for row in rows]
-
-    def count_daily_users(self):
-        """
-        Counts the number of users who used this homeserver in the last 24 hours.
-        """
-        yesterday = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24)
-        return self.runInteraction("count_daily_users", self._count_users, yesterday)
-
-    def count_monthly_users(self):
-        """
-        Counts the number of users who used this homeserver in the last 30 days.
-        Note this method is intended for phonehome metrics only and is different
-        from the mau figure in synapse.storage.monthly_active_users which,
-        amongst other things, includes a 3 day grace period before a user counts.
-        """
-        thirty_days_ago = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24 * 30)
-        return self.runInteraction(
-            "count_monthly_users", self._count_users, thirty_days_ago
-        )
-
-    def _count_users(self, txn, time_from):
-        """
-        Returns number of users seen in the past time_from period
-        """
-        sql = """
-            SELECT COALESCE(count(*), 0) FROM (
-                SELECT user_id FROM user_ips
-                WHERE last_seen > ?
-                GROUP BY user_id
-            ) u
-        """
-        txn.execute(sql, (time_from,))
-        count, = txn.fetchone()
-        return count
-
-    def count_r30_users(self):
-        """
-        Counts the number of 30 day retained users, defined as:-
-         * Users who have created their accounts more than 30 days ago
-         * Where last seen at most 30 days ago
-         * Where account creation and last_seen are > 30 days apart
-
-         Returns counts globaly for a given user as well as breaking
-         by platform
-        """
-
-        def _count_r30_users(txn):
-            thirty_days_in_secs = 86400 * 30
-            now = int(self._clock.time())
-            thirty_days_ago_in_secs = now - thirty_days_in_secs
-
-            sql = """
-                SELECT platform, COALESCE(count(*), 0) FROM (
-                     SELECT
-                        users.name, platform, users.creation_ts * 1000,
-                        MAX(uip.last_seen)
-                     FROM users
-                     INNER JOIN (
-                         SELECT
-                         user_id,
-                         last_seen,
-                         CASE
-                             WHEN user_agent LIKE '%%Android%%' THEN 'android'
-                             WHEN user_agent LIKE '%%iOS%%' THEN 'ios'
-                             WHEN user_agent LIKE '%%Electron%%' THEN 'electron'
-                             WHEN user_agent LIKE '%%Mozilla%%' THEN 'web'
-                             WHEN user_agent LIKE '%%Gecko%%' THEN 'web'
-                             ELSE 'unknown'
-                         END
-                         AS platform
-                         FROM user_ips
-                     ) uip
-                     ON users.name = uip.user_id
-                     AND users.appservice_id is NULL
-                     AND users.creation_ts < ?
-                     AND uip.last_seen/1000 > ?
-                     AND (uip.last_seen/1000) - users.creation_ts > 86400 * 30
-                     GROUP BY users.name, platform, users.creation_ts
-                ) u GROUP BY platform
-            """
-
-            results = {}
-            txn.execute(sql, (thirty_days_ago_in_secs, thirty_days_ago_in_secs))
-
-            for row in txn:
-                if row[0] == "unknown":
-                    pass
-                results[row[0]] = row[1]
-
-            sql = """
-                SELECT COALESCE(count(*), 0) FROM (
-                    SELECT users.name, users.creation_ts * 1000,
-                                                        MAX(uip.last_seen)
-                    FROM users
-                    INNER JOIN (
-                        SELECT
-                        user_id,
-                        last_seen
-                        FROM user_ips
-                    ) uip
-                    ON users.name = uip.user_id
-                    AND appservice_id is NULL
-                    AND users.creation_ts < ?
-                    AND uip.last_seen/1000 > ?
-                    AND (uip.last_seen/1000) - users.creation_ts > 86400 * 30
-                    GROUP BY users.name, users.creation_ts
-                ) u
-            """
-
-            txn.execute(sql, (thirty_days_ago_in_secs, thirty_days_ago_in_secs))
-
-            count, = txn.fetchone()
-            results["all"] = count
-
-            return results
-
-        return self.runInteraction("count_r30_users", _count_r30_users)
-
-    def _get_start_of_day(self):
-        """
-        Returns millisecond unixtime for start of UTC day.
-        """
-        now = time.gmtime()
-        today_start = calendar.timegm((now.tm_year, now.tm_mon, now.tm_mday, 0, 0, 0))
-        return today_start * 1000
-
-    def generate_user_daily_visits(self):
-        """
-        Generates daily visit data for use in cohort/ retention analysis
-        """
-
-        def _generate_user_daily_visits(txn):
-            logger.info("Calling _generate_user_daily_visits")
-            today_start = self._get_start_of_day()
-            a_day_in_milliseconds = 24 * 60 * 60 * 1000
-            now = self.clock.time_msec()
-
-            sql = """
-                INSERT INTO user_daily_visits (user_id, device_id, timestamp)
-                    SELECT u.user_id, u.device_id, ?
-                    FROM user_ips AS u
-                    LEFT JOIN (
-                      SELECT user_id, device_id, timestamp FROM user_daily_visits
-                      WHERE timestamp = ?
-                    ) udv
-                    ON u.user_id = udv.user_id AND u.device_id=udv.device_id
-                    INNER JOIN users ON users.name=u.user_id
-                    WHERE last_seen > ? AND last_seen <= ?
-                    AND udv.timestamp IS NULL AND users.is_guest=0
-                    AND users.appservice_id IS NULL
-                    GROUP BY u.user_id, u.device_id
-            """
-
-            # This means that the day has rolled over but there could still
-            # be entries from the previous day. There is an edge case
-            # where if the user logs in at 23:59 and overwrites their
-            # last_seen at 00:01 then they will not be counted in the
-            # previous day's stats - it is important that the query is run
-            # often to minimise this case.
-            if today_start > self._last_user_visit_update:
-                yesterday_start = today_start - a_day_in_milliseconds
-                txn.execute(
-                    sql,
-                    (
-                        yesterday_start,
-                        yesterday_start,
-                        self._last_user_visit_update,
-                        today_start,
-                    ),
-                )
-                self._last_user_visit_update = today_start
-
-            txn.execute(
-                sql, (today_start, today_start, self._last_user_visit_update, now)
-            )
-            # Update _last_user_visit_update to now. The reason to do this
-            # rather just clamping to the beginning of the day is to limit
-            # the size of the join - meaning that the query can be run more
-            # frequently
-            self._last_user_visit_update = now
-
-        return self.runInteraction(
-            "generate_user_daily_visits", _generate_user_daily_visits
-        )
-
-    def get_users(self):
-        """Function to reterive a list of users in users table.
-
-        Args:
-        Returns:
-            defer.Deferred: resolves to list[dict[str, Any]]
-        """
-        return self._simple_select_list(
-            table="users",
-            keyvalues={},
-            retcols=["name", "password_hash", "is_guest", "admin", "user_type"],
-            desc="get_users",
-        )
-
-    @defer.inlineCallbacks
-    def get_users_paginate(self, order, start, limit):
-        """Function to reterive a paginated list of users from
-        users list. This will return a json object, which contains
-        list of users and the total number of users in users table.
-
-        Args:
-            order (str): column name to order the select by this column
-            start (int): start number to begin the query from
-            limit (int): number of rows to reterive
-        Returns:
-            defer.Deferred: resolves to json object {list[dict[str, Any]], count}
-        """
-        users = yield self.runInteraction(
-            "get_users_paginate",
-            self._simple_select_list_paginate_txn,
-            table="users",
-            keyvalues={"is_guest": False},
-            orderby=order,
-            start=start,
-            limit=limit,
-            retcols=["name", "password_hash", "is_guest", "admin", "user_type"],
-        )
-        count = yield self.runInteraction("get_users_paginate", self.get_user_count_txn)
-        retval = {"users": users, "total": count}
-        return retval
-
-    def search_users(self, term):
-        """Function to search users list for one or more users with
-        the matched term.
-
-        Args:
-            term (str): search term
-            col (str): column to query term should be matched to
-        Returns:
-            defer.Deferred: resolves to list[dict[str, Any]]
-        """
-        return self._simple_search_list(
-            table="users",
-            term=term,
-            col="name",
-            retcols=["name", "password_hash", "is_guest", "admin", "user_type"],
-            desc="search_users",
-        )
+        self.persistence = EventsPersistenceStorage(hs, stores)
+        self.state = StateGroupStorage(hs, stores)
 
 
 def are_all_users_on_domain(txn, database_engine, domain):
diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py
index abe16334ec..1a2b7ebe25 100644
--- a/synapse/storage/_base.py
+++ b/synapse/storage/_base.py
@@ -20,6 +20,7 @@ import random
 import sys
 import threading
 import time
+from typing import Iterable, Tuple
 
 from six import PY2, iteritems, iterkeys, itervalues
 from six.moves import builtins, intern, range
@@ -30,7 +31,7 @@ from prometheus_client import Histogram
 from twisted.internet import defer
 
 from synapse.api.errors import StoreError
-from synapse.logging.context import LoggingContext, PreserveLoggingContext
+from synapse.logging.context import LoggingContext, make_deferred_yieldable
 from synapse.metrics.background_process_metrics import run_as_background_process
 from synapse.storage.engines import PostgresEngine, Sqlite3Engine
 from synapse.types import get_domain_from_id
@@ -493,7 +494,7 @@ class SQLBaseStore(object):
         exception_callbacks = []
 
         if LoggingContext.current_context() == LoggingContext.sentinel:
-            logger.warn("Starting db txn '%s' from sentinel context", desc)
+            logger.warning("Starting db txn '%s' from sentinel context", desc)
 
         try:
             result = yield self.runWithConnection(
@@ -531,7 +532,7 @@ class SQLBaseStore(object):
         """
         parent_context = LoggingContext.current_context()
         if parent_context == LoggingContext.sentinel:
-            logger.warn(
+            logger.warning(
                 "Starting db connection from sentinel context: metrics will be lost"
             )
             parent_context = None
@@ -550,8 +551,9 @@ class SQLBaseStore(object):
 
                 return func(conn, *args, **kwargs)
 
-        with PreserveLoggingContext():
-            result = yield self._db_pool.runWithConnection(inner_func, *args, **kwargs)
+        result = yield make_deferred_yieldable(
+            self._db_pool.runWithConnection(inner_func, *args, **kwargs)
+        )
 
         return result
 
@@ -717,7 +719,7 @@ class SQLBaseStore(object):
                     raise
 
                 # presumably we raced with another transaction: let's retry.
-                logger.warn(
+                logger.warning(
                     "IntegrityError when upserting into %s; retrying: %s", table, e
                 )
 
@@ -1162,19 +1164,18 @@ class SQLBaseStore(object):
         if not iterable:
             return []
 
-        sql = "SELECT %s FROM %s" % (", ".join(retcols), table)
-
-        clauses = []
-        values = []
-        clauses.append("%s IN (%s)" % (column, ",".join("?" for _ in iterable)))
-        values.extend(iterable)
+        clause, values = make_in_list_sql_clause(txn.database_engine, column, iterable)
+        clauses = [clause]
 
         for key, value in iteritems(keyvalues):
             clauses.append("%s = ?" % (key,))
             values.append(value)
 
-        if clauses:
-            sql = "%s WHERE %s" % (sql, " AND ".join(clauses))
+        sql = "SELECT %s FROM %s WHERE %s" % (
+            ", ".join(retcols),
+            table,
+            " AND ".join(clauses),
+        )
 
         txn.execute(sql, values)
         return cls.cursor_to_dict(txn)
@@ -1323,10 +1324,8 @@ class SQLBaseStore(object):
 
         sql = "DELETE FROM %s" % table
 
-        clauses = []
-        values = []
-        clauses.append("%s IN (%s)" % (column, ",".join("?" for _ in iterable)))
-        values.extend(iterable)
+        clause, values = make_in_list_sql_clause(txn.database_engine, column, iterable)
+        clauses = [clause]
 
         for key, value in iteritems(keyvalues):
             clauses.append("%s = ?" % (key,))
@@ -1693,3 +1692,30 @@ def db_to_json(db_content):
     except Exception:
         logging.warning("Tried to decode '%r' as JSON and failed", db_content)
         raise
+
+
+def make_in_list_sql_clause(
+    database_engine, column: str, iterable: Iterable
+) -> Tuple[str, Iterable]:
+    """Returns an SQL clause that checks the given column is in the iterable.
+
+    On SQLite this expands to `column IN (?, ?, ...)`, whereas on Postgres
+    it expands to `column = ANY(?)`. While both DBs support the `IN` form,
+    using the `ANY` form on postgres means that it views queries with
+    different length iterables as the same, helping the query stats.
+
+    Args:
+        database_engine
+        column: Name of the column
+        iterable: The values to check the column against.
+
+    Returns:
+        A tuple of SQL query and the args
+    """
+
+    if database_engine.supports_using_any_list:
+        # This should hopefully be faster, but also makes postgres query
+        # stats easier to understand.
+        return "%s = ANY(?)" % (column,), [list(iterable)]
+    else:
+        return "%s IN (%s)" % (column, ",".join("?" for _ in iterable)), list(iterable)
diff --git a/synapse/storage/background_updates.py b/synapse/storage/background_updates.py
index e5f0668f09..37d469ffd7 100644
--- a/synapse/storage/background_updates.py
+++ b/synapse/storage/background_updates.py
@@ -94,13 +94,16 @@ class BackgroundUpdateStore(SQLBaseStore):
         self._all_done = False
 
     def start_doing_background_updates(self):
-        run_as_background_process("background_updates", self._run_background_updates)
+        run_as_background_process("background_updates", self.run_background_updates)
 
     @defer.inlineCallbacks
-    def _run_background_updates(self):
+    def run_background_updates(self, sleep=True):
         logger.info("Starting background schema updates")
         while True:
-            yield self.hs.get_clock().sleep(self.BACKGROUND_UPDATE_INTERVAL_MS / 1000.0)
+            if sleep:
+                yield self.hs.get_clock().sleep(
+                    self.BACKGROUND_UPDATE_INTERVAL_MS / 1000.0
+                )
 
             try:
                 result = yield self.do_next_background_update(
@@ -140,7 +143,7 @@ class BackgroundUpdateStore(SQLBaseStore):
             "background_updates",
             keyvalues=None,
             retcol="1",
-            desc="check_background_updates",
+            desc="has_completed_background_updates",
         )
         if not updates:
             self._all_done = True
@@ -148,6 +151,26 @@ class BackgroundUpdateStore(SQLBaseStore):
 
         return False
 
+    async def has_completed_background_update(self, update_name) -> bool:
+        """Check if the given background update has finished running.
+        """
+
+        if self._all_done:
+            return True
+
+        if update_name in self._background_update_queue:
+            return False
+
+        update_exists = await self._simple_select_one_onecol(
+            "background_updates",
+            keyvalues={"update_name": update_name},
+            retcol="1",
+            desc="has_completed_background_update",
+            allow_none=True,
+        )
+
+        return not update_exists
+
     @defer.inlineCallbacks
     def do_next_background_update(self, desired_duration_ms):
         """Does some amount of work on the next queued background update
@@ -218,7 +241,7 @@ class BackgroundUpdateStore(SQLBaseStore):
         duration_ms = time_stop - time_start
 
         logger.info(
-            "Updating %r. Updated %r items in %rms."
+            "Running background update %r. Processed %r items in %rms."
             " (total_rate=%r/ms, current_rate=%r/ms, total_updated=%r, batch_size=%r)",
             update_name,
             items_updated,
diff --git a/synapse/storage/data_stores/__init__.py b/synapse/storage/data_stores/__init__.py
new file mode 100644
index 0000000000..cb184a98cc
--- /dev/null
+++ b/synapse/storage/data_stores/__init__.py
@@ -0,0 +1,26 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class DataStores(object):
+    """The various data stores.
+
+    These are low level interfaces to physical databases.
+    """
+
+    def __init__(self, main_store, db_conn, hs):
+        # Note we pass in the main store here as workers use a different main
+        # store.
+        self.main = main_store
diff --git a/synapse/storage/data_stores/main/__init__.py b/synapse/storage/data_stores/main/__init__.py
new file mode 100644
index 0000000000..10c940df1e
--- /dev/null
+++ b/synapse/storage/data_stores/main/__init__.py
@@ -0,0 +1,533 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2018 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import calendar
+import logging
+import time
+
+from twisted.internet import defer
+
+from synapse.api.constants import PresenceState
+from synapse.storage.engines import PostgresEngine
+from synapse.storage.util.id_generators import (
+    ChainedIdGenerator,
+    IdGenerator,
+    StreamIdGenerator,
+)
+from synapse.util.caches.stream_change_cache import StreamChangeCache
+
+from .account_data import AccountDataStore
+from .appservice import ApplicationServiceStore, ApplicationServiceTransactionStore
+from .client_ips import ClientIpStore
+from .deviceinbox import DeviceInboxStore
+from .devices import DeviceStore
+from .directory import DirectoryStore
+from .e2e_room_keys import EndToEndRoomKeyStore
+from .end_to_end_keys import EndToEndKeyStore
+from .event_federation import EventFederationStore
+from .event_push_actions import EventPushActionsStore
+from .events import EventsStore
+from .events_bg_updates import EventsBackgroundUpdatesStore
+from .filtering import FilteringStore
+from .group_server import GroupServerStore
+from .keys import KeyStore
+from .media_repository import MediaRepositoryStore
+from .monthly_active_users import MonthlyActiveUsersStore
+from .openid import OpenIdStore
+from .presence import PresenceStore, UserPresenceState
+from .profile import ProfileStore
+from .push_rule import PushRuleStore
+from .pusher import PusherStore
+from .receipts import ReceiptsStore
+from .registration import RegistrationStore
+from .rejections import RejectionsStore
+from .relations import RelationsStore
+from .room import RoomStore
+from .roommember import RoomMemberStore
+from .search import SearchStore
+from .signatures import SignatureStore
+from .state import StateStore
+from .stats import StatsStore
+from .stream import StreamStore
+from .tags import TagsStore
+from .transactions import TransactionStore
+from .user_directory import UserDirectoryStore
+from .user_erasure_store import UserErasureStore
+
+logger = logging.getLogger(__name__)
+
+
+class DataStore(
+    EventsBackgroundUpdatesStore,
+    RoomMemberStore,
+    RoomStore,
+    RegistrationStore,
+    StreamStore,
+    ProfileStore,
+    PresenceStore,
+    TransactionStore,
+    DirectoryStore,
+    KeyStore,
+    StateStore,
+    SignatureStore,
+    ApplicationServiceStore,
+    EventsStore,
+    EventFederationStore,
+    MediaRepositoryStore,
+    RejectionsStore,
+    FilteringStore,
+    PusherStore,
+    PushRuleStore,
+    ApplicationServiceTransactionStore,
+    ReceiptsStore,
+    EndToEndKeyStore,
+    EndToEndRoomKeyStore,
+    SearchStore,
+    TagsStore,
+    AccountDataStore,
+    EventPushActionsStore,
+    OpenIdStore,
+    ClientIpStore,
+    DeviceStore,
+    DeviceInboxStore,
+    UserDirectoryStore,
+    GroupServerStore,
+    UserErasureStore,
+    MonthlyActiveUsersStore,
+    StatsStore,
+    RelationsStore,
+):
+    def __init__(self, db_conn, hs):
+        self.hs = hs
+        self._clock = hs.get_clock()
+        self.database_engine = hs.database_engine
+
+        self._stream_id_gen = StreamIdGenerator(
+            db_conn,
+            "events",
+            "stream_ordering",
+            extra_tables=[("local_invites", "stream_id")],
+        )
+        self._backfill_id_gen = StreamIdGenerator(
+            db_conn,
+            "events",
+            "stream_ordering",
+            step=-1,
+            extra_tables=[("ex_outlier_stream", "event_stream_ordering")],
+        )
+        self._presence_id_gen = StreamIdGenerator(
+            db_conn, "presence_stream", "stream_id"
+        )
+        self._device_inbox_id_gen = StreamIdGenerator(
+            db_conn, "device_max_stream_id", "stream_id"
+        )
+        self._public_room_id_gen = StreamIdGenerator(
+            db_conn, "public_room_list_stream", "stream_id"
+        )
+        self._device_list_id_gen = StreamIdGenerator(
+            db_conn,
+            "device_lists_stream",
+            "stream_id",
+            extra_tables=[("user_signature_stream", "stream_id")],
+        )
+        self._cross_signing_id_gen = StreamIdGenerator(
+            db_conn, "e2e_cross_signing_keys", "stream_id"
+        )
+
+        self._access_tokens_id_gen = IdGenerator(db_conn, "access_tokens", "id")
+        self._event_reports_id_gen = IdGenerator(db_conn, "event_reports", "id")
+        self._push_rule_id_gen = IdGenerator(db_conn, "push_rules", "id")
+        self._push_rules_enable_id_gen = IdGenerator(db_conn, "push_rules_enable", "id")
+        self._push_rules_stream_id_gen = ChainedIdGenerator(
+            self._stream_id_gen, db_conn, "push_rules_stream", "stream_id"
+        )
+        self._pushers_id_gen = StreamIdGenerator(
+            db_conn, "pushers", "id", extra_tables=[("deleted_pushers", "stream_id")]
+        )
+        self._group_updates_id_gen = StreamIdGenerator(
+            db_conn, "local_group_updates", "stream_id"
+        )
+
+        if isinstance(self.database_engine, PostgresEngine):
+            self._cache_id_gen = StreamIdGenerator(
+                db_conn, "cache_invalidation_stream", "stream_id"
+            )
+        else:
+            self._cache_id_gen = None
+
+        self._presence_on_startup = self._get_active_presence(db_conn)
+
+        presence_cache_prefill, min_presence_val = self._get_cache_dict(
+            db_conn,
+            "presence_stream",
+            entity_column="user_id",
+            stream_column="stream_id",
+            max_value=self._presence_id_gen.get_current_token(),
+        )
+        self.presence_stream_cache = StreamChangeCache(
+            "PresenceStreamChangeCache",
+            min_presence_val,
+            prefilled_cache=presence_cache_prefill,
+        )
+
+        max_device_inbox_id = self._device_inbox_id_gen.get_current_token()
+        device_inbox_prefill, min_device_inbox_id = self._get_cache_dict(
+            db_conn,
+            "device_inbox",
+            entity_column="user_id",
+            stream_column="stream_id",
+            max_value=max_device_inbox_id,
+            limit=1000,
+        )
+        self._device_inbox_stream_cache = StreamChangeCache(
+            "DeviceInboxStreamChangeCache",
+            min_device_inbox_id,
+            prefilled_cache=device_inbox_prefill,
+        )
+        # The federation outbox and the local device inbox uses the same
+        # stream_id generator.
+        device_outbox_prefill, min_device_outbox_id = self._get_cache_dict(
+            db_conn,
+            "device_federation_outbox",
+            entity_column="destination",
+            stream_column="stream_id",
+            max_value=max_device_inbox_id,
+            limit=1000,
+        )
+        self._device_federation_outbox_stream_cache = StreamChangeCache(
+            "DeviceFederationOutboxStreamChangeCache",
+            min_device_outbox_id,
+            prefilled_cache=device_outbox_prefill,
+        )
+
+        device_list_max = self._device_list_id_gen.get_current_token()
+        self._device_list_stream_cache = StreamChangeCache(
+            "DeviceListStreamChangeCache", device_list_max
+        )
+        self._user_signature_stream_cache = StreamChangeCache(
+            "UserSignatureStreamChangeCache", device_list_max
+        )
+        self._device_list_federation_stream_cache = StreamChangeCache(
+            "DeviceListFederationStreamChangeCache", device_list_max
+        )
+
+        events_max = self._stream_id_gen.get_current_token()
+        curr_state_delta_prefill, min_curr_state_delta_id = self._get_cache_dict(
+            db_conn,
+            "current_state_delta_stream",
+            entity_column="room_id",
+            stream_column="stream_id",
+            max_value=events_max,  # As we share the stream id with events token
+            limit=1000,
+        )
+        self._curr_state_delta_stream_cache = StreamChangeCache(
+            "_curr_state_delta_stream_cache",
+            min_curr_state_delta_id,
+            prefilled_cache=curr_state_delta_prefill,
+        )
+
+        _group_updates_prefill, min_group_updates_id = self._get_cache_dict(
+            db_conn,
+            "local_group_updates",
+            entity_column="user_id",
+            stream_column="stream_id",
+            max_value=self._group_updates_id_gen.get_current_token(),
+            limit=1000,
+        )
+        self._group_updates_stream_cache = StreamChangeCache(
+            "_group_updates_stream_cache",
+            min_group_updates_id,
+            prefilled_cache=_group_updates_prefill,
+        )
+
+        self._stream_order_on_start = self.get_room_max_stream_ordering()
+        self._min_stream_order_on_start = self.get_room_min_stream_ordering()
+
+        # Used in _generate_user_daily_visits to keep track of progress
+        self._last_user_visit_update = self._get_start_of_day()
+
+        super(DataStore, self).__init__(db_conn, hs)
+
+    def take_presence_startup_info(self):
+        active_on_startup = self._presence_on_startup
+        self._presence_on_startup = None
+        return active_on_startup
+
+    def _get_active_presence(self, db_conn):
+        """Fetch non-offline presence from the database so that we can register
+        the appropriate time outs.
+        """
+
+        sql = (
+            "SELECT user_id, state, last_active_ts, last_federation_update_ts,"
+            " last_user_sync_ts, status_msg, currently_active FROM presence_stream"
+            " WHERE state != ?"
+        )
+        sql = self.database_engine.convert_param_style(sql)
+
+        txn = db_conn.cursor()
+        txn.execute(sql, (PresenceState.OFFLINE,))
+        rows = self.cursor_to_dict(txn)
+        txn.close()
+
+        for row in rows:
+            row["currently_active"] = bool(row["currently_active"])
+
+        return [UserPresenceState(**row) for row in rows]
+
+    def count_daily_users(self):
+        """
+        Counts the number of users who used this homeserver in the last 24 hours.
+        """
+        yesterday = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24)
+        return self.runInteraction("count_daily_users", self._count_users, yesterday)
+
+    def count_monthly_users(self):
+        """
+        Counts the number of users who used this homeserver in the last 30 days.
+        Note this method is intended for phonehome metrics only and is different
+        from the mau figure in synapse.storage.monthly_active_users which,
+        amongst other things, includes a 3 day grace period before a user counts.
+        """
+        thirty_days_ago = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24 * 30)
+        return self.runInteraction(
+            "count_monthly_users", self._count_users, thirty_days_ago
+        )
+
+    def _count_users(self, txn, time_from):
+        """
+        Returns number of users seen in the past time_from period
+        """
+        sql = """
+            SELECT COALESCE(count(*), 0) FROM (
+                SELECT user_id FROM user_ips
+                WHERE last_seen > ?
+                GROUP BY user_id
+            ) u
+        """
+        txn.execute(sql, (time_from,))
+        (count,) = txn.fetchone()
+        return count
+
+    def count_r30_users(self):
+        """
+        Counts the number of 30 day retained users, defined as:-
+         * Users who have created their accounts more than 30 days ago
+         * Where last seen at most 30 days ago
+         * Where account creation and last_seen are > 30 days apart
+
+         Returns counts globaly for a given user as well as breaking
+         by platform
+        """
+
+        def _count_r30_users(txn):
+            thirty_days_in_secs = 86400 * 30
+            now = int(self._clock.time())
+            thirty_days_ago_in_secs = now - thirty_days_in_secs
+
+            sql = """
+                SELECT platform, COALESCE(count(*), 0) FROM (
+                     SELECT
+                        users.name, platform, users.creation_ts * 1000,
+                        MAX(uip.last_seen)
+                     FROM users
+                     INNER JOIN (
+                         SELECT
+                         user_id,
+                         last_seen,
+                         CASE
+                             WHEN user_agent LIKE '%%Android%%' THEN 'android'
+                             WHEN user_agent LIKE '%%iOS%%' THEN 'ios'
+                             WHEN user_agent LIKE '%%Electron%%' THEN 'electron'
+                             WHEN user_agent LIKE '%%Mozilla%%' THEN 'web'
+                             WHEN user_agent LIKE '%%Gecko%%' THEN 'web'
+                             ELSE 'unknown'
+                         END
+                         AS platform
+                         FROM user_ips
+                     ) uip
+                     ON users.name = uip.user_id
+                     AND users.appservice_id is NULL
+                     AND users.creation_ts < ?
+                     AND uip.last_seen/1000 > ?
+                     AND (uip.last_seen/1000) - users.creation_ts > 86400 * 30
+                     GROUP BY users.name, platform, users.creation_ts
+                ) u GROUP BY platform
+            """
+
+            results = {}
+            txn.execute(sql, (thirty_days_ago_in_secs, thirty_days_ago_in_secs))
+
+            for row in txn:
+                if row[0] == "unknown":
+                    pass
+                results[row[0]] = row[1]
+
+            sql = """
+                SELECT COALESCE(count(*), 0) FROM (
+                    SELECT users.name, users.creation_ts * 1000,
+                                                        MAX(uip.last_seen)
+                    FROM users
+                    INNER JOIN (
+                        SELECT
+                        user_id,
+                        last_seen
+                        FROM user_ips
+                    ) uip
+                    ON users.name = uip.user_id
+                    AND appservice_id is NULL
+                    AND users.creation_ts < ?
+                    AND uip.last_seen/1000 > ?
+                    AND (uip.last_seen/1000) - users.creation_ts > 86400 * 30
+                    GROUP BY users.name, users.creation_ts
+                ) u
+            """
+
+            txn.execute(sql, (thirty_days_ago_in_secs, thirty_days_ago_in_secs))
+
+            (count,) = txn.fetchone()
+            results["all"] = count
+
+            return results
+
+        return self.runInteraction("count_r30_users", _count_r30_users)
+
+    def _get_start_of_day(self):
+        """
+        Returns millisecond unixtime for start of UTC day.
+        """
+        now = time.gmtime()
+        today_start = calendar.timegm((now.tm_year, now.tm_mon, now.tm_mday, 0, 0, 0))
+        return today_start * 1000
+
+    def generate_user_daily_visits(self):
+        """
+        Generates daily visit data for use in cohort/ retention analysis
+        """
+
+        def _generate_user_daily_visits(txn):
+            logger.info("Calling _generate_user_daily_visits")
+            today_start = self._get_start_of_day()
+            a_day_in_milliseconds = 24 * 60 * 60 * 1000
+            now = self.clock.time_msec()
+
+            sql = """
+                INSERT INTO user_daily_visits (user_id, device_id, timestamp)
+                    SELECT u.user_id, u.device_id, ?
+                    FROM user_ips AS u
+                    LEFT JOIN (
+                      SELECT user_id, device_id, timestamp FROM user_daily_visits
+                      WHERE timestamp = ?
+                    ) udv
+                    ON u.user_id = udv.user_id AND u.device_id=udv.device_id
+                    INNER JOIN users ON users.name=u.user_id
+                    WHERE last_seen > ? AND last_seen <= ?
+                    AND udv.timestamp IS NULL AND users.is_guest=0
+                    AND users.appservice_id IS NULL
+                    GROUP BY u.user_id, u.device_id
+            """
+
+            # This means that the day has rolled over but there could still
+            # be entries from the previous day. There is an edge case
+            # where if the user logs in at 23:59 and overwrites their
+            # last_seen at 00:01 then they will not be counted in the
+            # previous day's stats - it is important that the query is run
+            # often to minimise this case.
+            if today_start > self._last_user_visit_update:
+                yesterday_start = today_start - a_day_in_milliseconds
+                txn.execute(
+                    sql,
+                    (
+                        yesterday_start,
+                        yesterday_start,
+                        self._last_user_visit_update,
+                        today_start,
+                    ),
+                )
+                self._last_user_visit_update = today_start
+
+            txn.execute(
+                sql, (today_start, today_start, self._last_user_visit_update, now)
+            )
+            # Update _last_user_visit_update to now. The reason to do this
+            # rather just clamping to the beginning of the day is to limit
+            # the size of the join - meaning that the query can be run more
+            # frequently
+            self._last_user_visit_update = now
+
+        return self.runInteraction(
+            "generate_user_daily_visits", _generate_user_daily_visits
+        )
+
+    def get_users(self):
+        """Function to reterive a list of users in users table.
+
+        Args:
+        Returns:
+            defer.Deferred: resolves to list[dict[str, Any]]
+        """
+        return self._simple_select_list(
+            table="users",
+            keyvalues={},
+            retcols=["name", "password_hash", "is_guest", "admin", "user_type"],
+            desc="get_users",
+        )
+
+    @defer.inlineCallbacks
+    def get_users_paginate(self, order, start, limit):
+        """Function to reterive a paginated list of users from
+        users list. This will return a json object, which contains
+        list of users and the total number of users in users table.
+
+        Args:
+            order (str): column name to order the select by this column
+            start (int): start number to begin the query from
+            limit (int): number of rows to reterive
+        Returns:
+            defer.Deferred: resolves to json object {list[dict[str, Any]], count}
+        """
+        users = yield self.runInteraction(
+            "get_users_paginate",
+            self._simple_select_list_paginate_txn,
+            table="users",
+            keyvalues={"is_guest": False},
+            orderby=order,
+            start=start,
+            limit=limit,
+            retcols=["name", "password_hash", "is_guest", "admin", "user_type"],
+        )
+        count = yield self.runInteraction("get_users_paginate", self.get_user_count_txn)
+        retval = {"users": users, "total": count}
+        return retval
+
+    def search_users(self, term):
+        """Function to search users list for one or more users with
+        the matched term.
+
+        Args:
+            term (str): search term
+            col (str): column to query term should be matched to
+        Returns:
+            defer.Deferred: resolves to list[dict[str, Any]]
+        """
+        return self._simple_search_list(
+            table="users",
+            term=term,
+            col="name",
+            retcols=["name", "password_hash", "is_guest", "admin", "user_type"],
+            desc="search_users",
+        )
diff --git a/synapse/storage/account_data.py b/synapse/storage/data_stores/main/account_data.py
similarity index 100%
rename from synapse/storage/account_data.py
rename to synapse/storage/data_stores/main/account_data.py
diff --git a/synapse/storage/appservice.py b/synapse/storage/data_stores/main/appservice.py
similarity index 99%
rename from synapse/storage/appservice.py
rename to synapse/storage/data_stores/main/appservice.py
index 36657753cd..81babf2029 100644
--- a/synapse/storage/appservice.py
+++ b/synapse/storage/data_stores/main/appservice.py
@@ -22,9 +22,8 @@ from twisted.internet import defer
 
 from synapse.appservice import AppServiceTransaction
 from synapse.config.appservice import load_appservices
-from synapse.storage.events_worker import EventsWorkerStore
-
-from ._base import SQLBaseStore
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.data_stores.main.events_worker import EventsWorkerStore
 
 logger = logging.getLogger(__name__)
 
@@ -165,7 +164,6 @@ class ApplicationServiceTransactionWorkerStore(
         )
         if result:
             return result.get("state")
-            return
         return None
 
     def set_appservice_state(self, service, state):
diff --git a/synapse/storage/client_ips.py b/synapse/storage/data_stores/main/client_ips.py
similarity index 69%
rename from synapse/storage/client_ips.py
rename to synapse/storage/data_stores/main/client_ips.py
index 6db8c54077..706c6a1f3f 100644
--- a/synapse/storage/client_ips.py
+++ b/synapse/storage/data_stores/main/client_ips.py
@@ -19,12 +19,11 @@ from six import iteritems
 
 from twisted.internet import defer
 
-from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.metrics.background_process_metrics import wrap_as_background_process
+from synapse.storage import background_updates
+from synapse.storage._base import Cache
 from synapse.util.caches import CACHE_SIZE_FACTOR
 
-from . import background_updates
-from ._base import Cache
-
 logger = logging.getLogger(__name__)
 
 # Number of msec of granularity to store the user IP 'last seen' time. Smaller
@@ -33,14 +32,9 @@ logger = logging.getLogger(__name__)
 LAST_SEEN_GRANULARITY = 120 * 1000
 
 
-class ClientIpStore(background_updates.BackgroundUpdateStore):
+class ClientIpBackgroundUpdateStore(background_updates.BackgroundUpdateStore):
     def __init__(self, db_conn, hs):
-
-        self.client_ip_last_seen = Cache(
-            name="client_ip_last_seen", keylen=4, max_entries=50000 * CACHE_SIZE_FACTOR
-        )
-
-        super(ClientIpStore, self).__init__(db_conn, hs)
+        super(ClientIpBackgroundUpdateStore, self).__init__(db_conn, hs)
 
         self.register_background_index_update(
             "user_ips_device_index",
@@ -85,14 +79,9 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
             "user_ips_drop_nonunique_index", self._remove_user_ip_nonunique
         )
 
-        # (user_id, access_token, ip,) -> (user_agent, device_id, last_seen)
-        self._batch_row_update = {}
-
-        self._client_ip_looper = self._clock.looping_call(
-            self._update_client_ips_batch, 5 * 1000
-        )
-        self.hs.get_reactor().addSystemEventTrigger(
-            "before", "shutdown", self._update_client_ips_batch
+        # Update the last seen info in devices.
+        self.register_background_update_handler(
+            "devices_last_seen", self._devices_last_seen_update
         )
 
     @defer.inlineCallbacks
@@ -293,6 +282,110 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
 
         return batch_size
 
+    @defer.inlineCallbacks
+    def _devices_last_seen_update(self, progress, batch_size):
+        """Background update to insert last seen info into devices table
+        """
+
+        last_user_id = progress.get("last_user_id", "")
+        last_device_id = progress.get("last_device_id", "")
+
+        def _devices_last_seen_update_txn(txn):
+            # This consists of two queries:
+            #
+            #   1. The sub-query searches for the next N devices and joins
+            #      against user_ips to find the max last_seen associated with
+            #      that device.
+            #   2. The outer query then joins again against user_ips on
+            #      user/device/last_seen. This *should* hopefully only
+            #      return one row, but if it does return more than one then
+            #      we'll just end up updating the same device row multiple
+            #      times, which is fine.
+
+            if self.database_engine.supports_tuple_comparison:
+                where_clause = "(user_id, device_id) > (?, ?)"
+                where_args = [last_user_id, last_device_id]
+            else:
+                # We explicitly do a `user_id >= ? AND (...)` here to ensure
+                # that an index is used, as doing `user_id > ? OR (user_id = ? AND ...)`
+                # makes it hard for query optimiser to tell that it can use the
+                # index on user_id
+                where_clause = "user_id >= ? AND (user_id > ? OR device_id > ?)"
+                where_args = [last_user_id, last_user_id, last_device_id]
+
+            sql = """
+                SELECT
+                    last_seen, ip, user_agent, user_id, device_id
+                FROM (
+                    SELECT
+                        user_id, device_id, MAX(u.last_seen) AS last_seen
+                    FROM devices
+                    INNER JOIN user_ips AS u USING (user_id, device_id)
+                    WHERE %(where_clause)s
+                    GROUP BY user_id, device_id
+                    ORDER BY user_id ASC, device_id ASC
+                    LIMIT ?
+                ) c
+                INNER JOIN user_ips AS u USING (user_id, device_id, last_seen)
+            """ % {
+                "where_clause": where_clause
+            }
+            txn.execute(sql, where_args + [batch_size])
+
+            rows = txn.fetchall()
+            if not rows:
+                return 0
+
+            sql = """
+                UPDATE devices
+                SET last_seen = ?, ip = ?, user_agent = ?
+                WHERE user_id = ? AND device_id = ?
+            """
+            txn.execute_batch(sql, rows)
+
+            _, _, _, user_id, device_id = rows[-1]
+            self._background_update_progress_txn(
+                txn,
+                "devices_last_seen",
+                {"last_user_id": user_id, "last_device_id": device_id},
+            )
+
+            return len(rows)
+
+        updated = yield self.runInteraction(
+            "_devices_last_seen_update", _devices_last_seen_update_txn
+        )
+
+        if not updated:
+            yield self._end_background_update("devices_last_seen")
+
+        return updated
+
+
+class ClientIpStore(ClientIpBackgroundUpdateStore):
+    def __init__(self, db_conn, hs):
+
+        self.client_ip_last_seen = Cache(
+            name="client_ip_last_seen", keylen=4, max_entries=50000 * CACHE_SIZE_FACTOR
+        )
+
+        super(ClientIpStore, self).__init__(db_conn, hs)
+
+        self.user_ips_max_age = hs.config.user_ips_max_age
+
+        # (user_id, access_token, ip,) -> (user_agent, device_id, last_seen)
+        self._batch_row_update = {}
+
+        self._client_ip_looper = self._clock.looping_call(
+            self._update_client_ips_batch, 5 * 1000
+        )
+        self.hs.get_reactor().addSystemEventTrigger(
+            "before", "shutdown", self._update_client_ips_batch
+        )
+
+        if self.user_ips_max_age:
+            self._clock.looping_call(self._prune_old_user_ips, 5 * 1000)
+
     @defer.inlineCallbacks
     def insert_client_ip(
         self, user_id, access_token, ip, user_agent, device_id, now=None
@@ -314,20 +407,19 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
 
         self._batch_row_update[key] = (user_agent, device_id, now)
 
+    @wrap_as_background_process("update_client_ips")
     def _update_client_ips_batch(self):
 
         # If the DB pool has already terminated, don't try updating
         if not self.hs.get_db_pool().running:
             return
 
-        def update():
-            to_update = self._batch_row_update
-            self._batch_row_update = {}
-            return self.runInteraction(
-                "_update_client_ips_batch", self._update_client_ips_batch_txn, to_update
-            )
+        to_update = self._batch_row_update
+        self._batch_row_update = {}
 
-        return run_as_background_process("update_client_ips", update)
+        return self.runInteraction(
+            "_update_client_ips_batch", self._update_client_ips_batch_txn, to_update
+        )
 
     def _update_client_ips_batch_txn(self, txn, to_update):
         if "user_ips" in self._unsafe_to_upsert_tables or (
@@ -354,6 +446,21 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
                     },
                     lock=False,
                 )
+
+                # Technically an access token might not be associated with
+                # a device so we need to check.
+                if device_id:
+                    self._simple_upsert_txn(
+                        txn,
+                        table="devices",
+                        keyvalues={"user_id": user_id, "device_id": device_id},
+                        values={
+                            "user_agent": user_agent,
+                            "last_seen": last_seen,
+                            "ip": ip,
+                        },
+                        lock=False,
+                    )
             except Exception as e:
                 # Failed to upsert, log and continue
                 logger.error("Failed to insert client IP %r: %r", entry, e)
@@ -372,19 +479,14 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
             keys giving the column names
         """
 
-        res = yield self.runInteraction(
-            "get_last_client_ip_by_device",
-            self._get_last_client_ip_by_device_txn,
-            user_id,
-            device_id,
-            retcols=(
-                "user_id",
-                "access_token",
-                "ip",
-                "user_agent",
-                "device_id",
-                "last_seen",
-            ),
+        keyvalues = {"user_id": user_id}
+        if device_id is not None:
+            keyvalues["device_id"] = device_id
+
+        res = yield self._simple_select_list(
+            table="devices",
+            keyvalues=keyvalues,
+            retcols=("user_id", "ip", "user_agent", "device_id", "last_seen"),
         )
 
         ret = {(d["user_id"], d["device_id"]): d for d in res}
@@ -403,42 +505,6 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
                     }
         return ret
 
-    @classmethod
-    def _get_last_client_ip_by_device_txn(cls, txn, user_id, device_id, retcols):
-        where_clauses = []
-        bindings = []
-        if device_id is None:
-            where_clauses.append("user_id = ?")
-            bindings.extend((user_id,))
-        else:
-            where_clauses.append("(user_id = ? AND device_id = ?)")
-            bindings.extend((user_id, device_id))
-
-        if not where_clauses:
-            return []
-
-        inner_select = (
-            "SELECT MAX(last_seen) mls, user_id, device_id FROM user_ips "
-            "WHERE %(where)s "
-            "GROUP BY user_id, device_id"
-        ) % {"where": " OR ".join(where_clauses)}
-
-        sql = (
-            "SELECT %(retcols)s FROM user_ips "
-            "JOIN (%(inner_select)s) ips ON"
-            "    user_ips.last_seen = ips.mls AND"
-            "    user_ips.user_id = ips.user_id AND"
-            "    (user_ips.device_id = ips.device_id OR"
-            "         (user_ips.device_id IS NULL AND ips.device_id IS NULL)"
-            "    )"
-        ) % {
-            "retcols": ",".join("user_ips." + c for c in retcols),
-            "inner_select": inner_select,
-        }
-
-        txn.execute(sql, bindings)
-        return cls.cursor_to_dict(txn)
-
     @defer.inlineCallbacks
     def get_user_ip_and_agents(self, user):
         user_id = user.to_string()
@@ -470,3 +536,45 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
             }
             for (access_token, ip), (user_agent, last_seen) in iteritems(results)
         )
+
+    @wrap_as_background_process("prune_old_user_ips")
+    async def _prune_old_user_ips(self):
+        """Removes entries in user IPs older than the configured period.
+        """
+
+        if self.user_ips_max_age is None:
+            # Nothing to do
+            return
+
+        if not await self.has_completed_background_update("devices_last_seen"):
+            # Only start pruning if we have finished populating the devices
+            # last seen info.
+            return
+
+        # We do a slightly funky SQL delete to ensure we don't try and delete
+        # too much at once (as the table may be very large from before we
+        # started pruning).
+        #
+        # This works by finding the max last_seen that is less than the given
+        # time, but has no more than N rows before it, deleting all rows with
+        # a lesser last_seen time. (We COALESCE so that the sub-SELECT always
+        # returns exactly one row).
+        sql = """
+            DELETE FROM user_ips
+            WHERE last_seen <= (
+                SELECT COALESCE(MAX(last_seen), -1)
+                FROM (
+                    SELECT last_seen FROM user_ips
+                    WHERE last_seen <= ?
+                    ORDER BY last_seen ASC
+                    LIMIT 5000
+                ) AS u
+            )
+        """
+
+        timestamp = self.clock.time_msec() - self.user_ips_max_age
+
+        def _prune_old_user_ips_txn(txn):
+            txn.execute(sql, (timestamp,))
+
+        await self.runInteraction("_prune_old_user_ips", _prune_old_user_ips_txn)
diff --git a/synapse/storage/deviceinbox.py b/synapse/storage/data_stores/main/deviceinbox.py
similarity index 92%
rename from synapse/storage/deviceinbox.py
rename to synapse/storage/data_stores/main/deviceinbox.py
index 4dca9de617..f04aad0743 100644
--- a/synapse/storage/deviceinbox.py
+++ b/synapse/storage/data_stores/main/deviceinbox.py
@@ -19,7 +19,8 @@ from canonicaljson import json
 
 from twisted.internet import defer
 
-from synapse.storage._base import SQLBaseStore
+from synapse.logging.opentracing import log_kv, set_tag, trace
+from synapse.storage._base import SQLBaseStore, make_in_list_sql_clause
 from synapse.storage.background_updates import BackgroundUpdateStore
 from synapse.util.caches.expiringcache import ExpiringCache
 
@@ -72,6 +73,7 @@ class DeviceInboxWorkerStore(SQLBaseStore):
             "get_new_messages_for_device", get_new_messages_for_device_txn
         )
 
+    @trace
     @defer.inlineCallbacks
     def delete_messages_for_device(self, user_id, device_id, up_to_stream_id):
         """
@@ -87,11 +89,15 @@ class DeviceInboxWorkerStore(SQLBaseStore):
         last_deleted_stream_id = self._last_device_delete_cache.get(
             (user_id, device_id), None
         )
+
+        set_tag("last_deleted_stream_id", last_deleted_stream_id)
+
         if last_deleted_stream_id:
             has_changed = self._device_inbox_stream_cache.has_entity_changed(
                 user_id, last_deleted_stream_id
             )
             if not has_changed:
+                log_kv({"message": "No changes in cache since last check"})
                 return 0
 
         def delete_messages_for_device_txn(txn):
@@ -107,6 +113,10 @@ class DeviceInboxWorkerStore(SQLBaseStore):
             "delete_messages_for_device", delete_messages_for_device_txn
         )
 
+        log_kv(
+            {"message": "deleted {} messages for device".format(count), "count": count}
+        )
+
         # Update the cache, ensuring that we only ever increase the value
         last_deleted_stream_id = self._last_device_delete_cache.get(
             (user_id, device_id), 0
@@ -117,6 +127,7 @@ class DeviceInboxWorkerStore(SQLBaseStore):
 
         return count
 
+    @trace
     def get_new_device_msgs_for_remote(
         self, destination, last_stream_id, current_stream_id, limit
     ):
@@ -132,16 +143,23 @@ class DeviceInboxWorkerStore(SQLBaseStore):
                 in the stream the messages got to.
         """
 
+        set_tag("destination", destination)
+        set_tag("last_stream_id", last_stream_id)
+        set_tag("current_stream_id", current_stream_id)
+        set_tag("limit", limit)
+
         has_changed = self._device_federation_outbox_stream_cache.has_entity_changed(
             destination, last_stream_id
         )
         if not has_changed or last_stream_id == current_stream_id:
+            log_kv({"message": "No new messages in stream"})
             return defer.succeed(([], current_stream_id))
 
         if limit <= 0:
             # This can happen if we run out of room for EDUs in the transaction.
             return defer.succeed(([], last_stream_id))
 
+        @trace
         def get_new_messages_for_remote_destination_txn(txn):
             sql = (
                 "SELECT stream_id, messages_json FROM device_federation_outbox"
@@ -156,6 +174,7 @@ class DeviceInboxWorkerStore(SQLBaseStore):
                 stream_pos = row[0]
                 messages.append(json.loads(row[1]))
             if len(messages) < limit:
+                log_kv({"message": "Set stream position to current position"})
                 stream_pos = current_stream_id
             return messages, stream_pos
 
@@ -164,6 +183,7 @@ class DeviceInboxWorkerStore(SQLBaseStore):
             get_new_messages_for_remote_destination_txn,
         )
 
+    @trace
     def delete_device_msgs_for_remote(self, destination, up_to_stream_id):
         """Used to delete messages when the remote destination acknowledges
         their receipt.
@@ -188,11 +208,11 @@ class DeviceInboxWorkerStore(SQLBaseStore):
         )
 
 
-class DeviceInboxStore(DeviceInboxWorkerStore, BackgroundUpdateStore):
+class DeviceInboxBackgroundUpdateStore(BackgroundUpdateStore):
     DEVICE_INBOX_STREAM_ID = "device_inbox_stream_drop"
 
     def __init__(self, db_conn, hs):
-        super(DeviceInboxStore, self).__init__(db_conn, hs)
+        super(DeviceInboxBackgroundUpdateStore, self).__init__(db_conn, hs)
 
         self.register_background_index_update(
             "device_inbox_stream_index",
@@ -205,6 +225,26 @@ class DeviceInboxStore(DeviceInboxWorkerStore, BackgroundUpdateStore):
             self.DEVICE_INBOX_STREAM_ID, self._background_drop_index_device_inbox
         )
 
+    @defer.inlineCallbacks
+    def _background_drop_index_device_inbox(self, progress, batch_size):
+        def reindex_txn(conn):
+            txn = conn.cursor()
+            txn.execute("DROP INDEX IF EXISTS device_inbox_stream_id")
+            txn.close()
+
+        yield self.runWithConnection(reindex_txn)
+
+        yield self._end_background_update(self.DEVICE_INBOX_STREAM_ID)
+
+        return 1
+
+
+class DeviceInboxStore(DeviceInboxWorkerStore, DeviceInboxBackgroundUpdateStore):
+    DEVICE_INBOX_STREAM_ID = "device_inbox_stream_drop"
+
+    def __init__(self, db_conn, hs):
+        super(DeviceInboxStore, self).__init__(db_conn, hs)
+
         # Map of (user_id, device_id) to the last stream_id that has been
         # deleted up to. This is so that we can no op deletions.
         self._last_device_delete_cache = ExpiringCache(
@@ -214,6 +254,7 @@ class DeviceInboxStore(DeviceInboxWorkerStore, BackgroundUpdateStore):
             expiry_ms=30 * 60 * 1000,
         )
 
+    @trace
     @defer.inlineCallbacks
     def add_messages_to_device_inbox(
         self, local_messages_by_user_then_device, remote_messages_by_destination
@@ -337,15 +378,15 @@ class DeviceInboxStore(DeviceInboxWorkerStore, BackgroundUpdateStore):
             else:
                 if not devices:
                     continue
-                sql = (
-                    "SELECT device_id FROM devices"
-                    " WHERE user_id = ? AND device_id IN ("
-                    + ",".join("?" * len(devices))
-                    + ")"
+
+                clause, args = make_in_list_sql_clause(
+                    txn.database_engine, "device_id", devices
                 )
+                sql = "SELECT device_id FROM devices WHERE user_id = ? AND " + clause
+
                 # TODO: Maybe this needs to be done in batches if there are
                 # too many local devices for a given user.
-                txn.execute(sql, [user_id] + devices)
+                txn.execute(sql, [user_id] + list(args))
                 for row in txn:
                     # Only insert into the local inbox if the device exists on
                     # this server
@@ -414,16 +455,3 @@ class DeviceInboxStore(DeviceInboxWorkerStore, BackgroundUpdateStore):
         return self.runInteraction(
             "get_all_new_device_messages", get_all_new_device_messages_txn
         )
-
-    @defer.inlineCallbacks
-    def _background_drop_index_device_inbox(self, progress, batch_size):
-        def reindex_txn(conn):
-            txn = conn.cursor()
-            txn.execute("DROP INDEX IF EXISTS device_inbox_stream_id")
-            txn.close()
-
-        yield self.runWithConnection(reindex_txn)
-
-        yield self._end_background_update(self.DEVICE_INBOX_STREAM_ID)
-
-        return 1
diff --git a/synapse/storage/devices.py b/synapse/storage/data_stores/main/devices.py
similarity index 79%
rename from synapse/storage/devices.py
rename to synapse/storage/data_stores/main/devices.py
index 76542c512d..71f62036c0 100644
--- a/synapse/storage/devices.py
+++ b/synapse/storage/data_stores/main/devices.py
@@ -1,5 +1,7 @@
 # -*- coding: utf-8 -*-
 # Copyright 2016 OpenMarket Ltd
+# Copyright 2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -20,15 +22,22 @@ from canonicaljson import json
 
 from twisted.internet import defer
 
-from synapse.api.errors import StoreError
+from synapse.api.errors import Codes, StoreError
 from synapse.logging.opentracing import (
     get_active_span_text_map,
+    set_tag,
     trace,
     whitelisted_homeserver,
 )
 from synapse.metrics.background_process_metrics import run_as_background_process
-from synapse.storage._base import Cache, SQLBaseStore, db_to_json
+from synapse.storage._base import (
+    Cache,
+    SQLBaseStore,
+    db_to_json,
+    make_in_list_sql_clause,
+)
 from synapse.storage.background_updates import BackgroundUpdateStore
+from synapse.types import get_verify_key_from_cross_signing_key
 from synapse.util import batch_iter
 from synapse.util.caches.descriptors import cached, cachedInlineCallbacks, cachedList
 
@@ -41,7 +50,8 @@ DROP_DEVICE_LIST_STREAMS_NON_UNIQUE_INDEXES = (
 
 class DeviceWorkerStore(SQLBaseStore):
     def get_device(self, user_id, device_id):
-        """Retrieve a device.
+        """Retrieve a device. Only returns devices that are not marked as
+        hidden.
 
         Args:
             user_id (str): The ID of the user which owns the device
@@ -53,14 +63,15 @@ class DeviceWorkerStore(SQLBaseStore):
         """
         return self._simple_select_one(
             table="devices",
-            keyvalues={"user_id": user_id, "device_id": device_id},
+            keyvalues={"user_id": user_id, "device_id": device_id, "hidden": False},
             retcols=("user_id", "device_id", "display_name"),
             desc="get_device",
         )
 
     @defer.inlineCallbacks
     def get_devices_by_user(self, user_id):
-        """Retrieve all of a user's registered devices.
+        """Retrieve all of a user's registered devices. Only returns devices
+        that are not marked as hidden.
 
         Args:
             user_id (str):
@@ -71,7 +82,7 @@ class DeviceWorkerStore(SQLBaseStore):
         """
         devices = yield self._simple_select_list(
             table="devices",
-            keyvalues={"user_id": user_id},
+            keyvalues={"user_id": user_id, "hidden": False},
             retcols=("user_id", "device_id", "display_name"),
             desc="get_devices_by_user",
         )
@@ -80,13 +91,18 @@ class DeviceWorkerStore(SQLBaseStore):
 
     @trace
     @defer.inlineCallbacks
-    def get_devices_by_remote(self, destination, from_stream_id, limit):
-        """Get stream of updates to send to remote servers
+    def get_device_updates_by_remote(self, destination, from_stream_id, limit):
+        """Get a stream of device updates to send to the given remote server.
 
+        Args:
+            destination (str): The host the device updates are intended for
+            from_stream_id (int): The minimum stream_id to filter updates by, exclusive
+            limit (int): Maximum number of device updates to return
         Returns:
-            Deferred[tuple[int, list[dict]]]:
+            Deferred[tuple[int, list[tuple[string,dict]]]]:
                 current stream id (ie, the stream id of the last update included in the
-                response), and the list of updates
+                response), and the list of updates, where each update is a pair of EDU
+                type and EDU contents
         """
         now_stream_id = self._device_list_id_gen.get_current_token()
 
@@ -107,8 +123,8 @@ class DeviceWorkerStore(SQLBaseStore):
         # stream_id; the rationale being that such a large device list update
         # is likely an error.
         updates = yield self.runInteraction(
-            "get_devices_by_remote",
-            self._get_devices_by_remote_txn,
+            "get_device_updates_by_remote",
+            self._get_device_updates_by_remote_txn,
             destination,
             from_stream_id,
             now_stream_id,
@@ -119,6 +135,37 @@ class DeviceWorkerStore(SQLBaseStore):
         if not updates:
             return now_stream_id, []
 
+        # get the cross-signing keys of the users in the list, so that we can
+        # determine which of the device changes were cross-signing keys
+        users = set(r[0] for r in updates)
+        master_key_by_user = {}
+        self_signing_key_by_user = {}
+        for user in users:
+            cross_signing_key = yield self.get_e2e_cross_signing_key(user, "master")
+            if cross_signing_key:
+                key_id, verify_key = get_verify_key_from_cross_signing_key(
+                    cross_signing_key
+                )
+                # verify_key is a VerifyKey from signedjson, which uses
+                # .version to denote the portion of the key ID after the
+                # algorithm and colon, which is the device ID
+                master_key_by_user[user] = {
+                    "key_info": cross_signing_key,
+                    "device_id": verify_key.version,
+                }
+
+            cross_signing_key = yield self.get_e2e_cross_signing_key(
+                user, "self_signing"
+            )
+            if cross_signing_key:
+                key_id, verify_key = get_verify_key_from_cross_signing_key(
+                    cross_signing_key
+                )
+                self_signing_key_by_user[user] = {
+                    "key_info": cross_signing_key,
+                    "device_id": verify_key.version,
+                }
+
         # if we have exceeded the limit, we need to exclude any results with the
         # same stream_id as the last row.
         if len(updates) > limit:
@@ -143,20 +190,33 @@ class DeviceWorkerStore(SQLBaseStore):
         # context which created the Edu.
 
         query_map = {}
-        for update in updates:
-            if stream_id_cutoff is not None and update[2] >= stream_id_cutoff:
+        cross_signing_keys_by_user = {}
+        for user_id, device_id, update_stream_id, update_context in updates:
+            if stream_id_cutoff is not None and update_stream_id >= stream_id_cutoff:
                 # Stop processing updates
                 break
 
-            key = (update[0], update[1])
+            if (
+                user_id in master_key_by_user
+                and device_id == master_key_by_user[user_id]["device_id"]
+            ):
+                result = cross_signing_keys_by_user.setdefault(user_id, {})
+                result["master_key"] = master_key_by_user[user_id]["key_info"]
+            elif (
+                user_id in self_signing_key_by_user
+                and device_id == self_signing_key_by_user[user_id]["device_id"]
+            ):
+                result = cross_signing_keys_by_user.setdefault(user_id, {})
+                result["self_signing_key"] = self_signing_key_by_user[user_id][
+                    "key_info"
+                ]
+            else:
+                key = (user_id, device_id)
 
-            update_context = update[3]
-            update_stream_id = update[2]
+                previous_update_stream_id, _ = query_map.get(key, (0, None))
 
-            previous_update_stream_id, _ = query_map.get(key, (0, None))
-
-            if update_stream_id > previous_update_stream_id:
-                query_map[key] = (update_stream_id, update_context)
+                if update_stream_id > previous_update_stream_id:
+                    query_map[key] = (update_stream_id, update_context)
 
         # If we didn't find any updates with a stream_id lower than the cutoff, it
         # means that there are more than limit updates all of which have the same
@@ -166,16 +226,22 @@ class DeviceWorkerStore(SQLBaseStore):
         # devices, in which case E2E isn't going to work well anyway. We'll just
         # skip that stream_id and return an empty list, and continue with the next
         # stream_id next time.
-        if not query_map:
+        if not query_map and not cross_signing_keys_by_user:
             return stream_id_cutoff, []
 
         results = yield self._get_device_update_edus_by_remote(
             destination, from_stream_id, query_map
         )
 
+        # add the updated cross-signing keys to the results list
+        for user_id, result in iteritems(cross_signing_keys_by_user):
+            result["user_id"] = user_id
+            # FIXME: switch to m.signing_key_update when MSC1756 is merged into the spec
+            results.append(("org.matrix.signing_key_update", result))
+
         return now_stream_id, results
 
-    def _get_devices_by_remote_txn(
+    def _get_device_updates_by_remote_txn(
         self, txn, destination, from_stream_id, now_stream_id, limit
     ):
         """Return device update information for a given remote destination
@@ -190,6 +256,7 @@ class DeviceWorkerStore(SQLBaseStore):
         Returns:
             List: List of device updates
         """
+        # get the list of device updates that need to be sent
         sql = """
             SELECT user_id, device_id, stream_id, opentracing_context FROM device_lists_outbound_pokes
             WHERE destination = ? AND ? < stream_id AND stream_id <= ? AND sent = ?
@@ -215,12 +282,16 @@ class DeviceWorkerStore(SQLBaseStore):
             List[Dict]: List of objects representing an device update EDU
 
         """
-        devices = yield self.runInteraction(
-            "_get_e2e_device_keys_txn",
-            self._get_e2e_device_keys_txn,
-            query_map.keys(),
-            include_all_devices=True,
-            include_deleted_devices=True,
+        devices = (
+            yield self.runInteraction(
+                "_get_e2e_device_keys_txn",
+                self._get_e2e_device_keys_txn,
+                query_map.keys(),
+                include_all_devices=True,
+                include_deleted_devices=True,
+            )
+            if query_map
+            else {}
         )
 
         results = []
@@ -252,7 +323,7 @@ class DeviceWorkerStore(SQLBaseStore):
                 else:
                     result["deleted"] = True
 
-                results.append(result)
+                results.append(("m.device_list_update", result))
 
         return results
 
@@ -318,9 +389,45 @@ class DeviceWorkerStore(SQLBaseStore):
         """
         txn.execute(sql, (destination, stream_id))
 
+    @defer.inlineCallbacks
+    def add_user_signature_change_to_streams(self, from_user_id, user_ids):
+        """Persist that a user has made new signatures
+
+        Args:
+            from_user_id (str): the user who made the signatures
+            user_ids (list[str]): the users who were signed
+        """
+
+        with self._device_list_id_gen.get_next() as stream_id:
+            yield self.runInteraction(
+                "add_user_sig_change_to_streams",
+                self._add_user_signature_change_txn,
+                from_user_id,
+                user_ids,
+                stream_id,
+            )
+        return stream_id
+
+    def _add_user_signature_change_txn(self, txn, from_user_id, user_ids, stream_id):
+        txn.call_after(
+            self._user_signature_stream_cache.entity_has_changed,
+            from_user_id,
+            stream_id,
+        )
+        self._simple_insert_txn(
+            txn,
+            "user_signature_stream",
+            values={
+                "stream_id": stream_id,
+                "from_user_id": from_user_id,
+                "user_ids": json.dumps(user_ids),
+            },
+        )
+
     def get_device_stream_token(self):
         return self._device_list_id_gen.get_current_token()
 
+    @trace
     @defer.inlineCallbacks
     def get_user_devices_from_cache(self, query_list):
         """Get the devices (and keys if any) for remote users from the cache.
@@ -352,6 +459,9 @@ class DeviceWorkerStore(SQLBaseStore):
             else:
                 results[user_id] = yield self._get_cached_devices_for_user(user_id)
 
+        set_tag("in_cache", results)
+        set_tag("not_in_cache", user_ids_not_in_cache)
+
         return user_ids_not_in_cache, results
 
     @cachedInlineCallbacks(num_args=2, tree=True)
@@ -443,11 +553,14 @@ class DeviceWorkerStore(SQLBaseStore):
             sql = """
                 SELECT DISTINCT user_id FROM device_lists_stream
                 WHERE stream_id > ?
-                AND user_id IN (%s)
+                AND
             """
 
             for chunk in batch_iter(to_check, 100):
-                txn.execute(sql % (",".join("?" for _ in chunk),), (from_key,) + chunk)
+                clause, args = make_in_list_sql_clause(
+                    txn.database_engine, "user_id", chunk
+                )
+                txn.execute(sql + clause, (from_key,) + tuple(args))
                 changes.update(user_id for user_id, in txn)
 
             return changes
@@ -456,6 +569,28 @@ class DeviceWorkerStore(SQLBaseStore):
             "get_users_whose_devices_changed", _get_users_whose_devices_changed_txn
         )
 
+    @defer.inlineCallbacks
+    def get_users_whose_signatures_changed(self, user_id, from_key):
+        """Get the users who have new cross-signing signatures made by `user_id` since
+        `from_key`.
+
+        Args:
+            user_id (str): the user who made the signatures
+            from_key (str): The device lists stream token
+        """
+        from_key = int(from_key)
+        if self._user_signature_stream_cache.has_entity_changed(user_id, from_key):
+            sql = """
+                SELECT DISTINCT user_ids FROM user_signature_stream
+                WHERE from_user_id = ? AND stream_id > ?
+            """
+            rows = yield self._execute(
+                "get_users_whose_signatures_changed", None, sql, user_id, from_key
+            )
+            return set(user for row in rows for user in json.loads(row[0]))
+        else:
+            return set()
+
     def get_all_device_list_changes_for_remotes(self, from_key, to_key):
         """Return a list of `(stream_id, user_id, destination)` which is the
         combined list of changes to devices, and which destinations need to be
@@ -507,17 +642,9 @@ class DeviceWorkerStore(SQLBaseStore):
         return results
 
 
-class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
+class DeviceBackgroundUpdateStore(BackgroundUpdateStore):
     def __init__(self, db_conn, hs):
-        super(DeviceStore, self).__init__(db_conn, hs)
-
-        # Map of (user_id, device_id) -> bool. If there is an entry that implies
-        # the device exists.
-        self.device_id_exists_cache = Cache(
-            name="device_id_exists", keylen=2, max_entries=10000
-        )
-
-        self._clock.looping_call(self._prune_old_outbound_device_pokes, 60 * 60 * 1000)
+        super(DeviceBackgroundUpdateStore, self).__init__(db_conn, hs)
 
         self.register_background_index_update(
             "device_lists_stream_idx",
@@ -550,6 +677,31 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
             self._drop_device_list_streams_non_unique_indexes,
         )
 
+    @defer.inlineCallbacks
+    def _drop_device_list_streams_non_unique_indexes(self, progress, batch_size):
+        def f(conn):
+            txn = conn.cursor()
+            txn.execute("DROP INDEX IF EXISTS device_lists_remote_cache_id")
+            txn.execute("DROP INDEX IF EXISTS device_lists_remote_extremeties_id")
+            txn.close()
+
+        yield self.runWithConnection(f)
+        yield self._end_background_update(DROP_DEVICE_LIST_STREAMS_NON_UNIQUE_INDEXES)
+        return 1
+
+
+class DeviceStore(DeviceWorkerStore, DeviceBackgroundUpdateStore):
+    def __init__(self, db_conn, hs):
+        super(DeviceStore, self).__init__(db_conn, hs)
+
+        # Map of (user_id, device_id) -> bool. If there is an entry that implies
+        # the device exists.
+        self.device_id_exists_cache = Cache(
+            name="device_id_exists", keylen=2, max_entries=10000
+        )
+
+        self._clock.looping_call(self._prune_old_outbound_device_pokes, 60 * 60 * 1000)
+
     @defer.inlineCallbacks
     def store_device(self, user_id, device_id, initial_device_display_name):
         """Ensure the given device is known; add it to the store if not
@@ -562,6 +714,8 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
         Returns:
             defer.Deferred: boolean whether the device was inserted or an
                 existing device existed with that ID.
+        Raises:
+            StoreError: if the device is already in use
         """
         key = (user_id, device_id)
         if self.device_id_exists_cache.get(key, None):
@@ -574,12 +728,25 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
                     "user_id": user_id,
                     "device_id": device_id,
                     "display_name": initial_device_display_name,
+                    "hidden": False,
                 },
                 desc="store_device",
                 or_ignore=True,
             )
+            if not inserted:
+                # if the device already exists, check if it's a real device, or
+                # if the device ID is reserved by something else
+                hidden = yield self._simple_select_one_onecol(
+                    "devices",
+                    keyvalues={"user_id": user_id, "device_id": device_id},
+                    retcol="hidden",
+                )
+                if hidden:
+                    raise StoreError(400, "The device ID is in use", Codes.FORBIDDEN)
             self.device_id_exists_cache.prefill(key, True)
             return inserted
+        except StoreError:
+            raise
         except Exception as e:
             logger.error(
                 "store_device with device_id=%s(%r) user_id=%s(%r)"
@@ -606,7 +773,7 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
         """
         yield self._simple_delete_one(
             table="devices",
-            keyvalues={"user_id": user_id, "device_id": device_id},
+            keyvalues={"user_id": user_id, "device_id": device_id, "hidden": False},
             desc="delete_device",
         )
 
@@ -626,14 +793,15 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
             table="devices",
             column="device_id",
             iterable=device_ids,
-            keyvalues={"user_id": user_id},
+            keyvalues={"user_id": user_id, "hidden": False},
             desc="delete_devices",
         )
         for device_id in device_ids:
             self.device_id_exists_cache.invalidate((user_id, device_id))
 
     def update_device(self, user_id, device_id, new_display_name=None):
-        """Update a device.
+        """Update a device. Only updates the device if it is not marked as
+        hidden.
 
         Args:
             user_id (str): The ID of the user which owns the device
@@ -652,7 +820,7 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
             return defer.succeed(None)
         return self._simple_update_one(
             table="devices",
-            keyvalues={"user_id": user_id, "device_id": device_id},
+            keyvalues={"user_id": user_id, "device_id": device_id, "hidden": False},
             updatevalues=updates,
             desc="update_device",
         )
@@ -851,7 +1019,7 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
                     "ts": now,
                     "opentracing_context": json.dumps(context)
                     if whitelisted_homeserver(destination)
-                    else None,
+                    else "{}",
                 }
                 for destination in hosts
                 for device_id in device_ids
@@ -905,15 +1073,3 @@ class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
             "_prune_old_outbound_device_pokes",
             _prune_txn,
         )
-
-    @defer.inlineCallbacks
-    def _drop_device_list_streams_non_unique_indexes(self, progress, batch_size):
-        def f(conn):
-            txn = conn.cursor()
-            txn.execute("DROP INDEX IF EXISTS device_lists_remote_cache_id")
-            txn.execute("DROP INDEX IF EXISTS device_lists_remote_extremeties_id")
-            txn.close()
-
-        yield self.runWithConnection(f)
-        yield self._end_background_update(DROP_DEVICE_LIST_STREAMS_NON_UNIQUE_INDEXES)
-        return 1
diff --git a/synapse/storage/directory.py b/synapse/storage/data_stores/main/directory.py
similarity index 98%
rename from synapse/storage/directory.py
rename to synapse/storage/data_stores/main/directory.py
index e966a73f3d..297966d9f4 100644
--- a/synapse/storage/directory.py
+++ b/synapse/storage/data_stores/main/directory.py
@@ -18,10 +18,9 @@ from collections import namedtuple
 from twisted.internet import defer
 
 from synapse.api.errors import SynapseError
+from synapse.storage._base import SQLBaseStore
 from synapse.util.caches.descriptors import cached
 
-from ._base import SQLBaseStore
-
 RoomAliasMapping = namedtuple("RoomAliasMapping", ("room_id", "room_alias", "servers"))
 
 
@@ -47,7 +46,6 @@ class DirectoryWorkerStore(SQLBaseStore):
 
         if not room_id:
             return None
-            return
 
         servers = yield self._simple_select_onecol(
             "room_alias_servers",
@@ -58,7 +56,6 @@ class DirectoryWorkerStore(SQLBaseStore):
 
         if not servers:
             return None
-            return
 
         return RoomAliasMapping(room_id, room_alias.to_string(), servers)
 
diff --git a/synapse/storage/e2e_room_keys.py b/synapse/storage/data_stores/main/e2e_room_keys.py
similarity index 97%
rename from synapse/storage/e2e_room_keys.py
rename to synapse/storage/data_stores/main/e2e_room_keys.py
index cb641468da..2773616257 100644
--- a/synapse/storage/e2e_room_keys.py
+++ b/synapse/storage/data_stores/main/e2e_room_keys.py
@@ -20,8 +20,7 @@ from twisted.internet import defer
 
 from synapse.api.errors import StoreError
 from synapse.logging.opentracing import log_kv, trace
-
-from ._base import SQLBaseStore
+from synapse.storage._base import SQLBaseStore
 
 
 class EndToEndRoomKeyStore(SQLBaseStore):
@@ -413,9 +412,17 @@ class EndToEndRoomKeyStore(SQLBaseStore):
         def _delete_e2e_room_keys_version_txn(txn):
             if version is None:
                 this_version = self._get_current_version(txn, user_id)
+                if this_version is None:
+                    raise StoreError(404, "No current backup version")
             else:
                 this_version = version
 
+            self._simple_delete_txn(
+                txn,
+                table="e2e_room_keys",
+                keyvalues={"user_id": user_id, "version": this_version},
+            )
+
             return self._simple_update_one_txn(
                 txn,
                 table="e2e_room_keys_versions",
diff --git a/synapse/storage/end_to_end_keys.py b/synapse/storage/data_stores/main/end_to_end_keys.py
similarity index 55%
rename from synapse/storage/end_to_end_keys.py
rename to synapse/storage/data_stores/main/end_to_end_keys.py
index 33e3a84933..073412a78d 100644
--- a/synapse/storage/end_to_end_keys.py
+++ b/synapse/storage/data_stores/main/end_to_end_keys.py
@@ -1,5 +1,7 @@
 # -*- coding: utf-8 -*-
 # Copyright 2015, 2016 OpenMarket Ltd
+# Copyright 2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,15 +16,14 @@
 # limitations under the License.
 from six import iteritems
 
-from canonicaljson import encode_canonical_json
+from canonicaljson import encode_canonical_json, json
 
 from twisted.internet import defer
 
 from synapse.logging.opentracing import log_kv, set_tag, trace
+from synapse.storage._base import SQLBaseStore, db_to_json
 from synapse.util.caches.descriptors import cached
 
-from ._base import SQLBaseStore, db_to_json
-
 
 class EndToEndKeyWorkerStore(SQLBaseStore):
     @trace
@@ -40,7 +41,8 @@ class EndToEndKeyWorkerStore(SQLBaseStore):
                 This option only takes effect if include_all_devices is true.
         Returns:
             Dict mapping from user-id to dict mapping from device_id to
-            dict containing "key_json", "device_display_name".
+            key data.  The key data will be a dict in the same format as the
+            DeviceKeys type returned by POST /_matrix/client/r0/keys/query.
         """
         set_tag("query_list", query_list)
         if not query_list:
@@ -54,11 +56,25 @@ class EndToEndKeyWorkerStore(SQLBaseStore):
             include_deleted_devices,
         )
 
+        # Build the result structure, un-jsonify the results, and add the
+        # "unsigned" section
+        rv = {}
         for user_id, device_keys in iteritems(results):
+            rv[user_id] = {}
             for device_id, device_info in iteritems(device_keys):
-                device_info["keys"] = db_to_json(device_info.pop("key_json"))
+                r = db_to_json(device_info.pop("key_json"))
+                r["unsigned"] = {}
+                display_name = device_info["device_display_name"]
+                if display_name is not None:
+                    r["unsigned"]["device_display_name"] = display_name
+                if "signatures" in device_info:
+                    for sig_user_id, sigs in device_info["signatures"].items():
+                        r.setdefault("signatures", {}).setdefault(
+                            sig_user_id, {}
+                        ).update(sigs)
+                rv[user_id][device_id] = r
 
-        return results
+        return rv
 
     @trace
     def _get_e2e_device_keys_txn(
@@ -69,6 +85,8 @@ class EndToEndKeyWorkerStore(SQLBaseStore):
 
         query_clauses = []
         query_params = []
+        signature_query_clauses = []
+        signature_query_params = []
 
         if include_all_devices is False:
             include_deleted_devices = False
@@ -79,12 +97,20 @@ class EndToEndKeyWorkerStore(SQLBaseStore):
         for (user_id, device_id) in query_list:
             query_clause = "user_id = ?"
             query_params.append(user_id)
+            signature_query_clause = "target_user_id = ?"
+            signature_query_params.append(user_id)
 
             if device_id is not None:
                 query_clause += " AND device_id = ?"
                 query_params.append(device_id)
+                signature_query_clause += " AND target_device_id = ?"
+                signature_query_params.append(device_id)
+
+            signature_query_clause += " AND user_id = ?"
+            signature_query_params.append(user_id)
 
             query_clauses.append(query_clause)
+            signature_query_clauses.append(signature_query_clause)
 
         sql = (
             "SELECT user_id, device_id, "
@@ -92,7 +118,7 @@ class EndToEndKeyWorkerStore(SQLBaseStore):
             "    k.key_json"
             " FROM devices d"
             "    %s JOIN e2e_device_keys_json k USING (user_id, device_id)"
-            " WHERE %s"
+            " WHERE %s AND NOT d.hidden"
         ) % (
             "LEFT" if include_all_devices else "INNER",
             " OR ".join("(" + q + ")" for q in query_clauses),
@@ -111,6 +137,22 @@ class EndToEndKeyWorkerStore(SQLBaseStore):
             for user_id, device_id in deleted_devices:
                 result.setdefault(user_id, {})[device_id] = None
 
+        # get signatures on the device
+        signature_sql = (
+            "SELECT * " "  FROM e2e_cross_signing_signatures " " WHERE %s"
+        ) % (" OR ".join("(" + q + ")" for q in signature_query_clauses))
+
+        txn.execute(signature_sql, signature_query_params)
+        rows = self.cursor_to_dict(txn)
+
+        for row in rows:
+            target_user_id = row["target_user_id"]
+            target_device_id = row["target_device_id"]
+            if target_user_id in result and target_device_id in result[target_user_id]:
+                result[target_user_id][target_device_id].setdefault(
+                    "signatures", {}
+                ).setdefault(row["user_id"], {})[row["key_id"]] = row["signature"]
+
         log_kv(result)
         return result
 
@@ -206,6 +248,97 @@ class EndToEndKeyWorkerStore(SQLBaseStore):
 
         return self.runInteraction("count_e2e_one_time_keys", _count_e2e_one_time_keys)
 
+    def _get_e2e_cross_signing_key_txn(self, txn, user_id, key_type, from_user_id=None):
+        """Returns a user's cross-signing key.
+
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            user_id (str): the user whose key is being requested
+            key_type (str): the type of key that is being set: either 'master'
+                for a master key, 'self_signing' for a self-signing key, or
+                'user_signing' for a user-signing key
+            from_user_id (str): if specified, signatures made by this user on
+                the key will be included in the result
+
+        Returns:
+            dict of the key data or None if not found
+        """
+        sql = (
+            "SELECT keydata "
+            "  FROM e2e_cross_signing_keys "
+            " WHERE user_id = ? AND keytype = ? ORDER BY stream_id DESC LIMIT 1"
+        )
+        txn.execute(sql, (user_id, key_type))
+        row = txn.fetchone()
+        if not row:
+            return None
+        key = json.loads(row[0])
+
+        device_id = None
+        for k in key["keys"].values():
+            device_id = k
+
+        if from_user_id is not None:
+            sql = (
+                "SELECT key_id, signature "
+                "  FROM e2e_cross_signing_signatures "
+                " WHERE user_id = ? "
+                "   AND target_user_id = ? "
+                "   AND target_device_id = ? "
+            )
+            txn.execute(sql, (from_user_id, user_id, device_id))
+            row = txn.fetchone()
+            if row:
+                key.setdefault("signatures", {}).setdefault(from_user_id, {})[
+                    row[0]
+                ] = row[1]
+
+        return key
+
+    def get_e2e_cross_signing_key(self, user_id, key_type, from_user_id=None):
+        """Returns a user's cross-signing key.
+
+        Args:
+            user_id (str): the user whose self-signing key is being requested
+            key_type (str): the type of cross-signing key to get
+            from_user_id (str): if specified, signatures made by this user on
+                the self-signing key will be included in the result
+
+        Returns:
+            dict of the key data or None if not found
+        """
+        return self.runInteraction(
+            "get_e2e_cross_signing_key",
+            self._get_e2e_cross_signing_key_txn,
+            user_id,
+            key_type,
+            from_user_id,
+        )
+
+    def get_all_user_signature_changes_for_remotes(self, from_key, to_key):
+        """Return a list of changes from the user signature stream to notify remotes.
+        Note that the user signature stream represents when a user signs their
+        device with their user-signing key, which is not published to other
+        users or servers, so no `destination` is needed in the returned
+        list. However, this is needed to poke workers.
+
+        Args:
+            from_key (int): the stream ID to start at (exclusive)
+            to_key (int): the stream ID to end at (inclusive)
+
+        Returns:
+            Deferred[list[(int,str)]] a list of `(stream_id, user_id)`
+        """
+        sql = """
+            SELECT MAX(stream_id) AS stream_id, from_user_id AS user_id
+            FROM user_signature_stream
+            WHERE ? < stream_id AND stream_id <= ?
+            GROUP BY user_id
+        """
+        return self._execute(
+            "get_all_user_signature_changes_for_remotes", None, sql, from_key, to_key
+        )
+
 
 class EndToEndKeyStore(EndToEndKeyWorkerStore, SQLBaseStore):
     def set_e2e_device_keys(self, user_id, device_id, time_now, device_keys):
@@ -311,3 +444,97 @@ class EndToEndKeyStore(EndToEndKeyWorkerStore, SQLBaseStore):
         return self.runInteraction(
             "delete_e2e_keys_by_device", delete_e2e_keys_by_device_txn
         )
+
+    def _set_e2e_cross_signing_key_txn(self, txn, user_id, key_type, key):
+        """Set a user's cross-signing key.
+
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            user_id (str): the user to set the signing key for
+            key_type (str): the type of key that is being set: either 'master'
+                for a master key, 'self_signing' for a self-signing key, or
+                'user_signing' for a user-signing key
+            key (dict): the key data
+        """
+        # the cross-signing keys need to occupy the same namespace as devices,
+        # since signatures are identified by device ID.  So add an entry to the
+        # device table to make sure that we don't have a collision with device
+        # IDs
+
+        # the 'key' dict will look something like:
+        # {
+        #   "user_id": "@alice:example.com",
+        #   "usage": ["self_signing"],
+        #   "keys": {
+        #     "ed25519:base64+self+signing+public+key": "base64+self+signing+public+key",
+        #   },
+        #   "signatures": {
+        #     "@alice:example.com": {
+        #       "ed25519:base64+master+public+key": "base64+signature"
+        #     }
+        #   }
+        # }
+        # The "keys" property must only have one entry, which will be the public
+        # key, so we just grab the first value in there
+        pubkey = next(iter(key["keys"].values()))
+        self._simple_insert_txn(
+            txn,
+            "devices",
+            values={
+                "user_id": user_id,
+                "device_id": pubkey,
+                "display_name": key_type + " signing key",
+                "hidden": True,
+            },
+        )
+
+        # and finally, store the key itself
+        with self._cross_signing_id_gen.get_next() as stream_id:
+            self._simple_insert_txn(
+                txn,
+                "e2e_cross_signing_keys",
+                values={
+                    "user_id": user_id,
+                    "keytype": key_type,
+                    "keydata": json.dumps(key),
+                    "stream_id": stream_id,
+                },
+            )
+
+    def set_e2e_cross_signing_key(self, user_id, key_type, key):
+        """Set a user's cross-signing key.
+
+        Args:
+            user_id (str): the user to set the user-signing key for
+            key_type (str): the type of cross-signing key to set
+            key (dict): the key data
+        """
+        return self.runInteraction(
+            "add_e2e_cross_signing_key",
+            self._set_e2e_cross_signing_key_txn,
+            user_id,
+            key_type,
+            key,
+        )
+
+    def store_e2e_cross_signing_signatures(self, user_id, signatures):
+        """Stores cross-signing signatures.
+
+        Args:
+            user_id (str): the user who made the signatures
+            signatures (iterable[SignatureListItem]): signatures to add
+        """
+        return self._simple_insert_many(
+            "e2e_cross_signing_signatures",
+            [
+                {
+                    "user_id": user_id,
+                    "key_id": item.signing_key_id,
+                    "target_user_id": item.target_user_id,
+                    "target_device_id": item.target_device_id,
+                    "signature": item.signature,
+                }
+                for item in signatures
+            ],
+            "add_e2e_signing_key",
+        )
diff --git a/synapse/storage/event_federation.py b/synapse/storage/data_stores/main/event_federation.py
similarity index 94%
rename from synapse/storage/event_federation.py
rename to synapse/storage/data_stores/main/event_federation.py
index 4f500d893e..90bef0cd2c 100644
--- a/synapse/storage/event_federation.py
+++ b/synapse/storage/data_stores/main/event_federation.py
@@ -12,6 +12,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import itertools
 import logging
 import random
 
@@ -24,9 +25,9 @@ from twisted.internet import defer
 
 from synapse.api.errors import StoreError
 from synapse.metrics.background_process_metrics import run_as_background_process
-from synapse.storage._base import SQLBaseStore
-from synapse.storage.events_worker import EventsWorkerStore
-from synapse.storage.signatures import SignatureWorkerStore
+from synapse.storage._base import SQLBaseStore, make_in_list_sql_clause
+from synapse.storage.data_stores.main.events_worker import EventsWorkerStore
+from synapse.storage.data_stores.main.signatures import SignatureWorkerStore
 from synapse.util.caches.descriptors import cached
 
 logger = logging.getLogger(__name__)
@@ -67,7 +68,7 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas
         else:
             results = set()
 
-        base_sql = "SELECT auth_id FROM event_auth WHERE event_id IN (%s)"
+        base_sql = "SELECT auth_id FROM event_auth WHERE "
 
         front = set(event_ids)
         while front:
@@ -75,7 +76,10 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas
             front_list = list(front)
             chunks = [front_list[x : x + 100] for x in range(0, len(front), 100)]
             for chunk in chunks:
-                txn.execute(base_sql % (",".join(["?"] * len(chunk)),), chunk)
+                clause, args = make_in_list_sql_clause(
+                    txn.database_engine, "event_id", chunk
+                )
+                txn.execute(base_sql + clause, list(args))
                 new_front.update([r[0] for r in txn])
 
             new_front -= results
@@ -190,12 +194,13 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas
             room_id,
         )
 
-    def get_rooms_with_many_extremities(self, min_count, limit):
+    def get_rooms_with_many_extremities(self, min_count, limit, room_id_filter):
         """Get the top rooms with at least N extremities.
 
         Args:
             min_count (int): The minimum number of extremities
             limit (int): The maximum number of rooms to return.
+            room_id_filter (iterable[str]): room_ids to exclude from the results
 
         Returns:
             Deferred[list]: At most `limit` room IDs that have at least
@@ -203,15 +208,25 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas
         """
 
         def _get_rooms_with_many_extremities_txn(txn):
+            where_clause = "1=1"
+            if room_id_filter:
+                where_clause = "room_id NOT IN (%s)" % (
+                    ",".join("?" for _ in room_id_filter),
+                )
+
             sql = """
                 SELECT room_id FROM event_forward_extremities
+                WHERE %s
                 GROUP BY room_id
                 HAVING count(*) > ?
                 ORDER BY count(*) DESC
                 LIMIT ?
-            """
+            """ % (
+                where_clause,
+            )
 
-            txn.execute(sql, (min_count, limit))
+            query_args = list(itertools.chain(room_id_filter, [min_count, limit]))
+            txn.execute(sql, query_args)
             return [room_id for room_id, in txn]
 
         return self.runInteraction(
@@ -349,9 +364,7 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas
         )
 
     def _get_backfill_events(self, txn, room_id, event_list, limit):
-        logger.debug(
-            "_get_backfill_events: %s, %s, %s", room_id, repr(event_list), limit
-        )
+        logger.debug("_get_backfill_events: %s, %r, %s", room_id, event_list, limit)
 
         event_results = set()
 
diff --git a/synapse/storage/event_push_actions.py b/synapse/storage/data_stores/main/event_push_actions.py
similarity index 99%
rename from synapse/storage/event_push_actions.py
rename to synapse/storage/data_stores/main/event_push_actions.py
index 22025effbc..04ce21ac66 100644
--- a/synapse/storage/event_push_actions.py
+++ b/synapse/storage/data_stores/main/event_push_actions.py
@@ -863,7 +863,7 @@ class EventPushActionsStore(EventPushActionsWorkerStore):
         )
         stream_row = txn.fetchone()
         if stream_row:
-            offset_stream_ordering, = stream_row
+            (offset_stream_ordering,) = stream_row
             rotate_to_stream_ordering = min(
                 self.stream_ordering_day_ago, offset_stream_ordering
             )
diff --git a/synapse/storage/events.py b/synapse/storage/data_stores/main/events.py
similarity index 68%
rename from synapse/storage/events.py
rename to synapse/storage/data_stores/main/events.py
index 32050868ff..301f8ea128 100644
--- a/synapse/storage/events.py
+++ b/synapse/storage/data_stores/main/events.py
@@ -17,37 +17,35 @@
 
 import itertools
 import logging
-from collections import Counter as c_counter, OrderedDict, deque, namedtuple
+from collections import Counter as c_counter, OrderedDict, namedtuple
 from functools import wraps
 
 from six import iteritems, text_type
 from six.moves import range
 
 from canonicaljson import json
-from prometheus_client import Counter, Histogram
+from prometheus_client import Counter
 
 from twisted.internet import defer
 
 import synapse.metrics
-from synapse.api.constants import EventTypes
+from synapse.api.constants import EventContentFields, EventTypes
 from synapse.api.errors import SynapseError
 from synapse.events import EventBase  # noqa: F401
 from synapse.events.snapshot import EventContext  # noqa: F401
-from synapse.logging.context import PreserveLoggingContext, make_deferred_yieldable
+from synapse.events.utils import prune_event_dict
 from synapse.logging.utils import log_function
 from synapse.metrics import BucketCollector
 from synapse.metrics.background_process_metrics import run_as_background_process
-from synapse.state import StateResolutionStore
+from synapse.storage._base import make_in_list_sql_clause
 from synapse.storage.background_updates import BackgroundUpdateStore
-from synapse.storage.event_federation import EventFederationStore
-from synapse.storage.events_worker import EventsWorkerStore
-from synapse.storage.state import StateGroupWorkerStore
+from synapse.storage.data_stores.main.event_federation import EventFederationStore
+from synapse.storage.data_stores.main.events_worker import EventsWorkerStore
+from synapse.storage.data_stores.main.state import StateGroupWorkerStore
 from synapse.types import RoomStreamToken, get_domain_from_id
 from synapse.util import batch_iter
-from synapse.util.async_helpers import ObservableDeferred
 from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
 from synapse.util.frozenutils import frozendict_json_encoder
-from synapse.util.metrics import Measure
 
 logger = logging.getLogger(__name__)
 
@@ -58,37 +56,6 @@ event_counter = Counter(
     ["type", "origin_type", "origin_entity"],
 )
 
-# The number of times we are recalculating the current state
-state_delta_counter = Counter("synapse_storage_events_state_delta", "")
-
-# The number of times we are recalculating state when there is only a
-# single forward extremity
-state_delta_single_event_counter = Counter(
-    "synapse_storage_events_state_delta_single_event", ""
-)
-
-# The number of times we are reculating state when we could have resonably
-# calculated the delta when we calculated the state for an event we were
-# persisting.
-state_delta_reuse_delta_counter = Counter(
-    "synapse_storage_events_state_delta_reuse_delta", ""
-)
-
-# The number of forward extremities for each new event.
-forward_extremities_counter = Histogram(
-    "synapse_storage_events_forward_extremities_persisted",
-    "Number of forward extremities for each new event",
-    buckets=(1, 2, 3, 5, 7, 10, 15, 20, 50, 100, 200, 500, "+Inf"),
-)
-
-# The number of stale forward extremities for each new event. Stale extremities
-# are those that were in the previous set of extremities as well as the new.
-stale_forward_extremities_counter = Histogram(
-    "synapse_storage_events_stale_forward_extremities_persisted",
-    "Number of unchanged forward extremities for each new event",
-    buckets=(0, 1, 2, 3, 5, 7, 10, 15, 20, 50, 100, 200, 500, "+Inf"),
-)
-
 
 def encode_json(json_object):
     """
@@ -100,110 +67,6 @@ def encode_json(json_object):
     return out
 
 
-class _EventPeristenceQueue(object):
-    """Queues up events so that they can be persisted in bulk with only one
-    concurrent transaction per room.
-    """
-
-    _EventPersistQueueItem = namedtuple(
-        "_EventPersistQueueItem", ("events_and_contexts", "backfilled", "deferred")
-    )
-
-    def __init__(self):
-        self._event_persist_queues = {}
-        self._currently_persisting_rooms = set()
-
-    def add_to_queue(self, room_id, events_and_contexts, backfilled):
-        """Add events to the queue, with the given persist_event options.
-
-        NB: due to the normal usage pattern of this method, it does *not*
-        follow the synapse logcontext rules, and leaves the logcontext in
-        place whether or not the returned deferred is ready.
-
-        Args:
-            room_id (str):
-            events_and_contexts (list[(EventBase, EventContext)]):
-            backfilled (bool):
-
-        Returns:
-            defer.Deferred: a deferred which will resolve once the events are
-                persisted. Runs its callbacks *without* a logcontext.
-        """
-        queue = self._event_persist_queues.setdefault(room_id, deque())
-        if queue:
-            # if the last item in the queue has the same `backfilled` setting,
-            # we can just add these new events to that item.
-            end_item = queue[-1]
-            if end_item.backfilled == backfilled:
-                end_item.events_and_contexts.extend(events_and_contexts)
-                return end_item.deferred.observe()
-
-        deferred = ObservableDeferred(defer.Deferred(), consumeErrors=True)
-
-        queue.append(
-            self._EventPersistQueueItem(
-                events_and_contexts=events_and_contexts,
-                backfilled=backfilled,
-                deferred=deferred,
-            )
-        )
-
-        return deferred.observe()
-
-    def handle_queue(self, room_id, per_item_callback):
-        """Attempts to handle the queue for a room if not already being handled.
-
-        The given callback will be invoked with for each item in the queue,
-        of type _EventPersistQueueItem. The per_item_callback will continuously
-        be called with new items, unless the queue becomnes empty. The return
-        value of the function will be given to the deferreds waiting on the item,
-        exceptions will be passed to the deferreds as well.
-
-        This function should therefore be called whenever anything is added
-        to the queue.
-
-        If another callback is currently handling the queue then it will not be
-        invoked.
-        """
-
-        if room_id in self._currently_persisting_rooms:
-            return
-
-        self._currently_persisting_rooms.add(room_id)
-
-        @defer.inlineCallbacks
-        def handle_queue_loop():
-            try:
-                queue = self._get_drainining_queue(room_id)
-                for item in queue:
-                    try:
-                        ret = yield per_item_callback(item)
-                    except Exception:
-                        with PreserveLoggingContext():
-                            item.deferred.errback()
-                    else:
-                        with PreserveLoggingContext():
-                            item.deferred.callback(ret)
-            finally:
-                queue = self._event_persist_queues.pop(room_id, None)
-                if queue:
-                    self._event_persist_queues[room_id] = queue
-                self._currently_persisting_rooms.discard(room_id)
-
-        # set handle_queue_loop off in the background
-        run_as_background_process("persist_events", handle_queue_loop)
-
-    def _get_drainining_queue(self, room_id):
-        queue = self._event_persist_queues.setdefault(room_id, deque())
-
-        try:
-            while True:
-                yield queue.popleft()
-        except IndexError:
-            # Queue has been drained.
-            pass
-
-
 _EventCacheEntry = namedtuple("_EventCacheEntry", ("event", "redacted_event"))
 
 
@@ -219,7 +82,7 @@ def _retry_on_integrity_error(func):
     @defer.inlineCallbacks
     def f(self, *args, **kwargs):
         try:
-            res = yield func(self, *args, **kwargs)
+            res = yield func(self, *args, delete_existing=False, **kwargs)
         except self.database_engine.module.IntegrityError:
             logger.exception("IntegrityError, retrying.")
             res = yield func(self, *args, delete_existing=True, **kwargs)
@@ -239,9 +102,6 @@ class EventsStore(
     def __init__(self, db_conn, hs):
         super(EventsStore, self).__init__(db_conn, hs)
 
-        self._event_persist_queue = _EventPeristenceQueue()
-        self._state_resolution_handler = hs.get_state_resolution_handler()
-
         # Collect metrics on the number of forward extremities that exist.
         # Counter of number of extremities to count
         self._current_forward_extremities_amount = c_counter()
@@ -262,6 +122,14 @@ class EventsStore(
 
         hs.get_clock().looping_call(read_forward_extremities, 60 * 60 * 1000)
 
+        def _censor_redactions():
+            return run_as_background_process(
+                "_censor_redactions", self._censor_redactions
+            )
+
+        if self.hs.config.redaction_retention_period is not None:
+            hs.get_clock().looping_call(_censor_redactions, 5 * 60 * 1000)
+
     @defer.inlineCallbacks
     def _read_forward_extremities(self):
         def fetch(txn):
@@ -276,340 +144,106 @@ class EventsStore(
         res = yield self.runInteraction("read_forward_extremities", fetch)
         self._current_forward_extremities_amount = c_counter(list(x[0] for x in res))
 
-    @defer.inlineCallbacks
-    def persist_events(self, events_and_contexts, backfilled=False):
-        """
-        Write events to the database
-        Args:
-            events_and_contexts: list of tuples of (event, context)
-            backfilled (bool): Whether the results are retrieved from federation
-                via backfill or not. Used to determine if they're "new" events
-                which might update the current state etc.
-
-        Returns:
-            Deferred[int]: the stream ordering of the latest persisted event
-        """
-        partitioned = {}
-        for event, ctx in events_and_contexts:
-            partitioned.setdefault(event.room_id, []).append((event, ctx))
-
-        deferreds = []
-        for room_id, evs_ctxs in iteritems(partitioned):
-            d = self._event_persist_queue.add_to_queue(
-                room_id, evs_ctxs, backfilled=backfilled
-            )
-            deferreds.append(d)
-
-        for room_id in partitioned:
-            self._maybe_start_persisting(room_id)
-
-        yield make_deferred_yieldable(
-            defer.gatherResults(deferreds, consumeErrors=True)
-        )
-
-        max_persisted_id = yield self._stream_id_gen.get_current_token()
-
-        return max_persisted_id
-
-    @defer.inlineCallbacks
-    @log_function
-    def persist_event(self, event, context, backfilled=False):
-        """
-
-        Args:
-            event (EventBase):
-            context (EventContext):
-            backfilled (bool):
-
-        Returns:
-            Deferred: resolves to (int, int): the stream ordering of ``event``,
-            and the stream ordering of the latest persisted event
-        """
-        deferred = self._event_persist_queue.add_to_queue(
-            event.room_id, [(event, context)], backfilled=backfilled
-        )
-
-        self._maybe_start_persisting(event.room_id)
-
-        yield make_deferred_yieldable(deferred)
-
-        max_persisted_id = yield self._stream_id_gen.get_current_token()
-        return (event.internal_metadata.stream_ordering, max_persisted_id)
-
-    def _maybe_start_persisting(self, room_id):
-        @defer.inlineCallbacks
-        def persisting_queue(item):
-            with Measure(self._clock, "persist_events"):
-                yield self._persist_events(
-                    item.events_and_contexts, backfilled=item.backfilled
-                )
-
-        self._event_persist_queue.handle_queue(room_id, persisting_queue)
-
     @_retry_on_integrity_error
     @defer.inlineCallbacks
-    def _persist_events(
-        self, events_and_contexts, backfilled=False, delete_existing=False
+    def _persist_events_and_state_updates(
+        self,
+        events_and_contexts,
+        current_state_for_room,
+        state_delta_for_room,
+        new_forward_extremeties,
+        backfilled=False,
+        delete_existing=False,
     ):
-        """Persist events to db
+        """Persist a set of events alongside updates to the current state and
+        forward extremities tables.
 
         Args:
             events_and_contexts (list[(EventBase, EventContext)]):
-            backfilled (bool):
+            current_state_for_room (dict[str, dict]): Map from room_id to the
+                current state of the room based on forward extremities
+            state_delta_for_room (dict[str, tuple]): Map from room_id to tuple
+                of `(to_delete, to_insert)` where to_delete is a list
+                of type/state keys to remove from current state, and to_insert
+                is a map (type,key)->event_id giving the state delta in each
+                room.
+            new_forward_extremities (dict[str, list[str]]): Map from room_id
+                to list of event IDs that are the new forward extremities of
+                the room.
+            backfilled (bool)
             delete_existing (bool):
 
         Returns:
             Deferred: resolves when the events have been persisted
         """
-        if not events_and_contexts:
-            return
 
-        chunks = [
-            events_and_contexts[x : x + 100]
-            for x in range(0, len(events_and_contexts), 100)
-        ]
+        # We want to calculate the stream orderings as late as possible, as
+        # we only notify after all events with a lesser stream ordering have
+        # been persisted. I.e. if we spend 10s inside the with block then
+        # that will delay all subsequent events from being notified about.
+        # Hence why we do it down here rather than wrapping the entire
+        # function.
+        #
+        # Its safe to do this after calculating the state deltas etc as we
+        # only need to protect the *persistence* of the events. This is to
+        # ensure that queries of the form "fetch events since X" don't
+        # return events and stream positions after events that are still in
+        # flight, as otherwise subsequent requests "fetch event since Y"
+        # will not return those events.
+        #
+        # Note: Multiple instances of this function cannot be in flight at
+        # the same time for the same room.
+        if backfilled:
+            stream_ordering_manager = self._backfill_id_gen.get_next_mult(
+                len(events_and_contexts)
+            )
+        else:
+            stream_ordering_manager = self._stream_id_gen.get_next_mult(
+                len(events_and_contexts)
+            )
 
-        for chunk in chunks:
-            # We can't easily parallelize these since different chunks
-            # might contain the same event. :(
+        with stream_ordering_manager as stream_orderings:
+            for (event, context), stream in zip(events_and_contexts, stream_orderings):
+                event.internal_metadata.stream_ordering = stream
 
-            # NB: Assumes that we are only persisting events for one room
-            # at a time.
-
-            # map room_id->list[event_ids] giving the new forward
-            # extremities in each room
-            new_forward_extremeties = {}
-
-            # map room_id->(type,state_key)->event_id tracking the full
-            # state in each room after adding these events.
-            # This is simply used to prefill the get_current_state_ids
-            # cache
-            current_state_for_room = {}
-
-            # map room_id->(to_delete, to_insert) where to_delete is a list
-            # of type/state keys to remove from current state, and to_insert
-            # is a map (type,key)->event_id giving the state delta in each
-            # room
-            state_delta_for_room = {}
+            yield self.runInteraction(
+                "persist_events",
+                self._persist_events_txn,
+                events_and_contexts=events_and_contexts,
+                backfilled=backfilled,
+                delete_existing=delete_existing,
+                state_delta_for_room=state_delta_for_room,
+                new_forward_extremeties=new_forward_extremeties,
+            )
+            persist_event_counter.inc(len(events_and_contexts))
 
             if not backfilled:
-                with Measure(self._clock, "_calculate_state_and_extrem"):
-                    # Work out the new "current state" for each room.
-                    # We do this by working out what the new extremities are and then
-                    # calculating the state from that.
-                    events_by_room = {}
-                    for event, context in chunk:
-                        events_by_room.setdefault(event.room_id, []).append(
-                            (event, context)
-                        )
-
-                    for room_id, ev_ctx_rm in iteritems(events_by_room):
-                        latest_event_ids = yield self.get_latest_event_ids_in_room(
-                            room_id
-                        )
-                        new_latest_event_ids = yield self._calculate_new_extremities(
-                            room_id, ev_ctx_rm, latest_event_ids
-                        )
-
-                        latest_event_ids = set(latest_event_ids)
-                        if new_latest_event_ids == latest_event_ids:
-                            # No change in extremities, so no change in state
-                            continue
-
-                        # there should always be at least one forward extremity.
-                        # (except during the initial persistence of the send_join
-                        # results, in which case there will be no existing
-                        # extremities, so we'll `continue` above and skip this bit.)
-                        assert new_latest_event_ids, "No forward extremities left!"
-
-                        new_forward_extremeties[room_id] = new_latest_event_ids
-
-                        len_1 = (
-                            len(latest_event_ids) == 1
-                            and len(new_latest_event_ids) == 1
-                        )
-                        if len_1:
-                            all_single_prev_not_state = all(
-                                len(event.prev_event_ids()) == 1
-                                and not event.is_state()
-                                for event, ctx in ev_ctx_rm
-                            )
-                            # Don't bother calculating state if they're just
-                            # a long chain of single ancestor non-state events.
-                            if all_single_prev_not_state:
-                                continue
-
-                        state_delta_counter.inc()
-                        if len(new_latest_event_ids) == 1:
-                            state_delta_single_event_counter.inc()
-
-                            # This is a fairly handwavey check to see if we could
-                            # have guessed what the delta would have been when
-                            # processing one of these events.
-                            # What we're interested in is if the latest extremities
-                            # were the same when we created the event as they are
-                            # now. When this server creates a new event (as opposed
-                            # to receiving it over federation) it will use the
-                            # forward extremities as the prev_events, so we can
-                            # guess this by looking at the prev_events and checking
-                            # if they match the current forward extremities.
-                            for ev, _ in ev_ctx_rm:
-                                prev_event_ids = set(ev.prev_event_ids())
-                                if latest_event_ids == prev_event_ids:
-                                    state_delta_reuse_delta_counter.inc()
-                                    break
-
-                        logger.info("Calculating state delta for room %s", room_id)
-                        with Measure(
-                            self._clock, "persist_events.get_new_state_after_events"
-                        ):
-                            res = yield self._get_new_state_after_events(
-                                room_id,
-                                ev_ctx_rm,
-                                latest_event_ids,
-                                new_latest_event_ids,
-                            )
-                            current_state, delta_ids = res
-
-                        # If either are not None then there has been a change,
-                        # and we need to work out the delta (or use that
-                        # given)
-                        if delta_ids is not None:
-                            # If there is a delta we know that we've
-                            # only added or replaced state, never
-                            # removed keys entirely.
-                            state_delta_for_room[room_id] = ([], delta_ids)
-                        elif current_state is not None:
-                            with Measure(
-                                self._clock, "persist_events.calculate_state_delta"
-                            ):
-                                delta = yield self._calculate_state_delta(
-                                    room_id, current_state
-                                )
-                            state_delta_for_room[room_id] = delta
-
-                        # If we have the current_state then lets prefill
-                        # the cache with it.
-                        if current_state is not None:
-                            current_state_for_room[room_id] = current_state
-
-            # We want to calculate the stream orderings as late as possible, as
-            # we only notify after all events with a lesser stream ordering have
-            # been persisted. I.e. if we spend 10s inside the with block then
-            # that will delay all subsequent events from being notified about.
-            # Hence why we do it down here rather than wrapping the entire
-            # function.
-            #
-            # Its safe to do this after calculating the state deltas etc as we
-            # only need to protect the *persistence* of the events. This is to
-            # ensure that queries of the form "fetch events since X" don't
-            # return events and stream positions after events that are still in
-            # flight, as otherwise subsequent requests "fetch event since Y"
-            # will not return those events.
-            #
-            # Note: Multiple instances of this function cannot be in flight at
-            # the same time for the same room.
-            if backfilled:
-                stream_ordering_manager = self._backfill_id_gen.get_next_mult(
-                    len(chunk)
+                # backfilled events have negative stream orderings, so we don't
+                # want to set the event_persisted_position to that.
+                synapse.metrics.event_persisted_position.set(
+                    events_and_contexts[-1][0].internal_metadata.stream_ordering
                 )
-            else:
-                stream_ordering_manager = self._stream_id_gen.get_next_mult(len(chunk))
 
-            with stream_ordering_manager as stream_orderings:
-                for (event, context), stream in zip(chunk, stream_orderings):
-                    event.internal_metadata.stream_ordering = stream
+            for event, context in events_and_contexts:
+                if context.app_service:
+                    origin_type = "local"
+                    origin_entity = context.app_service.id
+                elif self.hs.is_mine_id(event.sender):
+                    origin_type = "local"
+                    origin_entity = "*client*"
+                else:
+                    origin_type = "remote"
+                    origin_entity = get_domain_from_id(event.sender)
 
-                yield self.runInteraction(
-                    "persist_events",
-                    self._persist_events_txn,
-                    events_and_contexts=chunk,
-                    backfilled=backfilled,
-                    delete_existing=delete_existing,
-                    state_delta_for_room=state_delta_for_room,
-                    new_forward_extremeties=new_forward_extremeties,
+                event_counter.labels(event.type, origin_type, origin_entity).inc()
+
+            for room_id, new_state in iteritems(current_state_for_room):
+                self.get_current_state_ids.prefill((room_id,), new_state)
+
+            for room_id, latest_event_ids in iteritems(new_forward_extremeties):
+                self.get_latest_event_ids_in_room.prefill(
+                    (room_id,), list(latest_event_ids)
                 )
-                persist_event_counter.inc(len(chunk))
-
-                if not backfilled:
-                    # backfilled events have negative stream orderings, so we don't
-                    # want to set the event_persisted_position to that.
-                    synapse.metrics.event_persisted_position.set(
-                        chunk[-1][0].internal_metadata.stream_ordering
-                    )
-
-                for event, context in chunk:
-                    if context.app_service:
-                        origin_type = "local"
-                        origin_entity = context.app_service.id
-                    elif self.hs.is_mine_id(event.sender):
-                        origin_type = "local"
-                        origin_entity = "*client*"
-                    else:
-                        origin_type = "remote"
-                        origin_entity = get_domain_from_id(event.sender)
-
-                    event_counter.labels(event.type, origin_type, origin_entity).inc()
-
-                for room_id, new_state in iteritems(current_state_for_room):
-                    self.get_current_state_ids.prefill((room_id,), new_state)
-
-                for room_id, latest_event_ids in iteritems(new_forward_extremeties):
-                    self.get_latest_event_ids_in_room.prefill(
-                        (room_id,), list(latest_event_ids)
-                    )
-
-    @defer.inlineCallbacks
-    def _calculate_new_extremities(self, room_id, event_contexts, latest_event_ids):
-        """Calculates the new forward extremities for a room given events to
-        persist.
-
-        Assumes that we are only persisting events for one room at a time.
-        """
-
-        # we're only interested in new events which aren't outliers and which aren't
-        # being rejected.
-        new_events = [
-            event
-            for event, ctx in event_contexts
-            if not event.internal_metadata.is_outlier()
-            and not ctx.rejected
-            and not event.internal_metadata.is_soft_failed()
-        ]
-
-        latest_event_ids = set(latest_event_ids)
-
-        # start with the existing forward extremities
-        result = set(latest_event_ids)
-
-        # add all the new events to the list
-        result.update(event.event_id for event in new_events)
-
-        # Now remove all events which are prev_events of any of the new events
-        result.difference_update(
-            e_id for event in new_events for e_id in event.prev_event_ids()
-        )
-
-        # Remove any events which are prev_events of any existing events.
-        existing_prevs = yield self._get_events_which_are_prevs(result)
-        result.difference_update(existing_prevs)
-
-        # Finally handle the case where the new events have soft-failed prev
-        # events. If they do we need to remove them and their prev events,
-        # otherwise we end up with dangling extremities.
-        existing_prevs = yield self._get_prevs_before_rejected(
-            e_id for event in new_events for e_id in event.prev_event_ids()
-        )
-        result.difference_update(existing_prevs)
-
-        # We only update metrics for events that change forward extremities
-        # (e.g. we ignore backfill/outliers/etc)
-        if result != latest_event_ids:
-            forward_extremities_counter.observe(len(result))
-            stale = latest_event_ids & result
-            stale_forward_extremities_counter.observe(len(stale))
-
-        return result
 
     @defer.inlineCallbacks
     def _get_events_which_are_prevs(self, event_ids):
@@ -632,14 +266,16 @@ class EventsStore(
                 LEFT JOIN rejections USING (event_id)
                 LEFT JOIN event_json USING (event_id)
             WHERE
-                prev_event_id IN (%s)
-                AND NOT events.outlier
+                NOT events.outlier
                 AND rejections.event_id IS NULL
-            """ % (
-                ",".join("?" for _ in batch),
+                AND
+            """
+
+            clause, args = make_in_list_sql_clause(
+                self.database_engine, "prev_event_id", batch
             )
 
-            txn.execute(sql, batch)
+            txn.execute(sql + clause, args)
             results.extend(r[0] for r in txn if not json.loads(r[1]).get("soft_failed"))
 
         for chunk in batch_iter(event_ids, 100):
@@ -686,13 +322,15 @@ class EventsStore(
                     LEFT JOIN rejections USING (event_id)
                     LEFT JOIN event_json USING (event_id)
                 WHERE
-                    event_id IN (%s)
-                    AND NOT events.outlier
-                """ % (
-                    ",".join("?" for _ in to_recursively_check),
+                    NOT events.outlier
+                    AND
+                """
+
+                clause, args = make_in_list_sql_clause(
+                    self.database_engine, "event_id", to_recursively_check
                 )
 
-                txn.execute(sql, to_recursively_check)
+                txn.execute(sql + clause, args)
                 to_recursively_check = []
 
                 for event_id, prev_event_id, metadata, rejected in txn:
@@ -711,188 +349,6 @@ class EventsStore(
 
         return existing_prevs
 
-    @defer.inlineCallbacks
-    def _get_new_state_after_events(
-        self, room_id, events_context, old_latest_event_ids, new_latest_event_ids
-    ):
-        """Calculate the current state dict after adding some new events to
-        a room
-
-        Args:
-            room_id (str):
-                room to which the events are being added. Used for logging etc
-
-            events_context (list[(EventBase, EventContext)]):
-                events and contexts which are being added to the room
-
-            old_latest_event_ids (iterable[str]):
-                the old forward extremities for the room.
-
-            new_latest_event_ids (iterable[str]):
-                the new forward extremities for the room.
-
-        Returns:
-            Deferred[tuple[dict[(str,str), str]|None, dict[(str,str), str]|None]]:
-            Returns a tuple of two state maps, the first being the full new current
-            state and the second being the delta to the existing current state.
-            If both are None then there has been no change.
-
-            If there has been a change then we only return the delta if its
-            already been calculated. Conversely if we do know the delta then
-            the new current state is only returned if we've already calculated
-            it.
-        """
-        # map from state_group to ((type, key) -> event_id) state map
-        state_groups_map = {}
-
-        # Map from (prev state group, new state group) -> delta state dict
-        state_group_deltas = {}
-
-        for ev, ctx in events_context:
-            if ctx.state_group is None:
-                # This should only happen for outlier events.
-                if not ev.internal_metadata.is_outlier():
-                    raise Exception(
-                        "Context for new event %s has no state "
-                        "group" % (ev.event_id,)
-                    )
-                continue
-
-            if ctx.state_group in state_groups_map:
-                continue
-
-            # We're only interested in pulling out state that has already
-            # been cached in the context. We'll pull stuff out of the DB later
-            # if necessary.
-            current_state_ids = ctx.get_cached_current_state_ids()
-            if current_state_ids is not None:
-                state_groups_map[ctx.state_group] = current_state_ids
-
-            if ctx.prev_group:
-                state_group_deltas[(ctx.prev_group, ctx.state_group)] = ctx.delta_ids
-
-        # We need to map the event_ids to their state groups. First, let's
-        # check if the event is one we're persisting, in which case we can
-        # pull the state group from its context.
-        # Otherwise we need to pull the state group from the database.
-
-        # Set of events we need to fetch groups for. (We know none of the old
-        # extremities are going to be in events_context).
-        missing_event_ids = set(old_latest_event_ids)
-
-        event_id_to_state_group = {}
-        for event_id in new_latest_event_ids:
-            # First search in the list of new events we're adding.
-            for ev, ctx in events_context:
-                if event_id == ev.event_id and ctx.state_group is not None:
-                    event_id_to_state_group[event_id] = ctx.state_group
-                    break
-            else:
-                # If we couldn't find it, then we'll need to pull
-                # the state from the database
-                missing_event_ids.add(event_id)
-
-        if missing_event_ids:
-            # Now pull out the state groups for any missing events from DB
-            event_to_groups = yield self._get_state_group_for_events(missing_event_ids)
-            event_id_to_state_group.update(event_to_groups)
-
-        # State groups of old_latest_event_ids
-        old_state_groups = set(
-            event_id_to_state_group[evid] for evid in old_latest_event_ids
-        )
-
-        # State groups of new_latest_event_ids
-        new_state_groups = set(
-            event_id_to_state_group[evid] for evid in new_latest_event_ids
-        )
-
-        # If they old and new groups are the same then we don't need to do
-        # anything.
-        if old_state_groups == new_state_groups:
-            return None, None
-
-        if len(new_state_groups) == 1 and len(old_state_groups) == 1:
-            # If we're going from one state group to another, lets check if
-            # we have a delta for that transition. If we do then we can just
-            # return that.
-
-            new_state_group = next(iter(new_state_groups))
-            old_state_group = next(iter(old_state_groups))
-
-            delta_ids = state_group_deltas.get((old_state_group, new_state_group), None)
-            if delta_ids is not None:
-                # We have a delta from the existing to new current state,
-                # so lets just return that. If we happen to already have
-                # the current state in memory then lets also return that,
-                # but it doesn't matter if we don't.
-                new_state = state_groups_map.get(new_state_group)
-                return new_state, delta_ids
-
-        # Now that we have calculated new_state_groups we need to get
-        # their state IDs so we can resolve to a single state set.
-        missing_state = new_state_groups - set(state_groups_map)
-        if missing_state:
-            group_to_state = yield self._get_state_for_groups(missing_state)
-            state_groups_map.update(group_to_state)
-
-        if len(new_state_groups) == 1:
-            # If there is only one state group, then we know what the current
-            # state is.
-            return state_groups_map[new_state_groups.pop()], None
-
-        # Ok, we need to defer to the state handler to resolve our state sets.
-
-        state_groups = {sg: state_groups_map[sg] for sg in new_state_groups}
-
-        events_map = {ev.event_id: ev for ev, _ in events_context}
-
-        # We need to get the room version, which is in the create event.
-        # Normally that'd be in the database, but its also possible that we're
-        # currently trying to persist it.
-        room_version = None
-        for ev, _ in events_context:
-            if ev.type == EventTypes.Create and ev.state_key == "":
-                room_version = ev.content.get("room_version", "1")
-                break
-
-        if not room_version:
-            room_version = yield self.get_room_version(room_id)
-
-        logger.debug("calling resolve_state_groups from preserve_events")
-        res = yield self._state_resolution_handler.resolve_state_groups(
-            room_id,
-            room_version,
-            state_groups,
-            events_map,
-            state_res_store=StateResolutionStore(self),
-        )
-
-        return res.state, None
-
-    @defer.inlineCallbacks
-    def _calculate_state_delta(self, room_id, current_state):
-        """Calculate the new state deltas for a room.
-
-        Assumes that we are only persisting events for one room at a time.
-
-        Returns:
-            tuple[list, dict] (to_delete, to_insert): where to_delete are the
-            type/state_keys to remove from current_state_events and `to_insert`
-            are the updates to current_state_events.
-        """
-        existing_state = yield self.get_current_state_ids(room_id)
-
-        to_delete = [key for key in existing_state if key not in current_state]
-
-        to_insert = {
-            key: ev_id
-            for key, ev_id in iteritems(current_state)
-            if ev_id != existing_state.get(key)
-        }
-
-        return to_delete, to_insert
-
     @log_function
     def _persist_events_txn(
         self,
@@ -1380,6 +836,18 @@ class EventsStore(
             ],
         )
 
+        for event, _ in events_and_contexts:
+            if not event.internal_metadata.is_redacted():
+                # If we're persisting an unredacted event we go and ensure
+                # that we mark any redactions that reference this event as
+                # requiring censoring.
+                self._simple_update_txn(
+                    txn,
+                    table="redactions",
+                    keyvalues={"redacts": event.event_id},
+                    updatevalues={"have_censored": False},
+                )
+
     def _store_rejected_events_txn(self, txn, events_and_contexts):
         """Add rows to the 'rejections' table for received events which were
         rejected
@@ -1464,6 +932,13 @@ class EventsStore(
 
             self._handle_event_relations(txn, event)
 
+            # Store the labels for this event.
+            labels = event.content.get(EventContentFields.LABELS)
+            if labels:
+                self.insert_labels_for_event_txn(
+                    txn, event.event_id, labels, event.room_id, event.depth
+                )
+
         # Insert into the room_memberships table.
         self._store_room_members_txn(
             txn,
@@ -1522,10 +997,14 @@ class EventsStore(
                 " FROM events as e"
                 " LEFT JOIN rejections as rej USING (event_id)"
                 " LEFT JOIN redactions as r ON e.event_id = r.redacts"
-                " WHERE e.event_id IN (%s)"
-            ) % (",".join(["?"] * len(ev_map)),)
+                " WHERE "
+            )
 
-            txn.execute(sql, list(ev_map))
+            clause, args = make_in_list_sql_clause(
+                self.database_engine, "e.event_id", list(ev_map)
+            )
+
+            txn.execute(sql + clause, args)
             rows = self.cursor_to_dict(txn)
             for row in rows:
                 event = ev_map[row["event_id"]]
@@ -1543,11 +1022,100 @@ class EventsStore(
     def _store_redaction(self, txn, event):
         # invalidate the cache for the redacted event
         txn.call_after(self._invalidate_get_event_cache, event.redacts)
-        txn.execute(
-            "INSERT INTO redactions (event_id, redacts) VALUES (?,?)",
-            (event.event_id, event.redacts),
+
+        self._simple_insert_txn(
+            txn,
+            table="redactions",
+            values={
+                "event_id": event.event_id,
+                "redacts": event.redacts,
+                "received_ts": self._clock.time_msec(),
+            },
         )
 
+    @defer.inlineCallbacks
+    def _censor_redactions(self):
+        """Censors all redactions older than the configured period that haven't
+        been censored yet.
+
+        By censor we mean update the event_json table with the redacted event.
+
+        Returns:
+            Deferred
+        """
+
+        if self.hs.config.redaction_retention_period is None:
+            return
+
+        before_ts = self._clock.time_msec() - self.hs.config.redaction_retention_period
+
+        # We fetch all redactions that:
+        #   1. point to an event we have,
+        #   2. has a received_ts from before the cut off, and
+        #   3. we haven't yet censored.
+        #
+        # This is limited to 100 events to ensure that we don't try and do too
+        # much at once. We'll get called again so this should eventually catch
+        # up.
+        sql = """
+            SELECT redactions.event_id, redacts FROM redactions
+            LEFT JOIN events AS original_event ON (
+                redacts = original_event.event_id
+            )
+            WHERE NOT have_censored
+            AND redactions.received_ts <= ?
+            ORDER BY redactions.received_ts ASC
+            LIMIT ?
+        """
+
+        rows = yield self._execute(
+            "_censor_redactions_fetch", None, sql, before_ts, 100
+        )
+
+        updates = []
+
+        for redaction_id, event_id in rows:
+            redaction_event = yield self.get_event(redaction_id, allow_none=True)
+            original_event = yield self.get_event(
+                event_id, allow_rejected=True, allow_none=True
+            )
+
+            # The SQL above ensures that we have both the redaction and
+            # original event, so if the `get_event` calls return None it
+            # means that the redaction wasn't allowed. Either way we know that
+            # the result won't change so we mark the fact that we've checked.
+            if (
+                redaction_event
+                and original_event
+                and original_event.internal_metadata.is_redacted()
+            ):
+                # Redaction was allowed
+                pruned_json = encode_json(prune_event_dict(original_event.get_dict()))
+            else:
+                # Redaction wasn't allowed
+                pruned_json = None
+
+            updates.append((redaction_id, event_id, pruned_json))
+
+        def _update_censor_txn(txn):
+            for redaction_id, event_id, pruned_json in updates:
+                if pruned_json:
+                    self._simple_update_one_txn(
+                        txn,
+                        table="event_json",
+                        keyvalues={"event_id": event_id},
+                        updatevalues={"json": pruned_json},
+                    )
+
+                self._simple_update_one_txn(
+                    txn,
+                    table="redactions",
+                    keyvalues={"event_id": redaction_id},
+                    updatevalues={"have_censored": True},
+                )
+
+        yield self.runInteraction("_update_censor_txn", _update_censor_txn)
+
     @defer.inlineCallbacks
     def count_daily_messages(self):
         """
@@ -1564,7 +1132,7 @@ class EventsStore(
                 AND stream_ordering > ?
             """
             txn.execute(sql, (self.stream_ordering_day_ago,))
-            count, = txn.fetchone()
+            (count,) = txn.fetchone()
             return count
 
         ret = yield self.runInteraction("count_messages", _count_messages)
@@ -1585,7 +1153,7 @@ class EventsStore(
             """
 
             txn.execute(sql, (like_clause, self.stream_ordering_day_ago))
-            count, = txn.fetchone()
+            (count,) = txn.fetchone()
             return count
 
         ret = yield self.runInteraction("count_daily_sent_messages", _count_messages)
@@ -1600,7 +1168,7 @@ class EventsStore(
                 AND stream_ordering > ?
             """
             txn.execute(sql, (self.stream_ordering_day_ago,))
-            count, = txn.fetchone()
+            (count,) = txn.fetchone()
             return count
 
         ret = yield self.runInteraction("count_daily_active_rooms", _count)
@@ -2085,7 +1653,7 @@ class EventsStore(
         """,
             (room_id,),
         )
-        min_depth, = txn.fetchone()
+        (min_depth,) = txn.fetchone()
 
         logger.info("[purge] updating room_depth to %d", min_depth)
 
@@ -2139,11 +1707,12 @@ class EventsStore(
             sql = """
                 SELECT DISTINCT state_group FROM event_to_state_groups
                 LEFT JOIN events_to_purge AS ep USING (event_id)
-                WHERE state_group IN (%s) AND ep.event_id IS NULL
-            """ % (
-                ",".join("?" for _ in current_search),
+                WHERE ep.event_id IS NULL AND
+            """
+            clause, args = make_in_list_sql_clause(
+                txn.database_engine, "state_group", current_search
             )
-            txn.execute(sql, list(current_search))
+            txn.execute(sql + clause, list(args))
 
             referenced = set(sg for sg, in txn)
             referenced_groups |= referenced
@@ -2270,12 +1839,12 @@ class EventsStore(
             "room_aliases",
             "room_depth",
             "room_memberships",
-            "room_state",
-            "room_stats",
+            "room_stats_state",
+            "room_stats_current",
+            "room_stats_historical",
             "room_stats_earliest_token",
             "rooms",
             "stream_ordering_to_exterm",
-            "topics",
             "users_in_public_rooms",
             "users_who_share_private_rooms",
             # no useful index, but let's clear them anyway
@@ -2318,12 +1887,11 @@ class EventsStore(
 
         logger.info("[purge] done")
 
-    @defer.inlineCallbacks
-    def is_event_after(self, event_id1, event_id2):
+    async def is_event_after(self, event_id1, event_id2):
         """Returns True if event_id1 is after event_id2 in the stream
         """
-        to_1, so_1 = yield self._get_event_ordering(event_id1)
-        to_2, so_2 = yield self._get_event_ordering(event_id2)
+        to_1, so_1 = await self._get_event_ordering(event_id1)
+        to_2, so_2 = await self._get_event_ordering(event_id2)
         return (to_1, so_1) > (to_2, so_2)
 
     @cachedInlineCallbacks(max_entries=5000)
@@ -2356,6 +1924,33 @@ class EventsStore(
             get_all_updated_current_state_deltas_txn,
         )
 
+    def insert_labels_for_event_txn(
+        self, txn, event_id, labels, room_id, topological_ordering
+    ):
+        """Store the mapping between an event's ID and its labels, with one row per
+        (event_id, label) tuple.
+
+        Args:
+            txn (LoggingTransaction): The transaction to execute.
+            event_id (str): The event's ID.
+            labels (list[str]): A list of text labels.
+            room_id (str): The ID of the room the event was sent to.
+            topological_ordering (int): The position of the event in the room's topology.
+        """
+        return self._simple_insert_many_txn(
+            txn=txn,
+            table="event_labels",
+            values=[
+                {
+                    "event_id": event_id,
+                    "label": label,
+                    "room_id": room_id,
+                    "topological_ordering": topological_ordering,
+                }
+                for label in labels
+            ],
+        )
+
 
 AllNewEventsResult = namedtuple(
     "AllNewEventsResult",
diff --git a/synapse/storage/events_bg_updates.py b/synapse/storage/data_stores/main/events_bg_updates.py
similarity index 79%
rename from synapse/storage/events_bg_updates.py
rename to synapse/storage/data_stores/main/events_bg_updates.py
index 6587f31e2b..51352b9966 100644
--- a/synapse/storage/events_bg_updates.py
+++ b/synapse/storage/data_stores/main/events_bg_updates.py
@@ -21,6 +21,7 @@ from canonicaljson import json
 
 from twisted.internet import defer
 
+from synapse.storage._base import make_in_list_sql_clause
 from synapse.storage.background_updates import BackgroundUpdateStore
 
 logger = logging.getLogger(__name__)
@@ -67,6 +68,23 @@ class EventsBackgroundUpdatesStore(BackgroundUpdateStore):
             self.DELETE_SOFT_FAILED_EXTREMITIES, self._cleanup_extremities_bg_update
         )
 
+        self.register_background_update_handler(
+            "redactions_received_ts", self._redactions_received_ts
+        )
+
+        # This index gets deleted in `event_fix_redactions_bytes` update
+        self.register_background_index_update(
+            "event_fix_redactions_bytes_create_index",
+            index_name="redactions_censored_redacts",
+            table="redactions",
+            columns=["redacts"],
+            where_clause="have_censored",
+        )
+
+        self.register_background_update_handler(
+            "event_fix_redactions_bytes", self._event_fix_redactions_bytes
+        )
+
     @defer.inlineCallbacks
     def _background_reindex_fields_sender(self, progress, batch_size):
         target_min_stream_id = progress["target_min_stream_id_inclusive"]
@@ -308,12 +326,13 @@ class EventsBackgroundUpdatesStore(BackgroundUpdateStore):
                     INNER JOIN event_json USING (event_id)
                     LEFT JOIN rejections USING (event_id)
                     WHERE
-                        prev_event_id IN (%s)
-                        AND NOT events.outlier
-                """ % (
-                    ",".join("?" for _ in to_check),
+                        NOT events.outlier
+                        AND
+                """
+                clause, args = make_in_list_sql_clause(
+                    self.database_engine, "prev_event_id", to_check
                 )
-                txn.execute(sql, to_check)
+                txn.execute(sql + clause, list(args))
 
                 for prev_event_id, event_id, metadata, rejected in txn:
                     if event_id in graph:
@@ -397,3 +416,90 @@ class EventsBackgroundUpdatesStore(BackgroundUpdateStore):
             )
 
         return num_handled
+
+    @defer.inlineCallbacks
+    def _redactions_received_ts(self, progress, batch_size):
+        """Handles filling out the `received_ts` column in redactions.
+        """
+        last_event_id = progress.get("last_event_id", "")
+
+        def _redactions_received_ts_txn(txn):
+            # Fetch the set of event IDs that we want to update
+            sql = """
+                SELECT event_id FROM redactions
+                WHERE event_id > ?
+                ORDER BY event_id ASC
+                LIMIT ?
+            """
+
+            txn.execute(sql, (last_event_id, batch_size))
+
+            rows = txn.fetchall()
+            if not rows:
+                return 0
+
+            (upper_event_id,) = rows[-1]
+
+            # Update the redactions with the received_ts.
+            #
+            # Note: Not all events have an associated received_ts, so we
+            # fallback to using origin_server_ts. If we for some reason don't
+            # have an origin_server_ts, lets just use the current timestamp.
+            #
+            # We don't want to leave it null, as then we'll never try and
+            # censor those redactions.
+            sql = """
+                UPDATE redactions
+                SET received_ts = (
+                    SELECT COALESCE(received_ts, origin_server_ts, ?) FROM events
+                    WHERE events.event_id = redactions.event_id
+                )
+                WHERE ? <= event_id AND event_id <= ?
+            """
+
+            txn.execute(sql, (self._clock.time_msec(), last_event_id, upper_event_id))
+
+            self._background_update_progress_txn(
+                txn, "redactions_received_ts", {"last_event_id": upper_event_id}
+            )
+
+            return len(rows)
+
+        count = yield self.runInteraction(
+            "_redactions_received_ts", _redactions_received_ts_txn
+        )
+
+        if not count:
+            yield self._end_background_update("redactions_received_ts")
+
+        return count
+
+    @defer.inlineCallbacks
+    def _event_fix_redactions_bytes(self, progress, batch_size):
+        """Undoes hex encoded censored redacted event JSON.
+        """
+
+        def _event_fix_redactions_bytes_txn(txn):
+            # This update is quite fast due to new index.
+            txn.execute(
+                """
+                UPDATE event_json
+                SET
+                    json = convert_from(json::bytea, 'utf8')
+                FROM redactions
+                WHERE
+                    redactions.have_censored
+                    AND event_json.event_id = redactions.redacts
+                    AND json NOT LIKE '{%';
+                """
+            )
+
+            txn.execute("DROP INDEX redactions_censored_redacts")
+
+        yield self.runInteraction(
+            "_event_fix_redactions_bytes", _event_fix_redactions_bytes_txn
+        )
+
+        yield self._end_background_update("event_fix_redactions_bytes")
+
+        return 1
diff --git a/synapse/storage/events_worker.py b/synapse/storage/data_stores/main/events_worker.py
similarity index 95%
rename from synapse/storage/events_worker.py
rename to synapse/storage/data_stores/main/events_worker.py
index c6fa7f82fd..4c4b76bd93 100644
--- a/synapse/storage/events_worker.py
+++ b/synapse/storage/data_stores/main/events_worker.py
@@ -31,12 +31,11 @@ from synapse.events.snapshot import EventContext  # noqa: F401
 from synapse.events.utils import prune_event
 from synapse.logging.context import LoggingContext, PreserveLoggingContext
 from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.storage._base import SQLBaseStore, make_in_list_sql_clause
 from synapse.types import get_domain_from_id
 from synapse.util import batch_iter
 from synapse.util.metrics import Measure
 
-from ._base import SQLBaseStore
-
 logger = logging.getLogger(__name__)
 
 
@@ -238,6 +237,20 @@ class EventsWorkerStore(SQLBaseStore):
             # we have to recheck auth now.
 
             if not allow_rejected and entry.event.type == EventTypes.Redaction:
+                if not hasattr(entry.event, "redacts"):
+                    # A redacted redaction doesn't have a `redacts` key, in
+                    # which case lets just withhold the event.
+                    #
+                    # Note: Most of the time if the redactions has been
+                    # redacted we still have the un-redacted event in the DB
+                    # and so we'll still see the `redacts` key. However, this
+                    # isn't always true e.g. if we have censored the event.
+                    logger.debug(
+                        "Withholding redaction event %s as we don't have redacts key",
+                        event_id,
+                    )
+                    continue
+
                 redacted_event_id = entry.event.redacts
                 event_map = yield self._get_events_from_cache_or_db([redacted_event_id])
                 original_event_entry = event_map.get(redacted_event_id)
@@ -609,10 +622,14 @@ class EventsWorkerStore(SQLBaseStore):
                 " rej.reason "
                 " FROM event_json as e"
                 " LEFT JOIN rejections as rej USING (event_id)"
-                " WHERE e.event_id IN (%s)"
-            ) % (",".join(["?"] * len(evs)),)
+                " WHERE "
+            )
 
-            txn.execute(sql, evs)
+            clause, args = make_in_list_sql_clause(
+                txn.database_engine, "e.event_id", evs
+            )
+
+            txn.execute(sql + clause, args)
 
             for row in txn:
                 event_id = row[0]
@@ -626,11 +643,11 @@ class EventsWorkerStore(SQLBaseStore):
                 }
 
             # check for redactions
-            redactions_sql = (
-                "SELECT event_id, redacts FROM redactions WHERE redacts IN (%s)"
-            ) % (",".join(["?"] * len(evs)),)
+            redactions_sql = "SELECT event_id, redacts FROM redactions WHERE "
 
-            txn.execute(redactions_sql, evs)
+            clause, args = make_in_list_sql_clause(txn.database_engine, "redacts", evs)
+
+            txn.execute(redactions_sql + clause, args)
 
             for (redacter, redacted) in txn:
                 d = event_dict.get(redacted)
@@ -739,10 +756,11 @@ class EventsWorkerStore(SQLBaseStore):
         results = set()
 
         def have_seen_events_txn(txn, chunk):
-            sql = "SELECT event_id FROM events as e WHERE e.event_id IN (%s)" % (
-                ",".join("?" * len(chunk)),
+            sql = "SELECT event_id FROM events as e WHERE "
+            clause, args = make_in_list_sql_clause(
+                txn.database_engine, "e.event_id", chunk
             )
-            txn.execute(sql, chunk)
+            txn.execute(sql + clause, args)
             for (event_id,) in txn:
                 results.add(event_id)
 
diff --git a/synapse/storage/filtering.py b/synapse/storage/data_stores/main/filtering.py
similarity index 92%
rename from synapse/storage/filtering.py
rename to synapse/storage/data_stores/main/filtering.py
index 23b48f6cea..a2a2a67927 100644
--- a/synapse/storage/filtering.py
+++ b/synapse/storage/data_stores/main/filtering.py
@@ -16,10 +16,9 @@
 from canonicaljson import encode_canonical_json
 
 from synapse.api.errors import Codes, SynapseError
+from synapse.storage._base import SQLBaseStore, db_to_json
 from synapse.util.caches.descriptors import cachedInlineCallbacks
 
-from ._base import SQLBaseStore, db_to_json
-
 
 class FilteringStore(SQLBaseStore):
     @cachedInlineCallbacks(num_args=2)
@@ -51,7 +50,7 @@ class FilteringStore(SQLBaseStore):
                 "SELECT filter_id FROM user_filters "
                 "WHERE user_id = ? AND filter_json = ?"
             )
-            txn.execute(sql, (user_localpart, def_json))
+            txn.execute(sql, (user_localpart, bytearray(def_json)))
             filter_id_response = txn.fetchone()
             if filter_id_response is not None:
                 return filter_id_response[0]
@@ -68,7 +67,7 @@ class FilteringStore(SQLBaseStore):
                 "INSERT INTO user_filters (user_id, filter_id, filter_json)"
                 "VALUES(?, ?, ?)"
             )
-            txn.execute(sql, (user_localpart, filter_id, def_json))
+            txn.execute(sql, (user_localpart, filter_id, bytearray(def_json)))
 
             return filter_id
 
diff --git a/synapse/storage/group_server.py b/synapse/storage/data_stores/main/group_server.py
similarity index 99%
rename from synapse/storage/group_server.py
rename to synapse/storage/data_stores/main/group_server.py
index 15b01c6958..b3a2771f1b 100644
--- a/synapse/storage/group_server.py
+++ b/synapse/storage/data_stores/main/group_server.py
@@ -19,8 +19,7 @@ from canonicaljson import json
 from twisted.internet import defer
 
 from synapse.api.errors import SynapseError
-
-from ._base import SQLBaseStore
+from synapse.storage._base import SQLBaseStore
 
 # The category ID for the "default" category. We don't store as null in the
 # database to avoid the fun of null != null
@@ -250,7 +249,7 @@ class GroupServerStore(SQLBaseStore):
                 WHERE group_id = ? AND category_id = ?
             """
             txn.execute(sql, (group_id, category_id))
-            order, = txn.fetchone()
+            (order,) = txn.fetchone()
 
         if existing:
             to_update = {}
@@ -510,7 +509,7 @@ class GroupServerStore(SQLBaseStore):
                 WHERE group_id = ? AND role_id = ?
             """
             txn.execute(sql, (group_id, role_id))
-            order, = txn.fetchone()
+            (order,) = txn.fetchone()
 
         if existing:
             to_update = {}
diff --git a/synapse/storage/data_stores/main/keys.py b/synapse/storage/data_stores/main/keys.py
new file mode 100644
index 0000000000..ebc7db3ed6
--- /dev/null
+++ b/synapse/storage/data_stores/main/keys.py
@@ -0,0 +1,214 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2019 New Vector Ltd.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import itertools
+import logging
+
+import six
+
+from signedjson.key import decode_verify_key_bytes
+
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.keys import FetchKeyResult
+from synapse.util import batch_iter
+from synapse.util.caches.descriptors import cached, cachedList
+
+logger = logging.getLogger(__name__)
+
+# py2 sqlite has buffer hardcoded as only binary type, so we must use it,
+# despite being deprecated and removed in favor of memoryview
+if six.PY2:
+    db_binary_type = six.moves.builtins.buffer
+else:
+    db_binary_type = memoryview
+
+
+class KeyStore(SQLBaseStore):
+    """Persistence for signature verification keys
+    """
+
+    @cached()
+    def _get_server_verify_key(self, server_name_and_key_id):
+        raise NotImplementedError()
+
+    @cachedList(
+        cached_method_name="_get_server_verify_key", list_name="server_name_and_key_ids"
+    )
+    def get_server_verify_keys(self, server_name_and_key_ids):
+        """
+        Args:
+            server_name_and_key_ids (iterable[Tuple[str, str]]):
+                iterable of (server_name, key-id) tuples to fetch keys for
+
+        Returns:
+            Deferred: resolves to dict[Tuple[str, str], FetchKeyResult|None]:
+                map from (server_name, key_id) -> FetchKeyResult, or None if the key is
+                unknown
+        """
+        keys = {}
+
+        def _get_keys(txn, batch):
+            """Processes a batch of keys to fetch, and adds the result to `keys`."""
+
+            # batch_iter always returns tuples so it's safe to do len(batch)
+            sql = (
+                "SELECT server_name, key_id, verify_key, ts_valid_until_ms "
+                "FROM server_signature_keys WHERE 1=0"
+            ) + " OR (server_name=? AND key_id=?)" * len(batch)
+
+            txn.execute(sql, tuple(itertools.chain.from_iterable(batch)))
+
+            for row in txn:
+                server_name, key_id, key_bytes, ts_valid_until_ms = row
+
+                if ts_valid_until_ms is None:
+                    # Old keys may be stored with a ts_valid_until_ms of null,
+                    # in which case we treat this as if it was set to `0`, i.e.
+                    # it won't match key requests that define a minimum
+                    # `ts_valid_until_ms`.
+                    ts_valid_until_ms = 0
+
+                res = FetchKeyResult(
+                    verify_key=decode_verify_key_bytes(key_id, bytes(key_bytes)),
+                    valid_until_ts=ts_valid_until_ms,
+                )
+                keys[(server_name, key_id)] = res
+
+        def _txn(txn):
+            for batch in batch_iter(server_name_and_key_ids, 50):
+                _get_keys(txn, batch)
+            return keys
+
+        return self.runInteraction("get_server_verify_keys", _txn)
+
+    def store_server_verify_keys(self, from_server, ts_added_ms, verify_keys):
+        """Stores NACL verification keys for remote servers.
+        Args:
+            from_server (str): Where the verification keys were looked up
+            ts_added_ms (int): The time to record that the key was added
+            verify_keys (iterable[tuple[str, str, FetchKeyResult]]):
+                keys to be stored. Each entry is a triplet of
+                (server_name, key_id, key).
+        """
+        key_values = []
+        value_values = []
+        invalidations = []
+        for server_name, key_id, fetch_result in verify_keys:
+            key_values.append((server_name, key_id))
+            value_values.append(
+                (
+                    from_server,
+                    ts_added_ms,
+                    fetch_result.valid_until_ts,
+                    db_binary_type(fetch_result.verify_key.encode()),
+                )
+            )
+            # invalidate takes a tuple corresponding to the params of
+            # _get_server_verify_key. _get_server_verify_key only takes one
+            # param, which is itself the 2-tuple (server_name, key_id).
+            invalidations.append((server_name, key_id))
+
+        def _invalidate(res):
+            f = self._get_server_verify_key.invalidate
+            for i in invalidations:
+                f((i,))
+            return res
+
+        return self.runInteraction(
+            "store_server_verify_keys",
+            self._simple_upsert_many_txn,
+            table="server_signature_keys",
+            key_names=("server_name", "key_id"),
+            key_values=key_values,
+            value_names=(
+                "from_server",
+                "ts_added_ms",
+                "ts_valid_until_ms",
+                "verify_key",
+            ),
+            value_values=value_values,
+        ).addCallback(_invalidate)
+
+    def store_server_keys_json(
+        self, server_name, key_id, from_server, ts_now_ms, ts_expires_ms, key_json_bytes
+    ):
+        """Stores the JSON bytes for a set of keys from a server
+        The JSON should be signed by the originating server, the intermediate
+        server, and by this server. Updates the value for the
+        (server_name, key_id, from_server) triplet if one already existed.
+        Args:
+            server_name (str): The name of the server.
+            key_id (str): The identifer of the key this JSON is for.
+            from_server (str): The server this JSON was fetched from.
+            ts_now_ms (int): The time now in milliseconds.
+            ts_valid_until_ms (int): The time when this json stops being valid.
+            key_json (bytes): The encoded JSON.
+        """
+        return self._simple_upsert(
+            table="server_keys_json",
+            keyvalues={
+                "server_name": server_name,
+                "key_id": key_id,
+                "from_server": from_server,
+            },
+            values={
+                "server_name": server_name,
+                "key_id": key_id,
+                "from_server": from_server,
+                "ts_added_ms": ts_now_ms,
+                "ts_valid_until_ms": ts_expires_ms,
+                "key_json": db_binary_type(key_json_bytes),
+            },
+            desc="store_server_keys_json",
+        )
+
+    def get_server_keys_json(self, server_keys):
+        """Retrive the key json for a list of server_keys and key ids.
+        If no keys are found for a given server, key_id and source then
+        that server, key_id, and source triplet entry will be an empty list.
+        The JSON is returned as a byte array so that it can be efficiently
+        used in an HTTP response.
+        Args:
+            server_keys (list): List of (server_name, key_id, source) triplets.
+        Returns:
+            Deferred[dict[Tuple[str, str, str|None], list[dict]]]:
+                Dict mapping (server_name, key_id, source) triplets to lists of dicts
+        """
+
+        def _get_server_keys_json_txn(txn):
+            results = {}
+            for server_name, key_id, from_server in server_keys:
+                keyvalues = {"server_name": server_name}
+                if key_id is not None:
+                    keyvalues["key_id"] = key_id
+                if from_server is not None:
+                    keyvalues["from_server"] = from_server
+                rows = self._simple_select_list_txn(
+                    txn,
+                    "server_keys_json",
+                    keyvalues=keyvalues,
+                    retcols=(
+                        "key_id",
+                        "from_server",
+                        "ts_added_ms",
+                        "ts_valid_until_ms",
+                        "key_json",
+                    ),
+                )
+                results[(server_name, key_id, from_server)] = rows
+            return results
+
+        return self.runInteraction("get_server_keys_json", _get_server_keys_json_txn)
diff --git a/synapse/storage/media_repository.py b/synapse/storage/data_stores/main/media_repository.py
similarity index 97%
rename from synapse/storage/media_repository.py
rename to synapse/storage/data_stores/main/media_repository.py
index 6b1238ce4a..84b5f3ad5e 100644
--- a/synapse/storage/media_repository.py
+++ b/synapse/storage/data_stores/main/media_repository.py
@@ -15,11 +15,9 @@
 from synapse.storage.background_updates import BackgroundUpdateStore
 
 
-class MediaRepositoryStore(BackgroundUpdateStore):
-    """Persistence for attachments and avatars"""
-
+class MediaRepositoryBackgroundUpdateStore(BackgroundUpdateStore):
     def __init__(self, db_conn, hs):
-        super(MediaRepositoryStore, self).__init__(db_conn, hs)
+        super(MediaRepositoryBackgroundUpdateStore, self).__init__(db_conn, hs)
 
         self.register_background_index_update(
             update_name="local_media_repository_url_idx",
@@ -29,6 +27,13 @@ class MediaRepositoryStore(BackgroundUpdateStore):
             where_clause="url_cache IS NOT NULL",
         )
 
+
+class MediaRepositoryStore(MediaRepositoryBackgroundUpdateStore):
+    """Persistence for attachments and avatars"""
+
+    def __init__(self, db_conn, hs):
+        super(MediaRepositoryStore, self).__init__(db_conn, hs)
+
     def get_local_media(self, media_id):
         """Get the metadata for a local piece of media
         Returns:
diff --git a/synapse/storage/monthly_active_users.py b/synapse/storage/data_stores/main/monthly_active_users.py
similarity index 79%
rename from synapse/storage/monthly_active_users.py
rename to synapse/storage/data_stores/main/monthly_active_users.py
index 752e9788a2..b41c3d317a 100644
--- a/synapse/storage/monthly_active_users.py
+++ b/synapse/storage/data_stores/main/monthly_active_users.py
@@ -16,10 +16,9 @@ import logging
 
 from twisted.internet import defer
 
+from synapse.storage._base import SQLBaseStore
 from synapse.util.caches.descriptors import cached
 
-from ._base import SQLBaseStore
-
 logger = logging.getLogger(__name__)
 
 # Number of msec of granularity to store the monthly_active_user timestamp
@@ -32,7 +31,6 @@ class MonthlyActiveUsersStore(SQLBaseStore):
         super(MonthlyActiveUsersStore, self).__init__(None, hs)
         self._clock = hs.get_clock()
         self.hs = hs
-        self.reserved_users = ()
         # Do not add more reserved users than the total allowable number
         self._new_transaction(
             dbconn,
@@ -51,7 +49,6 @@ class MonthlyActiveUsersStore(SQLBaseStore):
             txn (cursor):
             threepids (list[dict]): List of threepid dicts to reserve
         """
-        reserved_user_list = []
 
         for tp in threepids:
             user_id = self.get_user_id_by_threepid_txn(txn, tp["medium"], tp["address"])
@@ -60,10 +57,8 @@ class MonthlyActiveUsersStore(SQLBaseStore):
                 is_support = self.is_support_user_txn(txn, user_id)
                 if not is_support:
                     self.upsert_monthly_active_user_txn(txn, user_id)
-                    reserved_user_list.append(user_id)
             else:
                 logger.warning("mau limit reserved threepid %s not found in db" % tp)
-        self.reserved_users = tuple(reserved_user_list)
 
     @defer.inlineCallbacks
     def reap_monthly_active_users(self):
@@ -74,8 +69,11 @@ class MonthlyActiveUsersStore(SQLBaseStore):
             Deferred[]
         """
 
-        def _reap_users(txn):
-            # Purge stale users
+        def _reap_users(txn, reserved_users):
+            """
+            Args:
+                reserved_users (tuple): reserved users to preserve
+            """
 
             thirty_days_ago = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24 * 30)
             query_args = [thirty_days_ago]
@@ -83,20 +81,19 @@ class MonthlyActiveUsersStore(SQLBaseStore):
 
             # Need if/else since 'AND user_id NOT IN ({})' fails on Postgres
             # when len(reserved_users) == 0. Works fine on sqlite.
-            if len(self.reserved_users) > 0:
+            if len(reserved_users) > 0:
                 # questionmarks is a hack to overcome sqlite not supporting
                 # tuples in 'WHERE IN %s'
-                questionmarks = "?" * len(self.reserved_users)
+                question_marks = ",".join("?" * len(reserved_users))
 
-                query_args.extend(self.reserved_users)
-                sql = base_sql + """ AND user_id NOT IN ({})""".format(
-                    ",".join(questionmarks)
-                )
+                query_args.extend(reserved_users)
+                sql = base_sql + " AND user_id NOT IN ({})".format(question_marks)
             else:
                 sql = base_sql
 
             txn.execute(sql, query_args)
 
+            max_mau_value = self.hs.config.max_mau_value
             if self.hs.config.limit_usage_by_mau:
                 # If MAU user count still exceeds the MAU threshold, then delete on
                 # a least recently active basis.
@@ -106,31 +103,52 @@ class MonthlyActiveUsersStore(SQLBaseStore):
                 # While Postgres does not require 'LIMIT', but also does not support
                 # negative LIMIT values. So there is no way to write it that both can
                 # support
-                safe_guard = self.hs.config.max_mau_value - len(self.reserved_users)
-                # Must be greater than zero for postgres
-                safe_guard = safe_guard if safe_guard > 0 else 0
-                query_args = [safe_guard]
-
-                base_sql = """
-                    DELETE FROM monthly_active_users
-                    WHERE user_id NOT IN (
-                        SELECT user_id FROM monthly_active_users
-                        ORDER BY timestamp DESC
-                        LIMIT ?
+                if len(reserved_users) == 0:
+                    sql = """
+                        DELETE FROM monthly_active_users
+                        WHERE user_id NOT IN (
+                            SELECT user_id FROM monthly_active_users
+                            ORDER BY timestamp DESC
+                            LIMIT ?
                         )
-                    """
+                        """
+                    txn.execute(sql, (max_mau_value,))
                 # Need if/else since 'AND user_id NOT IN ({})' fails on Postgres
                 # when len(reserved_users) == 0. Works fine on sqlite.
-                if len(self.reserved_users) > 0:
-                    query_args.extend(self.reserved_users)
-                    sql = base_sql + """ AND user_id NOT IN ({})""".format(
-                        ",".join(questionmarks)
-                    )
                 else:
-                    sql = base_sql
-                txn.execute(sql, query_args)
+                    # Must be >= 0 for postgres
+                    num_of_non_reserved_users_to_remove = max(
+                        max_mau_value - len(reserved_users), 0
+                    )
 
-        yield self.runInteraction("reap_monthly_active_users", _reap_users)
+                    # It is important to filter reserved users twice to guard
+                    # against the case where the reserved user is present in the
+                    # SELECT, meaning that a legitmate mau is deleted.
+                    sql = """
+                        DELETE FROM monthly_active_users
+                        WHERE user_id NOT IN (
+                            SELECT user_id FROM monthly_active_users
+                            WHERE user_id NOT IN ({})
+                            ORDER BY timestamp DESC
+                            LIMIT ?
+                        )
+                        AND user_id NOT IN ({})
+                    """.format(
+                        question_marks, question_marks
+                    )
+
+                    query_args = [
+                        *reserved_users,
+                        num_of_non_reserved_users_to_remove,
+                        *reserved_users,
+                    ]
+
+                    txn.execute(sql, query_args)
+
+        reserved_users = yield self.get_registered_reserved_users()
+        yield self.runInteraction(
+            "reap_monthly_active_users", _reap_users, reserved_users
+        )
         # It seems poor to invalidate the whole cache, Postgres supports
         # 'Returning' which would allow me to invalidate only the
         # specific users, but sqlite has no way to do this and instead
@@ -153,27 +171,31 @@ class MonthlyActiveUsersStore(SQLBaseStore):
             sql = "SELECT COALESCE(count(*), 0) FROM monthly_active_users"
 
             txn.execute(sql)
-            count, = txn.fetchone()
+            (count,) = txn.fetchone()
             return count
 
         return self.runInteraction("count_users", _count_users)
 
     @defer.inlineCallbacks
-    def get_registered_reserved_users_count(self):
-        """Of the reserved threepids defined in config, how many are associated
+    def get_registered_reserved_users(self):
+        """Of the reserved threepids defined in config, which are associated
         with registered users?
 
         Returns:
-            Defered[int]: Number of real reserved users
+            Defered[list]: Real reserved users
         """
-        count = 0
-        for tp in self.hs.config.mau_limits_reserved_threepids:
+        users = []
+
+        for tp in self.hs.config.mau_limits_reserved_threepids[
+            : self.hs.config.max_mau_value
+        ]:
             user_id = yield self.hs.get_datastore().get_user_id_by_threepid(
                 tp["medium"], tp["address"]
             )
             if user_id:
-                count = count + 1
-        return count
+                users.append(user_id)
+
+        return users
 
     @defer.inlineCallbacks
     def upsert_monthly_active_user(self, user_id):
diff --git a/synapse/storage/openid.py b/synapse/storage/data_stores/main/openid.py
similarity index 95%
rename from synapse/storage/openid.py
rename to synapse/storage/data_stores/main/openid.py
index b3318045ee..79b40044d9 100644
--- a/synapse/storage/openid.py
+++ b/synapse/storage/data_stores/main/openid.py
@@ -1,4 +1,4 @@
-from ._base import SQLBaseStore
+from synapse.storage._base import SQLBaseStore
 
 
 class OpenIdStore(SQLBaseStore):
diff --git a/synapse/storage/data_stores/main/presence.py b/synapse/storage/data_stores/main/presence.py
new file mode 100644
index 0000000000..523ed6575e
--- /dev/null
+++ b/synapse/storage/data_stores/main/presence.py
@@ -0,0 +1,150 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from twisted.internet import defer
+
+from synapse.storage._base import SQLBaseStore, make_in_list_sql_clause
+from synapse.storage.presence import UserPresenceState
+from synapse.util import batch_iter
+from synapse.util.caches.descriptors import cached, cachedList
+
+
+class PresenceStore(SQLBaseStore):
+    @defer.inlineCallbacks
+    def update_presence(self, presence_states):
+        stream_ordering_manager = self._presence_id_gen.get_next_mult(
+            len(presence_states)
+        )
+
+        with stream_ordering_manager as stream_orderings:
+            yield self.runInteraction(
+                "update_presence",
+                self._update_presence_txn,
+                stream_orderings,
+                presence_states,
+            )
+
+        return stream_orderings[-1], self._presence_id_gen.get_current_token()
+
+    def _update_presence_txn(self, txn, stream_orderings, presence_states):
+        for stream_id, state in zip(stream_orderings, presence_states):
+            txn.call_after(
+                self.presence_stream_cache.entity_has_changed, state.user_id, stream_id
+            )
+            txn.call_after(self._get_presence_for_user.invalidate, (state.user_id,))
+
+        # Actually insert new rows
+        self._simple_insert_many_txn(
+            txn,
+            table="presence_stream",
+            values=[
+                {
+                    "stream_id": stream_id,
+                    "user_id": state.user_id,
+                    "state": state.state,
+                    "last_active_ts": state.last_active_ts,
+                    "last_federation_update_ts": state.last_federation_update_ts,
+                    "last_user_sync_ts": state.last_user_sync_ts,
+                    "status_msg": state.status_msg,
+                    "currently_active": state.currently_active,
+                }
+                for state in presence_states
+            ],
+        )
+
+        # Delete old rows to stop database from getting really big
+        sql = "DELETE FROM presence_stream WHERE stream_id < ? AND "
+
+        for states in batch_iter(presence_states, 50):
+            clause, args = make_in_list_sql_clause(
+                self.database_engine, "user_id", [s.user_id for s in states]
+            )
+            txn.execute(sql + clause, [stream_id] + list(args))
+
+    def get_all_presence_updates(self, last_id, current_id):
+        if last_id == current_id:
+            return defer.succeed([])
+
+        def get_all_presence_updates_txn(txn):
+            sql = (
+                "SELECT stream_id, user_id, state, last_active_ts,"
+                " last_federation_update_ts, last_user_sync_ts, status_msg,"
+                " currently_active"
+                " FROM presence_stream"
+                " WHERE ? < stream_id AND stream_id <= ?"
+            )
+            txn.execute(sql, (last_id, current_id))
+            return txn.fetchall()
+
+        return self.runInteraction(
+            "get_all_presence_updates", get_all_presence_updates_txn
+        )
+
+    @cached()
+    def _get_presence_for_user(self, user_id):
+        raise NotImplementedError()
+
+    @cachedList(
+        cached_method_name="_get_presence_for_user",
+        list_name="user_ids",
+        num_args=1,
+        inlineCallbacks=True,
+    )
+    def get_presence_for_users(self, user_ids):
+        rows = yield self._simple_select_many_batch(
+            table="presence_stream",
+            column="user_id",
+            iterable=user_ids,
+            keyvalues={},
+            retcols=(
+                "user_id",
+                "state",
+                "last_active_ts",
+                "last_federation_update_ts",
+                "last_user_sync_ts",
+                "status_msg",
+                "currently_active",
+            ),
+            desc="get_presence_for_users",
+        )
+
+        for row in rows:
+            row["currently_active"] = bool(row["currently_active"])
+
+        return {row["user_id"]: UserPresenceState(**row) for row in rows}
+
+    def get_current_presence_token(self):
+        return self._presence_id_gen.get_current_token()
+
+    def allow_presence_visible(self, observed_localpart, observer_userid):
+        return self._simple_insert(
+            table="presence_allow_inbound",
+            values={
+                "observed_user_id": observed_localpart,
+                "observer_user_id": observer_userid,
+            },
+            desc="allow_presence_visible",
+            or_ignore=True,
+        )
+
+    def disallow_presence_visible(self, observed_localpart, observer_userid):
+        return self._simple_delete_one(
+            table="presence_allow_inbound",
+            keyvalues={
+                "observed_user_id": observed_localpart,
+                "observer_user_id": observer_userid,
+            },
+            desc="disallow_presence_visible",
+        )
diff --git a/synapse/storage/profile.py b/synapse/storage/data_stores/main/profile.py
similarity index 98%
rename from synapse/storage/profile.py
rename to synapse/storage/data_stores/main/profile.py
index 8a5d8e9b18..e4e8a1c1d6 100644
--- a/synapse/storage/profile.py
+++ b/synapse/storage/data_stores/main/profile.py
@@ -16,9 +16,8 @@
 from twisted.internet import defer
 
 from synapse.api.errors import StoreError
-from synapse.storage.roommember import ProfileInfo
-
-from ._base import SQLBaseStore
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.data_stores.main.roommember import ProfileInfo
 
 
 class ProfileWorkerStore(SQLBaseStore):
@@ -35,7 +34,6 @@ class ProfileWorkerStore(SQLBaseStore):
             if e.code == 404:
                 # no match
                 return ProfileInfo(None, None)
-                return
             else:
                 raise
 
diff --git a/synapse/storage/data_stores/main/push_rule.py b/synapse/storage/data_stores/main/push_rule.py
new file mode 100644
index 0000000000..b520062d84
--- /dev/null
+++ b/synapse/storage/data_stores/main/push_rule.py
@@ -0,0 +1,713 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2018 New Vector Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+import logging
+
+from canonicaljson import json
+
+from twisted.internet import defer
+
+from synapse.push.baserules import list_with_base_rules
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.data_stores.main.appservice import ApplicationServiceWorkerStore
+from synapse.storage.data_stores.main.pusher import PusherWorkerStore
+from synapse.storage.data_stores.main.receipts import ReceiptsWorkerStore
+from synapse.storage.data_stores.main.roommember import RoomMemberWorkerStore
+from synapse.storage.push_rule import InconsistentRuleException, RuleNotFoundException
+from synapse.util.caches.descriptors import cachedInlineCallbacks, cachedList
+from synapse.util.caches.stream_change_cache import StreamChangeCache
+
+logger = logging.getLogger(__name__)
+
+
+def _load_rules(rawrules, enabled_map):
+    ruleslist = []
+    for rawrule in rawrules:
+        rule = dict(rawrule)
+        rule["conditions"] = json.loads(rawrule["conditions"])
+        rule["actions"] = json.loads(rawrule["actions"])
+        ruleslist.append(rule)
+
+    # We're going to be mutating this a lot, so do a deep copy
+    rules = list(list_with_base_rules(ruleslist))
+
+    for i, rule in enumerate(rules):
+        rule_id = rule["rule_id"]
+        if rule_id in enabled_map:
+            if rule.get("enabled", True) != bool(enabled_map[rule_id]):
+                # Rules are cached across users.
+                rule = dict(rule)
+                rule["enabled"] = bool(enabled_map[rule_id])
+                rules[i] = rule
+
+    return rules
+
+
+class PushRulesWorkerStore(
+    ApplicationServiceWorkerStore,
+    ReceiptsWorkerStore,
+    PusherWorkerStore,
+    RoomMemberWorkerStore,
+    SQLBaseStore,
+):
+    """This is an abstract base class where subclasses must implement
+    `get_max_push_rules_stream_id` which can be called in the initializer.
+    """
+
+    # This ABCMeta metaclass ensures that we cannot be instantiated without
+    # the abstract methods being implemented.
+    __metaclass__ = abc.ABCMeta
+
+    def __init__(self, db_conn, hs):
+        super(PushRulesWorkerStore, self).__init__(db_conn, hs)
+
+        push_rules_prefill, push_rules_id = self._get_cache_dict(
+            db_conn,
+            "push_rules_stream",
+            entity_column="user_id",
+            stream_column="stream_id",
+            max_value=self.get_max_push_rules_stream_id(),
+        )
+
+        self.push_rules_stream_cache = StreamChangeCache(
+            "PushRulesStreamChangeCache",
+            push_rules_id,
+            prefilled_cache=push_rules_prefill,
+        )
+
+    @abc.abstractmethod
+    def get_max_push_rules_stream_id(self):
+        """Get the position of the push rules stream.
+
+        Returns:
+            int
+        """
+        raise NotImplementedError()
+
+    @cachedInlineCallbacks(max_entries=5000)
+    def get_push_rules_for_user(self, user_id):
+        rows = yield self._simple_select_list(
+            table="push_rules",
+            keyvalues={"user_name": user_id},
+            retcols=(
+                "user_name",
+                "rule_id",
+                "priority_class",
+                "priority",
+                "conditions",
+                "actions",
+            ),
+            desc="get_push_rules_enabled_for_user",
+        )
+
+        rows.sort(key=lambda row: (-int(row["priority_class"]), -int(row["priority"])))
+
+        enabled_map = yield self.get_push_rules_enabled_for_user(user_id)
+
+        rules = _load_rules(rows, enabled_map)
+
+        return rules
+
+    @cachedInlineCallbacks(max_entries=5000)
+    def get_push_rules_enabled_for_user(self, user_id):
+        results = yield self._simple_select_list(
+            table="push_rules_enable",
+            keyvalues={"user_name": user_id},
+            retcols=("user_name", "rule_id", "enabled"),
+            desc="get_push_rules_enabled_for_user",
+        )
+        return {r["rule_id"]: False if r["enabled"] == 0 else True for r in results}
+
+    def have_push_rules_changed_for_user(self, user_id, last_id):
+        if not self.push_rules_stream_cache.has_entity_changed(user_id, last_id):
+            return defer.succeed(False)
+        else:
+
+            def have_push_rules_changed_txn(txn):
+                sql = (
+                    "SELECT COUNT(stream_id) FROM push_rules_stream"
+                    " WHERE user_id = ? AND ? < stream_id"
+                )
+                txn.execute(sql, (user_id, last_id))
+                (count,) = txn.fetchone()
+                return bool(count)
+
+            return self.runInteraction(
+                "have_push_rules_changed", have_push_rules_changed_txn
+            )
+
+    @cachedList(
+        cached_method_name="get_push_rules_for_user",
+        list_name="user_ids",
+        num_args=1,
+        inlineCallbacks=True,
+    )
+    def bulk_get_push_rules(self, user_ids):
+        if not user_ids:
+            return {}
+
+        results = {user_id: [] for user_id in user_ids}
+
+        rows = yield self._simple_select_many_batch(
+            table="push_rules",
+            column="user_name",
+            iterable=user_ids,
+            retcols=("*",),
+            desc="bulk_get_push_rules",
+        )
+
+        rows.sort(key=lambda row: (-int(row["priority_class"]), -int(row["priority"])))
+
+        for row in rows:
+            results.setdefault(row["user_name"], []).append(row)
+
+        enabled_map_by_user = yield self.bulk_get_push_rules_enabled(user_ids)
+
+        for user_id, rules in results.items():
+            results[user_id] = _load_rules(rules, enabled_map_by_user.get(user_id, {}))
+
+        return results
+
+    @defer.inlineCallbacks
+    def copy_push_rule_from_room_to_room(self, new_room_id, user_id, rule):
+        """Copy a single push rule from one room to another for a specific user.
+
+        Args:
+            new_room_id (str): ID of the new room.
+            user_id (str): ID of user the push rule belongs to.
+            rule (Dict): A push rule.
+        """
+        # Create new rule id
+        rule_id_scope = "/".join(rule["rule_id"].split("/")[:-1])
+        new_rule_id = rule_id_scope + "/" + new_room_id
+
+        # Change room id in each condition
+        for condition in rule.get("conditions", []):
+            if condition.get("key") == "room_id":
+                condition["pattern"] = new_room_id
+
+        # Add the rule for the new room
+        yield self.add_push_rule(
+            user_id=user_id,
+            rule_id=new_rule_id,
+            priority_class=rule["priority_class"],
+            conditions=rule["conditions"],
+            actions=rule["actions"],
+        )
+
+    @defer.inlineCallbacks
+    def copy_push_rules_from_room_to_room_for_user(
+        self, old_room_id, new_room_id, user_id
+    ):
+        """Copy all of the push rules from one room to another for a specific
+        user.
+
+        Args:
+            old_room_id (str): ID of the old room.
+            new_room_id (str): ID of the new room.
+            user_id (str): ID of user to copy push rules for.
+        """
+        # Retrieve push rules for this user
+        user_push_rules = yield self.get_push_rules_for_user(user_id)
+
+        # Get rules relating to the old room and copy them to the new room
+        for rule in user_push_rules:
+            conditions = rule.get("conditions", [])
+            if any(
+                (c.get("key") == "room_id" and c.get("pattern") == old_room_id)
+                for c in conditions
+            ):
+                yield self.copy_push_rule_from_room_to_room(new_room_id, user_id, rule)
+
+    @defer.inlineCallbacks
+    def bulk_get_push_rules_for_room(self, event, context):
+        state_group = context.state_group
+        if not state_group:
+            # If state_group is None it means it has yet to be assigned a
+            # state group, i.e. we need to make sure that calls with a state_group
+            # of None don't hit previous cached calls with a None state_group.
+            # To do this we set the state_group to a new object as object() != object()
+            state_group = object()
+
+        current_state_ids = yield context.get_current_state_ids(self)
+        result = yield self._bulk_get_push_rules_for_room(
+            event.room_id, state_group, current_state_ids, event=event
+        )
+        return result
+
+    @cachedInlineCallbacks(num_args=2, cache_context=True)
+    def _bulk_get_push_rules_for_room(
+        self, room_id, state_group, current_state_ids, cache_context, event=None
+    ):
+        # We don't use `state_group`, its there so that we can cache based
+        # on it. However, its important that its never None, since two current_state's
+        # with a state_group of None are likely to be different.
+        # See bulk_get_push_rules_for_room for how we work around this.
+        assert state_group is not None
+
+        # We also will want to generate notifs for other people in the room so
+        # their unread countss are correct in the event stream, but to avoid
+        # generating them for bot / AS users etc, we only do so for people who've
+        # sent a read receipt into the room.
+
+        users_in_room = yield self._get_joined_users_from_context(
+            room_id,
+            state_group,
+            current_state_ids,
+            on_invalidate=cache_context.invalidate,
+            event=event,
+        )
+
+        # We ignore app service users for now. This is so that we don't fill
+        # up the `get_if_users_have_pushers` cache with AS entries that we
+        # know don't have pushers, nor even read receipts.
+        local_users_in_room = set(
+            u
+            for u in users_in_room
+            if self.hs.is_mine_id(u)
+            and not self.get_if_app_services_interested_in_user(u)
+        )
+
+        # users in the room who have pushers need to get push rules run because
+        # that's how their pushers work
+        if_users_with_pushers = yield self.get_if_users_have_pushers(
+            local_users_in_room, on_invalidate=cache_context.invalidate
+        )
+        user_ids = set(
+            uid for uid, have_pusher in if_users_with_pushers.items() if have_pusher
+        )
+
+        users_with_receipts = yield self.get_users_with_read_receipts_in_room(
+            room_id, on_invalidate=cache_context.invalidate
+        )
+
+        # any users with pushers must be ours: they have pushers
+        for uid in users_with_receipts:
+            if uid in local_users_in_room:
+                user_ids.add(uid)
+
+        rules_by_user = yield self.bulk_get_push_rules(
+            user_ids, on_invalidate=cache_context.invalidate
+        )
+
+        rules_by_user = {k: v for k, v in rules_by_user.items() if v is not None}
+
+        return rules_by_user
+
+    @cachedList(
+        cached_method_name="get_push_rules_enabled_for_user",
+        list_name="user_ids",
+        num_args=1,
+        inlineCallbacks=True,
+    )
+    def bulk_get_push_rules_enabled(self, user_ids):
+        if not user_ids:
+            return {}
+
+        results = {user_id: {} for user_id in user_ids}
+
+        rows = yield self._simple_select_many_batch(
+            table="push_rules_enable",
+            column="user_name",
+            iterable=user_ids,
+            retcols=("user_name", "rule_id", "enabled"),
+            desc="bulk_get_push_rules_enabled",
+        )
+        for row in rows:
+            enabled = bool(row["enabled"])
+            results.setdefault(row["user_name"], {})[row["rule_id"]] = enabled
+        return results
+
+
+class PushRuleStore(PushRulesWorkerStore):
+    @defer.inlineCallbacks
+    def add_push_rule(
+        self,
+        user_id,
+        rule_id,
+        priority_class,
+        conditions,
+        actions,
+        before=None,
+        after=None,
+    ):
+        conditions_json = json.dumps(conditions)
+        actions_json = json.dumps(actions)
+        with self._push_rules_stream_id_gen.get_next() as ids:
+            stream_id, event_stream_ordering = ids
+            if before or after:
+                yield self.runInteraction(
+                    "_add_push_rule_relative_txn",
+                    self._add_push_rule_relative_txn,
+                    stream_id,
+                    event_stream_ordering,
+                    user_id,
+                    rule_id,
+                    priority_class,
+                    conditions_json,
+                    actions_json,
+                    before,
+                    after,
+                )
+            else:
+                yield self.runInteraction(
+                    "_add_push_rule_highest_priority_txn",
+                    self._add_push_rule_highest_priority_txn,
+                    stream_id,
+                    event_stream_ordering,
+                    user_id,
+                    rule_id,
+                    priority_class,
+                    conditions_json,
+                    actions_json,
+                )
+
+    def _add_push_rule_relative_txn(
+        self,
+        txn,
+        stream_id,
+        event_stream_ordering,
+        user_id,
+        rule_id,
+        priority_class,
+        conditions_json,
+        actions_json,
+        before,
+        after,
+    ):
+        # Lock the table since otherwise we'll have annoying races between the
+        # SELECT here and the UPSERT below.
+        self.database_engine.lock_table(txn, "push_rules")
+
+        relative_to_rule = before or after
+
+        res = self._simple_select_one_txn(
+            txn,
+            table="push_rules",
+            keyvalues={"user_name": user_id, "rule_id": relative_to_rule},
+            retcols=["priority_class", "priority"],
+            allow_none=True,
+        )
+
+        if not res:
+            raise RuleNotFoundException(
+                "before/after rule not found: %s" % (relative_to_rule,)
+            )
+
+        base_priority_class = res["priority_class"]
+        base_rule_priority = res["priority"]
+
+        if base_priority_class != priority_class:
+            raise InconsistentRuleException(
+                "Given priority class does not match class of relative rule"
+            )
+
+        if before:
+            # Higher priority rules are executed first, So adding a rule before
+            # a rule means giving it a higher priority than that rule.
+            new_rule_priority = base_rule_priority + 1
+        else:
+            # We increment the priority of the existing rules to make space for
+            # the new rule. Therefore if we want this rule to appear after
+            # an existing rule we give it the priority of the existing rule,
+            # and then increment the priority of the existing rule.
+            new_rule_priority = base_rule_priority
+
+        sql = (
+            "UPDATE push_rules SET priority = priority + 1"
+            " WHERE user_name = ? AND priority_class = ? AND priority >= ?"
+        )
+
+        txn.execute(sql, (user_id, priority_class, new_rule_priority))
+
+        self._upsert_push_rule_txn(
+            txn,
+            stream_id,
+            event_stream_ordering,
+            user_id,
+            rule_id,
+            priority_class,
+            new_rule_priority,
+            conditions_json,
+            actions_json,
+        )
+
+    def _add_push_rule_highest_priority_txn(
+        self,
+        txn,
+        stream_id,
+        event_stream_ordering,
+        user_id,
+        rule_id,
+        priority_class,
+        conditions_json,
+        actions_json,
+    ):
+        # Lock the table since otherwise we'll have annoying races between the
+        # SELECT here and the UPSERT below.
+        self.database_engine.lock_table(txn, "push_rules")
+
+        # find the highest priority rule in that class
+        sql = (
+            "SELECT COUNT(*), MAX(priority) FROM push_rules"
+            " WHERE user_name = ? and priority_class = ?"
+        )
+        txn.execute(sql, (user_id, priority_class))
+        res = txn.fetchall()
+        (how_many, highest_prio) = res[0]
+
+        new_prio = 0
+        if how_many > 0:
+            new_prio = highest_prio + 1
+
+        self._upsert_push_rule_txn(
+            txn,
+            stream_id,
+            event_stream_ordering,
+            user_id,
+            rule_id,
+            priority_class,
+            new_prio,
+            conditions_json,
+            actions_json,
+        )
+
+    def _upsert_push_rule_txn(
+        self,
+        txn,
+        stream_id,
+        event_stream_ordering,
+        user_id,
+        rule_id,
+        priority_class,
+        priority,
+        conditions_json,
+        actions_json,
+        update_stream=True,
+    ):
+        """Specialised version of _simple_upsert_txn that picks a push_rule_id
+        using the _push_rule_id_gen if it needs to insert the rule. It assumes
+        that the "push_rules" table is locked"""
+
+        sql = (
+            "UPDATE push_rules"
+            " SET priority_class = ?, priority = ?, conditions = ?, actions = ?"
+            " WHERE user_name = ? AND rule_id = ?"
+        )
+
+        txn.execute(
+            sql,
+            (priority_class, priority, conditions_json, actions_json, user_id, rule_id),
+        )
+
+        if txn.rowcount == 0:
+            # We didn't update a row with the given rule_id so insert one
+            push_rule_id = self._push_rule_id_gen.get_next()
+
+            self._simple_insert_txn(
+                txn,
+                table="push_rules",
+                values={
+                    "id": push_rule_id,
+                    "user_name": user_id,
+                    "rule_id": rule_id,
+                    "priority_class": priority_class,
+                    "priority": priority,
+                    "conditions": conditions_json,
+                    "actions": actions_json,
+                },
+            )
+
+        if update_stream:
+            self._insert_push_rules_update_txn(
+                txn,
+                stream_id,
+                event_stream_ordering,
+                user_id,
+                rule_id,
+                op="ADD",
+                data={
+                    "priority_class": priority_class,
+                    "priority": priority,
+                    "conditions": conditions_json,
+                    "actions": actions_json,
+                },
+            )
+
+    @defer.inlineCallbacks
+    def delete_push_rule(self, user_id, rule_id):
+        """
+        Delete a push rule. Args specify the row to be deleted and can be
+        any of the columns in the push_rule table, but below are the
+        standard ones
+
+        Args:
+            user_id (str): The matrix ID of the push rule owner
+            rule_id (str): The rule_id of the rule to be deleted
+        """
+
+        def delete_push_rule_txn(txn, stream_id, event_stream_ordering):
+            self._simple_delete_one_txn(
+                txn, "push_rules", {"user_name": user_id, "rule_id": rule_id}
+            )
+
+            self._insert_push_rules_update_txn(
+                txn, stream_id, event_stream_ordering, user_id, rule_id, op="DELETE"
+            )
+
+        with self._push_rules_stream_id_gen.get_next() as ids:
+            stream_id, event_stream_ordering = ids
+            yield self.runInteraction(
+                "delete_push_rule",
+                delete_push_rule_txn,
+                stream_id,
+                event_stream_ordering,
+            )
+
+    @defer.inlineCallbacks
+    def set_push_rule_enabled(self, user_id, rule_id, enabled):
+        with self._push_rules_stream_id_gen.get_next() as ids:
+            stream_id, event_stream_ordering = ids
+            yield self.runInteraction(
+                "_set_push_rule_enabled_txn",
+                self._set_push_rule_enabled_txn,
+                stream_id,
+                event_stream_ordering,
+                user_id,
+                rule_id,
+                enabled,
+            )
+
+    def _set_push_rule_enabled_txn(
+        self, txn, stream_id, event_stream_ordering, user_id, rule_id, enabled
+    ):
+        new_id = self._push_rules_enable_id_gen.get_next()
+        self._simple_upsert_txn(
+            txn,
+            "push_rules_enable",
+            {"user_name": user_id, "rule_id": rule_id},
+            {"enabled": 1 if enabled else 0},
+            {"id": new_id},
+        )
+
+        self._insert_push_rules_update_txn(
+            txn,
+            stream_id,
+            event_stream_ordering,
+            user_id,
+            rule_id,
+            op="ENABLE" if enabled else "DISABLE",
+        )
+
+    @defer.inlineCallbacks
+    def set_push_rule_actions(self, user_id, rule_id, actions, is_default_rule):
+        actions_json = json.dumps(actions)
+
+        def set_push_rule_actions_txn(txn, stream_id, event_stream_ordering):
+            if is_default_rule:
+                # Add a dummy rule to the rules table with the user specified
+                # actions.
+                priority_class = -1
+                priority = 1
+                self._upsert_push_rule_txn(
+                    txn,
+                    stream_id,
+                    event_stream_ordering,
+                    user_id,
+                    rule_id,
+                    priority_class,
+                    priority,
+                    "[]",
+                    actions_json,
+                    update_stream=False,
+                )
+            else:
+                self._simple_update_one_txn(
+                    txn,
+                    "push_rules",
+                    {"user_name": user_id, "rule_id": rule_id},
+                    {"actions": actions_json},
+                )
+
+            self._insert_push_rules_update_txn(
+                txn,
+                stream_id,
+                event_stream_ordering,
+                user_id,
+                rule_id,
+                op="ACTIONS",
+                data={"actions": actions_json},
+            )
+
+        with self._push_rules_stream_id_gen.get_next() as ids:
+            stream_id, event_stream_ordering = ids
+            yield self.runInteraction(
+                "set_push_rule_actions",
+                set_push_rule_actions_txn,
+                stream_id,
+                event_stream_ordering,
+            )
+
+    def _insert_push_rules_update_txn(
+        self, txn, stream_id, event_stream_ordering, user_id, rule_id, op, data=None
+    ):
+        values = {
+            "stream_id": stream_id,
+            "event_stream_ordering": event_stream_ordering,
+            "user_id": user_id,
+            "rule_id": rule_id,
+            "op": op,
+        }
+        if data is not None:
+            values.update(data)
+
+        self._simple_insert_txn(txn, "push_rules_stream", values=values)
+
+        txn.call_after(self.get_push_rules_for_user.invalidate, (user_id,))
+        txn.call_after(self.get_push_rules_enabled_for_user.invalidate, (user_id,))
+        txn.call_after(
+            self.push_rules_stream_cache.entity_has_changed, user_id, stream_id
+        )
+
+    def get_all_push_rule_updates(self, last_id, current_id, limit):
+        """Get all the push rules changes that have happend on the server"""
+        if last_id == current_id:
+            return defer.succeed([])
+
+        def get_all_push_rule_updates_txn(txn):
+            sql = (
+                "SELECT stream_id, event_stream_ordering, user_id, rule_id,"
+                " op, priority_class, priority, conditions, actions"
+                " FROM push_rules_stream"
+                " WHERE ? < stream_id AND stream_id <= ?"
+                " ORDER BY stream_id ASC LIMIT ?"
+            )
+            txn.execute(sql, (last_id, current_id, limit))
+            return txn.fetchall()
+
+        return self.runInteraction(
+            "get_all_push_rule_updates", get_all_push_rule_updates_txn
+        )
+
+    def get_push_rules_stream_token(self):
+        """Get the position of the push rules stream.
+        Returns a pair of a stream id for the push_rules stream and the
+        room stream ordering it corresponds to."""
+        return self._push_rules_stream_id_gen.get_current_token()
+
+    def get_max_push_rules_stream_id(self):
+        return self.get_push_rules_stream_token()[0]
diff --git a/synapse/storage/pusher.py b/synapse/storage/data_stores/main/pusher.py
similarity index 98%
rename from synapse/storage/pusher.py
rename to synapse/storage/data_stores/main/pusher.py
index 3e0e834a62..d76861cdc0 100644
--- a/synapse/storage/pusher.py
+++ b/synapse/storage/data_stores/main/pusher.py
@@ -22,10 +22,9 @@ from canonicaljson import encode_canonical_json, json
 
 from twisted.internet import defer
 
+from synapse.storage._base import SQLBaseStore
 from synapse.util.caches.descriptors import cachedInlineCallbacks, cachedList
 
-from ._base import SQLBaseStore
-
 logger = logging.getLogger(__name__)
 
 if six.PY2:
@@ -45,7 +44,7 @@ class PusherWorkerStore(SQLBaseStore):
 
                 r["data"] = json.loads(dataJson)
             except Exception as e:
-                logger.warn(
+                logger.warning(
                     "Invalid JSON in data for pusher %d: %s, %s",
                     r["id"],
                     dataJson,
@@ -241,7 +240,7 @@ class PusherStore(PusherWorkerStore):
                     "device_display_name": device_display_name,
                     "ts": pushkey_ts,
                     "lang": lang,
-                    "data": encode_canonical_json(data),
+                    "data": bytearray(encode_canonical_json(data)),
                     "last_stream_ordering": last_stream_ordering,
                     "profile_tag": profile_tag,
                     "id": stream_id,
diff --git a/synapse/storage/receipts.py b/synapse/storage/data_stores/main/receipts.py
similarity index 92%
rename from synapse/storage/receipts.py
rename to synapse/storage/data_stores/main/receipts.py
index 290ddb30e8..0c24430f28 100644
--- a/synapse/storage/receipts.py
+++ b/synapse/storage/data_stores/main/receipts.py
@@ -21,12 +21,11 @@ from canonicaljson import json
 
 from twisted.internet import defer
 
+from synapse.storage._base import SQLBaseStore, make_in_list_sql_clause
+from synapse.storage.util.id_generators import StreamIdGenerator
 from synapse.util.caches.descriptors import cached, cachedInlineCallbacks, cachedList
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 
-from ._base import SQLBaseStore
-from .util.id_generators import StreamIdGenerator
-
 logger = logging.getLogger(__name__)
 
 
@@ -217,24 +216,26 @@ class ReceiptsWorkerStore(SQLBaseStore):
 
         def f(txn):
             if from_key:
-                sql = (
-                    "SELECT * FROM receipts_linearized WHERE"
-                    " room_id IN (%s) AND stream_id > ? AND stream_id <= ?"
-                ) % (",".join(["?"] * len(room_ids)))
-                args = list(room_ids)
-                args.extend([from_key, to_key])
+                sql = """
+                    SELECT * FROM receipts_linearized WHERE
+                    stream_id > ? AND stream_id <= ? AND
+                """
+                clause, args = make_in_list_sql_clause(
+                    self.database_engine, "room_id", room_ids
+                )
 
-                txn.execute(sql, args)
+                txn.execute(sql + clause, [from_key, to_key] + list(args))
             else:
-                sql = (
-                    "SELECT * FROM receipts_linearized WHERE"
-                    " room_id IN (%s) AND stream_id <= ?"
-                ) % (",".join(["?"] * len(room_ids)))
+                sql = """
+                    SELECT * FROM receipts_linearized WHERE
+                    stream_id <= ? AND
+                """
 
-                args = list(room_ids)
-                args.append(to_key)
+                clause, args = make_in_list_sql_clause(
+                    self.database_engine, "room_id", room_ids
+                )
 
-                txn.execute(sql, args)
+                txn.execute(sql + clause, [to_key] + list(args))
 
             return self.cursor_to_dict(txn)
 
@@ -433,13 +434,19 @@ class ReceiptsStore(ReceiptsWorkerStore):
             # we need to points in graph -> linearized form.
             # TODO: Make this better.
             def graph_to_linear(txn):
-                query = (
-                    "SELECT event_id WHERE room_id = ? AND stream_ordering IN ("
-                    " SELECT max(stream_ordering) WHERE event_id IN (%s)"
-                    ")"
-                ) % (",".join(["?"] * len(event_ids)))
+                clause, args = make_in_list_sql_clause(
+                    self.database_engine, "event_id", event_ids
+                )
 
-                txn.execute(query, [room_id] + event_ids)
+                sql = """
+                    SELECT event_id WHERE room_id = ? AND stream_ordering IN (
+                        SELECT max(stream_ordering) WHERE %s
+                    )
+                """ % (
+                    clause,
+                )
+
+                txn.execute(sql, [room_id] + list(args))
                 rows = txn.fetchall()
                 if rows:
                     return rows[0][0]
diff --git a/synapse/storage/registration.py b/synapse/storage/data_stores/main/registration.py
similarity index 88%
rename from synapse/storage/registration.py
rename to synapse/storage/data_stores/main/registration.py
index 3f50324253..f70d41ecab 100644
--- a/synapse/storage/registration.py
+++ b/synapse/storage/data_stores/main/registration.py
@@ -22,9 +22,10 @@ from six import iterkeys
 from six.moves import range
 
 from twisted.internet import defer
+from twisted.internet.defer import Deferred
 
 from synapse.api.constants import UserTypes
-from synapse.api.errors import Codes, StoreError, ThreepidValidationError
+from synapse.api.errors import Codes, StoreError, SynapseError, ThreepidValidationError
 from synapse.metrics.background_process_metrics import run_as_background_process
 from synapse.storage import background_updates
 from synapse.storage._base import SQLBaseStore
@@ -322,6 +323,19 @@ class RegistrationWorkerStore(SQLBaseStore):
 
         return None
 
+    @cachedInlineCallbacks()
+    def is_real_user(self, user_id):
+        """Determines if the user is a real user, ie does not have a 'user_type'.
+
+        Args:
+            user_id (str): user id to test
+
+        Returns:
+            Deferred[bool]: True if user 'user_type' is null or empty string
+        """
+        res = yield self.runInteraction("is_real_user", self.is_real_user_txn, user_id)
+        return res
+
     @cachedInlineCallbacks()
     def is_support_user(self, user_id):
         """Determines if the user is of type UserTypes.SUPPORT
@@ -337,6 +351,16 @@ class RegistrationWorkerStore(SQLBaseStore):
         )
         return res
 
+    def is_real_user_txn(self, txn, user_id):
+        res = self._simple_select_one_onecol_txn(
+            txn=txn,
+            table="users",
+            keyvalues={"name": user_id},
+            retcol="user_type",
+            allow_none=True,
+        )
+        return res is None
+
     def is_support_user_txn(self, txn, user_id):
         res = self._simple_select_one_onecol_txn(
             txn=txn,
@@ -361,6 +385,26 @@ class RegistrationWorkerStore(SQLBaseStore):
 
         return self.runInteraction("get_users_by_id_case_insensitive", f)
 
+    async def get_user_by_external_id(
+        self, auth_provider: str, external_id: str
+    ) -> str:
+        """Look up a user by their external auth id
+
+        Args:
+            auth_provider: identifier for the remote auth provider
+            external_id: id on that system
+
+        Returns:
+            str|None: the mxid of the user, or None if they are not known
+        """
+        return await self._simple_select_one_onecol(
+            table="user_external_ids",
+            keyvalues={"auth_provider": auth_provider, "external_id": external_id},
+            retcol="user_id",
+            allow_none=True,
+            desc="get_user_by_external_id",
+        )
+
     @defer.inlineCallbacks
     def count_all_users(self):
         """Counts all users registered on the homeserver."""
@@ -415,12 +459,26 @@ class RegistrationWorkerStore(SQLBaseStore):
                 WHERE appservice_id IS NULL
             """
             )
-            count, = txn.fetchone()
+            (count,) = txn.fetchone()
             return count
 
         ret = yield self.runInteraction("count_users", _count_users)
         return ret
 
+    @defer.inlineCallbacks
+    def count_real_users(self):
+        """Counts all users without a special user_type registered on the homeserver."""
+
+        def _count_users(txn):
+            txn.execute("SELECT COUNT(*) AS users FROM users where user_type is null")
+            rows = self.cursor_to_dict(txn)
+            if rows:
+                return rows[0]["users"]
+            return 0
+
+        ret = yield self.runInteraction("count_real_users", _count_users)
+        return ret
+
     @defer.inlineCallbacks
     def find_next_generated_user_id_localpart(self):
         """
@@ -435,7 +493,9 @@ class RegistrationWorkerStore(SQLBaseStore):
         """
 
         def _find_next_generated_user_id(txn):
-            txn.execute("SELECT name FROM users")
+            # We bound between '@1' and '@a' to avoid pulling the entire table
+            # out.
+            txn.execute("SELECT name FROM users WHERE '@1' <= name AND name < '@a'")
 
             regex = re.compile(r"^@(\d+):")
 
@@ -458,7 +518,7 @@ class RegistrationWorkerStore(SQLBaseStore):
         )
 
     @defer.inlineCallbacks
-    def get_user_id_by_threepid(self, medium, address, require_verified=False):
+    def get_user_id_by_threepid(self, medium, address):
         """Returns user id from threepid
 
         Args:
@@ -549,6 +609,26 @@ class RegistrationWorkerStore(SQLBaseStore):
             desc="add_user_bound_threepid",
         )
 
+    def user_get_bound_threepids(self, user_id):
+        """Get the threepids that a user has bound to an identity server through the homeserver
+        The homeserver remembers where binds to an identity server occurred. Using this
+        method can retrieve those threepids.
+
+        Args:
+            user_id (str): The ID of the user to retrieve threepids for
+
+        Returns:
+            Deferred[list[dict]]: List of dictionaries containing the following:
+                medium (str): The medium of the threepid (e.g "email")
+                address (str): The address of the threepid (e.g "bob@example.com")
+        """
+        return self._simple_select_list(
+            table="user_threepid_id_server",
+            keyvalues={"user_id": user_id},
+            retcols=["medium", "address"],
+            desc="user_get_bound_threepids",
+        )
+
     def remove_user_bound_threepid(self, user_id, medium, address, id_server):
         """The server proxied an unbind request to the given identity server on
         behalf of the given user, so we remove the mapping of threepid to
@@ -614,14 +694,107 @@ class RegistrationWorkerStore(SQLBaseStore):
         # Convert the integer into a boolean.
         return res == 1
 
+    def get_threepid_validation_session(
+        self, medium, client_secret, address=None, sid=None, validated=True
+    ):
+        """Gets a session_id and last_send_attempt (if available) for a
+        combination of validation metadata
 
-class RegistrationStore(
+        Args:
+            medium (str|None): The medium of the 3PID
+            address (str|None): The address of the 3PID
+            sid (str|None): The ID of the validation session
+            client_secret (str): A unique string provided by the client to help identify this
+                validation attempt
+            validated (bool|None): Whether sessions should be filtered by
+                whether they have been validated already or not. None to
+                perform no filtering
+
+        Returns:
+            Deferred[dict|None]: A dict containing the following:
+                * address - address of the 3pid
+                * medium - medium of the 3pid
+                * client_secret - a secret provided by the client for this validation session
+                * session_id - ID of the validation session
+                * send_attempt - a number serving to dedupe send attempts for this session
+                * validated_at - timestamp of when this session was validated if so
+
+                Otherwise None if a validation session is not found
+        """
+        if not client_secret:
+            raise SynapseError(
+                400, "Missing parameter: client_secret", errcode=Codes.MISSING_PARAM
+            )
+
+        keyvalues = {"client_secret": client_secret}
+        if medium:
+            keyvalues["medium"] = medium
+        if address:
+            keyvalues["address"] = address
+        if sid:
+            keyvalues["session_id"] = sid
+
+        assert address or sid
+
+        def get_threepid_validation_session_txn(txn):
+            sql = """
+                SELECT address, session_id, medium, client_secret,
+                last_send_attempt, validated_at
+                FROM threepid_validation_session WHERE %s
+                """ % (
+                " AND ".join("%s = ?" % k for k in iterkeys(keyvalues)),
+            )
+
+            if validated is not None:
+                sql += " AND validated_at IS " + ("NOT NULL" if validated else "NULL")
+
+            sql += " LIMIT 1"
+
+            txn.execute(sql, list(keyvalues.values()))
+            rows = self.cursor_to_dict(txn)
+            if not rows:
+                return None
+
+            return rows[0]
+
+        return self.runInteraction(
+            "get_threepid_validation_session", get_threepid_validation_session_txn
+        )
+
+    def delete_threepid_session(self, session_id):
+        """Removes a threepid validation session from the database. This can
+        be done after validation has been performed and whatever action was
+        waiting on it has been carried out
+
+        Args:
+            session_id (str): The ID of the session to delete
+        """
+
+        def delete_threepid_session_txn(txn):
+            self._simple_delete_txn(
+                txn,
+                table="threepid_validation_token",
+                keyvalues={"session_id": session_id},
+            )
+            self._simple_delete_txn(
+                txn,
+                table="threepid_validation_session",
+                keyvalues={"session_id": session_id},
+            )
+
+        return self.runInteraction(
+            "delete_threepid_session", delete_threepid_session_txn
+        )
+
+
+class RegistrationBackgroundUpdateStore(
     RegistrationWorkerStore, background_updates.BackgroundUpdateStore
 ):
     def __init__(self, db_conn, hs):
-        super(RegistrationStore, self).__init__(db_conn, hs)
+        super(RegistrationBackgroundUpdateStore, self).__init__(db_conn, hs)
 
         self.clock = hs.get_clock()
+        self.config = hs.config
 
         self.register_background_index_update(
             "access_tokens_device_index",
@@ -637,8 +810,6 @@ class RegistrationStore(
             columns=["creation_ts"],
         )
 
-        self._account_validity = hs.config.account_validity
-
         # we no longer use refresh tokens, but it's possible that some people
         # might have a background update queued to build this index. Just
         # clear the background update.
@@ -652,17 +823,6 @@ class RegistrationStore(
             "users_set_deactivated_flag", self._background_update_set_deactivated_flag
         )
 
-        # Create a background job for culling expired 3PID validity tokens
-        def start_cull():
-            # run as a background process to make sure that the database transactions
-            # have a logcontext to report to
-            return run_as_background_process(
-                "cull_expired_threepid_validation_tokens",
-                self.cull_expired_threepid_validation_tokens,
-            )
-
-        hs.get_clock().looping_call(start_cull, THIRTY_MINUTES_IN_MS)
-
     @defer.inlineCallbacks
     def _background_update_set_deactivated_flag(self, progress, batch_size):
         """Retrieves a list of all deactivated users and sets the 'deactivated' flag to 1
@@ -695,7 +855,7 @@ class RegistrationStore(
             rows = self.cursor_to_dict(txn)
 
             if not rows:
-                return True
+                return True, 0
 
             rows_processed_nb = 0
 
@@ -711,18 +871,66 @@ class RegistrationStore(
             )
 
             if batch_size > len(rows):
-                return True
+                return True, len(rows)
             else:
-                return False
+                return False, len(rows)
 
-        end = yield self.runInteraction(
+        end, nb_processed = yield self.runInteraction(
             "users_set_deactivated_flag", _background_update_set_deactivated_flag_txn
         )
 
         if end:
             yield self._end_background_update("users_set_deactivated_flag")
 
-        return batch_size
+        return nb_processed
+
+    @defer.inlineCallbacks
+    def _bg_user_threepids_grandfather(self, progress, batch_size):
+        """We now track which identity servers a user binds their 3PID to, so
+        we need to handle the case of existing bindings where we didn't track
+        this.
+
+        We do this by grandfathering in existing user threepids assuming that
+        they used one of the server configured trusted identity servers.
+        """
+        id_servers = set(self.config.trusted_third_party_id_servers)
+
+        def _bg_user_threepids_grandfather_txn(txn):
+            sql = """
+                INSERT INTO user_threepid_id_server
+                    (user_id, medium, address, id_server)
+                SELECT user_id, medium, address, ?
+                FROM user_threepids
+            """
+
+            txn.executemany(sql, [(id_server,) for id_server in id_servers])
+
+        if id_servers:
+            yield self.runInteraction(
+                "_bg_user_threepids_grandfather", _bg_user_threepids_grandfather_txn
+            )
+
+        yield self._end_background_update("user_threepids_grandfather")
+
+        return 1
+
+
+class RegistrationStore(RegistrationBackgroundUpdateStore):
+    def __init__(self, db_conn, hs):
+        super(RegistrationStore, self).__init__(db_conn, hs)
+
+        self._account_validity = hs.config.account_validity
+
+        # Create a background job for culling expired 3PID validity tokens
+        def start_cull():
+            # run as a background process to make sure that the database transactions
+            # have a logcontext to report to
+            return run_as_background_process(
+                "cull_expired_threepid_validation_tokens",
+                self.cull_expired_threepid_validation_tokens,
+            )
+
+        hs.get_clock().looping_call(start_cull, THIRTY_MINUTES_IN_MS)
 
     @defer.inlineCallbacks
     def add_access_token_to_user(self, user_id, token, device_id, valid_until_ms):
@@ -869,9 +1077,40 @@ class RegistrationStore(
                 (user_id_obj.localpart, create_profile_with_displayname),
             )
 
+        if self.hs.config.stats_enabled:
+            # we create a new completed user statistics row
+
+            # we don't strictly need current_token since this user really can't
+            # have any state deltas before now (as it is a new user), but still,
+            # we include it for completeness.
+            current_token = self._get_max_stream_id_in_current_state_deltas_txn(txn)
+            self._update_stats_delta_txn(
+                txn, now, "user", user_id, {}, complete_with_stream_id=current_token
+            )
+
         self._invalidate_cache_and_stream(txn, self.get_user_by_id, (user_id,))
         txn.call_after(self.is_guest.invalidate, (user_id,))
 
+    def record_user_external_id(
+        self, auth_provider: str, external_id: str, user_id: str
+    ) -> Deferred:
+        """Record a mapping from an external user id to a mxid
+
+        Args:
+            auth_provider: identifier for the remote auth provider
+            external_id: id on that system
+            user_id: complete mxid that it is mapped to
+        """
+        return self._simple_insert(
+            table="user_external_ids",
+            values={
+                "auth_provider": auth_provider,
+                "external_id": external_id,
+                "user_id": user_id,
+            },
+            desc="record_user_external_id",
+        )
+
     def user_set_password_hash(self, user_id, password_hash):
         """
         NB. This does *not* evict any cache because the one use for this
@@ -1041,90 +1280,6 @@ class RegistrationStore(
             desc="get_users_pending_deactivation",
         )
 
-    @defer.inlineCallbacks
-    def _bg_user_threepids_grandfather(self, progress, batch_size):
-        """We now track which identity servers a user binds their 3PID to, so
-        we need to handle the case of existing bindings where we didn't track
-        this.
-
-        We do this by grandfathering in existing user threepids assuming that
-        they used one of the server configured trusted identity servers.
-        """
-        id_servers = set(self.config.trusted_third_party_id_servers)
-
-        def _bg_user_threepids_grandfather_txn(txn):
-            sql = """
-                INSERT INTO user_threepid_id_server
-                    (user_id, medium, address, id_server)
-                SELECT user_id, medium, address, ?
-                FROM user_threepids
-            """
-
-            txn.executemany(sql, [(id_server,) for id_server in id_servers])
-
-        if id_servers:
-            yield self.runInteraction(
-                "_bg_user_threepids_grandfather", _bg_user_threepids_grandfather_txn
-            )
-
-        yield self._end_background_update("user_threepids_grandfather")
-
-        return 1
-
-    def get_threepid_validation_session(
-        self, medium, client_secret, address=None, sid=None, validated=True
-    ):
-        """Gets a session_id and last_send_attempt (if available) for a
-        client_secret/medium/(address|session_id) combo
-
-        Args:
-            medium (str|None): The medium of the 3PID
-            address (str|None): The address of the 3PID
-            sid (str|None): The ID of the validation session
-            client_secret (str|None): A unique string provided by the client to
-                help identify this validation attempt
-            validated (bool|None): Whether sessions should be filtered by
-                whether they have been validated already or not. None to
-                perform no filtering
-
-        Returns:
-            deferred {str, int}|None: A dict containing the
-                latest session_id and send_attempt count for this 3PID.
-                Otherwise None if there hasn't been a previous attempt
-        """
-        keyvalues = {"medium": medium, "client_secret": client_secret}
-        if address:
-            keyvalues["address"] = address
-        if sid:
-            keyvalues["session_id"] = sid
-
-        assert address or sid
-
-        def get_threepid_validation_session_txn(txn):
-            sql = """
-                SELECT address, session_id, medium, client_secret,
-                last_send_attempt, validated_at
-                FROM threepid_validation_session WHERE %s
-                """ % (
-                " AND ".join("%s = ?" % k for k in iterkeys(keyvalues)),
-            )
-
-            if validated is not None:
-                sql += " AND validated_at IS " + ("NOT NULL" if validated else "NULL")
-
-            sql += " LIMIT 1"
-
-            txn.execute(sql, list(keyvalues.values()))
-            rows = self.cursor_to_dict(txn)
-            if not rows:
-                return None
-
-            return rows[0]
-
-        return self.runInteraction(
-            "get_threepid_validation_session", get_threepid_validation_session_txn
-        )
-
     def validate_threepid_session(self, session_id, client_secret, token, current_ts):
         """Attempt to validate a threepid session using a token
 
@@ -1136,10 +1291,15 @@ class RegistrationStore(
             current_ts (int): The current unix time in milliseconds. Used for
                 checking token expiry status
 
+        Raises:
+            ThreepidValidationError: if a matching validation token was not found or has
+                expired
+
         Returns:
             deferred str|None: A str representing a link to redirect the user
             to if there is one.
         """
+
         # Insert everything into a transaction in order to run atomically
         def validate_threepid_session_txn(txn):
             row = self._simple_select_one_txn(
@@ -1311,42 +1471,6 @@ class RegistrationStore(
             self.clock.time_msec(),
         )
 
-    def delete_threepid_session(self, session_id):
-        """Removes a threepid validation session from the database. This can
-        be done after validation has been performed and whatever action was
-        waiting on it has been carried out
-
-        Args:
-            session_id (str): The ID of the session to delete
-        """
-
-        def delete_threepid_session_txn(txn):
-            self._simple_delete_txn(
-                txn,
-                table="threepid_validation_token",
-                keyvalues={"session_id": session_id},
-            )
-            self._simple_delete_txn(
-                txn,
-                table="threepid_validation_session",
-                keyvalues={"session_id": session_id},
-            )
-
-        return self.runInteraction(
-            "delete_threepid_session", delete_threepid_session_txn
-        )
-
-    def set_user_deactivated_status_txn(self, txn, user_id, deactivated):
-        self._simple_update_one_txn(
-            txn=txn,
-            table="users",
-            keyvalues={"name": user_id},
-            updatevalues={"deactivated": 1 if deactivated else 0},
-        )
-        self._invalidate_cache_and_stream(
-            txn, self.get_user_deactivated_status, (user_id,)
-        )
-
     @defer.inlineCallbacks
     def set_user_deactivated_status(self, user_id, deactivated):
         """Set the `deactivated` property for the provided user to the provided value.
@@ -1362,3 +1486,14 @@ class RegistrationStore(
             user_id,
             deactivated,
         )
+
+    def set_user_deactivated_status_txn(self, txn, user_id, deactivated):
+        self._simple_update_one_txn(
+            txn=txn,
+            table="users",
+            keyvalues={"name": user_id},
+            updatevalues={"deactivated": 1 if deactivated else 0},
+        )
+        self._invalidate_cache_and_stream(
+            txn, self.get_user_deactivated_status, (user_id,)
+        )
diff --git a/synapse/storage/rejections.py b/synapse/storage/data_stores/main/rejections.py
similarity index 96%
rename from synapse/storage/rejections.py
rename to synapse/storage/data_stores/main/rejections.py
index f4c1c2a457..7d5de0ea2e 100644
--- a/synapse/storage/rejections.py
+++ b/synapse/storage/data_stores/main/rejections.py
@@ -15,7 +15,7 @@
 
 import logging
 
-from ._base import SQLBaseStore
+from synapse.storage._base import SQLBaseStore
 
 logger = logging.getLogger(__name__)
 
diff --git a/synapse/storage/data_stores/main/relations.py b/synapse/storage/data_stores/main/relations.py
new file mode 100644
index 0000000000..858f65582b
--- /dev/null
+++ b/synapse/storage/data_stores/main/relations.py
@@ -0,0 +1,385 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 New Vector Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+
+import attr
+
+from synapse.api.constants import RelationTypes
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.data_stores.main.stream import generate_pagination_where_clause
+from synapse.storage.relations import (
+    AggregationPaginationToken,
+    PaginationChunk,
+    RelationPaginationToken,
+)
+from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
+
+logger = logging.getLogger(__name__)
+
+
+class RelationsWorkerStore(SQLBaseStore):
+    @cached(tree=True)
+    def get_relations_for_event(
+        self,
+        event_id,
+        relation_type=None,
+        event_type=None,
+        aggregation_key=None,
+        limit=5,
+        direction="b",
+        from_token=None,
+        to_token=None,
+    ):
+        """Get a list of relations for an event, ordered by topological ordering.
+
+        Args:
+            event_id (str): Fetch events that relate to this event ID.
+            relation_type (str|None): Only fetch events with this relation
+                type, if given.
+            event_type (str|None): Only fetch events with this event type, if
+                given.
+            aggregation_key (str|None): Only fetch events with this aggregation
+                key, if given.
+            limit (int): Only fetch the most recent `limit` events.
+            direction (str): Whether to fetch the most recent first (`"b"`) or
+                the oldest first (`"f"`).
+            from_token (RelationPaginationToken|None): Fetch rows from the given
+                token, or from the start if None.
+            to_token (RelationPaginationToken|None): Fetch rows up to the given
+                token, or up to the end if None.
+
+        Returns:
+            Deferred[PaginationChunk]: List of event IDs that match relations
+            requested. The rows are of the form `{"event_id": "..."}`.
+        """
+
+        where_clause = ["relates_to_id = ?"]
+        where_args = [event_id]
+
+        if relation_type is not None:
+            where_clause.append("relation_type = ?")
+            where_args.append(relation_type)
+
+        if event_type is not None:
+            where_clause.append("type = ?")
+            where_args.append(event_type)
+
+        if aggregation_key:
+            where_clause.append("aggregation_key = ?")
+            where_args.append(aggregation_key)
+
+        pagination_clause = generate_pagination_where_clause(
+            direction=direction,
+            column_names=("topological_ordering", "stream_ordering"),
+            from_token=attr.astuple(from_token) if from_token else None,
+            to_token=attr.astuple(to_token) if to_token else None,
+            engine=self.database_engine,
+        )
+
+        if pagination_clause:
+            where_clause.append(pagination_clause)
+
+        if direction == "b":
+            order = "DESC"
+        else:
+            order = "ASC"
+
+        sql = """
+            SELECT event_id, topological_ordering, stream_ordering
+            FROM event_relations
+            INNER JOIN events USING (event_id)
+            WHERE %s
+            ORDER BY topological_ordering %s, stream_ordering %s
+            LIMIT ?
+        """ % (
+            " AND ".join(where_clause),
+            order,
+            order,
+        )
+
+        def _get_recent_references_for_event_txn(txn):
+            txn.execute(sql, where_args + [limit + 1])
+
+            last_topo_id = None
+            last_stream_id = None
+            events = []
+            for row in txn:
+                events.append({"event_id": row[0]})
+                last_topo_id = row[1]
+                last_stream_id = row[2]
+
+            next_batch = None
+            if len(events) > limit and last_topo_id and last_stream_id:
+                next_batch = RelationPaginationToken(last_topo_id, last_stream_id)
+
+            return PaginationChunk(
+                chunk=list(events[:limit]), next_batch=next_batch, prev_batch=from_token
+            )
+
+        return self.runInteraction(
+            "get_recent_references_for_event", _get_recent_references_for_event_txn
+        )
+
+    @cached(tree=True)
+    def get_aggregation_groups_for_event(
+        self,
+        event_id,
+        event_type=None,
+        limit=5,
+        direction="b",
+        from_token=None,
+        to_token=None,
+    ):
+        """Get a list of annotations on the event, grouped by event type and
+        aggregation key, sorted by count.
+
+        This is used e.g. to get the what and how many reactions have happend
+        on an event.
+
+        Args:
+            event_id (str): Fetch events that relate to this event ID.
+            event_type (str|None): Only fetch events with this event type, if
+                given.
+            limit (int): Only fetch the `limit` groups.
+            direction (str): Whether to fetch the highest count first (`"b"`) or
+                the lowest count first (`"f"`).
+            from_token (AggregationPaginationToken|None): Fetch rows from the
+                given token, or from the start if None.
+            to_token (AggregationPaginationToken|None): Fetch rows up to the
+                given token, or up to the end if None.
+
+
+        Returns:
+            Deferred[PaginationChunk]: List of groups of annotations that
+            match. Each row is a dict with `type`, `key` and `count` fields.
+        """
+
+        where_clause = ["relates_to_id = ?", "relation_type = ?"]
+        where_args = [event_id, RelationTypes.ANNOTATION]
+
+        if event_type:
+            where_clause.append("type = ?")
+            where_args.append(event_type)
+
+        having_clause = generate_pagination_where_clause(
+            direction=direction,
+            column_names=("COUNT(*)", "MAX(stream_ordering)"),
+            from_token=attr.astuple(from_token) if from_token else None,
+            to_token=attr.astuple(to_token) if to_token else None,
+            engine=self.database_engine,
+        )
+
+        if direction == "b":
+            order = "DESC"
+        else:
+            order = "ASC"
+
+        if having_clause:
+            having_clause = "HAVING " + having_clause
+        else:
+            having_clause = ""
+
+        sql = """
+            SELECT type, aggregation_key, COUNT(DISTINCT sender), MAX(stream_ordering)
+            FROM event_relations
+            INNER JOIN events USING (event_id)
+            WHERE {where_clause}
+            GROUP BY relation_type, type, aggregation_key
+            {having_clause}
+            ORDER BY COUNT(*) {order}, MAX(stream_ordering) {order}
+            LIMIT ?
+        """.format(
+            where_clause=" AND ".join(where_clause),
+            order=order,
+            having_clause=having_clause,
+        )
+
+        def _get_aggregation_groups_for_event_txn(txn):
+            txn.execute(sql, where_args + [limit + 1])
+
+            next_batch = None
+            events = []
+            for row in txn:
+                events.append({"type": row[0], "key": row[1], "count": row[2]})
+                next_batch = AggregationPaginationToken(row[2], row[3])
+
+            if len(events) <= limit:
+                next_batch = None
+
+            return PaginationChunk(
+                chunk=list(events[:limit]), next_batch=next_batch, prev_batch=from_token
+            )
+
+        return self.runInteraction(
+            "get_aggregation_groups_for_event", _get_aggregation_groups_for_event_txn
+        )
+
+    @cachedInlineCallbacks()
+    def get_applicable_edit(self, event_id):
+        """Get the most recent edit (if any) that has happened for the given
+        event.
+
+        Correctly handles checking whether edits were allowed to happen.
+
+        Args:
+            event_id (str): The original event ID
+
+        Returns:
+            Deferred[EventBase|None]: Returns the most recent edit, if any.
+        """
+
+        # We only allow edits for `m.room.message` events that have the same sender
+        # and event type. We can't assert these things during regular event auth so
+        # we have to do the checks post hoc.
+
+        # Fetches latest edit that has the same type and sender as the
+        # original, and is an `m.room.message`.
+        sql = """
+            SELECT edit.event_id FROM events AS edit
+            INNER JOIN event_relations USING (event_id)
+            INNER JOIN events AS original ON
+                original.event_id = relates_to_id
+                AND edit.type = original.type
+                AND edit.sender = original.sender
+            WHERE
+                relates_to_id = ?
+                AND relation_type = ?
+                AND edit.type = 'm.room.message'
+            ORDER by edit.origin_server_ts DESC, edit.event_id DESC
+            LIMIT 1
+        """
+
+        def _get_applicable_edit_txn(txn):
+            txn.execute(sql, (event_id, RelationTypes.REPLACE))
+            row = txn.fetchone()
+            if row:
+                return row[0]
+
+        edit_id = yield self.runInteraction(
+            "get_applicable_edit", _get_applicable_edit_txn
+        )
+
+        if not edit_id:
+            return
+
+        edit_event = yield self.get_event(edit_id, allow_none=True)
+        return edit_event
+
+    def has_user_annotated_event(self, parent_id, event_type, aggregation_key, sender):
+        """Check if a user has already annotated an event with the same key
+        (e.g. already liked an event).
+
+        Args:
+            parent_id (str): The event being annotated
+            event_type (str): The event type of the annotation
+            aggregation_key (str): The aggregation key of the annotation
+            sender (str): The sender of the annotation
+
+        Returns:
+            Deferred[bool]
+        """
+
+        sql = """
+            SELECT 1 FROM event_relations
+            INNER JOIN events USING (event_id)
+            WHERE
+                relates_to_id = ?
+                AND relation_type = ?
+                AND type = ?
+                AND sender = ?
+                AND aggregation_key = ?
+            LIMIT 1;
+        """
+
+        def _get_if_user_has_annotated_event(txn):
+            txn.execute(
+                sql,
+                (
+                    parent_id,
+                    RelationTypes.ANNOTATION,
+                    event_type,
+                    sender,
+                    aggregation_key,
+                ),
+            )
+
+            return bool(txn.fetchone())
+
+        return self.runInteraction(
+            "get_if_user_has_annotated_event", _get_if_user_has_annotated_event
+        )
+
+
+class RelationsStore(RelationsWorkerStore):
+    def _handle_event_relations(self, txn, event):
+        """Handles inserting relation data during peristence of events
+
+        Args:
+            txn
+            event (EventBase)
+        """
+        relation = event.content.get("m.relates_to")
+        if not relation:
+            # No relations
+            return
+
+        rel_type = relation.get("rel_type")
+        if rel_type not in (
+            RelationTypes.ANNOTATION,
+            RelationTypes.REFERENCE,
+            RelationTypes.REPLACE,
+        ):
+            # Unknown relation type
+            return
+
+        parent_id = relation.get("event_id")
+        if not parent_id:
+            # Invalid relation
+            return
+
+        aggregation_key = relation.get("key")
+
+        self._simple_insert_txn(
+            txn,
+            table="event_relations",
+            values={
+                "event_id": event.event_id,
+                "relates_to_id": parent_id,
+                "relation_type": rel_type,
+                "aggregation_key": aggregation_key,
+            },
+        )
+
+        txn.call_after(self.get_relations_for_event.invalidate_many, (parent_id,))
+        txn.call_after(
+            self.get_aggregation_groups_for_event.invalidate_many, (parent_id,)
+        )
+
+        if rel_type == RelationTypes.REPLACE:
+            txn.call_after(self.get_applicable_edit.invalidate, (parent_id,))
+
+    def _handle_redaction(self, txn, redacted_event_id):
+        """Handles receiving a redaction and checking whether we need to remove
+        any redacted relations from the database.
+
+        Args:
+            txn
+            redacted_event_id (str): The event that was redacted.
+        """
+
+        self._simple_delete_txn(
+            txn, table="event_relations", keyvalues={"event_id": redacted_event_id}
+        )
diff --git a/synapse/storage/room.py b/synapse/storage/data_stores/main/room.py
similarity index 71%
rename from synapse/storage/room.py
rename to synapse/storage/data_stores/main/room.py
index 08e13f3a3b..67bb1b6f60 100644
--- a/synapse/storage/room.py
+++ b/synapse/storage/data_stores/main/room.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -16,6 +17,7 @@
 import collections
 import logging
 import re
+from typing import Optional, Tuple
 
 from canonicaljson import json
 
@@ -23,7 +25,8 @@ from twisted.internet import defer
 
 from synapse.api.errors import StoreError
 from synapse.storage._base import SQLBaseStore
-from synapse.storage.search import SearchStore
+from synapse.storage.data_stores.main.search import SearchStore
+from synapse.types import ThirdPartyInstanceID
 from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
 
 logger = logging.getLogger(__name__)
@@ -63,103 +66,200 @@ class RoomWorkerStore(SQLBaseStore):
             desc="get_public_room_ids",
         )
 
-    @cached(num_args=2, max_entries=100)
-    def get_public_room_ids_at_stream_id(self, stream_id, network_tuple):
-        """Get pulbic rooms for a particular list, or across all lists.
+    def count_public_rooms(self, network_tuple, ignore_non_federatable):
+        """Counts the number of public rooms as tracked in the room_stats_current
+        and room_stats_state table.
 
         Args:
-            stream_id (int)
-            network_tuple (ThirdPartyInstanceID): The list to use (None, None)
-                means the main list, None means all lsits.
+            network_tuple (ThirdPartyInstanceID|None)
+            ignore_non_federatable (bool): If true filters out non-federatable rooms
         """
-        return self.runInteraction(
-            "get_public_room_ids_at_stream_id",
-            self.get_public_room_ids_at_stream_id_txn,
-            stream_id,
-            network_tuple=network_tuple,
-        )
 
-    def get_public_room_ids_at_stream_id_txn(self, txn, stream_id, network_tuple):
-        return {
-            rm
-            for rm, vis in self.get_published_at_stream_id_txn(
-                txn, stream_id, network_tuple=network_tuple
-            ).items()
-            if vis
-        }
+        def _count_public_rooms_txn(txn):
+            query_args = []
 
-    def get_published_at_stream_id_txn(self, txn, stream_id, network_tuple):
-        if network_tuple:
-            # We want to get from a particular list. No aggregation required.
-
-            sql = """
-                SELECT room_id, visibility FROM public_room_list_stream
-                INNER JOIN (
-                    SELECT room_id, max(stream_id) AS stream_id
-                    FROM public_room_list_stream
-                    WHERE stream_id <= ? %s
-                    GROUP BY room_id
-                ) grouped USING (room_id, stream_id)
+            if network_tuple:
+                if network_tuple.appservice_id:
+                    published_sql = """
+                        SELECT room_id from appservice_room_list
+                        WHERE appservice_id = ? AND network_id = ?
+                    """
+                    query_args.append(network_tuple.appservice_id)
+                    query_args.append(network_tuple.network_id)
+                else:
+                    published_sql = """
+                        SELECT room_id FROM rooms WHERE is_public
+                    """
+            else:
+                published_sql = """
+                    SELECT room_id FROM rooms WHERE is_public
+                    UNION SELECT room_id from appservice_room_list
             """
 
-            if network_tuple.appservice_id is not None:
-                txn.execute(
-                    sql % ("AND appservice_id = ? AND network_id = ?",),
-                    (stream_id, network_tuple.appservice_id, network_tuple.network_id),
+            sql = """
+                SELECT
+                    COALESCE(COUNT(*), 0)
+                FROM (
+                    %(published_sql)s
+                ) published
+                INNER JOIN room_stats_state USING (room_id)
+                INNER JOIN room_stats_current USING (room_id)
+                WHERE
+                    (
+                        join_rules = 'public' OR history_visibility = 'world_readable'
+                    )
+                    AND joined_members > 0
+            """ % {
+                "published_sql": published_sql
+            }
+
+            txn.execute(sql, query_args)
+            return txn.fetchone()[0]
+
+        return self.runInteraction("count_public_rooms", _count_public_rooms_txn)
+
+    @defer.inlineCallbacks
+    def get_largest_public_rooms(
+        self,
+        network_tuple: Optional[ThirdPartyInstanceID],
+        search_filter: Optional[dict],
+        limit: Optional[int],
+        bounds: Optional[Tuple[int, str]],
+        forwards: bool,
+        ignore_non_federatable: bool = False,
+    ):
+        """Gets the largest public rooms (where largest is in terms of joined
+        members, as tracked in the statistics table).
+
+        Args:
+            network_tuple
+            search_filter
+            limit: Maxmimum number of rows to return, unlimited otherwise.
+            bounds: An uppoer or lower bound to apply to result set if given,
+                consists of a joined member count and room_id (these are
+                excluded from result set).
+            forwards: true iff going forwards, going backwards otherwise
+            ignore_non_federatable: If true filters out non-federatable rooms.
+
+        Returns:
+            Rooms in order: biggest number of joined users first.
+            We then arbitrarily use the room_id as a tie breaker.
+
+        """
+
+        where_clauses = []
+        query_args = []
+
+        if network_tuple:
+            if network_tuple.appservice_id:
+                published_sql = """
+                    SELECT room_id from appservice_room_list
+                    WHERE appservice_id = ? AND network_id = ?
+                """
+                query_args.append(network_tuple.appservice_id)
+                query_args.append(network_tuple.network_id)
+            else:
+                published_sql = """
+                    SELECT room_id FROM rooms WHERE is_public
+                """
+        else:
+            published_sql = """
+                SELECT room_id FROM rooms WHERE is_public
+                UNION SELECT room_id from appservice_room_list
+            """
+
+        # Work out the bounds if we're given them, these bounds look slightly
+        # odd, but are designed to help query planner use indices by pulling
+        # out a common bound.
+        if bounds:
+            last_joined_members, last_room_id = bounds
+            if forwards:
+                where_clauses.append(
+                    """
+                        joined_members <= ? AND (
+                            joined_members < ? OR room_id < ?
+                        )
+                    """
                 )
             else:
-                txn.execute(sql % ("AND appservice_id IS NULL",), (stream_id,))
-            return dict(txn)
-        else:
-            # We want to get from all lists, so we need to aggregate the results
+                where_clauses.append(
+                    """
+                        joined_members >= ? AND (
+                            joined_members > ? OR room_id > ?
+                        )
+                    """
+                )
 
-            logger.info("Executing full list")
+            query_args += [last_joined_members, last_joined_members, last_room_id]
 
-            sql = """
-                SELECT room_id, visibility
-                FROM public_room_list_stream
-                INNER JOIN (
-                    SELECT
-                        room_id, max(stream_id) AS stream_id, appservice_id,
-                        network_id
-                    FROM public_room_list_stream
-                    WHERE stream_id <= ?
-                    GROUP BY room_id, appservice_id, network_id
-                ) grouped USING (room_id, stream_id)
+        if ignore_non_federatable:
+            where_clauses.append("is_federatable")
+
+        if search_filter and search_filter.get("generic_search_term", None):
+            search_term = "%" + search_filter["generic_search_term"] + "%"
+
+            where_clauses.append(
+                """
+                    (
+                        LOWER(name) LIKE ?
+                        OR LOWER(topic) LIKE ?
+                        OR LOWER(canonical_alias) LIKE ?
+                    )
+                """
+            )
+            query_args += [
+                search_term.lower(),
+                search_term.lower(),
+                search_term.lower(),
+            ]
+
+        where_clause = ""
+        if where_clauses:
+            where_clause = " AND " + " AND ".join(where_clauses)
+
+        sql = """
+            SELECT
+                room_id, name, topic, canonical_alias, joined_members,
+                avatar, history_visibility, joined_members, guest_access
+            FROM (
+                %(published_sql)s
+            ) published
+            INNER JOIN room_stats_state USING (room_id)
+            INNER JOIN room_stats_current USING (room_id)
+            WHERE
+                (
+                    join_rules = 'public' OR history_visibility = 'world_readable'
+                )
+                AND joined_members > 0
+                %(where_clause)s
+            ORDER BY joined_members %(dir)s, room_id %(dir)s
+        """ % {
+            "published_sql": published_sql,
+            "where_clause": where_clause,
+            "dir": "DESC" if forwards else "ASC",
+        }
+
+        if limit is not None:
+            query_args.append(limit)
+
+            sql += """
+                LIMIT ?
             """
 
-            txn.execute(sql, (stream_id,))
+        def _get_largest_public_rooms_txn(txn):
+            txn.execute(sql, query_args)
 
-            results = {}
-            # A room is visible if its visible on any list.
-            for room_id, visibility in txn:
-                results[room_id] = bool(visibility) or results.get(room_id, False)
+            results = self.cursor_to_dict(txn)
+
+            if not forwards:
+                results.reverse()
 
             return results
 
-    def get_public_room_changes(self, prev_stream_id, new_stream_id, network_tuple):
-        def get_public_room_changes_txn(txn):
-            then_rooms = self.get_public_room_ids_at_stream_id_txn(
-                txn, prev_stream_id, network_tuple
-            )
-
-            now_rooms_dict = self.get_published_at_stream_id_txn(
-                txn, new_stream_id, network_tuple
-            )
-
-            now_rooms_visible = set(rm for rm, vis in now_rooms_dict.items() if vis)
-            now_rooms_not_visible = set(
-                rm for rm, vis in now_rooms_dict.items() if not vis
-            )
-
-            newly_visible = now_rooms_visible - then_rooms
-            newly_unpublished = now_rooms_not_visible & then_rooms
-
-            return newly_visible, newly_unpublished
-
-        return self.runInteraction(
-            "get_public_room_changes", get_public_room_changes_txn
+        ret_val = yield self.runInteraction(
+            "get_largest_public_rooms", _get_largest_public_rooms_txn
         )
+        defer.returnValue(ret_val)
 
     @cached(max_entries=10000)
     def is_room_blocked(self, room_id):
diff --git a/synapse/storage/data_stores/main/roommember.py b/synapse/storage/data_stores/main/roommember.py
new file mode 100644
index 0000000000..2af24a20b7
--- /dev/null
+++ b/synapse/storage/data_stores/main/roommember.py
@@ -0,0 +1,1145 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2018 New Vector Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+
+from six import iteritems, itervalues
+
+from canonicaljson import json
+
+from twisted.internet import defer
+
+from synapse.api.constants import EventTypes, Membership
+from synapse.metrics import LaterGauge
+from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.storage._base import LoggingTransaction, make_in_list_sql_clause
+from synapse.storage.background_updates import BackgroundUpdateStore
+from synapse.storage.data_stores.main.events_worker import EventsWorkerStore
+from synapse.storage.engines import Sqlite3Engine
+from synapse.storage.roommember import (
+    GetRoomsForUserWithStreamOrdering,
+    MemberSummary,
+    ProfileInfo,
+    RoomsForUser,
+)
+from synapse.types import get_domain_from_id
+from synapse.util.async_helpers import Linearizer
+from synapse.util.caches import intern_string
+from synapse.util.caches.descriptors import cached, cachedInlineCallbacks, cachedList
+from synapse.util.metrics import Measure
+from synapse.util.stringutils import to_ascii
+
+logger = logging.getLogger(__name__)
+
+
+_MEMBERSHIP_PROFILE_UPDATE_NAME = "room_membership_profile_update"
+_CURRENT_STATE_MEMBERSHIP_UPDATE_NAME = "current_state_events_membership"
+
+
+class RoomMemberWorkerStore(EventsWorkerStore):
+    def __init__(self, db_conn, hs):
+        super(RoomMemberWorkerStore, self).__init__(db_conn, hs)
+
+        # Is the current_state_events.membership up to date? Or is the
+        # background update still running?
+        self._current_state_events_membership_up_to_date = False
+
+        txn = LoggingTransaction(
+            db_conn.cursor(),
+            name="_check_safe_current_state_events_membership_updated",
+            database_engine=self.database_engine,
+        )
+        self._check_safe_current_state_events_membership_updated_txn(txn)
+        txn.close()
+
+        if self.hs.config.metrics_flags.known_servers:
+            self._known_servers_count = 1
+            self.hs.get_clock().looping_call(
+                run_as_background_process,
+                60 * 1000,
+                "_count_known_servers",
+                self._count_known_servers,
+            )
+            self.hs.get_clock().call_later(
+                1000,
+                run_as_background_process,
+                "_count_known_servers",
+                self._count_known_servers,
+            )
+            LaterGauge(
+                "synapse_federation_known_servers",
+                "",
+                [],
+                lambda: self._known_servers_count,
+            )
+
+    @defer.inlineCallbacks
+    def _count_known_servers(self):
+        """
+        Count the servers that this server knows about.
+
+        The statistic is stored on the class for the
+        `synapse_federation_known_servers` LaterGauge to collect.
+        """
+
+        def _transact(txn):
+            if isinstance(self.database_engine, Sqlite3Engine):
+                query = """
+                    SELECT COUNT(DISTINCT substr(out.user_id, pos+1))
+                    FROM (
+                        SELECT rm.user_id as user_id, instr(rm.user_id, ':')
+                            AS pos FROM room_memberships as rm
+                        INNER JOIN current_state_events as c ON rm.event_id = c.event_id
+                        WHERE c.type = 'm.room.member'
+                    ) as out
+                """
+            else:
+                query = """
+                    SELECT COUNT(DISTINCT split_part(state_key, ':', 2))
+                    FROM current_state_events
+                    WHERE type = 'm.room.member' AND membership = 'join';
+                """
+            txn.execute(query)
+            return list(txn)[0][0]
+
+        count = yield self.runInteraction("get_known_servers", _transact)
+
+        # We always know about ourselves, even if we have nothing in
+        # room_memberships (for example, the server is new).
+        self._known_servers_count = max([count, 1])
+        return self._known_servers_count
+
+    def _check_safe_current_state_events_membership_updated_txn(self, txn):
+        """Checks if it is safe to assume the new current_state_events
+        membership column is up to date
+        """
+
+        pending_update = self._simple_select_one_txn(
+            txn,
+            table="background_updates",
+            keyvalues={"update_name": _CURRENT_STATE_MEMBERSHIP_UPDATE_NAME},
+            retcols=["update_name"],
+            allow_none=True,
+        )
+
+        self._current_state_events_membership_up_to_date = not pending_update
+
+        # If the update is still running, reschedule to run.
+        if pending_update:
+            self._clock.call_later(
+                15.0,
+                run_as_background_process,
+                "_check_safe_current_state_events_membership_updated",
+                self.runInteraction,
+                "_check_safe_current_state_events_membership_updated",
+                self._check_safe_current_state_events_membership_updated_txn,
+            )
+
+    @cachedInlineCallbacks(max_entries=100000, iterable=True, cache_context=True)
+    def get_hosts_in_room(self, room_id, cache_context):
+        """Returns the set of all hosts currently in the room
+        """
+        user_ids = yield self.get_users_in_room(
+            room_id, on_invalidate=cache_context.invalidate
+        )
+        hosts = frozenset(get_domain_from_id(user_id) for user_id in user_ids)
+        return hosts
+
+    @cached(max_entries=100000, iterable=True)
+    def get_users_in_room(self, room_id):
+        return self.runInteraction(
+            "get_users_in_room", self.get_users_in_room_txn, room_id
+        )
+
+    def get_users_in_room_txn(self, txn, room_id):
+        # If we can assume current_state_events.membership is up to date
+        # then we can avoid a join, which is a Very Good Thing given how
+        # frequently this function gets called.
+        if self._current_state_events_membership_up_to_date:
+            sql = """
+                SELECT state_key FROM current_state_events
+                WHERE type = 'm.room.member' AND room_id = ? AND membership = ?
+            """
+        else:
+            sql = """
+                SELECT state_key FROM room_memberships as m
+                INNER JOIN current_state_events as c
+                ON m.event_id = c.event_id
+                AND m.room_id = c.room_id
+                AND m.user_id = c.state_key
+                WHERE c.type = 'm.room.member' AND c.room_id = ? AND m.membership = ?
+            """
+
+        txn.execute(sql, (room_id, Membership.JOIN))
+        return [to_ascii(r[0]) for r in txn]
+
+    @cached(max_entries=100000)
+    def get_room_summary(self, room_id):
+        """ Get the details of a room roughly suitable for use by the room
+        summary extension to /sync. Useful when lazy loading room members.
+        Args:
+            room_id (str): The room ID to query
+        Returns:
+            Deferred[dict[str, MemberSummary]:
+                dict of membership states, pointing to a MemberSummary named tuple.
+        """
+
+        def _get_room_summary_txn(txn):
+            # first get counts.
+            # We do this all in one transaction to keep the cache small.
+            # FIXME: get rid of this when we have room_stats
+
+            # If we can assume current_state_events.membership is up to date
+            # then we can avoid a join, which is a Very Good Thing given how
+            # frequently this function gets called.
+            if self._current_state_events_membership_up_to_date:
+                # Note, rejected events will have a null membership field, so
+                # we we manually filter them out.
+                sql = """
+                    SELECT count(*), membership FROM current_state_events
+                    WHERE type = 'm.room.member' AND room_id = ?
+                        AND membership IS NOT NULL
+                    GROUP BY membership
+                """
+            else:
+                sql = """
+                    SELECT count(*), m.membership FROM room_memberships as m
+                    INNER JOIN current_state_events as c
+                    ON m.event_id = c.event_id
+                    AND m.room_id = c.room_id
+                    AND m.user_id = c.state_key
+                    WHERE c.type = 'm.room.member' AND c.room_id = ?
+                    GROUP BY m.membership
+                """
+
+            txn.execute(sql, (room_id,))
+            res = {}
+            for count, membership in txn:
+                summary = res.setdefault(to_ascii(membership), MemberSummary([], count))
+
+            # we order by membership and then fairly arbitrarily by event_id so
+            # heroes are consistent
+            if self._current_state_events_membership_up_to_date:
+                # Note, rejected events will have a null membership field, so
+                # we we manually filter them out.
+                sql = """
+                    SELECT state_key, membership, event_id
+                    FROM current_state_events
+                    WHERE type = 'm.room.member' AND room_id = ?
+                        AND membership IS NOT NULL
+                    ORDER BY
+                        CASE membership WHEN ? THEN 1 WHEN ? THEN 2 ELSE 3 END ASC,
+                        event_id ASC
+                    LIMIT ?
+                """
+            else:
+                sql = """
+                    SELECT c.state_key, m.membership, c.event_id
+                    FROM room_memberships as m
+                    INNER JOIN current_state_events as c USING (room_id, event_id)
+                    WHERE c.type = 'm.room.member' AND c.room_id = ?
+                    ORDER BY
+                        CASE m.membership WHEN ? THEN 1 WHEN ? THEN 2 ELSE 3 END ASC,
+                        c.event_id ASC
+                    LIMIT ?
+                """
+
+            # 6 is 5 (number of heroes) plus 1, in case one of them is the calling user.
+            txn.execute(sql, (room_id, Membership.JOIN, Membership.INVITE, 6))
+            for user_id, membership, event_id in txn:
+                summary = res[to_ascii(membership)]
+                # we will always have a summary for this membership type at this
+                # point given the summary currently contains the counts.
+                members = summary.members
+                members.append((to_ascii(user_id), to_ascii(event_id)))
+
+            return res
+
+        return self.runInteraction("get_room_summary", _get_room_summary_txn)
+
+    def _get_user_counts_in_room_txn(self, txn, room_id):
+        """
+        Get the user count in a room by membership.
+
+        Args:
+            room_id (str)
+            membership (Membership)
+
+        Returns:
+            Deferred[int]
+        """
+        sql = """
+        SELECT m.membership, count(*) FROM room_memberships as m
+            INNER JOIN current_state_events as c USING(event_id)
+            WHERE c.type = 'm.room.member' AND c.room_id = ?
+            GROUP BY m.membership
+        """
+
+        txn.execute(sql, (room_id,))
+        return {row[0]: row[1] for row in txn}
+
+    @cached()
+    def get_invited_rooms_for_user(self, user_id):
+        """ Get all the rooms the user is invited to
+        Args:
+            user_id (str): The user ID.
+        Returns:
+            A deferred list of RoomsForUser.
+        """
+
+        return self.get_rooms_for_user_where_membership_is(user_id, [Membership.INVITE])
+
+    @defer.inlineCallbacks
+    def get_invite_for_user_in_room(self, user_id, room_id):
+        """Gets the invite for the given user and room
+
+        Args:
+            user_id (str)
+            room_id (str)
+
+        Returns:
+            Deferred: Resolves to either a RoomsForUser or None if no invite was
+                found.
+        """
+        invites = yield self.get_invited_rooms_for_user(user_id)
+        for invite in invites:
+            if invite.room_id == room_id:
+                return invite
+        return None
+
+    @defer.inlineCallbacks
+    def get_rooms_for_user_where_membership_is(self, user_id, membership_list):
+        """ Get all the rooms for this user where the membership for this user
+        matches one in the membership list.
+
+        Filters out forgotten rooms.
+
+        Args:
+            user_id (str): The user ID.
+            membership_list (list): A list of synapse.api.constants.Membership
+            values which the user must be in.
+
+        Returns:
+            Deferred[list[RoomsForUser]]
+        """
+        if not membership_list:
+            return defer.succeed(None)
+
+        rooms = yield self.runInteraction(
+            "get_rooms_for_user_where_membership_is",
+            self._get_rooms_for_user_where_membership_is_txn,
+            user_id,
+            membership_list,
+        )
+
+        # Now we filter out forgotten rooms
+        forgotten_rooms = yield self.get_forgotten_rooms_for_user(user_id)
+        return [room for room in rooms if room.room_id not in forgotten_rooms]
+
+    def _get_rooms_for_user_where_membership_is_txn(
+        self, txn, user_id, membership_list
+    ):
+
+        do_invite = Membership.INVITE in membership_list
+        membership_list = [m for m in membership_list if m != Membership.INVITE]
+
+        results = []
+        if membership_list:
+            if self._current_state_events_membership_up_to_date:
+                clause, args = make_in_list_sql_clause(
+                    self.database_engine, "c.membership", membership_list
+                )
+                sql = """
+                    SELECT room_id, e.sender, c.membership, event_id, e.stream_ordering
+                    FROM current_state_events AS c
+                    INNER JOIN events AS e USING (room_id, event_id)
+                    WHERE
+                        c.type = 'm.room.member'
+                        AND state_key = ?
+                        AND %s
+                """ % (
+                    clause,
+                )
+            else:
+                clause, args = make_in_list_sql_clause(
+                    self.database_engine, "m.membership", membership_list
+                )
+                sql = """
+                    SELECT room_id, e.sender, m.membership, event_id, e.stream_ordering
+                    FROM current_state_events AS c
+                    INNER JOIN room_memberships AS m USING (room_id, event_id)
+                    INNER JOIN events AS e USING (room_id, event_id)
+                    WHERE
+                        c.type = 'm.room.member'
+                        AND state_key = ?
+                        AND %s
+                """ % (
+                    clause,
+                )
+
+            txn.execute(sql, (user_id, *args))
+            results = [RoomsForUser(**r) for r in self.cursor_to_dict(txn)]
+
+        if do_invite:
+            sql = (
+                "SELECT i.room_id, inviter, i.event_id, e.stream_ordering"
+                " FROM local_invites as i"
+                " INNER JOIN events as e USING (event_id)"
+                " WHERE invitee = ? AND locally_rejected is NULL"
+                " AND replaced_by is NULL"
+            )
+
+            txn.execute(sql, (user_id,))
+            results.extend(
+                RoomsForUser(
+                    room_id=r["room_id"],
+                    sender=r["inviter"],
+                    event_id=r["event_id"],
+                    stream_ordering=r["stream_ordering"],
+                    membership=Membership.INVITE,
+                )
+                for r in self.cursor_to_dict(txn)
+            )
+
+        return results
+
+    @cachedInlineCallbacks(max_entries=500000, iterable=True)
+    def get_rooms_for_user_with_stream_ordering(self, user_id):
+        """Returns a set of room_ids the user is currently joined to
+
+        Args:
+            user_id (str)
+
+        Returns:
+            Deferred[frozenset[GetRoomsForUserWithStreamOrdering]]: Returns
+            the rooms the user is in currently, along with the stream ordering
+            of the most recent join for that user and room.
+        """
+        rooms = yield self.get_rooms_for_user_where_membership_is(
+            user_id, membership_list=[Membership.JOIN]
+        )
+        return frozenset(
+            GetRoomsForUserWithStreamOrdering(r.room_id, r.stream_ordering)
+            for r in rooms
+        )
+
+    @defer.inlineCallbacks
+    def get_rooms_for_user(self, user_id, on_invalidate=None):
+        """Returns a set of room_ids the user is currently joined to
+        """
+        rooms = yield self.get_rooms_for_user_with_stream_ordering(
+            user_id, on_invalidate=on_invalidate
+        )
+        return frozenset(r.room_id for r in rooms)
+
+    @cachedInlineCallbacks(max_entries=500000, cache_context=True, iterable=True)
+    def get_users_who_share_room_with_user(self, user_id, cache_context):
+        """Returns the set of users who share a room with `user_id`
+        """
+        room_ids = yield self.get_rooms_for_user(
+            user_id, on_invalidate=cache_context.invalidate
+        )
+
+        user_who_share_room = set()
+        for room_id in room_ids:
+            user_ids = yield self.get_users_in_room(
+                room_id, on_invalidate=cache_context.invalidate
+            )
+            user_who_share_room.update(user_ids)
+
+        return user_who_share_room
+
+    @defer.inlineCallbacks
+    def get_joined_users_from_context(self, event, context):
+        state_group = context.state_group
+        if not state_group:
+            # If state_group is None it means it has yet to be assigned a
+            # state group, i.e. we need to make sure that calls with a state_group
+            # of None don't hit previous cached calls with a None state_group.
+            # To do this we set the state_group to a new object as object() != object()
+            state_group = object()
+
+        current_state_ids = yield context.get_current_state_ids(self)
+        result = yield self._get_joined_users_from_context(
+            event.room_id, state_group, current_state_ids, event=event, context=context
+        )
+        return result
+
+    @defer.inlineCallbacks
+    def get_joined_users_from_state(self, room_id, state_entry):
+        state_group = state_entry.state_group
+        if not state_group:
+            # If state_group is None it means it has yet to be assigned a
+            # state group, i.e. we need to make sure that calls with a state_group
+            # of None don't hit previous cached calls with a None state_group.
+            # To do this we set the state_group to a new object as object() != object()
+            state_group = object()
+
+        with Measure(self._clock, "get_joined_users_from_state"):
+            return (
+                yield self._get_joined_users_from_context(
+                    room_id, state_group, state_entry.state, context=state_entry
+                )
+            )
+
+    @cachedInlineCallbacks(
+        num_args=2, cache_context=True, iterable=True, max_entries=100000
+    )
+    def _get_joined_users_from_context(
+        self,
+        room_id,
+        state_group,
+        current_state_ids,
+        cache_context,
+        event=None,
+        context=None,
+    ):
+        # We don't use `state_group`, it's there so that we can cache based
+        # on it. However, it's important that it's never None, since two current_states
+        # with a state_group of None are likely to be different.
+        # See bulk_get_push_rules_for_room for how we work around this.
+        assert state_group is not None
+
+        users_in_room = {}
+        member_event_ids = [
+            e_id
+            for key, e_id in iteritems(current_state_ids)
+            if key[0] == EventTypes.Member
+        ]
+
+        if context is not None:
+            # If we have a context with a delta from a previous state group,
+            # check if we also have the result from the previous group in cache.
+            # If we do then we can reuse that result and simply update it with
+            # any membership changes in `delta_ids`
+            if context.prev_group and context.delta_ids:
+                prev_res = self._get_joined_users_from_context.cache.get(
+                    (room_id, context.prev_group), None
+                )
+                if prev_res and isinstance(prev_res, dict):
+                    users_in_room = dict(prev_res)
+                    member_event_ids = [
+                        e_id
+                        for key, e_id in iteritems(context.delta_ids)
+                        if key[0] == EventTypes.Member
+                    ]
+                    for etype, state_key in context.delta_ids:
+                        users_in_room.pop(state_key, None)
+
+        # We check if we have any of the member event ids in the event cache
+        # before we ask the DB
+
+        # We don't update the event cache hit ratio as it completely throws off
+        # the hit ratio counts. After all, we don't populate the cache if we
+        # miss it here
+        event_map = self._get_events_from_cache(
+            member_event_ids, allow_rejected=False, update_metrics=False
+        )
+
+        missing_member_event_ids = []
+        for event_id in member_event_ids:
+            ev_entry = event_map.get(event_id)
+            if ev_entry:
+                if ev_entry.event.membership == Membership.JOIN:
+                    users_in_room[to_ascii(ev_entry.event.state_key)] = ProfileInfo(
+                        display_name=to_ascii(
+                            ev_entry.event.content.get("displayname", None)
+                        ),
+                        avatar_url=to_ascii(
+                            ev_entry.event.content.get("avatar_url", None)
+                        ),
+                    )
+            else:
+                missing_member_event_ids.append(event_id)
+
+        if missing_member_event_ids:
+            event_to_memberships = yield self._get_joined_profiles_from_event_ids(
+                missing_member_event_ids
+            )
+            users_in_room.update((row for row in event_to_memberships.values() if row))
+
+        if event is not None and event.type == EventTypes.Member:
+            if event.membership == Membership.JOIN:
+                if event.event_id in member_event_ids:
+                    users_in_room[to_ascii(event.state_key)] = ProfileInfo(
+                        display_name=to_ascii(event.content.get("displayname", None)),
+                        avatar_url=to_ascii(event.content.get("avatar_url", None)),
+                    )
+
+        return users_in_room
+
+    @cached(max_entries=10000)
+    def _get_joined_profile_from_event_id(self, event_id):
+        raise NotImplementedError()
+
+    @cachedList(
+        cached_method_name="_get_joined_profile_from_event_id",
+        list_name="event_ids",
+        inlineCallbacks=True,
+    )
+    def _get_joined_profiles_from_event_ids(self, event_ids):
+        """For given set of member event_ids check if they point to a join
+        event and if so return the associated user and profile info.
+
+        Args:
+            event_ids (Iterable[str]): The member event IDs to lookup
+
+        Returns:
+            Deferred[dict[str, Tuple[str, ProfileInfo]|None]]: Map from event ID
+            to `user_id` and ProfileInfo (or None if not join event).
+        """
+
+        rows = yield self._simple_select_many_batch(
+            table="room_memberships",
+            column="event_id",
+            iterable=event_ids,
+            retcols=("user_id", "display_name", "avatar_url", "event_id"),
+            keyvalues={"membership": Membership.JOIN},
+            batch_size=500,
+            desc="_get_membership_from_event_ids",
+        )
+
+        return {
+            row["event_id"]: (
+                row["user_id"],
+                ProfileInfo(
+                    avatar_url=row["avatar_url"], display_name=row["display_name"]
+                ),
+            )
+            for row in rows
+        }
+
+    @cachedInlineCallbacks(max_entries=10000)
+    def is_host_joined(self, room_id, host):
+        if "%" in host or "_" in host:
+            raise Exception("Invalid host name")
+
+        sql = """
+            SELECT state_key FROM current_state_events AS c
+            INNER JOIN room_memberships AS m USING (event_id)
+            WHERE m.membership = 'join'
+                AND type = 'm.room.member'
+                AND c.room_id = ?
+                AND state_key LIKE ?
+            LIMIT 1
+        """
+
+        # We do need to be careful to ensure that host doesn't have any wild cards
+        # in it, but we checked above for known ones and we'll check below that
+        # the returned user actually has the correct domain.
+        like_clause = "%:" + host
+
+        rows = yield self._execute("is_host_joined", None, sql, room_id, like_clause)
+
+        if not rows:
+            return False
+
+        user_id = rows[0][0]
+        if get_domain_from_id(user_id) != host:
+            # This can only happen if the host name has something funky in it
+            raise Exception("Invalid host name")
+
+        return True
+
+    @cachedInlineCallbacks()
+    def was_host_joined(self, room_id, host):
+        """Check whether the server is or ever was in the room.
+
+        Args:
+            room_id (str)
+            host (str)
+
+        Returns:
+            Deferred: Resolves to True if the host is/was in the room, otherwise
+            False.
+        """
+        if "%" in host or "_" in host:
+            raise Exception("Invalid host name")
+
+        sql = """
+            SELECT user_id FROM room_memberships
+            WHERE room_id = ?
+                AND user_id LIKE ?
+                AND membership = 'join'
+            LIMIT 1
+        """
+
+        # We do need to be careful to ensure that host doesn't have any wild cards
+        # in it, but we checked above for known ones and we'll check below that
+        # the returned user actually has the correct domain.
+        like_clause = "%:" + host
+
+        rows = yield self._execute("was_host_joined", None, sql, room_id, like_clause)
+
+        if not rows:
+            return False
+
+        user_id = rows[0][0]
+        if get_domain_from_id(user_id) != host:
+            # This can only happen if the host name has something funky in it
+            raise Exception("Invalid host name")
+
+        return True
+
+    @defer.inlineCallbacks
+    def get_joined_hosts(self, room_id, state_entry):
+        state_group = state_entry.state_group
+        if not state_group:
+            # If state_group is None it means it has yet to be assigned a
+            # state group, i.e. we need to make sure that calls with a state_group
+            # of None don't hit previous cached calls with a None state_group.
+            # To do this we set the state_group to a new object as object() != object()
+            state_group = object()
+
+        with Measure(self._clock, "get_joined_hosts"):
+            return (
+                yield self._get_joined_hosts(
+                    room_id, state_group, state_entry.state, state_entry=state_entry
+                )
+            )
+
+    @cachedInlineCallbacks(num_args=2, max_entries=10000, iterable=True)
+    # @defer.inlineCallbacks
+    def _get_joined_hosts(self, room_id, state_group, current_state_ids, state_entry):
+        # We don't use `state_group`, its there so that we can cache based
+        # on it. However, its important that its never None, since two current_state's
+        # with a state_group of None are likely to be different.
+        # See bulk_get_push_rules_for_room for how we work around this.
+        assert state_group is not None
+
+        cache = yield self._get_joined_hosts_cache(room_id)
+        joined_hosts = yield cache.get_destinations(state_entry)
+
+        return joined_hosts
+
+    @cached(max_entries=10000)
+    def _get_joined_hosts_cache(self, room_id):
+        return _JoinedHostsCache(self, room_id)
+
+    @cachedInlineCallbacks(num_args=2)
+    def did_forget(self, user_id, room_id):
+        """Returns whether user_id has elected to discard history for room_id.
+
+        Returns False if they have since re-joined."""
+
+        def f(txn):
+            sql = (
+                "SELECT"
+                "  COUNT(*)"
+                " FROM"
+                "  room_memberships"
+                " WHERE"
+                "  user_id = ?"
+                " AND"
+                "  room_id = ?"
+                " AND"
+                "  forgotten = 0"
+            )
+            txn.execute(sql, (user_id, room_id))
+            rows = txn.fetchall()
+            return rows[0][0]
+
+        count = yield self.runInteraction("did_forget_membership", f)
+        return count == 0
+
+    @cached()
+    def get_forgotten_rooms_for_user(self, user_id):
+        """Gets all rooms the user has forgotten.
+
+        Args:
+            user_id (str)
+
+        Returns:
+            Deferred[set[str]]
+        """
+
+        def _get_forgotten_rooms_for_user_txn(txn):
+            # This is a slightly convoluted query that first looks up all rooms
+            # that the user has forgotten in the past, then rechecks that list
+            # to see if any have subsequently been updated. This is done so that
+            # we can use a partial index on `forgotten = 1` on the assumption
+            # that few users will actually forget many rooms.
+            #
+            # Note that a room is considered "forgotten" if *all* membership
+            # events for that user and room have the forgotten field set (as
+            # when a user forgets a room we update all rows for that user and
+            # room, not just the current one).
+            sql = """
+                SELECT room_id, (
+                    SELECT count(*) FROM room_memberships
+                    WHERE room_id = m.room_id AND user_id = m.user_id AND forgotten = 0
+                ) AS count
+                FROM room_memberships AS m
+                WHERE user_id = ? AND forgotten = 1
+                GROUP BY room_id, user_id;
+            """
+            txn.execute(sql, (user_id,))
+            return set(row[0] for row in txn if row[1] == 0)
+
+        return self.runInteraction(
+            "get_forgotten_rooms_for_user", _get_forgotten_rooms_for_user_txn
+        )
+
+    @defer.inlineCallbacks
+    def get_rooms_user_has_been_in(self, user_id):
+        """Get all rooms that the user has ever been in.
+
+        Args:
+            user_id (str)
+
+        Returns:
+            Deferred[set[str]]: Set of room IDs.
+        """
+
+        room_ids = yield self._simple_select_onecol(
+            table="room_memberships",
+            keyvalues={"membership": Membership.JOIN, "user_id": user_id},
+            retcol="room_id",
+            desc="get_rooms_user_has_been_in",
+        )
+
+        return set(room_ids)
+
+
+class RoomMemberBackgroundUpdateStore(BackgroundUpdateStore):
+    def __init__(self, db_conn, hs):
+        super(RoomMemberBackgroundUpdateStore, self).__init__(db_conn, hs)
+        self.register_background_update_handler(
+            _MEMBERSHIP_PROFILE_UPDATE_NAME, self._background_add_membership_profile
+        )
+        self.register_background_update_handler(
+            _CURRENT_STATE_MEMBERSHIP_UPDATE_NAME,
+            self._background_current_state_membership,
+        )
+        self.register_background_index_update(
+            "room_membership_forgotten_idx",
+            index_name="room_memberships_user_room_forgotten",
+            table="room_memberships",
+            columns=["user_id", "room_id"],
+            where_clause="forgotten = 1",
+        )
+
+    @defer.inlineCallbacks
+    def _background_add_membership_profile(self, progress, batch_size):
+        target_min_stream_id = progress.get(
+            "target_min_stream_id_inclusive", self._min_stream_order_on_start
+        )
+        max_stream_id = progress.get(
+            "max_stream_id_exclusive", self._stream_order_on_start + 1
+        )
+
+        INSERT_CLUMP_SIZE = 1000
+
+        def add_membership_profile_txn(txn):
+            sql = """
+                SELECT stream_ordering, event_id, events.room_id, event_json.json
+                FROM events
+                INNER JOIN event_json USING (event_id)
+                INNER JOIN room_memberships USING (event_id)
+                WHERE ? <= stream_ordering AND stream_ordering < ?
+                AND type = 'm.room.member'
+                ORDER BY stream_ordering DESC
+                LIMIT ?
+            """
+
+            txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
+
+            rows = self.cursor_to_dict(txn)
+            if not rows:
+                return 0
+
+            min_stream_id = rows[-1]["stream_ordering"]
+
+            to_update = []
+            for row in rows:
+                event_id = row["event_id"]
+                room_id = row["room_id"]
+                try:
+                    event_json = json.loads(row["json"])
+                    content = event_json["content"]
+                except Exception:
+                    continue
+
+                display_name = content.get("displayname", None)
+                avatar_url = content.get("avatar_url", None)
+
+                if display_name or avatar_url:
+                    to_update.append((display_name, avatar_url, event_id, room_id))
+
+            to_update_sql = """
+                UPDATE room_memberships SET display_name = ?, avatar_url = ?
+                WHERE event_id = ? AND room_id = ?
+            """
+            for index in range(0, len(to_update), INSERT_CLUMP_SIZE):
+                clump = to_update[index : index + INSERT_CLUMP_SIZE]
+                txn.executemany(to_update_sql, clump)
+
+            progress = {
+                "target_min_stream_id_inclusive": target_min_stream_id,
+                "max_stream_id_exclusive": min_stream_id,
+            }
+
+            self._background_update_progress_txn(
+                txn, _MEMBERSHIP_PROFILE_UPDATE_NAME, progress
+            )
+
+            return len(rows)
+
+        result = yield self.runInteraction(
+            _MEMBERSHIP_PROFILE_UPDATE_NAME, add_membership_profile_txn
+        )
+
+        if not result:
+            yield self._end_background_update(_MEMBERSHIP_PROFILE_UPDATE_NAME)
+
+        return result
+
+    @defer.inlineCallbacks
+    def _background_current_state_membership(self, progress, batch_size):
+        """Update the new membership column on current_state_events.
+
+        This works by iterating over all rooms in alphebetical order.
+        """
+
+        def _background_current_state_membership_txn(txn, last_processed_room):
+            processed = 0
+            while processed < batch_size:
+                txn.execute(
+                    """
+                        SELECT MIN(room_id) FROM current_state_events WHERE room_id > ?
+                    """,
+                    (last_processed_room,),
+                )
+                row = txn.fetchone()
+                if not row or not row[0]:
+                    return processed, True
+
+                (next_room,) = row
+
+                sql = """
+                    UPDATE current_state_events
+                    SET membership = (
+                        SELECT membership FROM room_memberships
+                        WHERE event_id = current_state_events.event_id
+                    )
+                    WHERE room_id = ?
+                """
+                txn.execute(sql, (next_room,))
+                processed += txn.rowcount
+
+                last_processed_room = next_room
+
+            self._background_update_progress_txn(
+                txn,
+                _CURRENT_STATE_MEMBERSHIP_UPDATE_NAME,
+                {"last_processed_room": last_processed_room},
+            )
+
+            return processed, False
+
+        # If we haven't got a last processed room then just use the empty
+        # string, which will compare before all room IDs correctly.
+        last_processed_room = progress.get("last_processed_room", "")
+
+        row_count, finished = yield self.runInteraction(
+            "_background_current_state_membership_update",
+            _background_current_state_membership_txn,
+            last_processed_room,
+        )
+
+        if finished:
+            yield self._end_background_update(_CURRENT_STATE_MEMBERSHIP_UPDATE_NAME)
+
+        return row_count
+
+
+class RoomMemberStore(RoomMemberWorkerStore, RoomMemberBackgroundUpdateStore):
+    def __init__(self, db_conn, hs):
+        super(RoomMemberStore, self).__init__(db_conn, hs)
+
+    def _store_room_members_txn(self, txn, events, backfilled):
+        """Store a room member in the database.
+        """
+        self._simple_insert_many_txn(
+            txn,
+            table="room_memberships",
+            values=[
+                {
+                    "event_id": event.event_id,
+                    "user_id": event.state_key,
+                    "sender": event.user_id,
+                    "room_id": event.room_id,
+                    "membership": event.membership,
+                    "display_name": event.content.get("displayname", None),
+                    "avatar_url": event.content.get("avatar_url", None),
+                }
+                for event in events
+            ],
+        )
+
+        for event in events:
+            txn.call_after(
+                self._membership_stream_cache.entity_has_changed,
+                event.state_key,
+                event.internal_metadata.stream_ordering,
+            )
+            txn.call_after(
+                self.get_invited_rooms_for_user.invalidate, (event.state_key,)
+            )
+
+            # We update the local_invites table only if the event is "current",
+            # i.e., its something that has just happened. If the event is an
+            # outlier it is only current if its an "out of band membership",
+            # like a remote invite or a rejection of a remote invite.
+            is_new_state = not backfilled and (
+                not event.internal_metadata.is_outlier()
+                or event.internal_metadata.is_out_of_band_membership()
+            )
+            is_mine = self.hs.is_mine_id(event.state_key)
+            if is_new_state and is_mine:
+                if event.membership == Membership.INVITE:
+                    self._simple_insert_txn(
+                        txn,
+                        table="local_invites",
+                        values={
+                            "event_id": event.event_id,
+                            "invitee": event.state_key,
+                            "inviter": event.sender,
+                            "room_id": event.room_id,
+                            "stream_id": event.internal_metadata.stream_ordering,
+                        },
+                    )
+                else:
+                    sql = (
+                        "UPDATE local_invites SET stream_id = ?, replaced_by = ? WHERE"
+                        " room_id = ? AND invitee = ? AND locally_rejected is NULL"
+                        " AND replaced_by is NULL"
+                    )
+
+                    txn.execute(
+                        sql,
+                        (
+                            event.internal_metadata.stream_ordering,
+                            event.event_id,
+                            event.room_id,
+                            event.state_key,
+                        ),
+                    )
+
+    @defer.inlineCallbacks
+    def locally_reject_invite(self, user_id, room_id):
+        sql = (
+            "UPDATE local_invites SET stream_id = ?, locally_rejected = ? WHERE"
+            " room_id = ? AND invitee = ? AND locally_rejected is NULL"
+            " AND replaced_by is NULL"
+        )
+
+        def f(txn, stream_ordering):
+            txn.execute(sql, (stream_ordering, True, room_id, user_id))
+
+        with self._stream_id_gen.get_next() as stream_ordering:
+            yield self.runInteraction("locally_reject_invite", f, stream_ordering)
+
+    def forget(self, user_id, room_id):
+        """Indicate that user_id wishes to discard history for room_id."""
+
+        def f(txn):
+            sql = (
+                "UPDATE"
+                "  room_memberships"
+                " SET"
+                "  forgotten = 1"
+                " WHERE"
+                "  user_id = ?"
+                " AND"
+                "  room_id = ?"
+            )
+            txn.execute(sql, (user_id, room_id))
+
+            self._invalidate_cache_and_stream(txn, self.did_forget, (user_id, room_id))
+            self._invalidate_cache_and_stream(
+                txn, self.get_forgotten_rooms_for_user, (user_id,)
+            )
+
+        return self.runInteraction("forget_membership", f)
+
+
+class _JoinedHostsCache(object):
+    """Cache for joined hosts in a room that is optimised to handle updates
+    via state deltas.
+    """
+
+    def __init__(self, store, room_id):
+        self.store = store
+        self.room_id = room_id
+
+        self.hosts_to_joined_users = {}
+
+        self.state_group = object()
+
+        self.linearizer = Linearizer("_JoinedHostsCache")
+
+        self._len = 0
+
+    @defer.inlineCallbacks
+    def get_destinations(self, state_entry):
+        """Get set of destinations for a state entry
+
+        Args:
+            state_entry(synapse.state._StateCacheEntry)
+        """
+        if state_entry.state_group == self.state_group:
+            return frozenset(self.hosts_to_joined_users)
+
+        with (yield self.linearizer.queue(())):
+            if state_entry.state_group == self.state_group:
+                pass
+            elif state_entry.prev_group == self.state_group:
+                for (typ, state_key), event_id in iteritems(state_entry.delta_ids):
+                    if typ != EventTypes.Member:
+                        continue
+
+                    host = intern_string(get_domain_from_id(state_key))
+                    user_id = state_key
+                    known_joins = self.hosts_to_joined_users.setdefault(host, set())
+
+                    event = yield self.store.get_event(event_id)
+                    if event.membership == Membership.JOIN:
+                        known_joins.add(user_id)
+                    else:
+                        known_joins.discard(user_id)
+
+                        if not known_joins:
+                            self.hosts_to_joined_users.pop(host, None)
+            else:
+                joined_users = yield self.store.get_joined_users_from_state(
+                    self.room_id, state_entry
+                )
+
+                self.hosts_to_joined_users = {}
+                for user_id in joined_users:
+                    host = intern_string(get_domain_from_id(user_id))
+                    self.hosts_to_joined_users.setdefault(host, set()).add(user_id)
+
+            if state_entry.state_group:
+                self.state_group = state_entry.state_group
+            else:
+                self.state_group = object()
+            self._len = sum(len(v) for v in itervalues(self.hosts_to_joined_users))
+        return frozenset(self.hosts_to_joined_users)
+
+    def __len__(self):
+        return self._len
diff --git a/synapse/storage/schema/delta/12/v12.sql b/synapse/storage/data_stores/main/schema/delta/12/v12.sql
similarity index 100%
rename from synapse/storage/schema/delta/12/v12.sql
rename to synapse/storage/data_stores/main/schema/delta/12/v12.sql
diff --git a/synapse/storage/schema/delta/13/v13.sql b/synapse/storage/data_stores/main/schema/delta/13/v13.sql
similarity index 100%
rename from synapse/storage/schema/delta/13/v13.sql
rename to synapse/storage/data_stores/main/schema/delta/13/v13.sql
diff --git a/synapse/storage/schema/delta/14/v14.sql b/synapse/storage/data_stores/main/schema/delta/14/v14.sql
similarity index 100%
rename from synapse/storage/schema/delta/14/v14.sql
rename to synapse/storage/data_stores/main/schema/delta/14/v14.sql
diff --git a/synapse/storage/schema/delta/15/appservice_txns.sql b/synapse/storage/data_stores/main/schema/delta/15/appservice_txns.sql
similarity index 100%
rename from synapse/storage/schema/delta/15/appservice_txns.sql
rename to synapse/storage/data_stores/main/schema/delta/15/appservice_txns.sql
diff --git a/synapse/storage/schema/delta/15/presence_indices.sql b/synapse/storage/data_stores/main/schema/delta/15/presence_indices.sql
similarity index 100%
rename from synapse/storage/schema/delta/15/presence_indices.sql
rename to synapse/storage/data_stores/main/schema/delta/15/presence_indices.sql
diff --git a/synapse/storage/schema/delta/15/v15.sql b/synapse/storage/data_stores/main/schema/delta/15/v15.sql
similarity index 100%
rename from synapse/storage/schema/delta/15/v15.sql
rename to synapse/storage/data_stores/main/schema/delta/15/v15.sql
diff --git a/synapse/storage/schema/delta/16/events_order_index.sql b/synapse/storage/data_stores/main/schema/delta/16/events_order_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/16/events_order_index.sql
rename to synapse/storage/data_stores/main/schema/delta/16/events_order_index.sql
diff --git a/synapse/storage/schema/delta/16/remote_media_cache_index.sql b/synapse/storage/data_stores/main/schema/delta/16/remote_media_cache_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/16/remote_media_cache_index.sql
rename to synapse/storage/data_stores/main/schema/delta/16/remote_media_cache_index.sql
diff --git a/synapse/storage/schema/delta/16/remove_duplicates.sql b/synapse/storage/data_stores/main/schema/delta/16/remove_duplicates.sql
similarity index 100%
rename from synapse/storage/schema/delta/16/remove_duplicates.sql
rename to synapse/storage/data_stores/main/schema/delta/16/remove_duplicates.sql
diff --git a/synapse/storage/schema/delta/16/room_alias_index.sql b/synapse/storage/data_stores/main/schema/delta/16/room_alias_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/16/room_alias_index.sql
rename to synapse/storage/data_stores/main/schema/delta/16/room_alias_index.sql
diff --git a/synapse/storage/schema/delta/16/unique_constraints.sql b/synapse/storage/data_stores/main/schema/delta/16/unique_constraints.sql
similarity index 100%
rename from synapse/storage/schema/delta/16/unique_constraints.sql
rename to synapse/storage/data_stores/main/schema/delta/16/unique_constraints.sql
diff --git a/synapse/storage/schema/delta/16/users.sql b/synapse/storage/data_stores/main/schema/delta/16/users.sql
similarity index 100%
rename from synapse/storage/schema/delta/16/users.sql
rename to synapse/storage/data_stores/main/schema/delta/16/users.sql
diff --git a/synapse/storage/schema/delta/17/drop_indexes.sql b/synapse/storage/data_stores/main/schema/delta/17/drop_indexes.sql
similarity index 100%
rename from synapse/storage/schema/delta/17/drop_indexes.sql
rename to synapse/storage/data_stores/main/schema/delta/17/drop_indexes.sql
diff --git a/synapse/storage/schema/delta/17/server_keys.sql b/synapse/storage/data_stores/main/schema/delta/17/server_keys.sql
similarity index 100%
rename from synapse/storage/schema/delta/17/server_keys.sql
rename to synapse/storage/data_stores/main/schema/delta/17/server_keys.sql
diff --git a/synapse/storage/schema/delta/17/user_threepids.sql b/synapse/storage/data_stores/main/schema/delta/17/user_threepids.sql
similarity index 100%
rename from synapse/storage/schema/delta/17/user_threepids.sql
rename to synapse/storage/data_stores/main/schema/delta/17/user_threepids.sql
diff --git a/synapse/storage/schema/delta/18/server_keys_bigger_ints.sql b/synapse/storage/data_stores/main/schema/delta/18/server_keys_bigger_ints.sql
similarity index 100%
rename from synapse/storage/schema/delta/18/server_keys_bigger_ints.sql
rename to synapse/storage/data_stores/main/schema/delta/18/server_keys_bigger_ints.sql
diff --git a/synapse/storage/schema/delta/19/event_index.sql b/synapse/storage/data_stores/main/schema/delta/19/event_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/19/event_index.sql
rename to synapse/storage/data_stores/main/schema/delta/19/event_index.sql
diff --git a/synapse/storage/schema/delta/20/dummy.sql b/synapse/storage/data_stores/main/schema/delta/20/dummy.sql
similarity index 100%
rename from synapse/storage/schema/delta/20/dummy.sql
rename to synapse/storage/data_stores/main/schema/delta/20/dummy.sql
diff --git a/synapse/storage/schema/delta/20/pushers.py b/synapse/storage/data_stores/main/schema/delta/20/pushers.py
similarity index 100%
rename from synapse/storage/schema/delta/20/pushers.py
rename to synapse/storage/data_stores/main/schema/delta/20/pushers.py
diff --git a/synapse/storage/schema/delta/21/end_to_end_keys.sql b/synapse/storage/data_stores/main/schema/delta/21/end_to_end_keys.sql
similarity index 100%
rename from synapse/storage/schema/delta/21/end_to_end_keys.sql
rename to synapse/storage/data_stores/main/schema/delta/21/end_to_end_keys.sql
diff --git a/synapse/storage/schema/delta/21/receipts.sql b/synapse/storage/data_stores/main/schema/delta/21/receipts.sql
similarity index 100%
rename from synapse/storage/schema/delta/21/receipts.sql
rename to synapse/storage/data_stores/main/schema/delta/21/receipts.sql
diff --git a/synapse/storage/schema/delta/22/receipts_index.sql b/synapse/storage/data_stores/main/schema/delta/22/receipts_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/22/receipts_index.sql
rename to synapse/storage/data_stores/main/schema/delta/22/receipts_index.sql
diff --git a/synapse/storage/schema/delta/22/user_threepids_unique.sql b/synapse/storage/data_stores/main/schema/delta/22/user_threepids_unique.sql
similarity index 100%
rename from synapse/storage/schema/delta/22/user_threepids_unique.sql
rename to synapse/storage/data_stores/main/schema/delta/22/user_threepids_unique.sql
diff --git a/synapse/storage/schema/delta/23/drop_state_index.sql b/synapse/storage/data_stores/main/schema/delta/23/drop_state_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/23/drop_state_index.sql
rename to synapse/storage/data_stores/main/schema/delta/23/drop_state_index.sql
diff --git a/synapse/storage/schema/delta/24/stats_reporting.sql b/synapse/storage/data_stores/main/schema/delta/24/stats_reporting.sql
similarity index 100%
rename from synapse/storage/schema/delta/24/stats_reporting.sql
rename to synapse/storage/data_stores/main/schema/delta/24/stats_reporting.sql
diff --git a/synapse/storage/schema/delta/25/fts.py b/synapse/storage/data_stores/main/schema/delta/25/fts.py
similarity index 100%
rename from synapse/storage/schema/delta/25/fts.py
rename to synapse/storage/data_stores/main/schema/delta/25/fts.py
diff --git a/synapse/storage/schema/delta/25/guest_access.sql b/synapse/storage/data_stores/main/schema/delta/25/guest_access.sql
similarity index 100%
rename from synapse/storage/schema/delta/25/guest_access.sql
rename to synapse/storage/data_stores/main/schema/delta/25/guest_access.sql
diff --git a/synapse/storage/schema/delta/25/history_visibility.sql b/synapse/storage/data_stores/main/schema/delta/25/history_visibility.sql
similarity index 100%
rename from synapse/storage/schema/delta/25/history_visibility.sql
rename to synapse/storage/data_stores/main/schema/delta/25/history_visibility.sql
diff --git a/synapse/storage/schema/delta/25/tags.sql b/synapse/storage/data_stores/main/schema/delta/25/tags.sql
similarity index 100%
rename from synapse/storage/schema/delta/25/tags.sql
rename to synapse/storage/data_stores/main/schema/delta/25/tags.sql
diff --git a/synapse/storage/schema/delta/26/account_data.sql b/synapse/storage/data_stores/main/schema/delta/26/account_data.sql
similarity index 100%
rename from synapse/storage/schema/delta/26/account_data.sql
rename to synapse/storage/data_stores/main/schema/delta/26/account_data.sql
diff --git a/synapse/storage/schema/delta/27/account_data.sql b/synapse/storage/data_stores/main/schema/delta/27/account_data.sql
similarity index 100%
rename from synapse/storage/schema/delta/27/account_data.sql
rename to synapse/storage/data_stores/main/schema/delta/27/account_data.sql
diff --git a/synapse/storage/schema/delta/27/forgotten_memberships.sql b/synapse/storage/data_stores/main/schema/delta/27/forgotten_memberships.sql
similarity index 100%
rename from synapse/storage/schema/delta/27/forgotten_memberships.sql
rename to synapse/storage/data_stores/main/schema/delta/27/forgotten_memberships.sql
diff --git a/synapse/storage/schema/delta/27/ts.py b/synapse/storage/data_stores/main/schema/delta/27/ts.py
similarity index 100%
rename from synapse/storage/schema/delta/27/ts.py
rename to synapse/storage/data_stores/main/schema/delta/27/ts.py
diff --git a/synapse/storage/schema/delta/28/event_push_actions.sql b/synapse/storage/data_stores/main/schema/delta/28/event_push_actions.sql
similarity index 100%
rename from synapse/storage/schema/delta/28/event_push_actions.sql
rename to synapse/storage/data_stores/main/schema/delta/28/event_push_actions.sql
diff --git a/synapse/storage/schema/delta/28/events_room_stream.sql b/synapse/storage/data_stores/main/schema/delta/28/events_room_stream.sql
similarity index 100%
rename from synapse/storage/schema/delta/28/events_room_stream.sql
rename to synapse/storage/data_stores/main/schema/delta/28/events_room_stream.sql
diff --git a/synapse/storage/schema/delta/28/public_roms_index.sql b/synapse/storage/data_stores/main/schema/delta/28/public_roms_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/28/public_roms_index.sql
rename to synapse/storage/data_stores/main/schema/delta/28/public_roms_index.sql
diff --git a/synapse/storage/schema/delta/28/receipts_user_id_index.sql b/synapse/storage/data_stores/main/schema/delta/28/receipts_user_id_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/28/receipts_user_id_index.sql
rename to synapse/storage/data_stores/main/schema/delta/28/receipts_user_id_index.sql
diff --git a/synapse/storage/schema/delta/28/upgrade_times.sql b/synapse/storage/data_stores/main/schema/delta/28/upgrade_times.sql
similarity index 100%
rename from synapse/storage/schema/delta/28/upgrade_times.sql
rename to synapse/storage/data_stores/main/schema/delta/28/upgrade_times.sql
diff --git a/synapse/storage/schema/delta/28/users_is_guest.sql b/synapse/storage/data_stores/main/schema/delta/28/users_is_guest.sql
similarity index 100%
rename from synapse/storage/schema/delta/28/users_is_guest.sql
rename to synapse/storage/data_stores/main/schema/delta/28/users_is_guest.sql
diff --git a/synapse/storage/schema/delta/29/push_actions.sql b/synapse/storage/data_stores/main/schema/delta/29/push_actions.sql
similarity index 100%
rename from synapse/storage/schema/delta/29/push_actions.sql
rename to synapse/storage/data_stores/main/schema/delta/29/push_actions.sql
diff --git a/synapse/storage/schema/delta/30/alias_creator.sql b/synapse/storage/data_stores/main/schema/delta/30/alias_creator.sql
similarity index 100%
rename from synapse/storage/schema/delta/30/alias_creator.sql
rename to synapse/storage/data_stores/main/schema/delta/30/alias_creator.sql
diff --git a/synapse/storage/schema/delta/30/as_users.py b/synapse/storage/data_stores/main/schema/delta/30/as_users.py
similarity index 100%
rename from synapse/storage/schema/delta/30/as_users.py
rename to synapse/storage/data_stores/main/schema/delta/30/as_users.py
diff --git a/synapse/storage/schema/delta/30/deleted_pushers.sql b/synapse/storage/data_stores/main/schema/delta/30/deleted_pushers.sql
similarity index 100%
rename from synapse/storage/schema/delta/30/deleted_pushers.sql
rename to synapse/storage/data_stores/main/schema/delta/30/deleted_pushers.sql
diff --git a/synapse/storage/schema/delta/30/presence_stream.sql b/synapse/storage/data_stores/main/schema/delta/30/presence_stream.sql
similarity index 100%
rename from synapse/storage/schema/delta/30/presence_stream.sql
rename to synapse/storage/data_stores/main/schema/delta/30/presence_stream.sql
diff --git a/synapse/storage/schema/delta/30/public_rooms.sql b/synapse/storage/data_stores/main/schema/delta/30/public_rooms.sql
similarity index 100%
rename from synapse/storage/schema/delta/30/public_rooms.sql
rename to synapse/storage/data_stores/main/schema/delta/30/public_rooms.sql
diff --git a/synapse/storage/schema/delta/30/push_rule_stream.sql b/synapse/storage/data_stores/main/schema/delta/30/push_rule_stream.sql
similarity index 100%
rename from synapse/storage/schema/delta/30/push_rule_stream.sql
rename to synapse/storage/data_stores/main/schema/delta/30/push_rule_stream.sql
diff --git a/synapse/storage/schema/delta/30/state_stream.sql b/synapse/storage/data_stores/main/schema/delta/30/state_stream.sql
similarity index 100%
rename from synapse/storage/schema/delta/30/state_stream.sql
rename to synapse/storage/data_stores/main/schema/delta/30/state_stream.sql
diff --git a/synapse/storage/schema/delta/30/threepid_guest_access_tokens.sql b/synapse/storage/data_stores/main/schema/delta/30/threepid_guest_access_tokens.sql
similarity index 100%
rename from synapse/storage/schema/delta/30/threepid_guest_access_tokens.sql
rename to synapse/storage/data_stores/main/schema/delta/30/threepid_guest_access_tokens.sql
diff --git a/synapse/storage/schema/delta/31/invites.sql b/synapse/storage/data_stores/main/schema/delta/31/invites.sql
similarity index 100%
rename from synapse/storage/schema/delta/31/invites.sql
rename to synapse/storage/data_stores/main/schema/delta/31/invites.sql
diff --git a/synapse/storage/schema/delta/31/local_media_repository_url_cache.sql b/synapse/storage/data_stores/main/schema/delta/31/local_media_repository_url_cache.sql
similarity index 100%
rename from synapse/storage/schema/delta/31/local_media_repository_url_cache.sql
rename to synapse/storage/data_stores/main/schema/delta/31/local_media_repository_url_cache.sql
diff --git a/synapse/storage/schema/delta/31/pushers.py b/synapse/storage/data_stores/main/schema/delta/31/pushers.py
similarity index 100%
rename from synapse/storage/schema/delta/31/pushers.py
rename to synapse/storage/data_stores/main/schema/delta/31/pushers.py
diff --git a/synapse/storage/schema/delta/31/pushers_index.sql b/synapse/storage/data_stores/main/schema/delta/31/pushers_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/31/pushers_index.sql
rename to synapse/storage/data_stores/main/schema/delta/31/pushers_index.sql
diff --git a/synapse/storage/schema/delta/31/search_update.py b/synapse/storage/data_stores/main/schema/delta/31/search_update.py
similarity index 100%
rename from synapse/storage/schema/delta/31/search_update.py
rename to synapse/storage/data_stores/main/schema/delta/31/search_update.py
diff --git a/synapse/storage/schema/delta/32/events.sql b/synapse/storage/data_stores/main/schema/delta/32/events.sql
similarity index 100%
rename from synapse/storage/schema/delta/32/events.sql
rename to synapse/storage/data_stores/main/schema/delta/32/events.sql
diff --git a/synapse/storage/schema/delta/32/openid.sql b/synapse/storage/data_stores/main/schema/delta/32/openid.sql
similarity index 100%
rename from synapse/storage/schema/delta/32/openid.sql
rename to synapse/storage/data_stores/main/schema/delta/32/openid.sql
diff --git a/synapse/storage/schema/delta/32/pusher_throttle.sql b/synapse/storage/data_stores/main/schema/delta/32/pusher_throttle.sql
similarity index 100%
rename from synapse/storage/schema/delta/32/pusher_throttle.sql
rename to synapse/storage/data_stores/main/schema/delta/32/pusher_throttle.sql
diff --git a/synapse/storage/schema/delta/32/remove_indices.sql b/synapse/storage/data_stores/main/schema/delta/32/remove_indices.sql
similarity index 100%
rename from synapse/storage/schema/delta/32/remove_indices.sql
rename to synapse/storage/data_stores/main/schema/delta/32/remove_indices.sql
diff --git a/synapse/storage/schema/delta/32/reports.sql b/synapse/storage/data_stores/main/schema/delta/32/reports.sql
similarity index 100%
rename from synapse/storage/schema/delta/32/reports.sql
rename to synapse/storage/data_stores/main/schema/delta/32/reports.sql
diff --git a/synapse/storage/schema/delta/33/access_tokens_device_index.sql b/synapse/storage/data_stores/main/schema/delta/33/access_tokens_device_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/33/access_tokens_device_index.sql
rename to synapse/storage/data_stores/main/schema/delta/33/access_tokens_device_index.sql
diff --git a/synapse/storage/schema/delta/33/devices.sql b/synapse/storage/data_stores/main/schema/delta/33/devices.sql
similarity index 100%
rename from synapse/storage/schema/delta/33/devices.sql
rename to synapse/storage/data_stores/main/schema/delta/33/devices.sql
diff --git a/synapse/storage/schema/delta/33/devices_for_e2e_keys.sql b/synapse/storage/data_stores/main/schema/delta/33/devices_for_e2e_keys.sql
similarity index 100%
rename from synapse/storage/schema/delta/33/devices_for_e2e_keys.sql
rename to synapse/storage/data_stores/main/schema/delta/33/devices_for_e2e_keys.sql
diff --git a/synapse/storage/schema/delta/33/devices_for_e2e_keys_clear_unknown_device.sql b/synapse/storage/data_stores/main/schema/delta/33/devices_for_e2e_keys_clear_unknown_device.sql
similarity index 100%
rename from synapse/storage/schema/delta/33/devices_for_e2e_keys_clear_unknown_device.sql
rename to synapse/storage/data_stores/main/schema/delta/33/devices_for_e2e_keys_clear_unknown_device.sql
diff --git a/synapse/storage/schema/delta/33/event_fields.py b/synapse/storage/data_stores/main/schema/delta/33/event_fields.py
similarity index 100%
rename from synapse/storage/schema/delta/33/event_fields.py
rename to synapse/storage/data_stores/main/schema/delta/33/event_fields.py
diff --git a/synapse/storage/schema/delta/33/remote_media_ts.py b/synapse/storage/data_stores/main/schema/delta/33/remote_media_ts.py
similarity index 100%
rename from synapse/storage/schema/delta/33/remote_media_ts.py
rename to synapse/storage/data_stores/main/schema/delta/33/remote_media_ts.py
diff --git a/synapse/storage/schema/delta/33/user_ips_index.sql b/synapse/storage/data_stores/main/schema/delta/33/user_ips_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/33/user_ips_index.sql
rename to synapse/storage/data_stores/main/schema/delta/33/user_ips_index.sql
diff --git a/synapse/storage/schema/delta/34/appservice_stream.sql b/synapse/storage/data_stores/main/schema/delta/34/appservice_stream.sql
similarity index 100%
rename from synapse/storage/schema/delta/34/appservice_stream.sql
rename to synapse/storage/data_stores/main/schema/delta/34/appservice_stream.sql
diff --git a/synapse/storage/schema/delta/34/cache_stream.py b/synapse/storage/data_stores/main/schema/delta/34/cache_stream.py
similarity index 100%
rename from synapse/storage/schema/delta/34/cache_stream.py
rename to synapse/storage/data_stores/main/schema/delta/34/cache_stream.py
diff --git a/synapse/storage/schema/delta/34/device_inbox.sql b/synapse/storage/data_stores/main/schema/delta/34/device_inbox.sql
similarity index 100%
rename from synapse/storage/schema/delta/34/device_inbox.sql
rename to synapse/storage/data_stores/main/schema/delta/34/device_inbox.sql
diff --git a/synapse/storage/schema/delta/34/push_display_name_rename.sql b/synapse/storage/data_stores/main/schema/delta/34/push_display_name_rename.sql
similarity index 100%
rename from synapse/storage/schema/delta/34/push_display_name_rename.sql
rename to synapse/storage/data_stores/main/schema/delta/34/push_display_name_rename.sql
diff --git a/synapse/storage/schema/delta/34/received_txn_purge.py b/synapse/storage/data_stores/main/schema/delta/34/received_txn_purge.py
similarity index 100%
rename from synapse/storage/schema/delta/34/received_txn_purge.py
rename to synapse/storage/data_stores/main/schema/delta/34/received_txn_purge.py
diff --git a/synapse/storage/schema/delta/35/add_state_index.sql b/synapse/storage/data_stores/main/schema/delta/35/add_state_index.sql
similarity index 92%
rename from synapse/storage/schema/delta/35/add_state_index.sql
rename to synapse/storage/data_stores/main/schema/delta/35/add_state_index.sql
index 0fce26345b..33980d02f0 100644
--- a/synapse/storage/schema/delta/35/add_state_index.sql
+++ b/synapse/storage/data_stores/main/schema/delta/35/add_state_index.sql
@@ -13,8 +13,5 @@
  * limitations under the License.
  */
 
-
-ALTER TABLE background_updates ADD COLUMN depends_on TEXT;
-
 INSERT into background_updates (update_name, progress_json, depends_on)
     VALUES ('state_group_state_type_index', '{}', 'state_group_state_deduplication');
diff --git a/synapse/storage/schema/delta/35/contains_url.sql b/synapse/storage/data_stores/main/schema/delta/35/contains_url.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/contains_url.sql
rename to synapse/storage/data_stores/main/schema/delta/35/contains_url.sql
diff --git a/synapse/storage/schema/delta/35/device_outbox.sql b/synapse/storage/data_stores/main/schema/delta/35/device_outbox.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/device_outbox.sql
rename to synapse/storage/data_stores/main/schema/delta/35/device_outbox.sql
diff --git a/synapse/storage/schema/delta/35/device_stream_id.sql b/synapse/storage/data_stores/main/schema/delta/35/device_stream_id.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/device_stream_id.sql
rename to synapse/storage/data_stores/main/schema/delta/35/device_stream_id.sql
diff --git a/synapse/storage/schema/delta/35/event_push_actions_index.sql b/synapse/storage/data_stores/main/schema/delta/35/event_push_actions_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/event_push_actions_index.sql
rename to synapse/storage/data_stores/main/schema/delta/35/event_push_actions_index.sql
diff --git a/synapse/storage/schema/delta/35/public_room_list_change_stream.sql b/synapse/storage/data_stores/main/schema/delta/35/public_room_list_change_stream.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/public_room_list_change_stream.sql
rename to synapse/storage/data_stores/main/schema/delta/35/public_room_list_change_stream.sql
diff --git a/synapse/storage/schema/delta/35/state.sql b/synapse/storage/data_stores/main/schema/delta/35/state.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/state.sql
rename to synapse/storage/data_stores/main/schema/delta/35/state.sql
diff --git a/synapse/storage/schema/delta/35/state_dedupe.sql b/synapse/storage/data_stores/main/schema/delta/35/state_dedupe.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/state_dedupe.sql
rename to synapse/storage/data_stores/main/schema/delta/35/state_dedupe.sql
diff --git a/synapse/storage/schema/delta/35/stream_order_to_extrem.sql b/synapse/storage/data_stores/main/schema/delta/35/stream_order_to_extrem.sql
similarity index 100%
rename from synapse/storage/schema/delta/35/stream_order_to_extrem.sql
rename to synapse/storage/data_stores/main/schema/delta/35/stream_order_to_extrem.sql
diff --git a/synapse/storage/schema/delta/36/readd_public_rooms.sql b/synapse/storage/data_stores/main/schema/delta/36/readd_public_rooms.sql
similarity index 100%
rename from synapse/storage/schema/delta/36/readd_public_rooms.sql
rename to synapse/storage/data_stores/main/schema/delta/36/readd_public_rooms.sql
diff --git a/synapse/storage/schema/delta/37/remove_auth_idx.py b/synapse/storage/data_stores/main/schema/delta/37/remove_auth_idx.py
similarity index 100%
rename from synapse/storage/schema/delta/37/remove_auth_idx.py
rename to synapse/storage/data_stores/main/schema/delta/37/remove_auth_idx.py
diff --git a/synapse/storage/schema/delta/37/user_threepids.sql b/synapse/storage/data_stores/main/schema/delta/37/user_threepids.sql
similarity index 100%
rename from synapse/storage/schema/delta/37/user_threepids.sql
rename to synapse/storage/data_stores/main/schema/delta/37/user_threepids.sql
diff --git a/synapse/storage/schema/delta/38/postgres_fts_gist.sql b/synapse/storage/data_stores/main/schema/delta/38/postgres_fts_gist.sql
similarity index 100%
rename from synapse/storage/schema/delta/38/postgres_fts_gist.sql
rename to synapse/storage/data_stores/main/schema/delta/38/postgres_fts_gist.sql
diff --git a/synapse/storage/schema/delta/39/appservice_room_list.sql b/synapse/storage/data_stores/main/schema/delta/39/appservice_room_list.sql
similarity index 100%
rename from synapse/storage/schema/delta/39/appservice_room_list.sql
rename to synapse/storage/data_stores/main/schema/delta/39/appservice_room_list.sql
diff --git a/synapse/storage/schema/delta/39/device_federation_stream_idx.sql b/synapse/storage/data_stores/main/schema/delta/39/device_federation_stream_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/39/device_federation_stream_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/39/device_federation_stream_idx.sql
diff --git a/synapse/storage/schema/delta/39/event_push_index.sql b/synapse/storage/data_stores/main/schema/delta/39/event_push_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/39/event_push_index.sql
rename to synapse/storage/data_stores/main/schema/delta/39/event_push_index.sql
diff --git a/synapse/storage/schema/delta/39/federation_out_position.sql b/synapse/storage/data_stores/main/schema/delta/39/federation_out_position.sql
similarity index 100%
rename from synapse/storage/schema/delta/39/federation_out_position.sql
rename to synapse/storage/data_stores/main/schema/delta/39/federation_out_position.sql
diff --git a/synapse/storage/schema/delta/39/membership_profile.sql b/synapse/storage/data_stores/main/schema/delta/39/membership_profile.sql
similarity index 100%
rename from synapse/storage/schema/delta/39/membership_profile.sql
rename to synapse/storage/data_stores/main/schema/delta/39/membership_profile.sql
diff --git a/synapse/storage/schema/delta/40/current_state_idx.sql b/synapse/storage/data_stores/main/schema/delta/40/current_state_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/40/current_state_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/40/current_state_idx.sql
diff --git a/synapse/storage/schema/delta/40/device_inbox.sql b/synapse/storage/data_stores/main/schema/delta/40/device_inbox.sql
similarity index 100%
rename from synapse/storage/schema/delta/40/device_inbox.sql
rename to synapse/storage/data_stores/main/schema/delta/40/device_inbox.sql
diff --git a/synapse/storage/schema/delta/40/device_list_streams.sql b/synapse/storage/data_stores/main/schema/delta/40/device_list_streams.sql
similarity index 100%
rename from synapse/storage/schema/delta/40/device_list_streams.sql
rename to synapse/storage/data_stores/main/schema/delta/40/device_list_streams.sql
diff --git a/synapse/storage/schema/delta/40/event_push_summary.sql b/synapse/storage/data_stores/main/schema/delta/40/event_push_summary.sql
similarity index 100%
rename from synapse/storage/schema/delta/40/event_push_summary.sql
rename to synapse/storage/data_stores/main/schema/delta/40/event_push_summary.sql
diff --git a/synapse/storage/schema/delta/40/pushers.sql b/synapse/storage/data_stores/main/schema/delta/40/pushers.sql
similarity index 100%
rename from synapse/storage/schema/delta/40/pushers.sql
rename to synapse/storage/data_stores/main/schema/delta/40/pushers.sql
diff --git a/synapse/storage/schema/delta/41/device_list_stream_idx.sql b/synapse/storage/data_stores/main/schema/delta/41/device_list_stream_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/41/device_list_stream_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/41/device_list_stream_idx.sql
diff --git a/synapse/storage/schema/delta/41/device_outbound_index.sql b/synapse/storage/data_stores/main/schema/delta/41/device_outbound_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/41/device_outbound_index.sql
rename to synapse/storage/data_stores/main/schema/delta/41/device_outbound_index.sql
diff --git a/synapse/storage/schema/delta/41/event_search_event_id_idx.sql b/synapse/storage/data_stores/main/schema/delta/41/event_search_event_id_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/41/event_search_event_id_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/41/event_search_event_id_idx.sql
diff --git a/synapse/storage/schema/delta/41/ratelimit.sql b/synapse/storage/data_stores/main/schema/delta/41/ratelimit.sql
similarity index 100%
rename from synapse/storage/schema/delta/41/ratelimit.sql
rename to synapse/storage/data_stores/main/schema/delta/41/ratelimit.sql
diff --git a/synapse/storage/schema/delta/42/current_state_delta.sql b/synapse/storage/data_stores/main/schema/delta/42/current_state_delta.sql
similarity index 100%
rename from synapse/storage/schema/delta/42/current_state_delta.sql
rename to synapse/storage/data_stores/main/schema/delta/42/current_state_delta.sql
diff --git a/synapse/storage/schema/delta/42/device_list_last_id.sql b/synapse/storage/data_stores/main/schema/delta/42/device_list_last_id.sql
similarity index 100%
rename from synapse/storage/schema/delta/42/device_list_last_id.sql
rename to synapse/storage/data_stores/main/schema/delta/42/device_list_last_id.sql
diff --git a/synapse/storage/schema/delta/42/event_auth_state_only.sql b/synapse/storage/data_stores/main/schema/delta/42/event_auth_state_only.sql
similarity index 100%
rename from synapse/storage/schema/delta/42/event_auth_state_only.sql
rename to synapse/storage/data_stores/main/schema/delta/42/event_auth_state_only.sql
diff --git a/synapse/storage/schema/delta/42/user_dir.py b/synapse/storage/data_stores/main/schema/delta/42/user_dir.py
similarity index 100%
rename from synapse/storage/schema/delta/42/user_dir.py
rename to synapse/storage/data_stores/main/schema/delta/42/user_dir.py
diff --git a/synapse/storage/schema/delta/43/blocked_rooms.sql b/synapse/storage/data_stores/main/schema/delta/43/blocked_rooms.sql
similarity index 100%
rename from synapse/storage/schema/delta/43/blocked_rooms.sql
rename to synapse/storage/data_stores/main/schema/delta/43/blocked_rooms.sql
diff --git a/synapse/storage/schema/delta/43/quarantine_media.sql b/synapse/storage/data_stores/main/schema/delta/43/quarantine_media.sql
similarity index 100%
rename from synapse/storage/schema/delta/43/quarantine_media.sql
rename to synapse/storage/data_stores/main/schema/delta/43/quarantine_media.sql
diff --git a/synapse/storage/schema/delta/43/url_cache.sql b/synapse/storage/data_stores/main/schema/delta/43/url_cache.sql
similarity index 100%
rename from synapse/storage/schema/delta/43/url_cache.sql
rename to synapse/storage/data_stores/main/schema/delta/43/url_cache.sql
diff --git a/synapse/storage/schema/delta/43/user_share.sql b/synapse/storage/data_stores/main/schema/delta/43/user_share.sql
similarity index 100%
rename from synapse/storage/schema/delta/43/user_share.sql
rename to synapse/storage/data_stores/main/schema/delta/43/user_share.sql
diff --git a/synapse/storage/schema/delta/44/expire_url_cache.sql b/synapse/storage/data_stores/main/schema/delta/44/expire_url_cache.sql
similarity index 100%
rename from synapse/storage/schema/delta/44/expire_url_cache.sql
rename to synapse/storage/data_stores/main/schema/delta/44/expire_url_cache.sql
diff --git a/synapse/storage/schema/delta/45/group_server.sql b/synapse/storage/data_stores/main/schema/delta/45/group_server.sql
similarity index 100%
rename from synapse/storage/schema/delta/45/group_server.sql
rename to synapse/storage/data_stores/main/schema/delta/45/group_server.sql
diff --git a/synapse/storage/schema/delta/45/profile_cache.sql b/synapse/storage/data_stores/main/schema/delta/45/profile_cache.sql
similarity index 100%
rename from synapse/storage/schema/delta/45/profile_cache.sql
rename to synapse/storage/data_stores/main/schema/delta/45/profile_cache.sql
diff --git a/synapse/storage/schema/delta/46/drop_refresh_tokens.sql b/synapse/storage/data_stores/main/schema/delta/46/drop_refresh_tokens.sql
similarity index 100%
rename from synapse/storage/schema/delta/46/drop_refresh_tokens.sql
rename to synapse/storage/data_stores/main/schema/delta/46/drop_refresh_tokens.sql
diff --git a/synapse/storage/schema/delta/46/drop_unique_deleted_pushers.sql b/synapse/storage/data_stores/main/schema/delta/46/drop_unique_deleted_pushers.sql
similarity index 100%
rename from synapse/storage/schema/delta/46/drop_unique_deleted_pushers.sql
rename to synapse/storage/data_stores/main/schema/delta/46/drop_unique_deleted_pushers.sql
diff --git a/synapse/storage/schema/delta/46/group_server.sql b/synapse/storage/data_stores/main/schema/delta/46/group_server.sql
similarity index 100%
rename from synapse/storage/schema/delta/46/group_server.sql
rename to synapse/storage/data_stores/main/schema/delta/46/group_server.sql
diff --git a/synapse/storage/schema/delta/46/local_media_repository_url_idx.sql b/synapse/storage/data_stores/main/schema/delta/46/local_media_repository_url_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/46/local_media_repository_url_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/46/local_media_repository_url_idx.sql
diff --git a/synapse/storage/schema/delta/46/user_dir_null_room_ids.sql b/synapse/storage/data_stores/main/schema/delta/46/user_dir_null_room_ids.sql
similarity index 100%
rename from synapse/storage/schema/delta/46/user_dir_null_room_ids.sql
rename to synapse/storage/data_stores/main/schema/delta/46/user_dir_null_room_ids.sql
diff --git a/synapse/storage/schema/delta/46/user_dir_typos.sql b/synapse/storage/data_stores/main/schema/delta/46/user_dir_typos.sql
similarity index 100%
rename from synapse/storage/schema/delta/46/user_dir_typos.sql
rename to synapse/storage/data_stores/main/schema/delta/46/user_dir_typos.sql
diff --git a/synapse/storage/schema/delta/47/last_access_media.sql b/synapse/storage/data_stores/main/schema/delta/47/last_access_media.sql
similarity index 100%
rename from synapse/storage/schema/delta/47/last_access_media.sql
rename to synapse/storage/data_stores/main/schema/delta/47/last_access_media.sql
diff --git a/synapse/storage/schema/delta/47/postgres_fts_gin.sql b/synapse/storage/data_stores/main/schema/delta/47/postgres_fts_gin.sql
similarity index 100%
rename from synapse/storage/schema/delta/47/postgres_fts_gin.sql
rename to synapse/storage/data_stores/main/schema/delta/47/postgres_fts_gin.sql
diff --git a/synapse/storage/schema/delta/47/push_actions_staging.sql b/synapse/storage/data_stores/main/schema/delta/47/push_actions_staging.sql
similarity index 100%
rename from synapse/storage/schema/delta/47/push_actions_staging.sql
rename to synapse/storage/data_stores/main/schema/delta/47/push_actions_staging.sql
diff --git a/synapse/storage/schema/delta/47/state_group_seq.py b/synapse/storage/data_stores/main/schema/delta/47/state_group_seq.py
similarity index 100%
rename from synapse/storage/schema/delta/47/state_group_seq.py
rename to synapse/storage/data_stores/main/schema/delta/47/state_group_seq.py
diff --git a/synapse/storage/schema/delta/48/add_user_consent.sql b/synapse/storage/data_stores/main/schema/delta/48/add_user_consent.sql
similarity index 100%
rename from synapse/storage/schema/delta/48/add_user_consent.sql
rename to synapse/storage/data_stores/main/schema/delta/48/add_user_consent.sql
diff --git a/synapse/storage/schema/delta/48/add_user_ips_last_seen_index.sql b/synapse/storage/data_stores/main/schema/delta/48/add_user_ips_last_seen_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/48/add_user_ips_last_seen_index.sql
rename to synapse/storage/data_stores/main/schema/delta/48/add_user_ips_last_seen_index.sql
diff --git a/synapse/storage/schema/delta/48/deactivated_users.sql b/synapse/storage/data_stores/main/schema/delta/48/deactivated_users.sql
similarity index 100%
rename from synapse/storage/schema/delta/48/deactivated_users.sql
rename to synapse/storage/data_stores/main/schema/delta/48/deactivated_users.sql
diff --git a/synapse/storage/schema/delta/48/group_unique_indexes.py b/synapse/storage/data_stores/main/schema/delta/48/group_unique_indexes.py
similarity index 100%
rename from synapse/storage/schema/delta/48/group_unique_indexes.py
rename to synapse/storage/data_stores/main/schema/delta/48/group_unique_indexes.py
diff --git a/synapse/storage/schema/delta/48/groups_joinable.sql b/synapse/storage/data_stores/main/schema/delta/48/groups_joinable.sql
similarity index 100%
rename from synapse/storage/schema/delta/48/groups_joinable.sql
rename to synapse/storage/data_stores/main/schema/delta/48/groups_joinable.sql
diff --git a/synapse/storage/schema/delta/49/add_user_consent_server_notice_sent.sql b/synapse/storage/data_stores/main/schema/delta/49/add_user_consent_server_notice_sent.sql
similarity index 100%
rename from synapse/storage/schema/delta/49/add_user_consent_server_notice_sent.sql
rename to synapse/storage/data_stores/main/schema/delta/49/add_user_consent_server_notice_sent.sql
diff --git a/synapse/storage/schema/delta/49/add_user_daily_visits.sql b/synapse/storage/data_stores/main/schema/delta/49/add_user_daily_visits.sql
similarity index 100%
rename from synapse/storage/schema/delta/49/add_user_daily_visits.sql
rename to synapse/storage/data_stores/main/schema/delta/49/add_user_daily_visits.sql
diff --git a/synapse/storage/schema/delta/49/add_user_ips_last_seen_only_index.sql b/synapse/storage/data_stores/main/schema/delta/49/add_user_ips_last_seen_only_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/49/add_user_ips_last_seen_only_index.sql
rename to synapse/storage/data_stores/main/schema/delta/49/add_user_ips_last_seen_only_index.sql
diff --git a/synapse/storage/schema/delta/50/add_creation_ts_users_index.sql b/synapse/storage/data_stores/main/schema/delta/50/add_creation_ts_users_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/50/add_creation_ts_users_index.sql
rename to synapse/storage/data_stores/main/schema/delta/50/add_creation_ts_users_index.sql
diff --git a/synapse/storage/schema/delta/50/erasure_store.sql b/synapse/storage/data_stores/main/schema/delta/50/erasure_store.sql
similarity index 100%
rename from synapse/storage/schema/delta/50/erasure_store.sql
rename to synapse/storage/data_stores/main/schema/delta/50/erasure_store.sql
diff --git a/synapse/storage/schema/delta/50/make_event_content_nullable.py b/synapse/storage/data_stores/main/schema/delta/50/make_event_content_nullable.py
similarity index 100%
rename from synapse/storage/schema/delta/50/make_event_content_nullable.py
rename to synapse/storage/data_stores/main/schema/delta/50/make_event_content_nullable.py
diff --git a/synapse/storage/schema/delta/51/e2e_room_keys.sql b/synapse/storage/data_stores/main/schema/delta/51/e2e_room_keys.sql
similarity index 100%
rename from synapse/storage/schema/delta/51/e2e_room_keys.sql
rename to synapse/storage/data_stores/main/schema/delta/51/e2e_room_keys.sql
diff --git a/synapse/storage/schema/delta/51/monthly_active_users.sql b/synapse/storage/data_stores/main/schema/delta/51/monthly_active_users.sql
similarity index 100%
rename from synapse/storage/schema/delta/51/monthly_active_users.sql
rename to synapse/storage/data_stores/main/schema/delta/51/monthly_active_users.sql
diff --git a/synapse/storage/schema/delta/52/add_event_to_state_group_index.sql b/synapse/storage/data_stores/main/schema/delta/52/add_event_to_state_group_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/52/add_event_to_state_group_index.sql
rename to synapse/storage/data_stores/main/schema/delta/52/add_event_to_state_group_index.sql
diff --git a/synapse/storage/schema/delta/52/device_list_streams_unique_idx.sql b/synapse/storage/data_stores/main/schema/delta/52/device_list_streams_unique_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/52/device_list_streams_unique_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/52/device_list_streams_unique_idx.sql
diff --git a/synapse/storage/schema/delta/52/e2e_room_keys.sql b/synapse/storage/data_stores/main/schema/delta/52/e2e_room_keys.sql
similarity index 100%
rename from synapse/storage/schema/delta/52/e2e_room_keys.sql
rename to synapse/storage/data_stores/main/schema/delta/52/e2e_room_keys.sql
diff --git a/synapse/storage/schema/delta/53/add_user_type_to_users.sql b/synapse/storage/data_stores/main/schema/delta/53/add_user_type_to_users.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/add_user_type_to_users.sql
rename to synapse/storage/data_stores/main/schema/delta/53/add_user_type_to_users.sql
diff --git a/synapse/storage/schema/delta/53/drop_sent_transactions.sql b/synapse/storage/data_stores/main/schema/delta/53/drop_sent_transactions.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/drop_sent_transactions.sql
rename to synapse/storage/data_stores/main/schema/delta/53/drop_sent_transactions.sql
diff --git a/synapse/storage/schema/delta/53/event_format_version.sql b/synapse/storage/data_stores/main/schema/delta/53/event_format_version.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/event_format_version.sql
rename to synapse/storage/data_stores/main/schema/delta/53/event_format_version.sql
diff --git a/synapse/storage/schema/delta/53/user_dir_populate.sql b/synapse/storage/data_stores/main/schema/delta/53/user_dir_populate.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/user_dir_populate.sql
rename to synapse/storage/data_stores/main/schema/delta/53/user_dir_populate.sql
diff --git a/synapse/storage/schema/delta/53/user_ips_index.sql b/synapse/storage/data_stores/main/schema/delta/53/user_ips_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/user_ips_index.sql
rename to synapse/storage/data_stores/main/schema/delta/53/user_ips_index.sql
diff --git a/synapse/storage/schema/delta/53/user_share.sql b/synapse/storage/data_stores/main/schema/delta/53/user_share.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/user_share.sql
rename to synapse/storage/data_stores/main/schema/delta/53/user_share.sql
diff --git a/synapse/storage/schema/delta/53/user_threepid_id.sql b/synapse/storage/data_stores/main/schema/delta/53/user_threepid_id.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/user_threepid_id.sql
rename to synapse/storage/data_stores/main/schema/delta/53/user_threepid_id.sql
diff --git a/synapse/storage/schema/delta/53/users_in_public_rooms.sql b/synapse/storage/data_stores/main/schema/delta/53/users_in_public_rooms.sql
similarity index 100%
rename from synapse/storage/schema/delta/53/users_in_public_rooms.sql
rename to synapse/storage/data_stores/main/schema/delta/53/users_in_public_rooms.sql
diff --git a/synapse/storage/schema/delta/54/account_validity_with_renewal.sql b/synapse/storage/data_stores/main/schema/delta/54/account_validity_with_renewal.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/account_validity_with_renewal.sql
rename to synapse/storage/data_stores/main/schema/delta/54/account_validity_with_renewal.sql
diff --git a/synapse/storage/schema/delta/54/add_validity_to_server_keys.sql b/synapse/storage/data_stores/main/schema/delta/54/add_validity_to_server_keys.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/add_validity_to_server_keys.sql
rename to synapse/storage/data_stores/main/schema/delta/54/add_validity_to_server_keys.sql
diff --git a/synapse/storage/schema/delta/54/delete_forward_extremities.sql b/synapse/storage/data_stores/main/schema/delta/54/delete_forward_extremities.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/delete_forward_extremities.sql
rename to synapse/storage/data_stores/main/schema/delta/54/delete_forward_extremities.sql
diff --git a/synapse/storage/schema/delta/54/drop_legacy_tables.sql b/synapse/storage/data_stores/main/schema/delta/54/drop_legacy_tables.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/drop_legacy_tables.sql
rename to synapse/storage/data_stores/main/schema/delta/54/drop_legacy_tables.sql
diff --git a/synapse/storage/schema/delta/54/drop_presence_list.sql b/synapse/storage/data_stores/main/schema/delta/54/drop_presence_list.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/drop_presence_list.sql
rename to synapse/storage/data_stores/main/schema/delta/54/drop_presence_list.sql
diff --git a/synapse/storage/schema/delta/54/relations.sql b/synapse/storage/data_stores/main/schema/delta/54/relations.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/relations.sql
rename to synapse/storage/data_stores/main/schema/delta/54/relations.sql
diff --git a/synapse/storage/schema/delta/54/stats.sql b/synapse/storage/data_stores/main/schema/delta/54/stats.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/stats.sql
rename to synapse/storage/data_stores/main/schema/delta/54/stats.sql
diff --git a/synapse/storage/schema/delta/54/stats2.sql b/synapse/storage/data_stores/main/schema/delta/54/stats2.sql
similarity index 100%
rename from synapse/storage/schema/delta/54/stats2.sql
rename to synapse/storage/data_stores/main/schema/delta/54/stats2.sql
diff --git a/synapse/storage/schema/delta/55/access_token_expiry.sql b/synapse/storage/data_stores/main/schema/delta/55/access_token_expiry.sql
similarity index 100%
rename from synapse/storage/schema/delta/55/access_token_expiry.sql
rename to synapse/storage/data_stores/main/schema/delta/55/access_token_expiry.sql
diff --git a/synapse/storage/schema/delta/55/track_threepid_validations.sql b/synapse/storage/data_stores/main/schema/delta/55/track_threepid_validations.sql
similarity index 100%
rename from synapse/storage/schema/delta/55/track_threepid_validations.sql
rename to synapse/storage/data_stores/main/schema/delta/55/track_threepid_validations.sql
diff --git a/synapse/storage/schema/delta/55/users_alter_deactivated.sql b/synapse/storage/data_stores/main/schema/delta/55/users_alter_deactivated.sql
similarity index 100%
rename from synapse/storage/schema/delta/55/users_alter_deactivated.sql
rename to synapse/storage/data_stores/main/schema/delta/55/users_alter_deactivated.sql
diff --git a/synapse/storage/schema/delta/56/add_spans_to_device_lists.sql b/synapse/storage/data_stores/main/schema/delta/56/add_spans_to_device_lists.sql
similarity index 100%
rename from synapse/storage/schema/delta/56/add_spans_to_device_lists.sql
rename to synapse/storage/data_stores/main/schema/delta/56/add_spans_to_device_lists.sql
diff --git a/synapse/storage/schema/delta/56/current_state_events_membership.sql b/synapse/storage/data_stores/main/schema/delta/56/current_state_events_membership.sql
similarity index 100%
rename from synapse/storage/schema/delta/56/current_state_events_membership.sql
rename to synapse/storage/data_stores/main/schema/delta/56/current_state_events_membership.sql
diff --git a/synapse/storage/schema/delta/56/current_state_events_membership_mk2.sql b/synapse/storage/data_stores/main/schema/delta/56/current_state_events_membership_mk2.sql
similarity index 100%
rename from synapse/storage/schema/delta/56/current_state_events_membership_mk2.sql
rename to synapse/storage/data_stores/main/schema/delta/56/current_state_events_membership_mk2.sql
diff --git a/synapse/storage/data_stores/main/schema/delta/56/delete_keys_from_deleted_backups.sql b/synapse/storage/data_stores/main/schema/delta/56/delete_keys_from_deleted_backups.sql
new file mode 100644
index 0000000000..1d2ddb1b1a
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/delete_keys_from_deleted_backups.sql
@@ -0,0 +1,25 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* delete room keys that belong to deleted room key version, or to room key
+ * versions that don't exist (anymore)
+ */
+DELETE FROM e2e_room_keys
+WHERE version NOT IN (
+  SELECT version
+  FROM e2e_room_keys_versions
+  WHERE e2e_room_keys.user_id = e2e_room_keys_versions.user_id
+  AND e2e_room_keys_versions.deleted = 0
+);
diff --git a/synapse/storage/data_stores/main/schema/delta/56/destinations_failure_ts.sql b/synapse/storage/data_stores/main/schema/delta/56/destinations_failure_ts.sql
new file mode 100644
index 0000000000..f00889290b
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/destinations_failure_ts.sql
@@ -0,0 +1,25 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/*
+ * Record the timestamp when a given server started failing
+ */
+ALTER TABLE destinations ADD failure_ts BIGINT;
+
+/* as a rough approximation, we assume that the server started failing at
+ * retry_interval before the last retry
+ */
+UPDATE destinations SET failure_ts = retry_last_ts - retry_interval
+    WHERE retry_last_ts > 0;
diff --git a/synapse/storage/data_stores/main/schema/delta/56/destinations_retry_interval_type.sql.postgres b/synapse/storage/data_stores/main/schema/delta/56/destinations_retry_interval_type.sql.postgres
new file mode 100644
index 0000000000..b9bbb18a91
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/destinations_retry_interval_type.sql.postgres
@@ -0,0 +1,18 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+-- We want to store large retry intervals so we upgrade the column from INT
+-- to BIGINT. We don't need to do this on SQLite.
+ALTER TABLE destinations ALTER retry_interval SET DATA TYPE BIGINT;
diff --git a/synapse/storage/data_stores/main/schema/delta/56/devices_last_seen.sql b/synapse/storage/data_stores/main/schema/delta/56/devices_last_seen.sql
new file mode 100644
index 0000000000..dfa902d0ba
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/devices_last_seen.sql
@@ -0,0 +1,24 @@
+/* Copyright 2019 Matrix.org Foundation CIC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+-- Track last seen information for a device in the devices table, rather
+-- than relying on it being in the user_ips table (which we want to be able
+-- to purge old entries from)
+ALTER TABLE devices ADD COLUMN last_seen BIGINT;
+ALTER TABLE devices ADD COLUMN ip TEXT;
+ALTER TABLE devices ADD COLUMN user_agent TEXT;
+
+INSERT INTO background_updates (update_name, progress_json) VALUES
+  ('devices_last_seen', '{}');
diff --git a/synapse/storage/schema/delta/56/drop_unused_event_tables.sql b/synapse/storage/data_stores/main/schema/delta/56/drop_unused_event_tables.sql
similarity index 100%
rename from synapse/storage/schema/delta/56/drop_unused_event_tables.sql
rename to synapse/storage/data_stores/main/schema/delta/56/drop_unused_event_tables.sql
diff --git a/synapse/storage/data_stores/main/schema/delta/56/event_labels.sql b/synapse/storage/data_stores/main/schema/delta/56/event_labels.sql
new file mode 100644
index 0000000000..5e29c1da19
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/event_labels.sql
@@ -0,0 +1,30 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+-- room_id and topoligical_ordering are denormalised from the events table in order to
+-- make the index work.
+CREATE TABLE IF NOT EXISTS event_labels (
+    event_id TEXT,
+    label TEXT,
+    room_id TEXT NOT NULL,
+    topological_ordering BIGINT NOT NULL,
+    PRIMARY KEY(event_id, label)
+);
+
+
+-- This index enables an event pagination looking for a particular label to index the
+-- event_labels table first, which is much quicker than scanning the events table and then
+-- filtering by label, if the label is rarely used relative to the size of the room.
+CREATE INDEX event_labels_room_id_label_idx ON event_labels(room_id, label, topological_ordering);
diff --git a/synapse/storage/schema/delta/56/fix_room_keys_index.sql b/synapse/storage/data_stores/main/schema/delta/56/fix_room_keys_index.sql
similarity index 100%
rename from synapse/storage/schema/delta/56/fix_room_keys_index.sql
rename to synapse/storage/data_stores/main/schema/delta/56/fix_room_keys_index.sql
diff --git a/synapse/storage/data_stores/main/schema/delta/56/hidden_devices.sql b/synapse/storage/data_stores/main/schema/delta/56/hidden_devices.sql
new file mode 100644
index 0000000000..67f8b20297
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/hidden_devices.sql
@@ -0,0 +1,18 @@
+/* Copyright 2019 New Vector Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+-- device list needs to know which ones are "real" devices, and which ones are
+-- just used to avoid collisions
+ALTER TABLE devices ADD COLUMN hidden BOOLEAN DEFAULT FALSE;
diff --git a/synapse/storage/data_stores/main/schema/delta/56/hidden_devices_fix.sql.sqlite b/synapse/storage/data_stores/main/schema/delta/56/hidden_devices_fix.sql.sqlite
new file mode 100644
index 0000000000..e8b1fd35d8
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/hidden_devices_fix.sql.sqlite
@@ -0,0 +1,42 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* Change the hidden column from a default value of FALSE to a default value of
+ * 0, because sqlite3 prior to 3.23.0 caused the hidden column to contain the
+ * string 'FALSE', which is truthy.
+ *
+ * Since sqlite doesn't allow us to just change the default value, we have to
+ * recreate the table, copy the data, fix the rows that have incorrect data, and
+ * replace the old table with the new table.
+ */
+
+CREATE TABLE IF NOT EXISTS devices2 (
+    user_id TEXT NOT NULL,
+    device_id TEXT NOT NULL,
+    display_name TEXT,
+    last_seen BIGINT,
+    ip TEXT,
+    user_agent TEXT,
+    hidden BOOLEAN DEFAULT 0,
+    CONSTRAINT device_uniqueness UNIQUE (user_id, device_id)
+);
+
+INSERT INTO devices2 SELECT * FROM devices;
+
+UPDATE devices2 SET hidden = 0 WHERE hidden = 'FALSE';
+
+DROP TABLE devices;
+
+ALTER TABLE devices2 RENAME TO devices;
diff --git a/synapse/storage/data_stores/main/schema/delta/56/public_room_list_idx.sql b/synapse/storage/data_stores/main/schema/delta/56/public_room_list_idx.sql
new file mode 100644
index 0000000000..7be31ffebb
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/public_room_list_idx.sql
@@ -0,0 +1,16 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+CREATE INDEX public_room_list_stream_network ON public_room_list_stream (appservice_id, network_id, room_id);
diff --git a/synapse/storage/data_stores/main/schema/delta/56/redaction_censor.sql b/synapse/storage/data_stores/main/schema/delta/56/redaction_censor.sql
new file mode 100644
index 0000000000..fe51b02309
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/redaction_censor.sql
@@ -0,0 +1,17 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+ALTER TABLE redactions ADD COLUMN have_censored BOOL NOT NULL DEFAULT false;
+CREATE INDEX redactions_have_censored ON redactions(event_id) WHERE not have_censored;
diff --git a/synapse/storage/data_stores/main/schema/delta/56/redaction_censor2.sql b/synapse/storage/data_stores/main/schema/delta/56/redaction_censor2.sql
new file mode 100644
index 0000000000..77a5eca499
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/redaction_censor2.sql
@@ -0,0 +1,20 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+ALTER TABLE redactions ADD COLUMN received_ts BIGINT;
+CREATE INDEX redactions_have_censored_ts ON redactions(received_ts) WHERE not have_censored;
+
+INSERT INTO background_updates (update_name, progress_json) VALUES
+  ('redactions_received_ts', '{}');
diff --git a/synapse/storage/data_stores/main/schema/delta/56/redaction_censor3_fix_update.sql.postgres b/synapse/storage/data_stores/main/schema/delta/56/redaction_censor3_fix_update.sql.postgres
new file mode 100644
index 0000000000..67471f3ef5
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/redaction_censor3_fix_update.sql.postgres
@@ -0,0 +1,25 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+-- There was a bug where we may have updated censored redactions as bytes,
+-- which can (somehow) cause json to be inserted hex encoded. These updates go
+-- and undoes any such hex encoded JSON.
+
+INSERT into background_updates (update_name, progress_json)
+  VALUES ('event_fix_redactions_bytes_create_index', '{}');
+
+INSERT into background_updates (update_name, progress_json, depends_on)
+  VALUES ('event_fix_redactions_bytes', '{}', 'event_fix_redactions_bytes_create_index');
diff --git a/synapse/storage/schema/delta/56/room_membership_idx.sql b/synapse/storage/data_stores/main/schema/delta/56/room_membership_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/56/room_membership_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/56/room_membership_idx.sql
diff --git a/synapse/storage/data_stores/main/schema/delta/56/signing_keys.sql b/synapse/storage/data_stores/main/schema/delta/56/signing_keys.sql
new file mode 100644
index 0000000000..27a96123e3
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/signing_keys.sql
@@ -0,0 +1,55 @@
+/* Copyright 2019 New Vector Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+-- cross-signing keys
+CREATE TABLE IF NOT EXISTS e2e_cross_signing_keys (
+    user_id TEXT NOT NULL,
+    -- the type of cross-signing key (master, user_signing, or self_signing)
+    keytype TEXT NOT NULL,
+    -- the full key information, as a json-encoded dict
+    keydata TEXT NOT NULL,
+    -- for keeping the keys in order, so that we can fetch the latest one
+    stream_id BIGINT NOT NULL
+);
+
+CREATE UNIQUE INDEX e2e_cross_signing_keys_idx ON e2e_cross_signing_keys(user_id, keytype, stream_id);
+
+-- cross-signing signatures
+CREATE TABLE IF NOT EXISTS e2e_cross_signing_signatures (
+    -- user who did the signing
+    user_id TEXT NOT NULL,
+    -- key used to sign
+    key_id TEXT NOT NULL,
+    -- user who was signed
+    target_user_id TEXT NOT NULL,
+    -- device/key that was signed
+    target_device_id TEXT NOT NULL,
+    -- the actual signature
+    signature TEXT NOT NULL
+);
+
+CREATE UNIQUE INDEX e2e_cross_signing_signatures_idx ON e2e_cross_signing_signatures(user_id, target_user_id, target_device_id);
+
+-- stream of user signature updates
+CREATE TABLE IF NOT EXISTS user_signature_stream (
+    -- uses the same stream ID as device list stream
+    stream_id BIGINT NOT NULL,
+    -- user who did the signing
+    from_user_id TEXT NOT NULL,
+    -- list of users who were signed, as a JSON array
+    user_ids TEXT NOT NULL
+);
+
+CREATE UNIQUE INDEX user_signature_stream_idx ON user_signature_stream(stream_id);
diff --git a/synapse/storage/data_stores/main/schema/delta/56/stats_separated.sql b/synapse/storage/data_stores/main/schema/delta/56/stats_separated.sql
new file mode 100644
index 0000000000..163529c071
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/stats_separated.sql
@@ -0,0 +1,152 @@
+/* Copyright 2018 New Vector Ltd
+ * Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+----- First clean up from previous versions of room stats.
+
+-- First remove old stats stuff
+DROP TABLE IF EXISTS room_stats;
+DROP TABLE IF EXISTS room_state;
+DROP TABLE IF EXISTS room_stats_state;
+DROP TABLE IF EXISTS user_stats;
+DROP TABLE IF EXISTS room_stats_earliest_tokens;
+DROP TABLE IF EXISTS _temp_populate_stats_position;
+DROP TABLE IF EXISTS _temp_populate_stats_rooms;
+DROP TABLE IF EXISTS stats_stream_pos;
+
+-- Unschedule old background updates if they're still scheduled
+DELETE FROM background_updates WHERE update_name IN (
+    'populate_stats_createtables',
+    'populate_stats_process_rooms',
+    'populate_stats_process_users',
+    'populate_stats_cleanup'
+);
+
+INSERT INTO background_updates (update_name, progress_json, depends_on) VALUES
+    ('populate_stats_process_rooms', '{}', '');
+
+INSERT INTO background_updates (update_name, progress_json, depends_on) VALUES
+    ('populate_stats_process_users', '{}', 'populate_stats_process_rooms');
+
+----- Create tables for our version of room stats.
+
+-- single-row table to track position of incremental updates
+DROP TABLE IF EXISTS stats_incremental_position;
+CREATE TABLE stats_incremental_position (
+    Lock CHAR(1) NOT NULL DEFAULT 'X' UNIQUE,  -- Makes sure this table only has one row.
+    stream_id  BIGINT NOT NULL,
+    CHECK (Lock='X')
+);
+
+-- insert a null row and make sure it is the only one.
+INSERT INTO stats_incremental_position (
+    stream_id
+) SELECT COALESCE(MAX(stream_ordering), 0) from events;
+
+-- represents PRESENT room statistics for a room
+-- only holds absolute fields
+DROP TABLE IF EXISTS room_stats_current;
+CREATE TABLE room_stats_current (
+    room_id TEXT NOT NULL PRIMARY KEY,
+
+    -- These are absolute counts
+    current_state_events INT NOT NULL,
+    joined_members INT NOT NULL,
+    invited_members INT NOT NULL,
+    left_members INT NOT NULL,
+    banned_members INT NOT NULL,
+
+    local_users_in_room INT NOT NULL,
+
+    -- The maximum delta stream position that this row takes into account.
+    completed_delta_stream_id BIGINT NOT NULL
+);
+
+
+-- represents HISTORICAL room statistics for a room
+DROP TABLE IF EXISTS room_stats_historical;
+CREATE TABLE room_stats_historical (
+    room_id TEXT NOT NULL,
+    -- These stats cover the time from (end_ts - bucket_size)...end_ts (in ms).
+    -- Note that end_ts is quantised.
+    end_ts BIGINT NOT NULL,
+    bucket_size BIGINT NOT NULL,
+
+    -- These stats are absolute counts
+    current_state_events BIGINT NOT NULL,
+    joined_members BIGINT NOT NULL,
+    invited_members BIGINT NOT NULL,
+    left_members BIGINT NOT NULL,
+    banned_members BIGINT NOT NULL,
+    local_users_in_room BIGINT NOT NULL,
+
+    -- These stats are per time slice
+    total_events BIGINT NOT NULL,
+    total_event_bytes BIGINT NOT NULL,
+
+    PRIMARY KEY (room_id, end_ts)
+);
+
+-- We use this index to speed up deletion of ancient room stats.
+CREATE INDEX room_stats_historical_end_ts ON room_stats_historical (end_ts);
+
+-- represents PRESENT statistics for a user
+-- only holds absolute fields
+DROP TABLE IF EXISTS user_stats_current;
+CREATE TABLE user_stats_current (
+    user_id TEXT NOT NULL PRIMARY KEY,
+
+    joined_rooms BIGINT NOT NULL,
+
+    -- The maximum delta stream position that this row takes into account.
+    completed_delta_stream_id BIGINT NOT NULL
+);
+
+-- represents HISTORICAL statistics for a user
+DROP TABLE IF EXISTS user_stats_historical;
+CREATE TABLE user_stats_historical (
+    user_id TEXT NOT NULL,
+    end_ts BIGINT NOT NULL,
+    bucket_size BIGINT NOT NULL,
+
+    joined_rooms BIGINT NOT NULL,
+
+    invites_sent BIGINT NOT NULL,
+    rooms_created BIGINT NOT NULL,
+    total_events BIGINT NOT NULL,
+    total_event_bytes BIGINT NOT NULL,
+
+    PRIMARY KEY (user_id, end_ts)
+);
+
+-- We use this index to speed up deletion of ancient user stats.
+CREATE INDEX user_stats_historical_end_ts ON user_stats_historical (end_ts);
+
+
+CREATE TABLE room_stats_state (
+    room_id TEXT NOT NULL,
+    name TEXT,
+    canonical_alias TEXT,
+    join_rules TEXT,
+    history_visibility TEXT,
+    encryption TEXT,
+    avatar TEXT,
+    guest_access TEXT,
+    is_federatable BOOLEAN,
+    topic TEXT
+);
+
+CREATE UNIQUE INDEX room_stats_state_room ON room_stats_state(room_id);
diff --git a/synapse/storage/data_stores/main/schema/delta/56/unique_user_filter_index.py b/synapse/storage/data_stores/main/schema/delta/56/unique_user_filter_index.py
new file mode 100644
index 0000000000..1de8b54961
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/unique_user_filter_index.py
@@ -0,0 +1,52 @@
+import logging
+
+from synapse.storage.engines import PostgresEngine
+
+logger = logging.getLogger(__name__)
+
+
+"""
+This migration updates the user_filters table as follows:
+
+ - drops any (user_id, filter_id) duplicates
+ - makes the columns NON-NULLable
+ - turns the index into a UNIQUE index
+"""
+
+
+def run_upgrade(cur, database_engine, *args, **kwargs):
+    pass
+
+
+def run_create(cur, database_engine, *args, **kwargs):
+    if isinstance(database_engine, PostgresEngine):
+        select_clause = """
+            SELECT DISTINCT ON (user_id, filter_id) user_id, filter_id, filter_json
+            FROM user_filters
+        """
+    else:
+        select_clause = """
+            SELECT * FROM user_filters GROUP BY user_id, filter_id
+        """
+    sql = """
+            DROP TABLE IF EXISTS user_filters_migration;
+            DROP INDEX IF EXISTS user_filters_unique;
+            CREATE TABLE user_filters_migration (
+                user_id TEXT NOT NULL,
+                filter_id BIGINT NOT NULL,
+                filter_json BYTEA NOT NULL
+            );
+            INSERT INTO user_filters_migration (user_id, filter_id, filter_json)
+                %s;
+            CREATE UNIQUE INDEX user_filters_unique ON user_filters_migration
+                (user_id, filter_id);
+            DROP TABLE user_filters;
+            ALTER TABLE user_filters_migration RENAME TO user_filters;
+        """ % (
+        select_clause,
+    )
+
+    if isinstance(database_engine, PostgresEngine):
+        cur.execute(sql)
+    else:
+        cur.executescript(sql)
diff --git a/synapse/storage/data_stores/main/schema/delta/56/user_external_ids.sql b/synapse/storage/data_stores/main/schema/delta/56/user_external_ids.sql
new file mode 100644
index 0000000000..91390c4527
--- /dev/null
+++ b/synapse/storage/data_stores/main/schema/delta/56/user_external_ids.sql
@@ -0,0 +1,24 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/*
+ * a table which records mappings from external auth providers to mxids
+ */
+CREATE TABLE IF NOT EXISTS user_external_ids (
+    auth_provider TEXT NOT NULL,
+    external_id TEXT NOT NULL,
+    user_id TEXT NOT NULL,
+    UNIQUE (auth_provider, external_id)
+);
diff --git a/synapse/storage/schema/delta/56/users_in_public_rooms_idx.sql b/synapse/storage/data_stores/main/schema/delta/56/users_in_public_rooms_idx.sql
similarity index 100%
rename from synapse/storage/schema/delta/56/users_in_public_rooms_idx.sql
rename to synapse/storage/data_stores/main/schema/delta/56/users_in_public_rooms_idx.sql
diff --git a/synapse/storage/schema/full_schemas/16/application_services.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/application_services.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/application_services.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/application_services.sql
diff --git a/synapse/storage/schema/full_schemas/16/event_edges.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/event_edges.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/event_edges.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/event_edges.sql
diff --git a/synapse/storage/schema/full_schemas/16/event_signatures.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/event_signatures.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/event_signatures.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/event_signatures.sql
diff --git a/synapse/storage/schema/full_schemas/16/im.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/im.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/im.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/im.sql
diff --git a/synapse/storage/schema/full_schemas/16/keys.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/keys.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/keys.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/keys.sql
diff --git a/synapse/storage/schema/full_schemas/16/media_repository.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/media_repository.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/media_repository.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/media_repository.sql
diff --git a/synapse/storage/schema/full_schemas/16/presence.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/presence.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/presence.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/presence.sql
diff --git a/synapse/storage/schema/full_schemas/16/profiles.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/profiles.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/profiles.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/profiles.sql
diff --git a/synapse/storage/schema/full_schemas/16/push.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/push.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/push.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/push.sql
diff --git a/synapse/storage/schema/full_schemas/16/redactions.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/redactions.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/redactions.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/redactions.sql
diff --git a/synapse/storage/schema/full_schemas/16/room_aliases.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/room_aliases.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/room_aliases.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/room_aliases.sql
diff --git a/synapse/storage/schema/full_schemas/16/state.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/state.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/state.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/state.sql
diff --git a/synapse/storage/schema/full_schemas/16/transactions.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/transactions.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/transactions.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/transactions.sql
diff --git a/synapse/storage/schema/full_schemas/16/users.sql b/synapse/storage/data_stores/main/schema/full_schemas/16/users.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/16/users.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/16/users.sql
diff --git a/synapse/storage/schema/full_schemas/54/full.sql.postgres b/synapse/storage/data_stores/main/schema/full_schemas/54/full.sql.postgres
similarity index 99%
rename from synapse/storage/schema/full_schemas/54/full.sql.postgres
rename to synapse/storage/data_stores/main/schema/full_schemas/54/full.sql.postgres
index 098434356f..4ad2929f32 100644
--- a/synapse/storage/schema/full_schemas/54/full.sql.postgres
+++ b/synapse/storage/data_stores/main/schema/full_schemas/54/full.sql.postgres
@@ -70,15 +70,6 @@ CREATE TABLE appservice_stream_position (
 );
 
 
-
-CREATE TABLE background_updates (
-    update_name text NOT NULL,
-    progress_json text NOT NULL,
-    depends_on text
-);
-
-
-
 CREATE TABLE blocked_rooms (
     room_id text NOT NULL,
     user_id text NOT NULL
@@ -1202,11 +1193,6 @@ ALTER TABLE ONLY appservice_stream_position
 
 
 
-ALTER TABLE ONLY background_updates
-    ADD CONSTRAINT background_updates_uniqueness UNIQUE (update_name);
-
-
-
 ALTER TABLE ONLY current_state_events
     ADD CONSTRAINT current_state_events_event_id_key UNIQUE (event_id);
 
@@ -2047,6 +2033,3 @@ CREATE INDEX users_who_share_private_rooms_r_idx ON users_who_share_private_room
 
 
 CREATE UNIQUE INDEX users_who_share_private_rooms_u_idx ON users_who_share_private_rooms USING btree (user_id, other_user_id, room_id);
-
-
-
diff --git a/synapse/storage/schema/full_schemas/54/full.sql.sqlite b/synapse/storage/data_stores/main/schema/full_schemas/54/full.sql.sqlite
similarity index 99%
rename from synapse/storage/schema/full_schemas/54/full.sql.sqlite
rename to synapse/storage/data_stores/main/schema/full_schemas/54/full.sql.sqlite
index be9295e4c9..bad33291e7 100644
--- a/synapse/storage/schema/full_schemas/54/full.sql.sqlite
+++ b/synapse/storage/data_stores/main/schema/full_schemas/54/full.sql.sqlite
@@ -67,7 +67,6 @@ CREATE INDEX receipts_linearized_id ON receipts_linearized( stream_id );
 CREATE INDEX receipts_linearized_room_stream ON receipts_linearized( room_id, stream_id );
 CREATE TABLE IF NOT EXISTS "user_threepids" ( user_id TEXT NOT NULL, medium TEXT NOT NULL, address TEXT NOT NULL, validated_at BIGINT NOT NULL, added_at BIGINT NOT NULL, CONSTRAINT medium_address UNIQUE (medium, address) );
 CREATE INDEX user_threepids_user_id ON user_threepids(user_id);
-CREATE TABLE background_updates( update_name TEXT NOT NULL, progress_json TEXT NOT NULL, depends_on TEXT, CONSTRAINT background_updates_uniqueness UNIQUE (update_name) );
 CREATE VIRTUAL TABLE event_search USING fts4 ( event_id, room_id, sender, key, value )
 /* event_search(event_id,room_id,sender,"key",value) */;
 CREATE TABLE IF NOT EXISTS 'event_search_content'(docid INTEGER PRIMARY KEY, 'c0event_id', 'c1room_id', 'c2sender', 'c3key', 'c4value');
diff --git a/synapse/storage/schema/full_schemas/54/stream_positions.sql b/synapse/storage/data_stores/main/schema/full_schemas/54/stream_positions.sql
similarity index 100%
rename from synapse/storage/schema/full_schemas/54/stream_positions.sql
rename to synapse/storage/data_stores/main/schema/full_schemas/54/stream_positions.sql
diff --git a/synapse/storage/schema/full_schemas/README.txt b/synapse/storage/data_stores/main/schema/full_schemas/README.txt
similarity index 100%
rename from synapse/storage/schema/full_schemas/README.txt
rename to synapse/storage/data_stores/main/schema/full_schemas/README.txt
diff --git a/synapse/storage/search.py b/synapse/storage/data_stores/main/search.py
similarity index 97%
rename from synapse/storage/search.py
rename to synapse/storage/data_stores/main/search.py
index df87ab6a6d..d1d7c6863d 100644
--- a/synapse/storage/search.py
+++ b/synapse/storage/data_stores/main/search.py
@@ -24,10 +24,10 @@ from canonicaljson import json
 from twisted.internet import defer
 
 from synapse.api.errors import SynapseError
+from synapse.storage._base import make_in_list_sql_clause
+from synapse.storage.background_updates import BackgroundUpdateStore
 from synapse.storage.engines import PostgresEngine, Sqlite3Engine
 
-from .background_updates import BackgroundUpdateStore
-
 logger = logging.getLogger(__name__)
 
 SearchEntry = namedtuple(
@@ -36,7 +36,7 @@ SearchEntry = namedtuple(
 )
 
 
-class SearchStore(BackgroundUpdateStore):
+class SearchBackgroundUpdateStore(BackgroundUpdateStore):
 
     EVENT_SEARCH_UPDATE_NAME = "event_search"
     EVENT_SEARCH_ORDER_UPDATE_NAME = "event_search_order"
@@ -44,7 +44,7 @@ class SearchStore(BackgroundUpdateStore):
     EVENT_SEARCH_USE_GIN_POSTGRES_NAME = "event_search_postgres_gin"
 
     def __init__(self, db_conn, hs):
-        super(SearchStore, self).__init__(db_conn, hs)
+        super(SearchBackgroundUpdateStore, self).__init__(db_conn, hs)
 
         if not hs.config.enable_search:
             return
@@ -196,7 +196,7 @@ class SearchStore(BackgroundUpdateStore):
                         " ON event_search USING GIN (vector)"
                     )
                 except psycopg2.ProgrammingError as e:
-                    logger.warn(
+                    logger.warning(
                         "Ignoring error %r when trying to switch from GIST to GIN", e
                     )
 
@@ -289,29 +289,6 @@ class SearchStore(BackgroundUpdateStore):
 
         return num_rows
 
-    def store_event_search_txn(self, txn, event, key, value):
-        """Add event to the search table
-
-        Args:
-            txn (cursor):
-            event (EventBase):
-            key (str):
-            value (str):
-        """
-        self.store_search_entries_txn(
-            txn,
-            (
-                SearchEntry(
-                    key=key,
-                    value=value,
-                    event_id=event.event_id,
-                    room_id=event.room_id,
-                    stream_ordering=event.internal_metadata.stream_ordering,
-                    origin_server_ts=event.origin_server_ts,
-                ),
-            ),
-        )
-
     def store_search_entries_txn(self, txn, entries):
         """Add entries to the search table
 
@@ -358,6 +335,34 @@ class SearchStore(BackgroundUpdateStore):
             # This should be unreachable.
             raise Exception("Unrecognized database engine")
 
+
+class SearchStore(SearchBackgroundUpdateStore):
+    def __init__(self, db_conn, hs):
+        super(SearchStore, self).__init__(db_conn, hs)
+
+    def store_event_search_txn(self, txn, event, key, value):
+        """Add event to the search table
+
+        Args:
+            txn (cursor):
+            event (EventBase):
+            key (str):
+            value (str):
+        """
+        self.store_search_entries_txn(
+            txn,
+            (
+                SearchEntry(
+                    key=key,
+                    value=value,
+                    event_id=event.event_id,
+                    room_id=event.room_id,
+                    stream_ordering=event.internal_metadata.stream_ordering,
+                    origin_server_ts=event.origin_server_ts,
+                ),
+            ),
+        )
+
     @defer.inlineCallbacks
     def search_msgs(self, room_ids, search_term, keys):
         """Performs a full text search over events with given keys.
@@ -380,8 +385,10 @@ class SearchStore(BackgroundUpdateStore):
         # Make sure we don't explode because the person is in too many rooms.
         # We filter the results below regardless.
         if len(room_ids) < 500:
-            clauses.append("room_id IN (%s)" % (",".join(["?"] * len(room_ids)),))
-            args.extend(room_ids)
+            clause, args = make_in_list_sql_clause(
+                self.database_engine, "room_id", room_ids
+            )
+            clauses = [clause]
 
         local_clauses = []
         for key in keys:
@@ -487,8 +494,10 @@ class SearchStore(BackgroundUpdateStore):
         # Make sure we don't explode because the person is in too many rooms.
         # We filter the results below regardless.
         if len(room_ids) < 500:
-            clauses.append("room_id IN (%s)" % (",".join(["?"] * len(room_ids)),))
-            args.extend(room_ids)
+            clause, args = make_in_list_sql_clause(
+                self.database_engine, "room_id", room_ids
+            )
+            clauses = [clause]
 
         local_clauses = []
         for key in keys:
@@ -663,7 +672,7 @@ class SearchStore(BackgroundUpdateStore):
                     )
                 )
                 txn.execute(query, (value, search_query))
-                headline, = txn.fetchall()[0]
+                (headline,) = txn.fetchall()[0]
 
                 # Now we need to pick the possible highlights out of the haedline
                 # result.
diff --git a/synapse/storage/signatures.py b/synapse/storage/data_stores/main/signatures.py
similarity index 98%
rename from synapse/storage/signatures.py
rename to synapse/storage/data_stores/main/signatures.py
index fb83218f90..556191b76f 100644
--- a/synapse/storage/signatures.py
+++ b/synapse/storage/data_stores/main/signatures.py
@@ -20,10 +20,9 @@ from unpaddedbase64 import encode_base64
 from twisted.internet import defer
 
 from synapse.crypto.event_signing import compute_event_reference_hash
+from synapse.storage._base import SQLBaseStore
 from synapse.util.caches.descriptors import cached, cachedList
 
-from ._base import SQLBaseStore
-
 # py2 sqlite has buffer hardcoded as only binary type, so we must use it,
 # despite being deprecated and removed in favor of memoryview
 if six.PY2:
diff --git a/synapse/storage/data_stores/main/state.py b/synapse/storage/data_stores/main/state.py
new file mode 100644
index 0000000000..9e1541988e
--- /dev/null
+++ b/synapse/storage/data_stores/main/state.py
@@ -0,0 +1,1251 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from collections import namedtuple
+from typing import Iterable, Tuple
+
+from six import iteritems, itervalues
+from six.moves import range
+
+from twisted.internet import defer
+
+from synapse.api.constants import EventTypes
+from synapse.api.errors import NotFoundError
+from synapse.events import EventBase
+from synapse.events.snapshot import EventContext
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.background_updates import BackgroundUpdateStore
+from synapse.storage.data_stores.main.events_worker import EventsWorkerStore
+from synapse.storage.engines import PostgresEngine
+from synapse.storage.state import StateFilter
+from synapse.util.caches import get_cache_factor_for, intern_string
+from synapse.util.caches.descriptors import cached, cachedList
+from synapse.util.caches.dictionary_cache import DictionaryCache
+from synapse.util.stringutils import to_ascii
+
+logger = logging.getLogger(__name__)
+
+
+MAX_STATE_DELTA_HOPS = 100
+
+
+class _GetStateGroupDelta(
+    namedtuple("_GetStateGroupDelta", ("prev_group", "delta_ids"))
+):
+    """Return type of get_state_group_delta that implements __len__, which lets
+    us use the itrable flag when caching
+    """
+
+    __slots__ = []
+
+    def __len__(self):
+        return len(self.delta_ids) if self.delta_ids else 0
+
+
+class StateGroupBackgroundUpdateStore(SQLBaseStore):
+    """Defines functions related to state groups needed to run the state backgroud
+    updates.
+    """
+
+    def _count_state_group_hops_txn(self, txn, state_group):
+        """Given a state group, count how many hops there are in the tree.
+
+        This is used to ensure the delta chains don't get too long.
+        """
+        if isinstance(self.database_engine, PostgresEngine):
+            sql = """
+                WITH RECURSIVE state(state_group) AS (
+                    VALUES(?::bigint)
+                    UNION ALL
+                    SELECT prev_state_group FROM state_group_edges e, state s
+                    WHERE s.state_group = e.state_group
+                )
+                SELECT count(*) FROM state;
+            """
+
+            txn.execute(sql, (state_group,))
+            row = txn.fetchone()
+            if row and row[0]:
+                return row[0]
+            else:
+                return 0
+        else:
+            # We don't use WITH RECURSIVE on sqlite3 as there are distributions
+            # that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
+            next_group = state_group
+            count = 0
+
+            while next_group:
+                next_group = self._simple_select_one_onecol_txn(
+                    txn,
+                    table="state_group_edges",
+                    keyvalues={"state_group": next_group},
+                    retcol="prev_state_group",
+                    allow_none=True,
+                )
+                if next_group:
+                    count += 1
+
+            return count
+
+    def _get_state_groups_from_groups_txn(
+        self, txn, groups, state_filter=StateFilter.all()
+    ):
+        results = {group: {} for group in groups}
+
+        where_clause, where_args = state_filter.make_sql_filter_clause()
+
+        # Unless the filter clause is empty, we're going to append it after an
+        # existing where clause
+        if where_clause:
+            where_clause = " AND (%s)" % (where_clause,)
+
+        if isinstance(self.database_engine, PostgresEngine):
+            # Temporarily disable sequential scans in this transaction. This is
+            # a temporary hack until we can add the right indices in
+            txn.execute("SET LOCAL enable_seqscan=off")
+
+            # The below query walks the state_group tree so that the "state"
+            # table includes all state_groups in the tree. It then joins
+            # against `state_groups_state` to fetch the latest state.
+            # It assumes that previous state groups are always numerically
+            # lesser.
+            # The PARTITION is used to get the event_id in the greatest state
+            # group for the given type, state_key.
+            # This may return multiple rows per (type, state_key), but last_value
+            # should be the same.
+            sql = """
+                WITH RECURSIVE state(state_group) AS (
+                    VALUES(?::bigint)
+                    UNION ALL
+                    SELECT prev_state_group FROM state_group_edges e, state s
+                    WHERE s.state_group = e.state_group
+                )
+                SELECT DISTINCT type, state_key, last_value(event_id) OVER (
+                    PARTITION BY type, state_key ORDER BY state_group ASC
+                    ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
+                ) AS event_id FROM state_groups_state
+                WHERE state_group IN (
+                    SELECT state_group FROM state
+                )
+            """
+
+            for group in groups:
+                args = [group]
+                args.extend(where_args)
+
+                txn.execute(sql + where_clause, args)
+                for row in txn:
+                    typ, state_key, event_id = row
+                    key = (typ, state_key)
+                    results[group][key] = event_id
+        else:
+            max_entries_returned = state_filter.max_entries_returned()
+
+            # We don't use WITH RECURSIVE on sqlite3 as there are distributions
+            # that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
+            for group in groups:
+                next_group = group
+
+                while next_group:
+                    # We did this before by getting the list of group ids, and
+                    # then passing that list to sqlite to get latest event for
+                    # each (type, state_key). However, that was terribly slow
+                    # without the right indices (which we can't add until
+                    # after we finish deduping state, which requires this func)
+                    args = [next_group]
+                    args.extend(where_args)
+
+                    txn.execute(
+                        "SELECT type, state_key, event_id FROM state_groups_state"
+                        " WHERE state_group = ? " + where_clause,
+                        args,
+                    )
+                    results[group].update(
+                        ((typ, state_key), event_id)
+                        for typ, state_key, event_id in txn
+                        if (typ, state_key) not in results[group]
+                    )
+
+                    # If the number of entries in the (type,state_key)->event_id dict
+                    # matches the number of (type,state_keys) types we were searching
+                    # for, then we must have found them all, so no need to go walk
+                    # further down the tree... UNLESS our types filter contained
+                    # wildcards (i.e. Nones) in which case we have to do an exhaustive
+                    # search
+                    if (
+                        max_entries_returned is not None
+                        and len(results[group]) == max_entries_returned
+                    ):
+                        break
+
+                    next_group = self._simple_select_one_onecol_txn(
+                        txn,
+                        table="state_group_edges",
+                        keyvalues={"state_group": next_group},
+                        retcol="prev_state_group",
+                        allow_none=True,
+                    )
+
+        return results
+
+
+# this inherits from EventsWorkerStore because it calls self.get_events
+class StateGroupWorkerStore(
+    EventsWorkerStore, StateGroupBackgroundUpdateStore, SQLBaseStore
+):
+    """The parts of StateGroupStore that can be called from workers.
+    """
+
+    STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
+    STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index"
+    CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx"
+
+    def __init__(self, db_conn, hs):
+        super(StateGroupWorkerStore, self).__init__(db_conn, hs)
+
+        # Originally the state store used a single DictionaryCache to cache the
+        # event IDs for the state types in a given state group to avoid hammering
+        # on the state_group* tables.
+        #
+        # The point of using a DictionaryCache is that it can cache a subset
+        # of the state events for a given state group (i.e. a subset of the keys for a
+        # given dict which is an entry in the cache for a given state group ID).
+        #
+        # However, this poses problems when performing complicated queries
+        # on the store - for instance: "give me all the state for this group, but
+        # limit members to this subset of users", as DictionaryCache's API isn't
+        # rich enough to say "please cache any of these fields, apart from this subset".
+        # This is problematic when lazy loading members, which requires this behaviour,
+        # as without it the cache has no choice but to speculatively load all
+        # state events for the group, which negates the efficiency being sought.
+        #
+        # Rather than overcomplicating DictionaryCache's API, we instead split the
+        # state_group_cache into two halves - one for tracking non-member events,
+        # and the other for tracking member_events.  This means that lazy loading
+        # queries can be made in a cache-friendly manner by querying both caches
+        # separately and then merging the result.  So for the example above, you
+        # would query the members cache for a specific subset of state keys
+        # (which DictionaryCache will handle efficiently and fine) and the non-members
+        # cache for all state (which DictionaryCache will similarly handle fine)
+        # and then just merge the results together.
+        #
+        # We size the non-members cache to be smaller than the members cache as the
+        # vast majority of state in Matrix (today) is member events.
+
+        self._state_group_cache = DictionaryCache(
+            "*stateGroupCache*",
+            # TODO: this hasn't been tuned yet
+            50000 * get_cache_factor_for("stateGroupCache"),
+        )
+        self._state_group_members_cache = DictionaryCache(
+            "*stateGroupMembersCache*",
+            500000 * get_cache_factor_for("stateGroupMembersCache"),
+        )
+
+    @defer.inlineCallbacks
+    def get_room_version(self, room_id):
+        """Get the room_version of a given room
+
+        Args:
+            room_id (str)
+
+        Returns:
+            Deferred[str]
+
+        Raises:
+            NotFoundError if the room is unknown
+        """
+        # for now we do this by looking at the create event. We may want to cache this
+        # more intelligently in future.
+
+        # Retrieve the room's create event
+        create_event = yield self.get_create_event_for_room(room_id)
+        return create_event.content.get("room_version", "1")
+
+    @defer.inlineCallbacks
+    def get_room_predecessor(self, room_id):
+        """Get the predecessor room of an upgraded room if one exists.
+        Otherwise return None.
+
+        Args:
+            room_id (str)
+
+        Returns:
+            Deferred[unicode|None]: predecessor room id
+
+        Raises:
+            NotFoundError if the room is unknown
+        """
+        # Retrieve the room's create event
+        create_event = yield self.get_create_event_for_room(room_id)
+
+        # Return predecessor if present
+        return create_event.content.get("predecessor", None)
+
+    @defer.inlineCallbacks
+    def get_create_event_for_room(self, room_id):
+        """Get the create state event for a room.
+
+        Args:
+            room_id (str)
+
+        Returns:
+            Deferred[EventBase]: The room creation event.
+
+        Raises:
+            NotFoundError if the room is unknown
+        """
+        state_ids = yield self.get_current_state_ids(room_id)
+        create_id = state_ids.get((EventTypes.Create, ""))
+
+        # If we can't find the create event, assume we've hit a dead end
+        if not create_id:
+            raise NotFoundError("Unknown room %s" % (room_id))
+
+        # Retrieve the room's create event and return
+        create_event = yield self.get_event(create_id)
+        return create_event
+
+    @cached(max_entries=100000, iterable=True)
+    def get_current_state_ids(self, room_id):
+        """Get the current state event ids for a room based on the
+        current_state_events table.
+
+        Args:
+            room_id (str)
+
+        Returns:
+            deferred: dict of (type, state_key) -> event_id
+        """
+
+        def _get_current_state_ids_txn(txn):
+            txn.execute(
+                """SELECT type, state_key, event_id FROM current_state_events
+                WHERE room_id = ?
+                """,
+                (room_id,),
+            )
+
+            return {
+                (intern_string(r[0]), intern_string(r[1])): to_ascii(r[2]) for r in txn
+            }
+
+        return self.runInteraction("get_current_state_ids", _get_current_state_ids_txn)
+
+    # FIXME: how should this be cached?
+    def get_filtered_current_state_ids(self, room_id, state_filter=StateFilter.all()):
+        """Get the current state event of a given type for a room based on the
+        current_state_events table.  This may not be as up-to-date as the result
+        of doing a fresh state resolution as per state_handler.get_current_state
+
+        Args:
+            room_id (str)
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns:
+            Deferred[dict[tuple[str, str], str]]: Map from type/state_key to
+            event ID.
+        """
+
+        where_clause, where_args = state_filter.make_sql_filter_clause()
+
+        if not where_clause:
+            # We delegate to the cached version
+            return self.get_current_state_ids(room_id)
+
+        def _get_filtered_current_state_ids_txn(txn):
+            results = {}
+            sql = """
+                SELECT type, state_key, event_id FROM current_state_events
+                WHERE room_id = ?
+            """
+
+            if where_clause:
+                sql += " AND (%s)" % (where_clause,)
+
+            args = [room_id]
+            args.extend(where_args)
+            txn.execute(sql, args)
+            for row in txn:
+                typ, state_key, event_id = row
+                key = (intern_string(typ), intern_string(state_key))
+                results[key] = event_id
+
+            return results
+
+        return self.runInteraction(
+            "get_filtered_current_state_ids", _get_filtered_current_state_ids_txn
+        )
+
+    @defer.inlineCallbacks
+    def get_canonical_alias_for_room(self, room_id):
+        """Get canonical alias for room, if any
+
+        Args:
+            room_id (str)
+
+        Returns:
+            Deferred[str|None]: The canonical alias, if any
+        """
+
+        state = yield self.get_filtered_current_state_ids(
+            room_id, StateFilter.from_types([(EventTypes.CanonicalAlias, "")])
+        )
+
+        event_id = state.get((EventTypes.CanonicalAlias, ""))
+        if not event_id:
+            return
+
+        event = yield self.get_event(event_id, allow_none=True)
+        if not event:
+            return
+
+        return event.content.get("canonical_alias")
+
+    @cached(max_entries=10000, iterable=True)
+    def get_state_group_delta(self, state_group):
+        """Given a state group try to return a previous group and a delta between
+        the old and the new.
+
+        Returns:
+            (prev_group, delta_ids), where both may be None.
+        """
+
+        def _get_state_group_delta_txn(txn):
+            prev_group = self._simple_select_one_onecol_txn(
+                txn,
+                table="state_group_edges",
+                keyvalues={"state_group": state_group},
+                retcol="prev_state_group",
+                allow_none=True,
+            )
+
+            if not prev_group:
+                return _GetStateGroupDelta(None, None)
+
+            delta_ids = self._simple_select_list_txn(
+                txn,
+                table="state_groups_state",
+                keyvalues={"state_group": state_group},
+                retcols=("type", "state_key", "event_id"),
+            )
+
+            return _GetStateGroupDelta(
+                prev_group,
+                {(row["type"], row["state_key"]): row["event_id"] for row in delta_ids},
+            )
+
+        return self.runInteraction("get_state_group_delta", _get_state_group_delta_txn)
+
+    @defer.inlineCallbacks
+    def get_state_groups_ids(self, _room_id, event_ids):
+        """Get the event IDs of all the state for the state groups for the given events
+
+        Args:
+            _room_id (str): id of the room for these events
+            event_ids (iterable[str]): ids of the events
+
+        Returns:
+            Deferred[dict[int, dict[tuple[str, str], str]]]:
+                dict of state_group_id -> (dict of (type, state_key) -> event id)
+        """
+        if not event_ids:
+            return {}
+
+        event_to_groups = yield self._get_state_group_for_events(event_ids)
+
+        groups = set(itervalues(event_to_groups))
+        group_to_state = yield self._get_state_for_groups(groups)
+
+        return group_to_state
+
+    @defer.inlineCallbacks
+    def get_state_ids_for_group(self, state_group):
+        """Get the event IDs of all the state in the given state group
+
+        Args:
+            state_group (int)
+
+        Returns:
+            Deferred[dict]: Resolves to a map of (type, state_key) -> event_id
+        """
+        group_to_state = yield self._get_state_for_groups((state_group,))
+
+        return group_to_state[state_group]
+
+    @defer.inlineCallbacks
+    def get_state_groups(self, room_id, event_ids):
+        """ Get the state groups for the given list of event_ids
+
+        Returns:
+            Deferred[dict[int, list[EventBase]]]:
+                dict of state_group_id -> list of state events.
+        """
+        if not event_ids:
+            return {}
+
+        group_to_ids = yield self.get_state_groups_ids(room_id, event_ids)
+
+        state_event_map = yield self.get_events(
+            [
+                ev_id
+                for group_ids in itervalues(group_to_ids)
+                for ev_id in itervalues(group_ids)
+            ],
+            get_prev_content=False,
+        )
+
+        return {
+            group: [
+                state_event_map[v]
+                for v in itervalues(event_id_map)
+                if v in state_event_map
+            ]
+            for group, event_id_map in iteritems(group_to_ids)
+        }
+
+    @defer.inlineCallbacks
+    def _get_state_groups_from_groups(self, groups, state_filter):
+        """Returns the state groups for a given set of groups, filtering on
+        types of state events.
+
+        Args:
+            groups(list[int]): list of state group IDs to query
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+        Returns:
+            Deferred[dict[int, dict[tuple[str, str], str]]]:
+                dict of state_group_id -> (dict of (type, state_key) -> event id)
+        """
+        results = {}
+
+        chunks = [groups[i : i + 100] for i in range(0, len(groups), 100)]
+        for chunk in chunks:
+            res = yield self.runInteraction(
+                "_get_state_groups_from_groups",
+                self._get_state_groups_from_groups_txn,
+                chunk,
+                state_filter,
+            )
+            results.update(res)
+
+        return results
+
+    @defer.inlineCallbacks
+    def get_state_for_events(self, event_ids, state_filter=StateFilter.all()):
+        """Given a list of event_ids and type tuples, return a list of state
+        dicts for each event.
+
+        Args:
+            event_ids (list[string])
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns:
+            deferred: A dict of (event_id) -> (type, state_key) -> [state_events]
+        """
+        event_to_groups = yield self._get_state_group_for_events(event_ids)
+
+        groups = set(itervalues(event_to_groups))
+        group_to_state = yield self._get_state_for_groups(groups, state_filter)
+
+        state_event_map = yield self.get_events(
+            [ev_id for sd in itervalues(group_to_state) for ev_id in itervalues(sd)],
+            get_prev_content=False,
+        )
+
+        event_to_state = {
+            event_id: {
+                k: state_event_map[v]
+                for k, v in iteritems(group_to_state[group])
+                if v in state_event_map
+            }
+            for event_id, group in iteritems(event_to_groups)
+        }
+
+        return {event: event_to_state[event] for event in event_ids}
+
+    @defer.inlineCallbacks
+    def get_state_ids_for_events(self, event_ids, state_filter=StateFilter.all()):
+        """
+        Get the state dicts corresponding to a list of events, containing the event_ids
+        of the state events (as opposed to the events themselves)
+
+        Args:
+            event_ids(list(str)): events whose state should be returned
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns:
+            A deferred dict from event_id -> (type, state_key) -> event_id
+        """
+        event_to_groups = yield self._get_state_group_for_events(event_ids)
+
+        groups = set(itervalues(event_to_groups))
+        group_to_state = yield self._get_state_for_groups(groups, state_filter)
+
+        event_to_state = {
+            event_id: group_to_state[group]
+            for event_id, group in iteritems(event_to_groups)
+        }
+
+        return {event: event_to_state[event] for event in event_ids}
+
+    @defer.inlineCallbacks
+    def get_state_for_event(self, event_id, state_filter=StateFilter.all()):
+        """
+        Get the state dict corresponding to a particular event
+
+        Args:
+            event_id(str): event whose state should be returned
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns:
+            A deferred dict from (type, state_key) -> state_event
+        """
+        state_map = yield self.get_state_for_events([event_id], state_filter)
+        return state_map[event_id]
+
+    @defer.inlineCallbacks
+    def get_state_ids_for_event(self, event_id, state_filter=StateFilter.all()):
+        """
+        Get the state dict corresponding to a particular event
+
+        Args:
+            event_id(str): event whose state should be returned
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns:
+            A deferred dict from (type, state_key) -> state_event
+        """
+        state_map = yield self.get_state_ids_for_events([event_id], state_filter)
+        return state_map[event_id]
+
+    @cached(max_entries=50000)
+    def _get_state_group_for_event(self, event_id):
+        return self._simple_select_one_onecol(
+            table="event_to_state_groups",
+            keyvalues={"event_id": event_id},
+            retcol="state_group",
+            allow_none=True,
+            desc="_get_state_group_for_event",
+        )
+
+    @cachedList(
+        cached_method_name="_get_state_group_for_event",
+        list_name="event_ids",
+        num_args=1,
+        inlineCallbacks=True,
+    )
+    def _get_state_group_for_events(self, event_ids):
+        """Returns mapping event_id -> state_group
+        """
+        rows = yield self._simple_select_many_batch(
+            table="event_to_state_groups",
+            column="event_id",
+            iterable=event_ids,
+            keyvalues={},
+            retcols=("event_id", "state_group"),
+            desc="_get_state_group_for_events",
+        )
+
+        return {row["event_id"]: row["state_group"] for row in rows}
+
+    def _get_state_for_group_using_cache(self, cache, group, state_filter):
+        """Checks if group is in cache. See `_get_state_for_groups`
+
+        Args:
+            cache(DictionaryCache): the state group cache to use
+            group(int): The state group to lookup
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns 2-tuple (`state_dict`, `got_all`).
+        `got_all` is a bool indicating if we successfully retrieved all
+        requests state from the cache, if False we need to query the DB for the
+        missing state.
+        """
+        is_all, known_absent, state_dict_ids = cache.get(group)
+
+        if is_all or state_filter.is_full():
+            # Either we have everything or want everything, either way
+            # `is_all` tells us whether we've gotten everything.
+            return state_filter.filter_state(state_dict_ids), is_all
+
+        # tracks whether any of our requested types are missing from the cache
+        missing_types = False
+
+        if state_filter.has_wildcards():
+            # We don't know if we fetched all the state keys for the types in
+            # the filter that are wildcards, so we have to assume that we may
+            # have missed some.
+            missing_types = True
+        else:
+            # There aren't any wild cards, so `concrete_types()` returns the
+            # complete list of event types we're wanting.
+            for key in state_filter.concrete_types():
+                if key not in state_dict_ids and key not in known_absent:
+                    missing_types = True
+                    break
+
+        return state_filter.filter_state(state_dict_ids), not missing_types
+
+    @defer.inlineCallbacks
+    def _get_state_for_groups(self, groups, state_filter=StateFilter.all()):
+        """Gets the state at each of a list of state groups, optionally
+        filtering by type/state_key
+
+        Args:
+            groups (iterable[int]): list of state groups for which we want
+                to get the state.
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+        Returns:
+            Deferred[dict[int, dict[tuple[str, str], str]]]:
+                dict of state_group_id -> (dict of (type, state_key) -> event id)
+        """
+
+        member_filter, non_member_filter = state_filter.get_member_split()
+
+        # Now we look them up in the member and non-member caches
+        (
+            non_member_state,
+            incomplete_groups_nm,
+        ) = yield self._get_state_for_groups_using_cache(
+            groups, self._state_group_cache, state_filter=non_member_filter
+        )
+
+        (
+            member_state,
+            incomplete_groups_m,
+        ) = yield self._get_state_for_groups_using_cache(
+            groups, self._state_group_members_cache, state_filter=member_filter
+        )
+
+        state = dict(non_member_state)
+        for group in groups:
+            state[group].update(member_state[group])
+
+        # Now fetch any missing groups from the database
+
+        incomplete_groups = incomplete_groups_m | incomplete_groups_nm
+
+        if not incomplete_groups:
+            return state
+
+        cache_sequence_nm = self._state_group_cache.sequence
+        cache_sequence_m = self._state_group_members_cache.sequence
+
+        # Help the cache hit ratio by expanding the filter a bit
+        db_state_filter = state_filter.return_expanded()
+
+        group_to_state_dict = yield self._get_state_groups_from_groups(
+            list(incomplete_groups), state_filter=db_state_filter
+        )
+
+        # Now lets update the caches
+        self._insert_into_cache(
+            group_to_state_dict,
+            db_state_filter,
+            cache_seq_num_members=cache_sequence_m,
+            cache_seq_num_non_members=cache_sequence_nm,
+        )
+
+        # And finally update the result dict, by filtering out any extra
+        # stuff we pulled out of the database.
+        for group, group_state_dict in iteritems(group_to_state_dict):
+            # We just replace any existing entries, as we will have loaded
+            # everything we need from the database anyway.
+            state[group] = state_filter.filter_state(group_state_dict)
+
+        return state
+
+    def _get_state_for_groups_using_cache(self, groups, cache, state_filter):
+        """Gets the state at each of a list of state groups, optionally
+        filtering by type/state_key, querying from a specific cache.
+
+        Args:
+            groups (iterable[int]): list of state groups for which we want
+                to get the state.
+            cache (DictionaryCache): the cache of group ids to state dicts which
+                we will pass through - either the normal state cache or the specific
+                members state cache.
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns:
+            tuple[dict[int, dict[tuple[str, str], str]], set[int]]: Tuple of
+            dict of state_group_id -> (dict of (type, state_key) -> event id)
+            of entries in the cache, and the state group ids either missing
+            from the cache or incomplete.
+        """
+        results = {}
+        incomplete_groups = set()
+        for group in set(groups):
+            state_dict_ids, got_all = self._get_state_for_group_using_cache(
+                cache, group, state_filter
+            )
+            results[group] = state_dict_ids
+
+            if not got_all:
+                incomplete_groups.add(group)
+
+        return results, incomplete_groups
+
+    def _insert_into_cache(
+        self,
+        group_to_state_dict,
+        state_filter,
+        cache_seq_num_members,
+        cache_seq_num_non_members,
+    ):
+        """Inserts results from querying the database into the relevant cache.
+
+        Args:
+            group_to_state_dict (dict): The new entries pulled from database.
+                Map from state group to state dict
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+            cache_seq_num_members (int): Sequence number of member cache since
+                last lookup in cache
+            cache_seq_num_non_members (int): Sequence number of member cache since
+                last lookup in cache
+        """
+
+        # We need to work out which types we've fetched from the DB for the
+        # member vs non-member caches. This should be as accurate as possible,
+        # but can be an underestimate (e.g. when we have wild cards)
+
+        member_filter, non_member_filter = state_filter.get_member_split()
+        if member_filter.is_full():
+            # We fetched all member events
+            member_types = None
+        else:
+            # `concrete_types()` will only return a subset when there are wild
+            # cards in the filter, but that's fine.
+            member_types = member_filter.concrete_types()
+
+        if non_member_filter.is_full():
+            # We fetched all non member events
+            non_member_types = None
+        else:
+            non_member_types = non_member_filter.concrete_types()
+
+        for group, group_state_dict in iteritems(group_to_state_dict):
+            state_dict_members = {}
+            state_dict_non_members = {}
+
+            for k, v in iteritems(group_state_dict):
+                if k[0] == EventTypes.Member:
+                    state_dict_members[k] = v
+                else:
+                    state_dict_non_members[k] = v
+
+            self._state_group_members_cache.update(
+                cache_seq_num_members,
+                key=group,
+                value=state_dict_members,
+                fetched_keys=member_types,
+            )
+
+            self._state_group_cache.update(
+                cache_seq_num_non_members,
+                key=group,
+                value=state_dict_non_members,
+                fetched_keys=non_member_types,
+            )
+
+    def store_state_group(
+        self, event_id, room_id, prev_group, delta_ids, current_state_ids
+    ):
+        """Store a new set of state, returning a newly assigned state group.
+
+        Args:
+            event_id (str): The event ID for which the state was calculated
+            room_id (str)
+            prev_group (int|None): A previous state group for the room, optional.
+            delta_ids (dict|None): The delta between state at `prev_group` and
+                `current_state_ids`, if `prev_group` was given. Same format as
+                `current_state_ids`.
+            current_state_ids (dict): The state to store. Map of (type, state_key)
+                to event_id.
+
+        Returns:
+            Deferred[int]: The state group ID
+        """
+
+        def _store_state_group_txn(txn):
+            if current_state_ids is None:
+                # AFAIK, this can never happen
+                raise Exception("current_state_ids cannot be None")
+
+            state_group = self.database_engine.get_next_state_group_id(txn)
+
+            self._simple_insert_txn(
+                txn,
+                table="state_groups",
+                values={"id": state_group, "room_id": room_id, "event_id": event_id},
+            )
+
+            # We persist as a delta if we can, while also ensuring the chain
+            # of deltas isn't tooo long, as otherwise read performance degrades.
+            if prev_group:
+                is_in_db = self._simple_select_one_onecol_txn(
+                    txn,
+                    table="state_groups",
+                    keyvalues={"id": prev_group},
+                    retcol="id",
+                    allow_none=True,
+                )
+                if not is_in_db:
+                    raise Exception(
+                        "Trying to persist state with unpersisted prev_group: %r"
+                        % (prev_group,)
+                    )
+
+                potential_hops = self._count_state_group_hops_txn(txn, prev_group)
+            if prev_group and potential_hops < MAX_STATE_DELTA_HOPS:
+                self._simple_insert_txn(
+                    txn,
+                    table="state_group_edges",
+                    values={"state_group": state_group, "prev_state_group": prev_group},
+                )
+
+                self._simple_insert_many_txn(
+                    txn,
+                    table="state_groups_state",
+                    values=[
+                        {
+                            "state_group": state_group,
+                            "room_id": room_id,
+                            "type": key[0],
+                            "state_key": key[1],
+                            "event_id": state_id,
+                        }
+                        for key, state_id in iteritems(delta_ids)
+                    ],
+                )
+            else:
+                self._simple_insert_many_txn(
+                    txn,
+                    table="state_groups_state",
+                    values=[
+                        {
+                            "state_group": state_group,
+                            "room_id": room_id,
+                            "type": key[0],
+                            "state_key": key[1],
+                            "event_id": state_id,
+                        }
+                        for key, state_id in iteritems(current_state_ids)
+                    ],
+                )
+
+            # Prefill the state group caches with this group.
+            # It's fine to use the sequence like this as the state group map
+            # is immutable. (If the map wasn't immutable then this prefill could
+            # race with another update)
+
+            current_member_state_ids = {
+                s: ev
+                for (s, ev) in iteritems(current_state_ids)
+                if s[0] == EventTypes.Member
+            }
+            txn.call_after(
+                self._state_group_members_cache.update,
+                self._state_group_members_cache.sequence,
+                key=state_group,
+                value=dict(current_member_state_ids),
+            )
+
+            current_non_member_state_ids = {
+                s: ev
+                for (s, ev) in iteritems(current_state_ids)
+                if s[0] != EventTypes.Member
+            }
+            txn.call_after(
+                self._state_group_cache.update,
+                self._state_group_cache.sequence,
+                key=state_group,
+                value=dict(current_non_member_state_ids),
+            )
+
+            return state_group
+
+        return self.runInteraction("store_state_group", _store_state_group_txn)
+
+
+class StateBackgroundUpdateStore(
+    StateGroupBackgroundUpdateStore, BackgroundUpdateStore
+):
+
+    STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
+    STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index"
+    CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx"
+    EVENT_STATE_GROUP_INDEX_UPDATE_NAME = "event_to_state_groups_sg_index"
+
+    def __init__(self, db_conn, hs):
+        super(StateBackgroundUpdateStore, self).__init__(db_conn, hs)
+        self.register_background_update_handler(
+            self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME,
+            self._background_deduplicate_state,
+        )
+        self.register_background_update_handler(
+            self.STATE_GROUP_INDEX_UPDATE_NAME, self._background_index_state
+        )
+        self.register_background_index_update(
+            self.CURRENT_STATE_INDEX_UPDATE_NAME,
+            index_name="current_state_events_member_index",
+            table="current_state_events",
+            columns=["state_key"],
+            where_clause="type='m.room.member'",
+        )
+        self.register_background_index_update(
+            self.EVENT_STATE_GROUP_INDEX_UPDATE_NAME,
+            index_name="event_to_state_groups_sg_index",
+            table="event_to_state_groups",
+            columns=["state_group"],
+        )
+
+    @defer.inlineCallbacks
+    def _background_deduplicate_state(self, progress, batch_size):
+        """This background update will slowly deduplicate state by reencoding
+        them as deltas.
+        """
+        last_state_group = progress.get("last_state_group", 0)
+        rows_inserted = progress.get("rows_inserted", 0)
+        max_group = progress.get("max_group", None)
+
+        BATCH_SIZE_SCALE_FACTOR = 100
+
+        batch_size = max(1, int(batch_size / BATCH_SIZE_SCALE_FACTOR))
+
+        if max_group is None:
+            rows = yield self._execute(
+                "_background_deduplicate_state",
+                None,
+                "SELECT coalesce(max(id), 0) FROM state_groups",
+            )
+            max_group = rows[0][0]
+
+        def reindex_txn(txn):
+            new_last_state_group = last_state_group
+            for count in range(batch_size):
+                txn.execute(
+                    "SELECT id, room_id FROM state_groups"
+                    " WHERE ? < id AND id <= ?"
+                    " ORDER BY id ASC"
+                    " LIMIT 1",
+                    (new_last_state_group, max_group),
+                )
+                row = txn.fetchone()
+                if row:
+                    state_group, room_id = row
+
+                if not row or not state_group:
+                    return True, count
+
+                txn.execute(
+                    "SELECT state_group FROM state_group_edges"
+                    " WHERE state_group = ?",
+                    (state_group,),
+                )
+
+                # If we reach a point where we've already started inserting
+                # edges we should stop.
+                if txn.fetchall():
+                    return True, count
+
+                txn.execute(
+                    "SELECT coalesce(max(id), 0) FROM state_groups"
+                    " WHERE id < ? AND room_id = ?",
+                    (state_group, room_id),
+                )
+                (prev_group,) = txn.fetchone()
+                new_last_state_group = state_group
+
+                if prev_group:
+                    potential_hops = self._count_state_group_hops_txn(txn, prev_group)
+                    if potential_hops >= MAX_STATE_DELTA_HOPS:
+                        # We want to ensure chains are at most this long,#
+                        # otherwise read performance degrades.
+                        continue
+
+                    prev_state = self._get_state_groups_from_groups_txn(
+                        txn, [prev_group]
+                    )
+                    prev_state = prev_state[prev_group]
+
+                    curr_state = self._get_state_groups_from_groups_txn(
+                        txn, [state_group]
+                    )
+                    curr_state = curr_state[state_group]
+
+                    if not set(prev_state.keys()) - set(curr_state.keys()):
+                        # We can only do a delta if the current has a strict super set
+                        # of keys
+
+                        delta_state = {
+                            key: value
+                            for key, value in iteritems(curr_state)
+                            if prev_state.get(key, None) != value
+                        }
+
+                        self._simple_delete_txn(
+                            txn,
+                            table="state_group_edges",
+                            keyvalues={"state_group": state_group},
+                        )
+
+                        self._simple_insert_txn(
+                            txn,
+                            table="state_group_edges",
+                            values={
+                                "state_group": state_group,
+                                "prev_state_group": prev_group,
+                            },
+                        )
+
+                        self._simple_delete_txn(
+                            txn,
+                            table="state_groups_state",
+                            keyvalues={"state_group": state_group},
+                        )
+
+                        self._simple_insert_many_txn(
+                            txn,
+                            table="state_groups_state",
+                            values=[
+                                {
+                                    "state_group": state_group,
+                                    "room_id": room_id,
+                                    "type": key[0],
+                                    "state_key": key[1],
+                                    "event_id": state_id,
+                                }
+                                for key, state_id in iteritems(delta_state)
+                            ],
+                        )
+
+            progress = {
+                "last_state_group": state_group,
+                "rows_inserted": rows_inserted + batch_size,
+                "max_group": max_group,
+            }
+
+            self._background_update_progress_txn(
+                txn, self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, progress
+            )
+
+            return False, batch_size
+
+        finished, result = yield self.runInteraction(
+            self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, reindex_txn
+        )
+
+        if finished:
+            yield self._end_background_update(
+                self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME
+            )
+
+        return result * BATCH_SIZE_SCALE_FACTOR
+
+    @defer.inlineCallbacks
+    def _background_index_state(self, progress, batch_size):
+        def reindex_txn(conn):
+            conn.rollback()
+            if isinstance(self.database_engine, PostgresEngine):
+                # postgres insists on autocommit for the index
+                conn.set_session(autocommit=True)
+                try:
+                    txn = conn.cursor()
+                    txn.execute(
+                        "CREATE INDEX CONCURRENTLY state_groups_state_type_idx"
+                        " ON state_groups_state(state_group, type, state_key)"
+                    )
+                    txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
+                finally:
+                    conn.set_session(autocommit=False)
+            else:
+                txn = conn.cursor()
+                txn.execute(
+                    "CREATE INDEX state_groups_state_type_idx"
+                    " ON state_groups_state(state_group, type, state_key)"
+                )
+                txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
+
+        yield self.runWithConnection(reindex_txn)
+
+        yield self._end_background_update(self.STATE_GROUP_INDEX_UPDATE_NAME)
+
+        return 1
+
+
+class StateStore(StateGroupWorkerStore, StateBackgroundUpdateStore):
+    """ Keeps track of the state at a given event.
+
+    This is done by the concept of `state groups`. Every event is a assigned
+    a state group (identified by an arbitrary string), which references a
+    collection of state events. The current state of an event is then the
+    collection of state events referenced by the event's state group.
+
+    Hence, every change in the current state causes a new state group to be
+    generated. However, if no change happens (e.g., if we get a message event
+    with only one parent it inherits the state group from its parent.)
+
+    There are three tables:
+      * `state_groups`: Stores group name, first event with in the group and
+        room id.
+      * `event_to_state_groups`: Maps events to state groups.
+      * `state_groups_state`: Maps state group to state events.
+    """
+
+    def __init__(self, db_conn, hs):
+        super(StateStore, self).__init__(db_conn, hs)
+
+    def _store_event_state_mappings_txn(
+        self, txn, events_and_contexts: Iterable[Tuple[EventBase, EventContext]]
+    ):
+        state_groups = {}
+        for event, context in events_and_contexts:
+            if event.internal_metadata.is_outlier():
+                continue
+
+            # if the event was rejected, just give it the same state as its
+            # predecessor.
+            if context.rejected:
+                state_groups[event.event_id] = context.state_group_before_event
+                continue
+
+            state_groups[event.event_id] = context.state_group
+
+        self._simple_insert_many_txn(
+            txn,
+            table="event_to_state_groups",
+            values=[
+                {"state_group": state_group_id, "event_id": event_id}
+                for event_id, state_group_id in iteritems(state_groups)
+            ],
+        )
+
+        for event_id, state_group_id in iteritems(state_groups):
+            txn.call_after(
+                self._get_state_group_for_event.prefill, (event_id,), state_group_id
+            )
diff --git a/synapse/storage/state_deltas.py b/synapse/storage/data_stores/main/state_deltas.py
similarity index 69%
rename from synapse/storage/state_deltas.py
rename to synapse/storage/data_stores/main/state_deltas.py
index 5fdb442104..28f33ec18f 100644
--- a/synapse/storage/state_deltas.py
+++ b/synapse/storage/data_stores/main/state_deltas.py
@@ -21,7 +21,7 @@ logger = logging.getLogger(__name__)
 
 
 class StateDeltasStore(SQLBaseStore):
-    def get_current_state_deltas(self, prev_stream_id):
+    def get_current_state_deltas(self, prev_stream_id: int, max_stream_id: int):
         """Fetch a list of room state changes since the given stream id
 
         Each entry in the result contains the following fields:
@@ -36,15 +36,27 @@ class StateDeltasStore(SQLBaseStore):
 
         Args:
             prev_stream_id (int): point to get changes since (exclusive)
+            max_stream_id (int): the point that we know has been correctly persisted
+               - ie, an upper limit to return changes from.
 
         Returns:
-            Deferred[list[dict]]: results
+            Deferred[tuple[int, list[dict]]: A tuple consisting of:
+               - the stream id which these results go up to
+               - list of current_state_delta_stream rows. If it is empty, we are
+                 up to date.
         """
         prev_stream_id = int(prev_stream_id)
+
+        # check we're not going backwards
+        assert prev_stream_id <= max_stream_id
+
         if not self._curr_state_delta_stream_cache.has_any_entity_changed(
             prev_stream_id
         ):
-            return []
+            # if the CSDs haven't changed between prev_stream_id and now, we
+            # know for certain that they haven't changed between prev_stream_id and
+            # max_stream_id.
+            return max_stream_id, []
 
         def get_current_state_deltas_txn(txn):
             # First we calculate the max stream id that will give us less than
@@ -54,21 +66,29 @@ class StateDeltasStore(SQLBaseStore):
             sql = """
                 SELECT stream_id, count(*)
                 FROM current_state_delta_stream
-                WHERE stream_id > ?
+                WHERE stream_id > ? AND stream_id <= ?
                 GROUP BY stream_id
                 ORDER BY stream_id ASC
                 LIMIT 100
             """
-            txn.execute(sql, (prev_stream_id,))
+            txn.execute(sql, (prev_stream_id, max_stream_id))
 
             total = 0
-            max_stream_id = prev_stream_id
-            for max_stream_id, count in txn:
+
+            for stream_id, count in txn:
                 total += count
                 if total > 100:
                     # We arbitarily limit to 100 entries to ensure we don't
                     # select toooo many.
+                    logger.debug(
+                        "Clipping current_state_delta_stream rows to stream_id %i",
+                        stream_id,
+                    )
+                    clipped_stream_id = stream_id
                     break
+            else:
+                # if there's no problem, we may as well go right up to the max_stream_id
+                clipped_stream_id = max_stream_id
 
             # Now actually get the deltas
             sql = """
@@ -77,8 +97,8 @@ class StateDeltasStore(SQLBaseStore):
                 WHERE ? < stream_id AND stream_id <= ?
                 ORDER BY stream_id ASC
             """
-            txn.execute(sql, (prev_stream_id, max_stream_id))
-            return self.cursor_to_dict(txn)
+            txn.execute(sql, (prev_stream_id, clipped_stream_id))
+            return clipped_stream_id, self.cursor_to_dict(txn)
 
         return self.runInteraction(
             "get_current_state_deltas", get_current_state_deltas_txn
diff --git a/synapse/storage/data_stores/main/stats.py b/synapse/storage/data_stores/main/stats.py
new file mode 100644
index 0000000000..45b3de7d56
--- /dev/null
+++ b/synapse/storage/data_stores/main/stats.py
@@ -0,0 +1,881 @@
+# -*- coding: utf-8 -*-
+# Copyright 2018, 2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from itertools import chain
+
+from twisted.internet import defer
+from twisted.internet.defer import DeferredLock
+
+from synapse.api.constants import EventTypes, Membership
+from synapse.storage.data_stores.main.state_deltas import StateDeltasStore
+from synapse.storage.engines import PostgresEngine
+from synapse.util.caches.descriptors import cached
+
+logger = logging.getLogger(__name__)
+
+# these fields track absolutes (e.g. total number of rooms on the server)
+# You can think of these as Prometheus Gauges.
+# You can draw these stats on a line graph.
+# Example: number of users in a room
+ABSOLUTE_STATS_FIELDS = {
+    "room": (
+        "current_state_events",
+        "joined_members",
+        "invited_members",
+        "left_members",
+        "banned_members",
+        "local_users_in_room",
+    ),
+    "user": ("joined_rooms",),
+}
+
+# these fields are per-timeslice and so should be reset to 0 upon a new slice
+# You can draw these stats on a histogram.
+# Example: number of events sent locally during a time slice
+PER_SLICE_FIELDS = {
+    "room": ("total_events", "total_event_bytes"),
+    "user": ("invites_sent", "rooms_created", "total_events", "total_event_bytes"),
+}
+
+TYPE_TO_TABLE = {"room": ("room_stats", "room_id"), "user": ("user_stats", "user_id")}
+
+# these are the tables (& ID columns) which contain our actual subjects
+TYPE_TO_ORIGIN_TABLE = {"room": ("rooms", "room_id"), "user": ("users", "name")}
+
+
+class StatsStore(StateDeltasStore):
+    def __init__(self, db_conn, hs):
+        super(StatsStore, self).__init__(db_conn, hs)
+
+        self.server_name = hs.hostname
+        self.clock = self.hs.get_clock()
+        self.stats_enabled = hs.config.stats_enabled
+        self.stats_bucket_size = hs.config.stats_bucket_size
+
+        self.stats_delta_processing_lock = DeferredLock()
+
+        self.register_background_update_handler(
+            "populate_stats_process_rooms", self._populate_stats_process_rooms
+        )
+        self.register_background_update_handler(
+            "populate_stats_process_users", self._populate_stats_process_users
+        )
+        # we no longer need to perform clean-up, but we will give ourselves
+        # the potential to reintroduce it in the future – so documentation
+        # will still encourage the use of this no-op handler.
+        self.register_noop_background_update("populate_stats_cleanup")
+        self.register_noop_background_update("populate_stats_prepare")
+
+    def quantise_stats_time(self, ts):
+        """
+        Quantises a timestamp to be a multiple of the bucket size.
+
+        Args:
+            ts (int): the timestamp to quantise, in milliseconds since the Unix
+                Epoch
+
+        Returns:
+            int: a timestamp which
+              - is divisible by the bucket size;
+              - is no later than `ts`; and
+              - is the largest such timestamp.
+        """
+        return (ts // self.stats_bucket_size) * self.stats_bucket_size
+
+    @defer.inlineCallbacks
+    def _populate_stats_process_users(self, progress, batch_size):
+        """
+        This is a background update which regenerates statistics for users.
+        """
+        if not self.stats_enabled:
+            yield self._end_background_update("populate_stats_process_users")
+            return 1
+
+        last_user_id = progress.get("last_user_id", "")
+
+        def _get_next_batch(txn):
+            sql = """
+                    SELECT DISTINCT name FROM users
+                    WHERE name > ?
+                    ORDER BY name ASC
+                    LIMIT ?
+                """
+            txn.execute(sql, (last_user_id, batch_size))
+            return [r for r, in txn]
+
+        users_to_work_on = yield self.runInteraction(
+            "_populate_stats_process_users", _get_next_batch
+        )
+
+        # No more rooms -- complete the transaction.
+        if not users_to_work_on:
+            yield self._end_background_update("populate_stats_process_users")
+            return 1
+
+        for user_id in users_to_work_on:
+            yield self._calculate_and_set_initial_state_for_user(user_id)
+            progress["last_user_id"] = user_id
+
+        yield self.runInteraction(
+            "populate_stats_process_users",
+            self._background_update_progress_txn,
+            "populate_stats_process_users",
+            progress,
+        )
+
+        return len(users_to_work_on)
+
+    @defer.inlineCallbacks
+    def _populate_stats_process_rooms(self, progress, batch_size):
+        """
+        This is a background update which regenerates statistics for rooms.
+        """
+        if not self.stats_enabled:
+            yield self._end_background_update("populate_stats_process_rooms")
+            return 1
+
+        last_room_id = progress.get("last_room_id", "")
+
+        def _get_next_batch(txn):
+            sql = """
+                    SELECT DISTINCT room_id FROM current_state_events
+                    WHERE room_id > ?
+                    ORDER BY room_id ASC
+                    LIMIT ?
+                """
+            txn.execute(sql, (last_room_id, batch_size))
+            return [r for r, in txn]
+
+        rooms_to_work_on = yield self.runInteraction(
+            "populate_stats_rooms_get_batch", _get_next_batch
+        )
+
+        # No more rooms -- complete the transaction.
+        if not rooms_to_work_on:
+            yield self._end_background_update("populate_stats_process_rooms")
+            return 1
+
+        for room_id in rooms_to_work_on:
+            yield self._calculate_and_set_initial_state_for_room(room_id)
+            progress["last_room_id"] = room_id
+
+        yield self.runInteraction(
+            "_populate_stats_process_rooms",
+            self._background_update_progress_txn,
+            "populate_stats_process_rooms",
+            progress,
+        )
+
+        return len(rooms_to_work_on)
+
+    def get_stats_positions(self):
+        """
+        Returns the stats processor positions.
+        """
+        return self._simple_select_one_onecol(
+            table="stats_incremental_position",
+            keyvalues={},
+            retcol="stream_id",
+            desc="stats_incremental_position",
+        )
+
+    def update_room_state(self, room_id, fields):
+        """
+        Args:
+            room_id (str)
+            fields (dict[str:Any])
+        """
+
+        # For whatever reason some of the fields may contain null bytes, which
+        # postgres isn't a fan of, so we replace those fields with null.
+        for col in (
+            "join_rules",
+            "history_visibility",
+            "encryption",
+            "name",
+            "topic",
+            "avatar",
+            "canonical_alias",
+        ):
+            field = fields.get(col)
+            if field and "\0" in field:
+                fields[col] = None
+
+        return self._simple_upsert(
+            table="room_stats_state",
+            keyvalues={"room_id": room_id},
+            values=fields,
+            desc="update_room_state",
+        )
+
+    def get_statistics_for_subject(self, stats_type, stats_id, start, size=100):
+        """
+        Get statistics for a given subject.
+
+        Args:
+            stats_type (str): The type of subject
+            stats_id (str): The ID of the subject (e.g. room_id or user_id)
+            start (int): Pagination start. Number of entries, not timestamp.
+            size (int): How many entries to return.
+
+        Returns:
+            Deferred[list[dict]], where the dict has the keys of
+            ABSOLUTE_STATS_FIELDS[stats_type],  and "bucket_size" and "end_ts".
+        """
+        return self.runInteraction(
+            "get_statistics_for_subject",
+            self._get_statistics_for_subject_txn,
+            stats_type,
+            stats_id,
+            start,
+            size,
+        )
+
+    def _get_statistics_for_subject_txn(
+        self, txn, stats_type, stats_id, start, size=100
+    ):
+        """
+        Transaction-bound version of L{get_statistics_for_subject}.
+        """
+
+        table, id_col = TYPE_TO_TABLE[stats_type]
+        selected_columns = list(
+            ABSOLUTE_STATS_FIELDS[stats_type] + PER_SLICE_FIELDS[stats_type]
+        )
+
+        slice_list = self._simple_select_list_paginate_txn(
+            txn,
+            table + "_historical",
+            {id_col: stats_id},
+            "end_ts",
+            start,
+            size,
+            retcols=selected_columns + ["bucket_size", "end_ts"],
+            order_direction="DESC",
+        )
+
+        return slice_list
+
+    def get_room_stats_state(self, room_id):
+        """
+        Returns the current room_stats_state for a room.
+
+        Args:
+            room_id (str): The ID of the room to return state for.
+
+        Returns (dict):
+            Dictionary containing these keys:
+                "name", "topic", "canonical_alias", "avatar", "join_rules",
+                "history_visibility"
+        """
+        return self._simple_select_one(
+            "room_stats_state",
+            {"room_id": room_id},
+            retcols=(
+                "name",
+                "topic",
+                "canonical_alias",
+                "avatar",
+                "join_rules",
+                "history_visibility",
+            ),
+        )
+
+    @cached()
+    def get_earliest_token_for_stats(self, stats_type, id):
+        """
+        Fetch the "earliest token". This is used by the room stats delta
+        processor to ignore deltas that have been processed between the
+        start of the background task and any particular room's stats
+        being calculated.
+
+        Returns:
+            Deferred[int]
+        """
+        table, id_col = TYPE_TO_TABLE[stats_type]
+
+        return self._simple_select_one_onecol(
+            "%s_current" % (table,),
+            keyvalues={id_col: id},
+            retcol="completed_delta_stream_id",
+            allow_none=True,
+        )
+
+    def bulk_update_stats_delta(self, ts, updates, stream_id):
+        """Bulk update stats tables for a given stream_id and updates the stats
+        incremental position.
+
+        Args:
+            ts (int): Current timestamp in ms
+            updates(dict[str, dict[str, dict[str, Counter]]]): The updates to
+                commit as a mapping stats_type -> stats_id -> field -> delta.
+            stream_id (int): Current position.
+
+        Returns:
+            Deferred
+        """
+
+        def _bulk_update_stats_delta_txn(txn):
+            for stats_type, stats_updates in updates.items():
+                for stats_id, fields in stats_updates.items():
+                    logger.debug(
+                        "Updating %s stats for %s: %s", stats_type, stats_id, fields
+                    )
+                    self._update_stats_delta_txn(
+                        txn,
+                        ts=ts,
+                        stats_type=stats_type,
+                        stats_id=stats_id,
+                        fields=fields,
+                        complete_with_stream_id=stream_id,
+                    )
+
+            self._simple_update_one_txn(
+                txn,
+                table="stats_incremental_position",
+                keyvalues={},
+                updatevalues={"stream_id": stream_id},
+            )
+
+        return self.runInteraction(
+            "bulk_update_stats_delta", _bulk_update_stats_delta_txn
+        )
+
+    def update_stats_delta(
+        self,
+        ts,
+        stats_type,
+        stats_id,
+        fields,
+        complete_with_stream_id,
+        absolute_field_overrides=None,
+    ):
+        """
+        Updates the statistics for a subject, with a delta (difference/relative
+        change).
+
+        Args:
+            ts (int): timestamp of the change
+            stats_type (str): "room" or "user" – the kind of subject
+            stats_id (str): the subject's ID (room ID or user ID)
+            fields (dict[str, int]): Deltas of stats values.
+            complete_with_stream_id (int, optional):
+                If supplied, converts an incomplete row into a complete row,
+                with the supplied stream_id marked as the stream_id where the
+                row was completed.
+            absolute_field_overrides (dict[str, int]): Current stats values
+                (i.e. not deltas) of absolute fields.
+                Does not work with per-slice fields.
+        """
+
+        return self.runInteraction(
+            "update_stats_delta",
+            self._update_stats_delta_txn,
+            ts,
+            stats_type,
+            stats_id,
+            fields,
+            complete_with_stream_id=complete_with_stream_id,
+            absolute_field_overrides=absolute_field_overrides,
+        )
+
+    def _update_stats_delta_txn(
+        self,
+        txn,
+        ts,
+        stats_type,
+        stats_id,
+        fields,
+        complete_with_stream_id,
+        absolute_field_overrides=None,
+    ):
+        if absolute_field_overrides is None:
+            absolute_field_overrides = {}
+
+        table, id_col = TYPE_TO_TABLE[stats_type]
+
+        quantised_ts = self.quantise_stats_time(int(ts))
+        end_ts = quantised_ts + self.stats_bucket_size
+
+        # Lets be paranoid and check that all the given field names are known
+        abs_field_names = ABSOLUTE_STATS_FIELDS[stats_type]
+        slice_field_names = PER_SLICE_FIELDS[stats_type]
+        for field in chain(fields.keys(), absolute_field_overrides.keys()):
+            if field not in abs_field_names and field not in slice_field_names:
+                # guard against potential SQL injection dodginess
+                raise ValueError(
+                    "%s is not a recognised field"
+                    " for stats type %s" % (field, stats_type)
+                )
+
+        # Per slice fields do not get added to the _current table
+
+        # This calculates the deltas (`field = field + ?` values)
+        # for absolute fields,
+        # * defaulting to 0 if not specified
+        #     (required for the INSERT part of upserting to work)
+        # * omitting overrides specified in `absolute_field_overrides`
+        deltas_of_absolute_fields = {
+            key: fields.get(key, 0)
+            for key in abs_field_names
+            if key not in absolute_field_overrides
+        }
+
+        # Keep the delta stream ID field up to date
+        absolute_field_overrides = absolute_field_overrides.copy()
+        absolute_field_overrides["completed_delta_stream_id"] = complete_with_stream_id
+
+        # first upsert the `_current` table
+        self._upsert_with_additive_relatives_txn(
+            txn=txn,
+            table=table + "_current",
+            keyvalues={id_col: stats_id},
+            absolutes=absolute_field_overrides,
+            additive_relatives=deltas_of_absolute_fields,
+        )
+
+        per_slice_additive_relatives = {
+            key: fields.get(key, 0) for key in slice_field_names
+        }
+        self._upsert_copy_from_table_with_additive_relatives_txn(
+            txn=txn,
+            into_table=table + "_historical",
+            keyvalues={id_col: stats_id},
+            extra_dst_insvalues={"bucket_size": self.stats_bucket_size},
+            extra_dst_keyvalues={"end_ts": end_ts},
+            additive_relatives=per_slice_additive_relatives,
+            src_table=table + "_current",
+            copy_columns=abs_field_names,
+        )
+
+    def _upsert_with_additive_relatives_txn(
+        self, txn, table, keyvalues, absolutes, additive_relatives
+    ):
+        """Used to update values in the stats tables.
+
+        This is basically a slightly convoluted upsert that *adds* to any
+        existing rows.
+
+        Args:
+            txn
+            table (str): Table name
+            keyvalues (dict[str, any]): Row-identifying key values
+            absolutes (dict[str, any]): Absolute (set) fields
+            additive_relatives (dict[str, int]): Fields that will be added onto
+                if existing row present.
+        """
+        if self.database_engine.can_native_upsert:
+            absolute_updates = [
+                "%(field)s = EXCLUDED.%(field)s" % {"field": field}
+                for field in absolutes.keys()
+            ]
+
+            relative_updates = [
+                "%(field)s = EXCLUDED.%(field)s + %(table)s.%(field)s"
+                % {"table": table, "field": field}
+                for field in additive_relatives.keys()
+            ]
+
+            insert_cols = []
+            qargs = []
+
+            for (key, val) in chain(
+                keyvalues.items(), absolutes.items(), additive_relatives.items()
+            ):
+                insert_cols.append(key)
+                qargs.append(val)
+
+            sql = """
+                INSERT INTO %(table)s (%(insert_cols_cs)s)
+                VALUES (%(insert_vals_qs)s)
+                ON CONFLICT (%(key_columns)s) DO UPDATE SET %(updates)s
+            """ % {
+                "table": table,
+                "insert_cols_cs": ", ".join(insert_cols),
+                "insert_vals_qs": ", ".join(
+                    ["?"] * (len(keyvalues) + len(absolutes) + len(additive_relatives))
+                ),
+                "key_columns": ", ".join(keyvalues),
+                "updates": ", ".join(chain(absolute_updates, relative_updates)),
+            }
+
+            txn.execute(sql, qargs)
+        else:
+            self.database_engine.lock_table(txn, table)
+            retcols = list(chain(absolutes.keys(), additive_relatives.keys()))
+            current_row = self._simple_select_one_txn(
+                txn, table, keyvalues, retcols, allow_none=True
+            )
+            if current_row is None:
+                merged_dict = {**keyvalues, **absolutes, **additive_relatives}
+                self._simple_insert_txn(txn, table, merged_dict)
+            else:
+                for (key, val) in additive_relatives.items():
+                    current_row[key] += val
+                current_row.update(absolutes)
+                self._simple_update_one_txn(txn, table, keyvalues, current_row)
+
+    def _upsert_copy_from_table_with_additive_relatives_txn(
+        self,
+        txn,
+        into_table,
+        keyvalues,
+        extra_dst_keyvalues,
+        extra_dst_insvalues,
+        additive_relatives,
+        src_table,
+        copy_columns,
+    ):
+        """Updates the historic stats table with latest updates.
+
+        This involves copying "absolute" fields from the `_current` table, and
+        adding relative fields to any existing values.
+
+        Args:
+             txn: Transaction
+             into_table (str): The destination table to UPSERT the row into
+             keyvalues (dict[str, any]): Row-identifying key values
+             extra_dst_keyvalues (dict[str, any]): Additional keyvalues
+                for `into_table`.
+             extra_dst_insvalues (dict[str, any]): Additional values to insert
+                on new row creation for `into_table`.
+             additive_relatives (dict[str, any]): Fields that will be added onto
+                if existing row present. (Must be disjoint from copy_columns.)
+             src_table (str): The source table to copy from
+             copy_columns (iterable[str]): The list of columns to copy
+        """
+        if self.database_engine.can_native_upsert:
+            ins_columns = chain(
+                keyvalues,
+                copy_columns,
+                additive_relatives,
+                extra_dst_keyvalues,
+                extra_dst_insvalues,
+            )
+            sel_exprs = chain(
+                keyvalues,
+                copy_columns,
+                (
+                    "?"
+                    for _ in chain(
+                        additive_relatives, extra_dst_keyvalues, extra_dst_insvalues
+                    )
+                ),
+            )
+            keyvalues_where = ("%s = ?" % f for f in keyvalues)
+
+            sets_cc = ("%s = EXCLUDED.%s" % (f, f) for f in copy_columns)
+            sets_ar = (
+                "%s = EXCLUDED.%s + %s.%s" % (f, f, into_table, f)
+                for f in additive_relatives
+            )
+
+            sql = """
+                INSERT INTO %(into_table)s (%(ins_columns)s)
+                SELECT %(sel_exprs)s
+                FROM %(src_table)s
+                WHERE %(keyvalues_where)s
+                ON CONFLICT (%(keyvalues)s)
+                DO UPDATE SET %(sets)s
+            """ % {
+                "into_table": into_table,
+                "ins_columns": ", ".join(ins_columns),
+                "sel_exprs": ", ".join(sel_exprs),
+                "keyvalues_where": " AND ".join(keyvalues_where),
+                "src_table": src_table,
+                "keyvalues": ", ".join(
+                    chain(keyvalues.keys(), extra_dst_keyvalues.keys())
+                ),
+                "sets": ", ".join(chain(sets_cc, sets_ar)),
+            }
+
+            qargs = list(
+                chain(
+                    additive_relatives.values(),
+                    extra_dst_keyvalues.values(),
+                    extra_dst_insvalues.values(),
+                    keyvalues.values(),
+                )
+            )
+            txn.execute(sql, qargs)
+        else:
+            self.database_engine.lock_table(txn, into_table)
+            src_row = self._simple_select_one_txn(
+                txn, src_table, keyvalues, copy_columns
+            )
+            all_dest_keyvalues = {**keyvalues, **extra_dst_keyvalues}
+            dest_current_row = self._simple_select_one_txn(
+                txn,
+                into_table,
+                keyvalues=all_dest_keyvalues,
+                retcols=list(chain(additive_relatives.keys(), copy_columns)),
+                allow_none=True,
+            )
+
+            if dest_current_row is None:
+                merged_dict = {
+                    **keyvalues,
+                    **extra_dst_keyvalues,
+                    **extra_dst_insvalues,
+                    **src_row,
+                    **additive_relatives,
+                }
+                self._simple_insert_txn(txn, into_table, merged_dict)
+            else:
+                for (key, val) in additive_relatives.items():
+                    src_row[key] = dest_current_row[key] + val
+                self._simple_update_txn(txn, into_table, all_dest_keyvalues, src_row)
+
+    def get_changes_room_total_events_and_bytes(self, min_pos, max_pos):
+        """Fetches the counts of events in the given range of stream IDs.
+
+        Args:
+            min_pos (int)
+            max_pos (int)
+
+        Returns:
+            Deferred[dict[str, dict[str, int]]]: Mapping of room ID to field
+            changes.
+        """
+
+        return self.runInteraction(
+            "stats_incremental_total_events_and_bytes",
+            self.get_changes_room_total_events_and_bytes_txn,
+            min_pos,
+            max_pos,
+        )
+
+    def get_changes_room_total_events_and_bytes_txn(self, txn, low_pos, high_pos):
+        """Gets the total_events and total_event_bytes counts for rooms and
+        senders, in a range of stream_orderings (including backfilled events).
+
+        Args:
+            txn
+            low_pos (int): Low stream ordering
+            high_pos (int): High stream ordering
+
+        Returns:
+            tuple[dict[str, dict[str, int]], dict[str, dict[str, int]]]: The
+            room and user deltas for total_events/total_event_bytes in the
+            format of `stats_id` -> fields
+        """
+
+        if low_pos >= high_pos:
+            # nothing to do here.
+            return {}, {}
+
+        if isinstance(self.database_engine, PostgresEngine):
+            new_bytes_expression = "OCTET_LENGTH(json)"
+        else:
+            new_bytes_expression = "LENGTH(CAST(json AS BLOB))"
+
+        sql = """
+            SELECT events.room_id, COUNT(*) AS new_events, SUM(%s) AS new_bytes
+            FROM events INNER JOIN event_json USING (event_id)
+            WHERE (? < stream_ordering AND stream_ordering <= ?)
+                OR (? <= stream_ordering AND stream_ordering <= ?)
+            GROUP BY events.room_id
+        """ % (
+            new_bytes_expression,
+        )
+
+        txn.execute(sql, (low_pos, high_pos, -high_pos, -low_pos))
+
+        room_deltas = {
+            room_id: {"total_events": new_events, "total_event_bytes": new_bytes}
+            for room_id, new_events, new_bytes in txn
+        }
+
+        sql = """
+            SELECT events.sender, COUNT(*) AS new_events, SUM(%s) AS new_bytes
+            FROM events INNER JOIN event_json USING (event_id)
+            WHERE (? < stream_ordering AND stream_ordering <= ?)
+                OR (? <= stream_ordering AND stream_ordering <= ?)
+            GROUP BY events.sender
+        """ % (
+            new_bytes_expression,
+        )
+
+        txn.execute(sql, (low_pos, high_pos, -high_pos, -low_pos))
+
+        user_deltas = {
+            user_id: {"total_events": new_events, "total_event_bytes": new_bytes}
+            for user_id, new_events, new_bytes in txn
+            if self.hs.is_mine_id(user_id)
+        }
+
+        return room_deltas, user_deltas
+
+    @defer.inlineCallbacks
+    def _calculate_and_set_initial_state_for_room(self, room_id):
+        """Calculate and insert an entry into room_stats_current.
+
+        Args:
+            room_id (str)
+
+        Returns:
+            Deferred[tuple[dict, dict, int]]: A tuple of room state, membership
+            counts and stream position.
+        """
+
+        def _fetch_current_state_stats(txn):
+            pos = self.get_room_max_stream_ordering()
+
+            rows = self._simple_select_many_txn(
+                txn,
+                table="current_state_events",
+                column="type",
+                iterable=[
+                    EventTypes.Create,
+                    EventTypes.JoinRules,
+                    EventTypes.RoomHistoryVisibility,
+                    EventTypes.Encryption,
+                    EventTypes.Name,
+                    EventTypes.Topic,
+                    EventTypes.RoomAvatar,
+                    EventTypes.CanonicalAlias,
+                ],
+                keyvalues={"room_id": room_id, "state_key": ""},
+                retcols=["event_id"],
+            )
+
+            event_ids = [row["event_id"] for row in rows]
+
+            txn.execute(
+                """
+                    SELECT membership, count(*) FROM current_state_events
+                    WHERE room_id = ? AND type = 'm.room.member'
+                    GROUP BY membership
+                """,
+                (room_id,),
+            )
+            membership_counts = {membership: cnt for membership, cnt in txn}
+
+            txn.execute(
+                """
+                    SELECT COALESCE(count(*), 0) FROM current_state_events
+                    WHERE room_id = ?
+                """,
+                (room_id,),
+            )
+
+            (current_state_events_count,) = txn.fetchone()
+
+            users_in_room = self.get_users_in_room_txn(txn, room_id)
+
+            return (
+                event_ids,
+                membership_counts,
+                current_state_events_count,
+                users_in_room,
+                pos,
+            )
+
+        (
+            event_ids,
+            membership_counts,
+            current_state_events_count,
+            users_in_room,
+            pos,
+        ) = yield self.runInteraction(
+            "get_initial_state_for_room", _fetch_current_state_stats
+        )
+
+        state_event_map = yield self.get_events(event_ids, get_prev_content=False)
+
+        room_state = {
+            "join_rules": None,
+            "history_visibility": None,
+            "encryption": None,
+            "name": None,
+            "topic": None,
+            "avatar": None,
+            "canonical_alias": None,
+            "is_federatable": True,
+        }
+
+        for event in state_event_map.values():
+            if event.type == EventTypes.JoinRules:
+                room_state["join_rules"] = event.content.get("join_rule")
+            elif event.type == EventTypes.RoomHistoryVisibility:
+                room_state["history_visibility"] = event.content.get(
+                    "history_visibility"
+                )
+            elif event.type == EventTypes.Encryption:
+                room_state["encryption"] = event.content.get("algorithm")
+            elif event.type == EventTypes.Name:
+                room_state["name"] = event.content.get("name")
+            elif event.type == EventTypes.Topic:
+                room_state["topic"] = event.content.get("topic")
+            elif event.type == EventTypes.RoomAvatar:
+                room_state["avatar"] = event.content.get("url")
+            elif event.type == EventTypes.CanonicalAlias:
+                room_state["canonical_alias"] = event.content.get("alias")
+            elif event.type == EventTypes.Create:
+                room_state["is_federatable"] = (
+                    event.content.get("m.federate", True) is True
+                )
+
+        yield self.update_room_state(room_id, room_state)
+
+        local_users_in_room = [u for u in users_in_room if self.hs.is_mine_id(u)]
+
+        yield self.update_stats_delta(
+            ts=self.clock.time_msec(),
+            stats_type="room",
+            stats_id=room_id,
+            fields={},
+            complete_with_stream_id=pos,
+            absolute_field_overrides={
+                "current_state_events": current_state_events_count,
+                "joined_members": membership_counts.get(Membership.JOIN, 0),
+                "invited_members": membership_counts.get(Membership.INVITE, 0),
+                "left_members": membership_counts.get(Membership.LEAVE, 0),
+                "banned_members": membership_counts.get(Membership.BAN, 0),
+                "local_users_in_room": len(local_users_in_room),
+            },
+        )
+
+    @defer.inlineCallbacks
+    def _calculate_and_set_initial_state_for_user(self, user_id):
+        def _calculate_and_set_initial_state_for_user_txn(txn):
+            pos = self._get_max_stream_id_in_current_state_deltas_txn(txn)
+
+            txn.execute(
+                """
+                SELECT COUNT(distinct room_id) FROM current_state_events
+                    WHERE type = 'm.room.member' AND state_key = ?
+                        AND membership = 'join'
+                """,
+                (user_id,),
+            )
+            (count,) = txn.fetchone()
+            return count, pos
+
+        joined_rooms, pos = yield self.runInteraction(
+            "calculate_and_set_initial_state_for_user",
+            _calculate_and_set_initial_state_for_user_txn,
+        )
+
+        yield self.update_stats_delta(
+            ts=self.clock.time_msec(),
+            stats_type="user",
+            stats_id=user_id,
+            fields={},
+            complete_with_stream_id=pos,
+            absolute_field_overrides={"joined_rooms": joined_rooms},
+        )
diff --git a/synapse/storage/stream.py b/synapse/storage/data_stores/main/stream.py
similarity index 97%
rename from synapse/storage/stream.py
rename to synapse/storage/data_stores/main/stream.py
index 490454f19a..616ef91d4e 100644
--- a/synapse/storage/stream.py
+++ b/synapse/storage/data_stores/main/stream.py
@@ -43,8 +43,8 @@ from twisted.internet import defer
 
 from synapse.logging.context import make_deferred_yieldable, run_in_background
 from synapse.storage._base import SQLBaseStore
+from synapse.storage.data_stores.main.events_worker import EventsWorkerStore
 from synapse.storage.engines import PostgresEngine
-from synapse.storage.events_worker import EventsWorkerStore
 from synapse.types import RoomStreamToken
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 
@@ -229,6 +229,14 @@ def filter_to_clause(event_filter):
         clauses.append("contains_url = ?")
         args.append(event_filter.contains_url)
 
+    # We're only applying the "labels" filter on the database query, because applying the
+    # "not_labels" filter via a SQL query is non-trivial. Instead, we let
+    # event_filter.check_fields apply it, which is not as efficient but makes the
+    # implementation simpler.
+    if event_filter.labels:
+        clauses.append("(%s)" % " OR ".join("label = ?" for _ in event_filter.labels))
+        args.extend(event_filter.labels)
+
     return " AND ".join(clauses), args
 
 
@@ -864,8 +872,9 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore):
         args.append(int(limit))
 
         sql = (
-            "SELECT event_id, topological_ordering, stream_ordering"
+            "SELECT DISTINCT event_id, topological_ordering, stream_ordering"
             " FROM events"
+            " LEFT JOIN event_labels USING (event_id, room_id, topological_ordering)"
             " WHERE outlier = ? AND room_id = ? AND %(bounds)s"
             " ORDER BY topological_ordering %(order)s,"
             " stream_ordering %(order)s LIMIT ?"
diff --git a/synapse/storage/tags.py b/synapse/storage/data_stores/main/tags.py
similarity index 99%
rename from synapse/storage/tags.py
rename to synapse/storage/data_stores/main/tags.py
index 20dd6bd53d..10d1887f75 100644
--- a/synapse/storage/tags.py
+++ b/synapse/storage/data_stores/main/tags.py
@@ -22,7 +22,7 @@ from canonicaljson import json
 
 from twisted.internet import defer
 
-from synapse.storage.account_data import AccountDataWorkerStore
+from synapse.storage.data_stores.main.account_data import AccountDataWorkerStore
 from synapse.util.caches.descriptors import cached
 
 logger = logging.getLogger(__name__)
diff --git a/synapse/storage/transactions.py b/synapse/storage/data_stores/main/transactions.py
similarity index 87%
rename from synapse/storage/transactions.py
rename to synapse/storage/data_stores/main/transactions.py
index b3c3bf55bc..01b1be5e14 100644
--- a/synapse/storage/transactions.py
+++ b/synapse/storage/data_stores/main/transactions.py
@@ -23,10 +23,9 @@ from canonicaljson import encode_canonical_json
 from twisted.internet import defer
 
 from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.storage._base import SQLBaseStore, db_to_json
 from synapse.util.caches.expiringcache import ExpiringCache
 
-from ._base import SQLBaseStore, db_to_json
-
 # py2 sqlite has buffer hardcoded as only binary type, so we must use it,
 # despite being deprecated and removed in favor of memoryview
 if six.PY2:
@@ -165,7 +164,7 @@ class TransactionStore(SQLBaseStore):
             txn,
             table="destinations",
             keyvalues={"destination": destination},
-            retcols=("destination", "retry_last_ts", "retry_interval"),
+            retcols=("destination", "failure_ts", "retry_last_ts", "retry_interval"),
             allow_none=True,
         )
 
@@ -174,12 +173,15 @@ class TransactionStore(SQLBaseStore):
         else:
             return None
 
-    def set_destination_retry_timings(self, destination, retry_last_ts, retry_interval):
+    def set_destination_retry_timings(
+        self, destination, failure_ts, retry_last_ts, retry_interval
+    ):
         """Sets the current retry timings for a given destination.
         Both timings should be zero if retrying is no longer occuring.
 
         Args:
             destination (str)
+            failure_ts (int|None) - when the server started failing (ms since epoch)
             retry_last_ts (int) - time of last retry attempt in unix epoch ms
             retry_interval (int) - how long until next retry in ms
         """
@@ -189,12 +191,13 @@ class TransactionStore(SQLBaseStore):
             "set_destination_retry_timings",
             self._set_destination_retry_timings,
             destination,
+            failure_ts,
             retry_last_ts,
             retry_interval,
         )
 
     def _set_destination_retry_timings(
-        self, txn, destination, retry_last_ts, retry_interval
+        self, txn, destination, failure_ts, retry_last_ts, retry_interval
     ):
 
         if self.database_engine.can_native_upsert:
@@ -202,9 +205,12 @@ class TransactionStore(SQLBaseStore):
             # resetting it) or greater than the existing retry interval.
 
             sql = """
-                INSERT INTO destinations (destination, retry_last_ts, retry_interval)
-                    VALUES (?, ?, ?)
+                INSERT INTO destinations (
+                    destination, failure_ts, retry_last_ts, retry_interval
+                )
+                    VALUES (?, ?, ?, ?)
                 ON CONFLICT (destination) DO UPDATE SET
+                        failure_ts = EXCLUDED.failure_ts,
                         retry_last_ts = EXCLUDED.retry_last_ts,
                         retry_interval = EXCLUDED.retry_interval
                     WHERE
@@ -212,7 +218,7 @@ class TransactionStore(SQLBaseStore):
                         OR destinations.retry_interval < EXCLUDED.retry_interval
             """
 
-            txn.execute(sql, (destination, retry_last_ts, retry_interval))
+            txn.execute(sql, (destination, failure_ts, retry_last_ts, retry_interval))
 
             return
 
@@ -225,7 +231,7 @@ class TransactionStore(SQLBaseStore):
             txn,
             table="destinations",
             keyvalues={"destination": destination},
-            retcols=("retry_last_ts", "retry_interval"),
+            retcols=("failure_ts", "retry_last_ts", "retry_interval"),
             allow_none=True,
         )
 
@@ -235,6 +241,7 @@ class TransactionStore(SQLBaseStore):
                 table="destinations",
                 values={
                     "destination": destination,
+                    "failure_ts": failure_ts,
                     "retry_last_ts": retry_last_ts,
                     "retry_interval": retry_interval,
                 },
@@ -245,31 +252,12 @@ class TransactionStore(SQLBaseStore):
                 "destinations",
                 keyvalues={"destination": destination},
                 updatevalues={
+                    "failure_ts": failure_ts,
                     "retry_last_ts": retry_last_ts,
                     "retry_interval": retry_interval,
                 },
             )
 
-    def get_destinations_needing_retry(self):
-        """Get all destinations which are due a retry for sending a transaction.
-
-        Returns:
-            list: A list of dicts
-        """
-
-        return self.runInteraction(
-            "get_destinations_needing_retry", self._get_destinations_needing_retry
-        )
-
-    def _get_destinations_needing_retry(self, txn):
-        query = (
-            "SELECT * FROM destinations"
-            " WHERE retry_last_ts > 0 and retry_next_ts < ?"
-        )
-
-        txn.execute(query, (self._clock.time_msec(),))
-        return self.cursor_to_dict(txn)
-
     def _start_cleanup_transactions(self):
         return run_as_background_process(
             "cleanup_transactions", self._cleanup_transactions
diff --git a/synapse/storage/user_directory.py b/synapse/storage/data_stores/main/user_directory.py
similarity index 98%
rename from synapse/storage/user_directory.py
rename to synapse/storage/data_stores/main/user_directory.py
index b5188d9bee..652abe0e6a 100644
--- a/synapse/storage/user_directory.py
+++ b/synapse/storage/data_stores/main/user_directory.py
@@ -20,9 +20,9 @@ from twisted.internet import defer
 
 from synapse.api.constants import EventTypes, JoinRules
 from synapse.storage.background_updates import BackgroundUpdateStore
+from synapse.storage.data_stores.main.state import StateFilter
+from synapse.storage.data_stores.main.state_deltas import StateDeltasStore
 from synapse.storage.engines import PostgresEngine, Sqlite3Engine
-from synapse.storage.state import StateFilter
-from synapse.storage.state_deltas import StateDeltasStore
 from synapse.types import get_domain_from_id, get_localpart_from_id
 from synapse.util.caches.descriptors import cached
 
@@ -32,14 +32,14 @@ logger = logging.getLogger(__name__)
 TEMP_TABLE = "_temp_populate_user_directory"
 
 
-class UserDirectoryStore(StateDeltasStore, BackgroundUpdateStore):
+class UserDirectoryBackgroundUpdateStore(StateDeltasStore, BackgroundUpdateStore):
 
     # How many records do we calculate before sending it to
     # add_users_who_share_private_rooms?
     SHARE_PRIVATE_WORKING_SET = 500
 
     def __init__(self, db_conn, hs):
-        super(UserDirectoryStore, self).__init__(db_conn, hs)
+        super(UserDirectoryBackgroundUpdateStore, self).__init__(db_conn, hs)
 
         self.server_name = hs.hostname
 
@@ -452,55 +452,6 @@ class UserDirectoryStore(StateDeltasStore, BackgroundUpdateStore):
             "update_profile_in_user_dir", _update_profile_in_user_dir_txn
         )
 
-    def remove_from_user_dir(self, user_id):
-        def _remove_from_user_dir_txn(txn):
-            self._simple_delete_txn(
-                txn, table="user_directory", keyvalues={"user_id": user_id}
-            )
-            self._simple_delete_txn(
-                txn, table="user_directory_search", keyvalues={"user_id": user_id}
-            )
-            self._simple_delete_txn(
-                txn, table="users_in_public_rooms", keyvalues={"user_id": user_id}
-            )
-            self._simple_delete_txn(
-                txn,
-                table="users_who_share_private_rooms",
-                keyvalues={"user_id": user_id},
-            )
-            self._simple_delete_txn(
-                txn,
-                table="users_who_share_private_rooms",
-                keyvalues={"other_user_id": user_id},
-            )
-            txn.call_after(self.get_user_in_directory.invalidate, (user_id,))
-
-        return self.runInteraction("remove_from_user_dir", _remove_from_user_dir_txn)
-
-    @defer.inlineCallbacks
-    def get_users_in_dir_due_to_room(self, room_id):
-        """Get all user_ids that are in the room directory because they're
-        in the given room_id
-        """
-        user_ids_share_pub = yield self._simple_select_onecol(
-            table="users_in_public_rooms",
-            keyvalues={"room_id": room_id},
-            retcol="user_id",
-            desc="get_users_in_dir_due_to_room",
-        )
-
-        user_ids_share_priv = yield self._simple_select_onecol(
-            table="users_who_share_private_rooms",
-            keyvalues={"room_id": room_id},
-            retcol="other_user_id",
-            desc="get_users_in_dir_due_to_room",
-        )
-
-        user_ids = set(user_ids_share_pub)
-        user_ids.update(user_ids_share_priv)
-
-        return user_ids
-
     def add_users_who_share_private_room(self, room_id, user_id_tuples):
         """Insert entries into the users_who_share_private_rooms table. The first
         user should be a local user.
@@ -551,6 +502,98 @@ class UserDirectoryStore(StateDeltasStore, BackgroundUpdateStore):
             "add_users_in_public_rooms", _add_users_in_public_rooms_txn
         )
 
+    def delete_all_from_user_dir(self):
+        """Delete the entire user directory
+        """
+
+        def _delete_all_from_user_dir_txn(txn):
+            txn.execute("DELETE FROM user_directory")
+            txn.execute("DELETE FROM user_directory_search")
+            txn.execute("DELETE FROM users_in_public_rooms")
+            txn.execute("DELETE FROM users_who_share_private_rooms")
+            txn.call_after(self.get_user_in_directory.invalidate_all)
+
+        return self.runInteraction(
+            "delete_all_from_user_dir", _delete_all_from_user_dir_txn
+        )
+
+    @cached()
+    def get_user_in_directory(self, user_id):
+        return self._simple_select_one(
+            table="user_directory",
+            keyvalues={"user_id": user_id},
+            retcols=("display_name", "avatar_url"),
+            allow_none=True,
+            desc="get_user_in_directory",
+        )
+
+    def update_user_directory_stream_pos(self, stream_id):
+        return self._simple_update_one(
+            table="user_directory_stream_pos",
+            keyvalues={},
+            updatevalues={"stream_id": stream_id},
+            desc="update_user_directory_stream_pos",
+        )
+
+
+class UserDirectoryStore(UserDirectoryBackgroundUpdateStore):
+
+    # How many records do we calculate before sending it to
+    # add_users_who_share_private_rooms?
+    SHARE_PRIVATE_WORKING_SET = 500
+
+    def __init__(self, db_conn, hs):
+        super(UserDirectoryStore, self).__init__(db_conn, hs)
+
+    def remove_from_user_dir(self, user_id):
+        def _remove_from_user_dir_txn(txn):
+            self._simple_delete_txn(
+                txn, table="user_directory", keyvalues={"user_id": user_id}
+            )
+            self._simple_delete_txn(
+                txn, table="user_directory_search", keyvalues={"user_id": user_id}
+            )
+            self._simple_delete_txn(
+                txn, table="users_in_public_rooms", keyvalues={"user_id": user_id}
+            )
+            self._simple_delete_txn(
+                txn,
+                table="users_who_share_private_rooms",
+                keyvalues={"user_id": user_id},
+            )
+            self._simple_delete_txn(
+                txn,
+                table="users_who_share_private_rooms",
+                keyvalues={"other_user_id": user_id},
+            )
+            txn.call_after(self.get_user_in_directory.invalidate, (user_id,))
+
+        return self.runInteraction("remove_from_user_dir", _remove_from_user_dir_txn)
+
+    @defer.inlineCallbacks
+    def get_users_in_dir_due_to_room(self, room_id):
+        """Get all user_ids that are in the room directory because they're
+        in the given room_id
+        """
+        user_ids_share_pub = yield self._simple_select_onecol(
+            table="users_in_public_rooms",
+            keyvalues={"room_id": room_id},
+            retcol="user_id",
+            desc="get_users_in_dir_due_to_room",
+        )
+
+        user_ids_share_priv = yield self._simple_select_onecol(
+            table="users_who_share_private_rooms",
+            keyvalues={"room_id": room_id},
+            retcol="other_user_id",
+            desc="get_users_in_dir_due_to_room",
+        )
+
+        user_ids = set(user_ids_share_pub)
+        user_ids.update(user_ids_share_priv)
+
+        return user_ids
+
     def remove_user_who_share_room(self, user_id, room_id):
         """
         Deletes entries in the users_who_share_*_rooms table. The first
@@ -637,31 +680,6 @@ class UserDirectoryStore(StateDeltasStore, BackgroundUpdateStore):
 
         return [room_id for room_id, in rows]
 
-    def delete_all_from_user_dir(self):
-        """Delete the entire user directory
-        """
-
-        def _delete_all_from_user_dir_txn(txn):
-            txn.execute("DELETE FROM user_directory")
-            txn.execute("DELETE FROM user_directory_search")
-            txn.execute("DELETE FROM users_in_public_rooms")
-            txn.execute("DELETE FROM users_who_share_private_rooms")
-            txn.call_after(self.get_user_in_directory.invalidate_all)
-
-        return self.runInteraction(
-            "delete_all_from_user_dir", _delete_all_from_user_dir_txn
-        )
-
-    @cached()
-    def get_user_in_directory(self, user_id):
-        return self._simple_select_one(
-            table="user_directory",
-            keyvalues={"user_id": user_id},
-            retcols=("display_name", "avatar_url"),
-            allow_none=True,
-            desc="get_user_in_directory",
-        )
-
     def get_user_directory_stream_pos(self):
         return self._simple_select_one_onecol(
             table="user_directory_stream_pos",
@@ -670,14 +688,6 @@ class UserDirectoryStore(StateDeltasStore, BackgroundUpdateStore):
             desc="get_user_directory_stream_pos",
         )
 
-    def update_user_directory_stream_pos(self, stream_id):
-        return self._simple_update_one(
-            table="user_directory_stream_pos",
-            keyvalues={},
-            updatevalues={"stream_id": stream_id},
-            desc="update_user_directory_stream_pos",
-        )
-
     @defer.inlineCallbacks
     def search_user_dir(self, user_id, search_term, limit):
         """Searches for users in directory
diff --git a/synapse/storage/user_erasure_store.py b/synapse/storage/data_stores/main/user_erasure_store.py
similarity index 88%
rename from synapse/storage/user_erasure_store.py
rename to synapse/storage/data_stores/main/user_erasure_store.py
index 05cabc2282..aa4f0da5f0 100644
--- a/synapse/storage/user_erasure_store.py
+++ b/synapse/storage/data_stores/main/user_erasure_store.py
@@ -56,15 +56,15 @@ class UserErasureWorkerStore(SQLBaseStore):
         # iterate it multiple times, and (b) avoiding duplicates.
         user_ids = tuple(set(user_ids))
 
-        def _get_erased_users(txn):
-            txn.execute(
-                "SELECT user_id FROM erased_users WHERE user_id IN (%s)"
-                % (",".join("?" * len(user_ids))),
-                user_ids,
-            )
-            return set(r[0] for r in txn)
+        rows = yield self._simple_select_many_batch(
+            table="erased_users",
+            column="user_id",
+            iterable=user_ids,
+            retcols=("user_id",),
+            desc="are_users_erased",
+        )
+        erased_users = set(row["user_id"] for row in rows)
 
-        erased_users = yield self.runInteraction("are_users_erased", _get_erased_users)
         res = dict((u, u in erased_users) for u in user_ids)
         return res
 
diff --git a/synapse/storage/engines/postgres.py b/synapse/storage/engines/postgres.py
index 289b6bc281..b7c4eda338 100644
--- a/synapse/storage/engines/postgres.py
+++ b/synapse/storage/engines/postgres.py
@@ -22,6 +22,13 @@ class PostgresEngine(object):
     def __init__(self, database_module, database_config):
         self.module = database_module
         self.module.extensions.register_type(self.module.extensions.UNICODE)
+
+        # Disables passing `bytes` to txn.execute, c.f. #6186. If you do
+        # actually want to use bytes than wrap it in `bytearray`.
+        def _disable_bytes_adapter(_):
+            raise Exception("Passing bytes to DB is disabled.")
+
+        self.module.extensions.register_adapter(bytes, _disable_bytes_adapter)
         self.synchronous_commit = database_config.get("synchronous_commit", True)
         self._version = None  # unknown as yet
 
@@ -72,6 +79,19 @@ class PostgresEngine(object):
         """
         return True
 
+    @property
+    def supports_tuple_comparison(self):
+        """
+        Do we support comparing tuples, i.e. `(a, b) > (c, d)`?
+        """
+        return True
+
+    @property
+    def supports_using_any_list(self):
+        """Do we support using `a = ANY(?)` and passing a list
+        """
+        return True
+
     def is_deadlock(self, error):
         if isinstance(error, self.module.DatabaseError):
             # https://www.postgresql.org/docs/current/static/errcodes-appendix.html
diff --git a/synapse/storage/engines/sqlite.py b/synapse/storage/engines/sqlite.py
index e9b9caa49a..ddad17dc5a 100644
--- a/synapse/storage/engines/sqlite.py
+++ b/synapse/storage/engines/sqlite.py
@@ -38,6 +38,20 @@ class Sqlite3Engine(object):
         """
         return self.module.sqlite_version_info >= (3, 24, 0)
 
+    @property
+    def supports_tuple_comparison(self):
+        """
+        Do we support comparing tuples, i.e. `(a, b) > (c, d)`? This requires
+        SQLite 3.15+.
+        """
+        return self.module.sqlite_version_info >= (3, 15, 0)
+
+    @property
+    def supports_using_any_list(self):
+        """Do we support using `a = ANY(?)` and passing a list
+        """
+        return False
+
     def check_database(self, txn):
         pass
 
diff --git a/synapse/storage/keys.py b/synapse/storage/keys.py
index e72f89e446..4769b21529 100644
--- a/synapse/storage/keys.py
+++ b/synapse/storage/keys.py
@@ -14,208 +14,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import itertools
 import logging
 
-import six
-
 import attr
-from signedjson.key import decode_verify_key_bytes
-
-from synapse.util import batch_iter
-from synapse.util.caches.descriptors import cached, cachedList
-
-from ._base import SQLBaseStore
 
 logger = logging.getLogger(__name__)
 
-# py2 sqlite has buffer hardcoded as only binary type, so we must use it,
-# despite being deprecated and removed in favor of memoryview
-if six.PY2:
-    db_binary_type = six.moves.builtins.buffer
-else:
-    db_binary_type = memoryview
-
 
 @attr.s(slots=True, frozen=True)
 class FetchKeyResult(object):
     verify_key = attr.ib()  # VerifyKey: the key itself
     valid_until_ts = attr.ib()  # int: how long we can use this key for
-
-
-class KeyStore(SQLBaseStore):
-    """Persistence for signature verification keys
-    """
-
-    @cached()
-    def _get_server_verify_key(self, server_name_and_key_id):
-        raise NotImplementedError()
-
-    @cachedList(
-        cached_method_name="_get_server_verify_key", list_name="server_name_and_key_ids"
-    )
-    def get_server_verify_keys(self, server_name_and_key_ids):
-        """
-        Args:
-            server_name_and_key_ids (iterable[Tuple[str, str]]):
-                iterable of (server_name, key-id) tuples to fetch keys for
-
-        Returns:
-            Deferred: resolves to dict[Tuple[str, str], FetchKeyResult|None]:
-                map from (server_name, key_id) -> FetchKeyResult, or None if the key is
-                unknown
-        """
-        keys = {}
-
-        def _get_keys(txn, batch):
-            """Processes a batch of keys to fetch, and adds the result to `keys`."""
-
-            # batch_iter always returns tuples so it's safe to do len(batch)
-            sql = (
-                "SELECT server_name, key_id, verify_key, ts_valid_until_ms "
-                "FROM server_signature_keys WHERE 1=0"
-            ) + " OR (server_name=? AND key_id=?)" * len(batch)
-
-            txn.execute(sql, tuple(itertools.chain.from_iterable(batch)))
-
-            for row in txn:
-                server_name, key_id, key_bytes, ts_valid_until_ms = row
-
-                if ts_valid_until_ms is None:
-                    # Old keys may be stored with a ts_valid_until_ms of null,
-                    # in which case we treat this as if it was set to `0`, i.e.
-                    # it won't match key requests that define a minimum
-                    # `ts_valid_until_ms`.
-                    ts_valid_until_ms = 0
-
-                res = FetchKeyResult(
-                    verify_key=decode_verify_key_bytes(key_id, bytes(key_bytes)),
-                    valid_until_ts=ts_valid_until_ms,
-                )
-                keys[(server_name, key_id)] = res
-
-        def _txn(txn):
-            for batch in batch_iter(server_name_and_key_ids, 50):
-                _get_keys(txn, batch)
-            return keys
-
-        return self.runInteraction("get_server_verify_keys", _txn)
-
-    def store_server_verify_keys(self, from_server, ts_added_ms, verify_keys):
-        """Stores NACL verification keys for remote servers.
-        Args:
-            from_server (str): Where the verification keys were looked up
-            ts_added_ms (int): The time to record that the key was added
-            verify_keys (iterable[tuple[str, str, FetchKeyResult]]):
-                keys to be stored. Each entry is a triplet of
-                (server_name, key_id, key).
-        """
-        key_values = []
-        value_values = []
-        invalidations = []
-        for server_name, key_id, fetch_result in verify_keys:
-            key_values.append((server_name, key_id))
-            value_values.append(
-                (
-                    from_server,
-                    ts_added_ms,
-                    fetch_result.valid_until_ts,
-                    db_binary_type(fetch_result.verify_key.encode()),
-                )
-            )
-            # invalidate takes a tuple corresponding to the params of
-            # _get_server_verify_key. _get_server_verify_key only takes one
-            # param, which is itself the 2-tuple (server_name, key_id).
-            invalidations.append((server_name, key_id))
-
-        def _invalidate(res):
-            f = self._get_server_verify_key.invalidate
-            for i in invalidations:
-                f((i,))
-            return res
-
-        return self.runInteraction(
-            "store_server_verify_keys",
-            self._simple_upsert_many_txn,
-            table="server_signature_keys",
-            key_names=("server_name", "key_id"),
-            key_values=key_values,
-            value_names=(
-                "from_server",
-                "ts_added_ms",
-                "ts_valid_until_ms",
-                "verify_key",
-            ),
-            value_values=value_values,
-        ).addCallback(_invalidate)
-
-    def store_server_keys_json(
-        self, server_name, key_id, from_server, ts_now_ms, ts_expires_ms, key_json_bytes
-    ):
-        """Stores the JSON bytes for a set of keys from a server
-        The JSON should be signed by the originating server, the intermediate
-        server, and by this server. Updates the value for the
-        (server_name, key_id, from_server) triplet if one already existed.
-        Args:
-            server_name (str): The name of the server.
-            key_id (str): The identifer of the key this JSON is for.
-            from_server (str): The server this JSON was fetched from.
-            ts_now_ms (int): The time now in milliseconds.
-            ts_valid_until_ms (int): The time when this json stops being valid.
-            key_json (bytes): The encoded JSON.
-        """
-        return self._simple_upsert(
-            table="server_keys_json",
-            keyvalues={
-                "server_name": server_name,
-                "key_id": key_id,
-                "from_server": from_server,
-            },
-            values={
-                "server_name": server_name,
-                "key_id": key_id,
-                "from_server": from_server,
-                "ts_added_ms": ts_now_ms,
-                "ts_valid_until_ms": ts_expires_ms,
-                "key_json": db_binary_type(key_json_bytes),
-            },
-            desc="store_server_keys_json",
-        )
-
-    def get_server_keys_json(self, server_keys):
-        """Retrive the key json for a list of server_keys and key ids.
-        If no keys are found for a given server, key_id and source then
-        that server, key_id, and source triplet entry will be an empty list.
-        The JSON is returned as a byte array so that it can be efficiently
-        used in an HTTP response.
-        Args:
-            server_keys (list): List of (server_name, key_id, source) triplets.
-        Returns:
-            Deferred[dict[Tuple[str, str, str|None], list[dict]]]:
-                Dict mapping (server_name, key_id, source) triplets to lists of dicts
-        """
-
-        def _get_server_keys_json_txn(txn):
-            results = {}
-            for server_name, key_id, from_server in server_keys:
-                keyvalues = {"server_name": server_name}
-                if key_id is not None:
-                    keyvalues["key_id"] = key_id
-                if from_server is not None:
-                    keyvalues["from_server"] = from_server
-                rows = self._simple_select_list_txn(
-                    txn,
-                    "server_keys_json",
-                    keyvalues=keyvalues,
-                    retcols=(
-                        "key_id",
-                        "from_server",
-                        "ts_added_ms",
-                        "ts_valid_until_ms",
-                        "key_json",
-                    ),
-                )
-                results[(server_name, key_id, from_server)] = rows
-            return results
-
-        return self.runInteraction("get_server_keys_json", _get_server_keys_json_txn)
diff --git a/synapse/storage/persist_events.py b/synapse/storage/persist_events.py
new file mode 100644
index 0000000000..fa03ca9ff7
--- /dev/null
+++ b/synapse/storage/persist_events.py
@@ -0,0 +1,649 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2018-2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from collections import deque, namedtuple
+
+from six import iteritems
+from six.moves import range
+
+from prometheus_client import Counter, Histogram
+
+from twisted.internet import defer
+
+from synapse.api.constants import EventTypes
+from synapse.logging.context import PreserveLoggingContext, make_deferred_yieldable
+from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.state import StateResolutionStore
+from synapse.storage.data_stores import DataStores
+from synapse.util.async_helpers import ObservableDeferred
+from synapse.util.metrics import Measure
+
+logger = logging.getLogger(__name__)
+
+# The number of times we are recalculating the current state
+state_delta_counter = Counter("synapse_storage_events_state_delta", "")
+
+# The number of times we are recalculating state when there is only a
+# single forward extremity
+state_delta_single_event_counter = Counter(
+    "synapse_storage_events_state_delta_single_event", ""
+)
+
+# The number of times we are reculating state when we could have resonably
+# calculated the delta when we calculated the state for an event we were
+# persisting.
+state_delta_reuse_delta_counter = Counter(
+    "synapse_storage_events_state_delta_reuse_delta", ""
+)
+
+# The number of forward extremities for each new event.
+forward_extremities_counter = Histogram(
+    "synapse_storage_events_forward_extremities_persisted",
+    "Number of forward extremities for each new event",
+    buckets=(1, 2, 3, 5, 7, 10, 15, 20, 50, 100, 200, 500, "+Inf"),
+)
+
+# The number of stale forward extremities for each new event. Stale extremities
+# are those that were in the previous set of extremities as well as the new.
+stale_forward_extremities_counter = Histogram(
+    "synapse_storage_events_stale_forward_extremities_persisted",
+    "Number of unchanged forward extremities for each new event",
+    buckets=(0, 1, 2, 3, 5, 7, 10, 15, 20, 50, 100, 200, 500, "+Inf"),
+)
+
+
+class _EventPeristenceQueue(object):
+    """Queues up events so that they can be persisted in bulk with only one
+    concurrent transaction per room.
+    """
+
+    _EventPersistQueueItem = namedtuple(
+        "_EventPersistQueueItem", ("events_and_contexts", "backfilled", "deferred")
+    )
+
+    def __init__(self):
+        self._event_persist_queues = {}
+        self._currently_persisting_rooms = set()
+
+    def add_to_queue(self, room_id, events_and_contexts, backfilled):
+        """Add events to the queue, with the given persist_event options.
+
+        NB: due to the normal usage pattern of this method, it does *not*
+        follow the synapse logcontext rules, and leaves the logcontext in
+        place whether or not the returned deferred is ready.
+
+        Args:
+            room_id (str):
+            events_and_contexts (list[(EventBase, EventContext)]):
+            backfilled (bool):
+
+        Returns:
+            defer.Deferred: a deferred which will resolve once the events are
+                persisted. Runs its callbacks *without* a logcontext.
+        """
+        queue = self._event_persist_queues.setdefault(room_id, deque())
+        if queue:
+            # if the last item in the queue has the same `backfilled` setting,
+            # we can just add these new events to that item.
+            end_item = queue[-1]
+            if end_item.backfilled == backfilled:
+                end_item.events_and_contexts.extend(events_and_contexts)
+                return end_item.deferred.observe()
+
+        deferred = ObservableDeferred(defer.Deferred(), consumeErrors=True)
+
+        queue.append(
+            self._EventPersistQueueItem(
+                events_and_contexts=events_and_contexts,
+                backfilled=backfilled,
+                deferred=deferred,
+            )
+        )
+
+        return deferred.observe()
+
+    def handle_queue(self, room_id, per_item_callback):
+        """Attempts to handle the queue for a room if not already being handled.
+
+        The given callback will be invoked with for each item in the queue,
+        of type _EventPersistQueueItem. The per_item_callback will continuously
+        be called with new items, unless the queue becomnes empty. The return
+        value of the function will be given to the deferreds waiting on the item,
+        exceptions will be passed to the deferreds as well.
+
+        This function should therefore be called whenever anything is added
+        to the queue.
+
+        If another callback is currently handling the queue then it will not be
+        invoked.
+        """
+
+        if room_id in self._currently_persisting_rooms:
+            return
+
+        self._currently_persisting_rooms.add(room_id)
+
+        @defer.inlineCallbacks
+        def handle_queue_loop():
+            try:
+                queue = self._get_drainining_queue(room_id)
+                for item in queue:
+                    try:
+                        ret = yield per_item_callback(item)
+                    except Exception:
+                        with PreserveLoggingContext():
+                            item.deferred.errback()
+                    else:
+                        with PreserveLoggingContext():
+                            item.deferred.callback(ret)
+            finally:
+                queue = self._event_persist_queues.pop(room_id, None)
+                if queue:
+                    self._event_persist_queues[room_id] = queue
+                self._currently_persisting_rooms.discard(room_id)
+
+        # set handle_queue_loop off in the background
+        run_as_background_process("persist_events", handle_queue_loop)
+
+    def _get_drainining_queue(self, room_id):
+        queue = self._event_persist_queues.setdefault(room_id, deque())
+
+        try:
+            while True:
+                yield queue.popleft()
+        except IndexError:
+            # Queue has been drained.
+            pass
+
+
+class EventsPersistenceStorage(object):
+    """High level interface for handling persisting newly received events.
+
+    Takes care of batching up events by room, and calculating the necessary
+    current state and forward extremity changes.
+    """
+
+    def __init__(self, hs, stores: DataStores):
+        # We ultimately want to split out the state store from the main store,
+        # so we use separate variables here even though they point to the same
+        # store for now.
+        self.main_store = stores.main
+        self.state_store = stores.main
+
+        self._clock = hs.get_clock()
+        self.is_mine_id = hs.is_mine_id
+        self._event_persist_queue = _EventPeristenceQueue()
+        self._state_resolution_handler = hs.get_state_resolution_handler()
+
+    @defer.inlineCallbacks
+    def persist_events(self, events_and_contexts, backfilled=False):
+        """
+        Write events to the database
+        Args:
+            events_and_contexts: list of tuples of (event, context)
+            backfilled (bool): Whether the results are retrieved from federation
+                via backfill or not. Used to determine if they're "new" events
+                which might update the current state etc.
+
+        Returns:
+            Deferred[int]: the stream ordering of the latest persisted event
+        """
+        partitioned = {}
+        for event, ctx in events_and_contexts:
+            partitioned.setdefault(event.room_id, []).append((event, ctx))
+
+        deferreds = []
+        for room_id, evs_ctxs in iteritems(partitioned):
+            d = self._event_persist_queue.add_to_queue(
+                room_id, evs_ctxs, backfilled=backfilled
+            )
+            deferreds.append(d)
+
+        for room_id in partitioned:
+            self._maybe_start_persisting(room_id)
+
+        yield make_deferred_yieldable(
+            defer.gatherResults(deferreds, consumeErrors=True)
+        )
+
+        max_persisted_id = yield self.main_store.get_current_events_token()
+
+        return max_persisted_id
+
+    @defer.inlineCallbacks
+    def persist_event(self, event, context, backfilled=False):
+        """
+
+        Args:
+            event (EventBase):
+            context (EventContext):
+            backfilled (bool):
+
+        Returns:
+            Deferred: resolves to (int, int): the stream ordering of ``event``,
+            and the stream ordering of the latest persisted event
+        """
+        deferred = self._event_persist_queue.add_to_queue(
+            event.room_id, [(event, context)], backfilled=backfilled
+        )
+
+        self._maybe_start_persisting(event.room_id)
+
+        yield make_deferred_yieldable(deferred)
+
+        max_persisted_id = yield self.main_store.get_current_events_token()
+        return (event.internal_metadata.stream_ordering, max_persisted_id)
+
+    def _maybe_start_persisting(self, room_id):
+        @defer.inlineCallbacks
+        def persisting_queue(item):
+            with Measure(self._clock, "persist_events"):
+                yield self._persist_events(
+                    item.events_and_contexts, backfilled=item.backfilled
+                )
+
+        self._event_persist_queue.handle_queue(room_id, persisting_queue)
+
+    @defer.inlineCallbacks
+    def _persist_events(self, events_and_contexts, backfilled=False):
+        """Calculates the change to current state and forward extremities, and
+        persists the given events and with those updates.
+
+        Args:
+            events_and_contexts (list[(EventBase, EventContext)]):
+            backfilled (bool):
+            delete_existing (bool):
+
+        Returns:
+            Deferred: resolves when the events have been persisted
+        """
+        if not events_and_contexts:
+            return
+
+        chunks = [
+            events_and_contexts[x : x + 100]
+            for x in range(0, len(events_and_contexts), 100)
+        ]
+
+        for chunk in chunks:
+            # We can't easily parallelize these since different chunks
+            # might contain the same event. :(
+
+            # NB: Assumes that we are only persisting events for one room
+            # at a time.
+
+            # map room_id->list[event_ids] giving the new forward
+            # extremities in each room
+            new_forward_extremeties = {}
+
+            # map room_id->(type,state_key)->event_id tracking the full
+            # state in each room after adding these events.
+            # This is simply used to prefill the get_current_state_ids
+            # cache
+            current_state_for_room = {}
+
+            # map room_id->(to_delete, to_insert) where to_delete is a list
+            # of type/state keys to remove from current state, and to_insert
+            # is a map (type,key)->event_id giving the state delta in each
+            # room
+            state_delta_for_room = {}
+
+            if not backfilled:
+                with Measure(self._clock, "_calculate_state_and_extrem"):
+                    # Work out the new "current state" for each room.
+                    # We do this by working out what the new extremities are and then
+                    # calculating the state from that.
+                    events_by_room = {}
+                    for event, context in chunk:
+                        events_by_room.setdefault(event.room_id, []).append(
+                            (event, context)
+                        )
+
+                    for room_id, ev_ctx_rm in iteritems(events_by_room):
+                        latest_event_ids = yield self.main_store.get_latest_event_ids_in_room(
+                            room_id
+                        )
+                        new_latest_event_ids = yield self._calculate_new_extremities(
+                            room_id, ev_ctx_rm, latest_event_ids
+                        )
+
+                        latest_event_ids = set(latest_event_ids)
+                        if new_latest_event_ids == latest_event_ids:
+                            # No change in extremities, so no change in state
+                            continue
+
+                        # there should always be at least one forward extremity.
+                        # (except during the initial persistence of the send_join
+                        # results, in which case there will be no existing
+                        # extremities, so we'll `continue` above and skip this bit.)
+                        assert new_latest_event_ids, "No forward extremities left!"
+
+                        new_forward_extremeties[room_id] = new_latest_event_ids
+
+                        len_1 = (
+                            len(latest_event_ids) == 1
+                            and len(new_latest_event_ids) == 1
+                        )
+                        if len_1:
+                            all_single_prev_not_state = all(
+                                len(event.prev_event_ids()) == 1
+                                and not event.is_state()
+                                for event, ctx in ev_ctx_rm
+                            )
+                            # Don't bother calculating state if they're just
+                            # a long chain of single ancestor non-state events.
+                            if all_single_prev_not_state:
+                                continue
+
+                        state_delta_counter.inc()
+                        if len(new_latest_event_ids) == 1:
+                            state_delta_single_event_counter.inc()
+
+                            # This is a fairly handwavey check to see if we could
+                            # have guessed what the delta would have been when
+                            # processing one of these events.
+                            # What we're interested in is if the latest extremities
+                            # were the same when we created the event as they are
+                            # now. When this server creates a new event (as opposed
+                            # to receiving it over federation) it will use the
+                            # forward extremities as the prev_events, so we can
+                            # guess this by looking at the prev_events and checking
+                            # if they match the current forward extremities.
+                            for ev, _ in ev_ctx_rm:
+                                prev_event_ids = set(ev.prev_event_ids())
+                                if latest_event_ids == prev_event_ids:
+                                    state_delta_reuse_delta_counter.inc()
+                                    break
+
+                        logger.info("Calculating state delta for room %s", room_id)
+                        with Measure(
+                            self._clock, "persist_events.get_new_state_after_events"
+                        ):
+                            res = yield self._get_new_state_after_events(
+                                room_id,
+                                ev_ctx_rm,
+                                latest_event_ids,
+                                new_latest_event_ids,
+                            )
+                            current_state, delta_ids = res
+
+                        # If either are not None then there has been a change,
+                        # and we need to work out the delta (or use that
+                        # given)
+                        if delta_ids is not None:
+                            # If there is a delta we know that we've
+                            # only added or replaced state, never
+                            # removed keys entirely.
+                            state_delta_for_room[room_id] = ([], delta_ids)
+                        elif current_state is not None:
+                            with Measure(
+                                self._clock, "persist_events.calculate_state_delta"
+                            ):
+                                delta = yield self._calculate_state_delta(
+                                    room_id, current_state
+                                )
+                            state_delta_for_room[room_id] = delta
+
+                        # If we have the current_state then lets prefill
+                        # the cache with it.
+                        if current_state is not None:
+                            current_state_for_room[room_id] = current_state
+
+            yield self.main_store._persist_events_and_state_updates(
+                chunk,
+                current_state_for_room=current_state_for_room,
+                state_delta_for_room=state_delta_for_room,
+                new_forward_extremeties=new_forward_extremeties,
+                backfilled=backfilled,
+            )
+
+    @defer.inlineCallbacks
+    def _calculate_new_extremities(self, room_id, event_contexts, latest_event_ids):
+        """Calculates the new forward extremities for a room given events to
+        persist.
+
+        Assumes that we are only persisting events for one room at a time.
+        """
+
+        # we're only interested in new events which aren't outliers and which aren't
+        # being rejected.
+        new_events = [
+            event
+            for event, ctx in event_contexts
+            if not event.internal_metadata.is_outlier()
+            and not ctx.rejected
+            and not event.internal_metadata.is_soft_failed()
+        ]
+
+        latest_event_ids = set(latest_event_ids)
+
+        # start with the existing forward extremities
+        result = set(latest_event_ids)
+
+        # add all the new events to the list
+        result.update(event.event_id for event in new_events)
+
+        # Now remove all events which are prev_events of any of the new events
+        result.difference_update(
+            e_id for event in new_events for e_id in event.prev_event_ids()
+        )
+
+        # Remove any events which are prev_events of any existing events.
+        existing_prevs = yield self.main_store._get_events_which_are_prevs(result)
+        result.difference_update(existing_prevs)
+
+        # Finally handle the case where the new events have soft-failed prev
+        # events. If they do we need to remove them and their prev events,
+        # otherwise we end up with dangling extremities.
+        existing_prevs = yield self.main_store._get_prevs_before_rejected(
+            e_id for event in new_events for e_id in event.prev_event_ids()
+        )
+        result.difference_update(existing_prevs)
+
+        # We only update metrics for events that change forward extremities
+        # (e.g. we ignore backfill/outliers/etc)
+        if result != latest_event_ids:
+            forward_extremities_counter.observe(len(result))
+            stale = latest_event_ids & result
+            stale_forward_extremities_counter.observe(len(stale))
+
+        return result
+
+    @defer.inlineCallbacks
+    def _get_new_state_after_events(
+        self, room_id, events_context, old_latest_event_ids, new_latest_event_ids
+    ):
+        """Calculate the current state dict after adding some new events to
+        a room
+
+        Args:
+            room_id (str):
+                room to which the events are being added. Used for logging etc
+
+            events_context (list[(EventBase, EventContext)]):
+                events and contexts which are being added to the room
+
+            old_latest_event_ids (iterable[str]):
+                the old forward extremities for the room.
+
+            new_latest_event_ids (iterable[str]):
+                the new forward extremities for the room.
+
+        Returns:
+            Deferred[tuple[dict[(str,str), str]|None, dict[(str,str), str]|None]]:
+            Returns a tuple of two state maps, the first being the full new current
+            state and the second being the delta to the existing current state.
+            If both are None then there has been no change.
+
+            If there has been a change then we only return the delta if its
+            already been calculated. Conversely if we do know the delta then
+            the new current state is only returned if we've already calculated
+            it.
+        """
+        # map from state_group to ((type, key) -> event_id) state map
+        state_groups_map = {}
+
+        # Map from (prev state group, new state group) -> delta state dict
+        state_group_deltas = {}
+
+        for ev, ctx in events_context:
+            if ctx.state_group is None:
+                # This should only happen for outlier events.
+                if not ev.internal_metadata.is_outlier():
+                    raise Exception(
+                        "Context for new event %s has no state "
+                        "group" % (ev.event_id,)
+                    )
+                continue
+
+            if ctx.state_group in state_groups_map:
+                continue
+
+            # We're only interested in pulling out state that has already
+            # been cached in the context. We'll pull stuff out of the DB later
+            # if necessary.
+            current_state_ids = ctx.get_cached_current_state_ids()
+            if current_state_ids is not None:
+                state_groups_map[ctx.state_group] = current_state_ids
+
+            if ctx.prev_group:
+                state_group_deltas[(ctx.prev_group, ctx.state_group)] = ctx.delta_ids
+
+        # We need to map the event_ids to their state groups. First, let's
+        # check if the event is one we're persisting, in which case we can
+        # pull the state group from its context.
+        # Otherwise we need to pull the state group from the database.
+
+        # Set of events we need to fetch groups for. (We know none of the old
+        # extremities are going to be in events_context).
+        missing_event_ids = set(old_latest_event_ids)
+
+        event_id_to_state_group = {}
+        for event_id in new_latest_event_ids:
+            # First search in the list of new events we're adding.
+            for ev, ctx in events_context:
+                if event_id == ev.event_id and ctx.state_group is not None:
+                    event_id_to_state_group[event_id] = ctx.state_group
+                    break
+            else:
+                # If we couldn't find it, then we'll need to pull
+                # the state from the database
+                missing_event_ids.add(event_id)
+
+        if missing_event_ids:
+            # Now pull out the state groups for any missing events from DB
+            event_to_groups = yield self.main_store._get_state_group_for_events(
+                missing_event_ids
+            )
+            event_id_to_state_group.update(event_to_groups)
+
+        # State groups of old_latest_event_ids
+        old_state_groups = set(
+            event_id_to_state_group[evid] for evid in old_latest_event_ids
+        )
+
+        # State groups of new_latest_event_ids
+        new_state_groups = set(
+            event_id_to_state_group[evid] for evid in new_latest_event_ids
+        )
+
+        # If they old and new groups are the same then we don't need to do
+        # anything.
+        if old_state_groups == new_state_groups:
+            return None, None
+
+        if len(new_state_groups) == 1 and len(old_state_groups) == 1:
+            # If we're going from one state group to another, lets check if
+            # we have a delta for that transition. If we do then we can just
+            # return that.
+
+            new_state_group = next(iter(new_state_groups))
+            old_state_group = next(iter(old_state_groups))
+
+            delta_ids = state_group_deltas.get((old_state_group, new_state_group), None)
+            if delta_ids is not None:
+                # We have a delta from the existing to new current state,
+                # so lets just return that. If we happen to already have
+                # the current state in memory then lets also return that,
+                # but it doesn't matter if we don't.
+                new_state = state_groups_map.get(new_state_group)
+                return new_state, delta_ids
+
+        # Now that we have calculated new_state_groups we need to get
+        # their state IDs so we can resolve to a single state set.
+        missing_state = new_state_groups - set(state_groups_map)
+        if missing_state:
+            group_to_state = yield self.state_store._get_state_for_groups(missing_state)
+            state_groups_map.update(group_to_state)
+
+        if len(new_state_groups) == 1:
+            # If there is only one state group, then we know what the current
+            # state is.
+            return state_groups_map[new_state_groups.pop()], None
+
+        # Ok, we need to defer to the state handler to resolve our state sets.
+
+        state_groups = {sg: state_groups_map[sg] for sg in new_state_groups}
+
+        events_map = {ev.event_id: ev for ev, _ in events_context}
+
+        # We need to get the room version, which is in the create event.
+        # Normally that'd be in the database, but its also possible that we're
+        # currently trying to persist it.
+        room_version = None
+        for ev, _ in events_context:
+            if ev.type == EventTypes.Create and ev.state_key == "":
+                room_version = ev.content.get("room_version", "1")
+                break
+
+        if not room_version:
+            room_version = yield self.main_store.get_room_version(room_id)
+
+        logger.debug("calling resolve_state_groups from preserve_events")
+        res = yield self._state_resolution_handler.resolve_state_groups(
+            room_id,
+            room_version,
+            state_groups,
+            events_map,
+            state_res_store=StateResolutionStore(self.main_store),
+        )
+
+        return res.state, None
+
+    @defer.inlineCallbacks
+    def _calculate_state_delta(self, room_id, current_state):
+        """Calculate the new state deltas for a room.
+
+        Assumes that we are only persisting events for one room at a time.
+
+        Returns:
+            tuple[list, dict] (to_delete, to_insert): where to_delete are the
+            type/state_keys to remove from current_state_events and `to_insert`
+            are the updates to current_state_events.
+        """
+        existing_state = yield self.main_store.get_current_state_ids(room_id)
+
+        to_delete = [key for key in existing_state if key not in current_state]
+
+        to_insert = {
+            key: ev_id
+            for key, ev_id in iteritems(current_state)
+            if ev_id != existing_state.get(key)
+        }
+
+        return to_delete, to_insert
diff --git a/synapse/storage/prepare_database.py b/synapse/storage/prepare_database.py
index e96eed8a6d..2e7753820e 100644
--- a/synapse/storage/prepare_database.py
+++ b/synapse/storage/prepare_database.py
@@ -14,12 +14,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import fnmatch
 import imp
 import logging
 import os
 import re
 
+import attr
+
 from synapse.storage.engines.postgres import PostgresEngine
 
 logger = logging.getLogger(__name__)
@@ -54,6 +55,10 @@ def prepare_database(db_conn, database_engine, config):
             application config, or None if we are connecting to an existing
             database which we expect to be configured already
     """
+
+    # For now we only have the one datastore.
+    data_stores = ["main"]
+
     try:
         cur = db_conn.cursor()
         version_info = _get_or_create_schema_state(cur, database_engine)
@@ -68,10 +73,16 @@ def prepare_database(db_conn, database_engine, config):
                     raise UpgradeDatabaseException("Database needs to be upgraded")
             else:
                 _upgrade_existing_database(
-                    cur, user_version, delta_files, upgraded, database_engine, config
+                    cur,
+                    user_version,
+                    delta_files,
+                    upgraded,
+                    database_engine,
+                    config,
+                    data_stores=data_stores,
                 )
         else:
-            _setup_new_database(cur, database_engine)
+            _setup_new_database(cur, database_engine, data_stores=data_stores)
 
         # check if any of our configured dynamic modules want a database
         if config is not None:
@@ -84,9 +95,10 @@ def prepare_database(db_conn, database_engine, config):
         raise
 
 
-def _setup_new_database(cur, database_engine):
+def _setup_new_database(cur, database_engine, data_stores):
     """Sets up the database by finding a base set of "full schemas" and then
-    applying any necessary deltas.
+    applying any necessary deltas, including schemas from the given data
+    stores.
 
     The "full_schemas" directory has subdirectories named after versions. This
     function searches for the highest version less than or equal to
@@ -111,52 +123,78 @@ def _setup_new_database(cur, database_engine):
 
     In the example foo.sql and bar.sql would be run, and then any delta files
     for versions strictly greater than 11.
+
+    Note: we apply the full schemas and deltas from the top level `schema/`
+    folder as well those in the data stores specified.
+
+    Args:
+        cur (Cursor): a database cursor
+        database_engine (DatabaseEngine)
+        data_stores (list[str]): The names of the data stores to instantiate
+            on the given database.
     """
     current_dir = os.path.join(dir_path, "schema", "full_schemas")
     directory_entries = os.listdir(current_dir)
 
-    valid_dirs = []
-    pattern = re.compile(r"^\d+(\.sql)?$")
+    # First we find the highest full schema version we have
+    valid_versions = []
+
+    for filename in directory_entries:
+        try:
+            ver = int(filename)
+        except ValueError:
+            continue
+
+        if ver <= SCHEMA_VERSION:
+            valid_versions.append(ver)
+
+    if not valid_versions:
+        raise PrepareDatabaseException(
+            "Could not find a suitable base set of full schemas"
+        )
+
+    max_current_ver = max(valid_versions)
+
+    logger.debug("Initialising schema v%d", max_current_ver)
+
+    # Now lets find all the full schema files, both in the global schema and
+    # in data store schemas.
+    directories = [os.path.join(current_dir, str(max_current_ver))]
+    directories.extend(
+        os.path.join(
+            dir_path,
+            "data_stores",
+            data_store,
+            "schema",
+            "full_schemas",
+            str(max_current_ver),
+        )
+        for data_store in data_stores
+    )
+
+    directory_entries = []
+    for directory in directories:
+        directory_entries.extend(
+            _DirectoryListing(file_name, os.path.join(directory, file_name))
+            for file_name in os.listdir(directory)
+        )
 
     if isinstance(database_engine, PostgresEngine):
         specific = "postgres"
     else:
         specific = "sqlite"
 
-    specific_pattern = re.compile(r"^\d+(\.sql." + specific + r")?$")
-
-    for filename in directory_entries:
-        match = pattern.match(filename) or specific_pattern.match(filename)
-        abs_path = os.path.join(current_dir, filename)
-        if match and os.path.isdir(abs_path):
-            ver = int(match.group(0))
-            if ver <= SCHEMA_VERSION:
-                valid_dirs.append((ver, abs_path))
-        else:
-            logger.debug("Ignoring entry '%s' in 'full_schemas'", filename)
-
-    if not valid_dirs:
-        raise PrepareDatabaseException(
-            "Could not find a suitable base set of full schemas"
-        )
-
-    max_current_ver, sql_dir = max(valid_dirs, key=lambda x: x[0])
-
-    logger.debug("Initialising schema v%d", max_current_ver)
-
-    directory_entries = os.listdir(sql_dir)
-
-    for filename in sorted(
-        fnmatch.filter(directory_entries, "*.sql")
-        + fnmatch.filter(directory_entries, "*.sql." + specific)
-    ):
-        sql_loc = os.path.join(sql_dir, filename)
-        logger.debug("Applying schema %s", sql_loc)
-        executescript(cur, sql_loc)
+    directory_entries.sort()
+    for entry in directory_entries:
+        if entry.file_name.endswith(".sql") or entry.file_name.endswith(
+            ".sql." + specific
+        ):
+            logger.debug("Applying schema %s", entry.absolute_path)
+            executescript(cur, entry.absolute_path)
 
     cur.execute(
         database_engine.convert_param_style(
-            "INSERT INTO schema_version (version, upgraded)" " VALUES (?,?)"
+            "INSERT INTO schema_version (version, upgraded) VALUES (?,?)"
         ),
         (max_current_ver, False),
     )
@@ -168,6 +206,7 @@ def _setup_new_database(cur, database_engine):
         upgraded=False,
         database_engine=database_engine,
         config=None,
+        data_stores=data_stores,
         is_empty=True,
     )
 
@@ -179,6 +218,7 @@ def _upgrade_existing_database(
     upgraded,
     database_engine,
     config,
+    data_stores,
     is_empty=False,
 ):
     """Upgrades an existing database.
@@ -215,6 +255,10 @@ def _upgrade_existing_database(
     only if `upgraded` is True. Then `foo.sql` and `bar.py` would be run in
     some arbitrary order.
 
+    Note: we apply the delta files from the specified data stores as well as
+    those in the top-level schema. We apply all delta files across data stores
+    for a version before applying those in the next version.
+
     Args:
         cur (Cursor)
         current_version (int): The current version of the schema.
@@ -224,6 +268,14 @@ def _upgrade_existing_database(
             applied deltas or from full schema file. If `True` the function
             will never apply delta files for the given `current_version`, since
             the current_version wasn't generated by applying those delta files.
+        database_engine (DatabaseEngine)
+        config (synapse.config.homeserver.HomeServerConfig|None):
+            application config, or None if we are connecting to an existing
+            database which we expect to be configured already
+        data_stores (list[str]): The names of the data stores to instantiate
+            on the given database.
+        is_empty (bool): Is this a blank database? I.e. do we need to run the
+            upgrade portions of the delta scripts.
     """
 
     if current_version > SCHEMA_VERSION:
@@ -248,24 +300,49 @@ def _upgrade_existing_database(
     for v in range(start_ver, SCHEMA_VERSION + 1):
         logger.info("Upgrading schema to v%d", v)
 
-        delta_dir = os.path.join(dir_path, "schema", "delta", str(v))
+        # We need to search both the global and per data store schema
+        # directories for schema updates.
 
-        try:
-            directory_entries = os.listdir(delta_dir)
-        except OSError:
-            logger.exception("Could not open delta dir for version %d", v)
-            raise UpgradeDatabaseException(
-                "Could not open delta dir for version %d" % (v,)
+        # First we find the directories to search in
+        delta_dir = os.path.join(dir_path, "schema", "delta", str(v))
+        directories = [delta_dir]
+        for data_store in data_stores:
+            directories.append(
+                os.path.join(
+                    dir_path, "data_stores", data_store, "schema", "delta", str(v)
+                )
             )
 
+        # Now find which directories have anything of interest.
+        directory_entries = []
+        for directory in directories:
+            logger.debug("Looking for schema deltas in %s", directory)
+            try:
+                file_names = os.listdir(directory)
+                directory_entries.extend(
+                    _DirectoryListing(file_name, os.path.join(directory, file_name))
+                    for file_name in file_names
+                )
+            except FileNotFoundError:
+                # Data stores can have empty entries for a given version delta.
+                pass
+            except OSError:
+                raise UpgradeDatabaseException(
+                    "Could not open delta dir for version %d: %s" % (v, directory)
+                )
+
+        # We sort to ensure that we apply the delta files in a consistent
+        # order (to avoid bugs caused by inconsistent directory listing order)
         directory_entries.sort()
-        for file_name in directory_entries:
+        for entry in directory_entries:
+            file_name = entry.file_name
             relative_path = os.path.join(str(v), file_name)
-            logger.debug("Found file: %s", relative_path)
+            absolute_path = entry.absolute_path
+
+            logger.debug("Found file: %s (%s)", relative_path, absolute_path)
             if relative_path in applied_delta_files:
                 continue
 
-            absolute_path = os.path.join(dir_path, "schema", "delta", relative_path)
             root_name, ext = os.path.splitext(file_name)
             if ext == ".py":
                 # This is a python upgrade module. We need to import into some
@@ -448,3 +525,16 @@ def _get_or_create_schema_state(txn, database_engine):
         return current_version, applied_deltas, upgraded
 
     return None
+
+
+@attr.s()
+class _DirectoryListing(object):
+    """Helper class to store schema file name and the
+    absolute path to it.
+
+    These entries get sorted, so for consistency we want to ensure that
+    `file_name` attr is kept first.
+    """
+
+    file_name = attr.ib()
+    absolute_path = attr.ib()
diff --git a/synapse/storage/presence.py b/synapse/storage/presence.py
index 5db6f2d84a..18a462f0ee 100644
--- a/synapse/storage/presence.py
+++ b/synapse/storage/presence.py
@@ -15,13 +15,7 @@
 
 from collections import namedtuple
 
-from twisted.internet import defer
-
 from synapse.api.constants import PresenceState
-from synapse.util import batch_iter
-from synapse.util.caches.descriptors import cached, cachedList
-
-from ._base import SQLBaseStore
 
 
 class UserPresenceState(
@@ -73,133 +67,3 @@ class UserPresenceState(
             status_msg=None,
             currently_active=False,
         )
-
-
-class PresenceStore(SQLBaseStore):
-    @defer.inlineCallbacks
-    def update_presence(self, presence_states):
-        stream_ordering_manager = self._presence_id_gen.get_next_mult(
-            len(presence_states)
-        )
-
-        with stream_ordering_manager as stream_orderings:
-            yield self.runInteraction(
-                "update_presence",
-                self._update_presence_txn,
-                stream_orderings,
-                presence_states,
-            )
-
-        return stream_orderings[-1], self._presence_id_gen.get_current_token()
-
-    def _update_presence_txn(self, txn, stream_orderings, presence_states):
-        for stream_id, state in zip(stream_orderings, presence_states):
-            txn.call_after(
-                self.presence_stream_cache.entity_has_changed, state.user_id, stream_id
-            )
-            txn.call_after(self._get_presence_for_user.invalidate, (state.user_id,))
-
-        # Actually insert new rows
-        self._simple_insert_many_txn(
-            txn,
-            table="presence_stream",
-            values=[
-                {
-                    "stream_id": stream_id,
-                    "user_id": state.user_id,
-                    "state": state.state,
-                    "last_active_ts": state.last_active_ts,
-                    "last_federation_update_ts": state.last_federation_update_ts,
-                    "last_user_sync_ts": state.last_user_sync_ts,
-                    "status_msg": state.status_msg,
-                    "currently_active": state.currently_active,
-                }
-                for state in presence_states
-            ],
-        )
-
-        # Delete old rows to stop database from getting really big
-        sql = (
-            "DELETE FROM presence_stream WHERE" " stream_id < ?" " AND user_id IN (%s)"
-        )
-
-        for states in batch_iter(presence_states, 50):
-            args = [stream_id]
-            args.extend(s.user_id for s in states)
-            txn.execute(sql % (",".join("?" for _ in states),), args)
-
-    def get_all_presence_updates(self, last_id, current_id):
-        if last_id == current_id:
-            return defer.succeed([])
-
-        def get_all_presence_updates_txn(txn):
-            sql = (
-                "SELECT stream_id, user_id, state, last_active_ts,"
-                " last_federation_update_ts, last_user_sync_ts, status_msg,"
-                " currently_active"
-                " FROM presence_stream"
-                " WHERE ? < stream_id AND stream_id <= ?"
-            )
-            txn.execute(sql, (last_id, current_id))
-            return txn.fetchall()
-
-        return self.runInteraction(
-            "get_all_presence_updates", get_all_presence_updates_txn
-        )
-
-    @cached()
-    def _get_presence_for_user(self, user_id):
-        raise NotImplementedError()
-
-    @cachedList(
-        cached_method_name="_get_presence_for_user",
-        list_name="user_ids",
-        num_args=1,
-        inlineCallbacks=True,
-    )
-    def get_presence_for_users(self, user_ids):
-        rows = yield self._simple_select_many_batch(
-            table="presence_stream",
-            column="user_id",
-            iterable=user_ids,
-            keyvalues={},
-            retcols=(
-                "user_id",
-                "state",
-                "last_active_ts",
-                "last_federation_update_ts",
-                "last_user_sync_ts",
-                "status_msg",
-                "currently_active",
-            ),
-            desc="get_presence_for_users",
-        )
-
-        for row in rows:
-            row["currently_active"] = bool(row["currently_active"])
-
-        return {row["user_id"]: UserPresenceState(**row) for row in rows}
-
-    def get_current_presence_token(self):
-        return self._presence_id_gen.get_current_token()
-
-    def allow_presence_visible(self, observed_localpart, observer_userid):
-        return self._simple_insert(
-            table="presence_allow_inbound",
-            values={
-                "observed_user_id": observed_localpart,
-                "observer_user_id": observer_userid,
-            },
-            desc="allow_presence_visible",
-            or_ignore=True,
-        )
-
-    def disallow_presence_visible(self, observed_localpart, observer_userid):
-        return self._simple_delete_one(
-            table="presence_allow_inbound",
-            keyvalues={
-                "observed_user_id": observed_localpart,
-                "observer_user_id": observer_userid,
-            },
-            desc="disallow_presence_visible",
-        )
diff --git a/synapse/storage/push_rule.py b/synapse/storage/push_rule.py
index a6517c4cf3..f47cec0d86 100644
--- a/synapse/storage/push_rule.py
+++ b/synapse/storage/push_rule.py
@@ -14,708 +14,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import abc
-import logging
-
-from canonicaljson import json
-
-from twisted.internet import defer
-
-from synapse.push.baserules import list_with_base_rules
-from synapse.storage.appservice import ApplicationServiceWorkerStore
-from synapse.storage.pusher import PusherWorkerStore
-from synapse.storage.receipts import ReceiptsWorkerStore
-from synapse.storage.roommember import RoomMemberWorkerStore
-from synapse.util.caches.descriptors import cachedInlineCallbacks, cachedList
-from synapse.util.caches.stream_change_cache import StreamChangeCache
-
-from ._base import SQLBaseStore
-
-logger = logging.getLogger(__name__)
-
-
-def _load_rules(rawrules, enabled_map):
-    ruleslist = []
-    for rawrule in rawrules:
-        rule = dict(rawrule)
-        rule["conditions"] = json.loads(rawrule["conditions"])
-        rule["actions"] = json.loads(rawrule["actions"])
-        ruleslist.append(rule)
-
-    # We're going to be mutating this a lot, so do a deep copy
-    rules = list(list_with_base_rules(ruleslist))
-
-    for i, rule in enumerate(rules):
-        rule_id = rule["rule_id"]
-        if rule_id in enabled_map:
-            if rule.get("enabled", True) != bool(enabled_map[rule_id]):
-                # Rules are cached across users.
-                rule = dict(rule)
-                rule["enabled"] = bool(enabled_map[rule_id])
-                rules[i] = rule
-
-    return rules
-
-
-class PushRulesWorkerStore(
-    ApplicationServiceWorkerStore,
-    ReceiptsWorkerStore,
-    PusherWorkerStore,
-    RoomMemberWorkerStore,
-    SQLBaseStore,
-):
-    """This is an abstract base class where subclasses must implement
-    `get_max_push_rules_stream_id` which can be called in the initializer.
-    """
-
-    # This ABCMeta metaclass ensures that we cannot be instantiated without
-    # the abstract methods being implemented.
-    __metaclass__ = abc.ABCMeta
-
-    def __init__(self, db_conn, hs):
-        super(PushRulesWorkerStore, self).__init__(db_conn, hs)
-
-        push_rules_prefill, push_rules_id = self._get_cache_dict(
-            db_conn,
-            "push_rules_stream",
-            entity_column="user_id",
-            stream_column="stream_id",
-            max_value=self.get_max_push_rules_stream_id(),
-        )
-
-        self.push_rules_stream_cache = StreamChangeCache(
-            "PushRulesStreamChangeCache",
-            push_rules_id,
-            prefilled_cache=push_rules_prefill,
-        )
-
-    @abc.abstractmethod
-    def get_max_push_rules_stream_id(self):
-        """Get the position of the push rules stream.
-
-        Returns:
-            int
-        """
-        raise NotImplementedError()
-
-    @cachedInlineCallbacks(max_entries=5000)
-    def get_push_rules_for_user(self, user_id):
-        rows = yield self._simple_select_list(
-            table="push_rules",
-            keyvalues={"user_name": user_id},
-            retcols=(
-                "user_name",
-                "rule_id",
-                "priority_class",
-                "priority",
-                "conditions",
-                "actions",
-            ),
-            desc="get_push_rules_enabled_for_user",
-        )
-
-        rows.sort(key=lambda row: (-int(row["priority_class"]), -int(row["priority"])))
-
-        enabled_map = yield self.get_push_rules_enabled_for_user(user_id)
-
-        rules = _load_rules(rows, enabled_map)
-
-        return rules
-
-    @cachedInlineCallbacks(max_entries=5000)
-    def get_push_rules_enabled_for_user(self, user_id):
-        results = yield self._simple_select_list(
-            table="push_rules_enable",
-            keyvalues={"user_name": user_id},
-            retcols=("user_name", "rule_id", "enabled"),
-            desc="get_push_rules_enabled_for_user",
-        )
-        return {r["rule_id"]: False if r["enabled"] == 0 else True for r in results}
-
-    def have_push_rules_changed_for_user(self, user_id, last_id):
-        if not self.push_rules_stream_cache.has_entity_changed(user_id, last_id):
-            return defer.succeed(False)
-        else:
-
-            def have_push_rules_changed_txn(txn):
-                sql = (
-                    "SELECT COUNT(stream_id) FROM push_rules_stream"
-                    " WHERE user_id = ? AND ? < stream_id"
-                )
-                txn.execute(sql, (user_id, last_id))
-                count, = txn.fetchone()
-                return bool(count)
-
-            return self.runInteraction(
-                "have_push_rules_changed", have_push_rules_changed_txn
-            )
-
-    @cachedList(
-        cached_method_name="get_push_rules_for_user",
-        list_name="user_ids",
-        num_args=1,
-        inlineCallbacks=True,
-    )
-    def bulk_get_push_rules(self, user_ids):
-        if not user_ids:
-            return {}
-
-        results = {user_id: [] for user_id in user_ids}
-
-        rows = yield self._simple_select_many_batch(
-            table="push_rules",
-            column="user_name",
-            iterable=user_ids,
-            retcols=("*",),
-            desc="bulk_get_push_rules",
-        )
-
-        rows.sort(key=lambda row: (-int(row["priority_class"]), -int(row["priority"])))
-
-        for row in rows:
-            results.setdefault(row["user_name"], []).append(row)
-
-        enabled_map_by_user = yield self.bulk_get_push_rules_enabled(user_ids)
-
-        for user_id, rules in results.items():
-            results[user_id] = _load_rules(rules, enabled_map_by_user.get(user_id, {}))
-
-        return results
-
-    @defer.inlineCallbacks
-    def move_push_rule_from_room_to_room(self, new_room_id, user_id, rule):
-        """Move a single push rule from one room to another for a specific user.
-
-        Args:
-            new_room_id (str): ID of the new room.
-            user_id (str): ID of user the push rule belongs to.
-            rule (Dict): A push rule.
-        """
-        # Create new rule id
-        rule_id_scope = "/".join(rule["rule_id"].split("/")[:-1])
-        new_rule_id = rule_id_scope + "/" + new_room_id
-
-        # Change room id in each condition
-        for condition in rule.get("conditions", []):
-            if condition.get("key") == "room_id":
-                condition["pattern"] = new_room_id
-
-        # Add the rule for the new room
-        yield self.add_push_rule(
-            user_id=user_id,
-            rule_id=new_rule_id,
-            priority_class=rule["priority_class"],
-            conditions=rule["conditions"],
-            actions=rule["actions"],
-        )
-
-        # Delete push rule for the old room
-        yield self.delete_push_rule(user_id, rule["rule_id"])
-
-    @defer.inlineCallbacks
-    def move_push_rules_from_room_to_room_for_user(
-        self, old_room_id, new_room_id, user_id
-    ):
-        """Move all of the push rules from one room to another for a specific
-        user.
-
-        Args:
-            old_room_id (str): ID of the old room.
-            new_room_id (str): ID of the new room.
-            user_id (str): ID of user to copy push rules for.
-        """
-        # Retrieve push rules for this user
-        user_push_rules = yield self.get_push_rules_for_user(user_id)
-
-        # Get rules relating to the old room, move them to the new room, then
-        # delete them from the old room
-        for rule in user_push_rules:
-            conditions = rule.get("conditions", [])
-            if any(
-                (c.get("key") == "room_id" and c.get("pattern") == old_room_id)
-                for c in conditions
-            ):
-                self.move_push_rule_from_room_to_room(new_room_id, user_id, rule)
-
-    @defer.inlineCallbacks
-    def bulk_get_push_rules_for_room(self, event, context):
-        state_group = context.state_group
-        if not state_group:
-            # If state_group is None it means it has yet to be assigned a
-            # state group, i.e. we need to make sure that calls with a state_group
-            # of None don't hit previous cached calls with a None state_group.
-            # To do this we set the state_group to a new object as object() != object()
-            state_group = object()
-
-        current_state_ids = yield context.get_current_state_ids(self)
-        result = yield self._bulk_get_push_rules_for_room(
-            event.room_id, state_group, current_state_ids, event=event
-        )
-        return result
-
-    @cachedInlineCallbacks(num_args=2, cache_context=True)
-    def _bulk_get_push_rules_for_room(
-        self, room_id, state_group, current_state_ids, cache_context, event=None
-    ):
-        # We don't use `state_group`, its there so that we can cache based
-        # on it. However, its important that its never None, since two current_state's
-        # with a state_group of None are likely to be different.
-        # See bulk_get_push_rules_for_room for how we work around this.
-        assert state_group is not None
-
-        # We also will want to generate notifs for other people in the room so
-        # their unread countss are correct in the event stream, but to avoid
-        # generating them for bot / AS users etc, we only do so for people who've
-        # sent a read receipt into the room.
-
-        users_in_room = yield self._get_joined_users_from_context(
-            room_id,
-            state_group,
-            current_state_ids,
-            on_invalidate=cache_context.invalidate,
-            event=event,
-        )
-
-        # We ignore app service users for now. This is so that we don't fill
-        # up the `get_if_users_have_pushers` cache with AS entries that we
-        # know don't have pushers, nor even read receipts.
-        local_users_in_room = set(
-            u
-            for u in users_in_room
-            if self.hs.is_mine_id(u)
-            and not self.get_if_app_services_interested_in_user(u)
-        )
-
-        # users in the room who have pushers need to get push rules run because
-        # that's how their pushers work
-        if_users_with_pushers = yield self.get_if_users_have_pushers(
-            local_users_in_room, on_invalidate=cache_context.invalidate
-        )
-        user_ids = set(
-            uid for uid, have_pusher in if_users_with_pushers.items() if have_pusher
-        )
-
-        users_with_receipts = yield self.get_users_with_read_receipts_in_room(
-            room_id, on_invalidate=cache_context.invalidate
-        )
-
-        # any users with pushers must be ours: they have pushers
-        for uid in users_with_receipts:
-            if uid in local_users_in_room:
-                user_ids.add(uid)
-
-        rules_by_user = yield self.bulk_get_push_rules(
-            user_ids, on_invalidate=cache_context.invalidate
-        )
-
-        rules_by_user = {k: v for k, v in rules_by_user.items() if v is not None}
-
-        return rules_by_user
-
-    @cachedList(
-        cached_method_name="get_push_rules_enabled_for_user",
-        list_name="user_ids",
-        num_args=1,
-        inlineCallbacks=True,
-    )
-    def bulk_get_push_rules_enabled(self, user_ids):
-        if not user_ids:
-            return {}
-
-        results = {user_id: {} for user_id in user_ids}
-
-        rows = yield self._simple_select_many_batch(
-            table="push_rules_enable",
-            column="user_name",
-            iterable=user_ids,
-            retcols=("user_name", "rule_id", "enabled"),
-            desc="bulk_get_push_rules_enabled",
-        )
-        for row in rows:
-            enabled = bool(row["enabled"])
-            results.setdefault(row["user_name"], {})[row["rule_id"]] = enabled
-        return results
-
-
-class PushRuleStore(PushRulesWorkerStore):
-    @defer.inlineCallbacks
-    def add_push_rule(
-        self,
-        user_id,
-        rule_id,
-        priority_class,
-        conditions,
-        actions,
-        before=None,
-        after=None,
-    ):
-        conditions_json = json.dumps(conditions)
-        actions_json = json.dumps(actions)
-        with self._push_rules_stream_id_gen.get_next() as ids:
-            stream_id, event_stream_ordering = ids
-            if before or after:
-                yield self.runInteraction(
-                    "_add_push_rule_relative_txn",
-                    self._add_push_rule_relative_txn,
-                    stream_id,
-                    event_stream_ordering,
-                    user_id,
-                    rule_id,
-                    priority_class,
-                    conditions_json,
-                    actions_json,
-                    before,
-                    after,
-                )
-            else:
-                yield self.runInteraction(
-                    "_add_push_rule_highest_priority_txn",
-                    self._add_push_rule_highest_priority_txn,
-                    stream_id,
-                    event_stream_ordering,
-                    user_id,
-                    rule_id,
-                    priority_class,
-                    conditions_json,
-                    actions_json,
-                )
-
-    def _add_push_rule_relative_txn(
-        self,
-        txn,
-        stream_id,
-        event_stream_ordering,
-        user_id,
-        rule_id,
-        priority_class,
-        conditions_json,
-        actions_json,
-        before,
-        after,
-    ):
-        # Lock the table since otherwise we'll have annoying races between the
-        # SELECT here and the UPSERT below.
-        self.database_engine.lock_table(txn, "push_rules")
-
-        relative_to_rule = before or after
-
-        res = self._simple_select_one_txn(
-            txn,
-            table="push_rules",
-            keyvalues={"user_name": user_id, "rule_id": relative_to_rule},
-            retcols=["priority_class", "priority"],
-            allow_none=True,
-        )
-
-        if not res:
-            raise RuleNotFoundException(
-                "before/after rule not found: %s" % (relative_to_rule,)
-            )
-
-        base_priority_class = res["priority_class"]
-        base_rule_priority = res["priority"]
-
-        if base_priority_class != priority_class:
-            raise InconsistentRuleException(
-                "Given priority class does not match class of relative rule"
-            )
-
-        if before:
-            # Higher priority rules are executed first, So adding a rule before
-            # a rule means giving it a higher priority than that rule.
-            new_rule_priority = base_rule_priority + 1
-        else:
-            # We increment the priority of the existing rules to make space for
-            # the new rule. Therefore if we want this rule to appear after
-            # an existing rule we give it the priority of the existing rule,
-            # and then increment the priority of the existing rule.
-            new_rule_priority = base_rule_priority
-
-        sql = (
-            "UPDATE push_rules SET priority = priority + 1"
-            " WHERE user_name = ? AND priority_class = ? AND priority >= ?"
-        )
-
-        txn.execute(sql, (user_id, priority_class, new_rule_priority))
-
-        self._upsert_push_rule_txn(
-            txn,
-            stream_id,
-            event_stream_ordering,
-            user_id,
-            rule_id,
-            priority_class,
-            new_rule_priority,
-            conditions_json,
-            actions_json,
-        )
-
-    def _add_push_rule_highest_priority_txn(
-        self,
-        txn,
-        stream_id,
-        event_stream_ordering,
-        user_id,
-        rule_id,
-        priority_class,
-        conditions_json,
-        actions_json,
-    ):
-        # Lock the table since otherwise we'll have annoying races between the
-        # SELECT here and the UPSERT below.
-        self.database_engine.lock_table(txn, "push_rules")
-
-        # find the highest priority rule in that class
-        sql = (
-            "SELECT COUNT(*), MAX(priority) FROM push_rules"
-            " WHERE user_name = ? and priority_class = ?"
-        )
-        txn.execute(sql, (user_id, priority_class))
-        res = txn.fetchall()
-        (how_many, highest_prio) = res[0]
-
-        new_prio = 0
-        if how_many > 0:
-            new_prio = highest_prio + 1
-
-        self._upsert_push_rule_txn(
-            txn,
-            stream_id,
-            event_stream_ordering,
-            user_id,
-            rule_id,
-            priority_class,
-            new_prio,
-            conditions_json,
-            actions_json,
-        )
-
-    def _upsert_push_rule_txn(
-        self,
-        txn,
-        stream_id,
-        event_stream_ordering,
-        user_id,
-        rule_id,
-        priority_class,
-        priority,
-        conditions_json,
-        actions_json,
-        update_stream=True,
-    ):
-        """Specialised version of _simple_upsert_txn that picks a push_rule_id
-        using the _push_rule_id_gen if it needs to insert the rule. It assumes
-        that the "push_rules" table is locked"""
-
-        sql = (
-            "UPDATE push_rules"
-            " SET priority_class = ?, priority = ?, conditions = ?, actions = ?"
-            " WHERE user_name = ? AND rule_id = ?"
-        )
-
-        txn.execute(
-            sql,
-            (priority_class, priority, conditions_json, actions_json, user_id, rule_id),
-        )
-
-        if txn.rowcount == 0:
-            # We didn't update a row with the given rule_id so insert one
-            push_rule_id = self._push_rule_id_gen.get_next()
-
-            self._simple_insert_txn(
-                txn,
-                table="push_rules",
-                values={
-                    "id": push_rule_id,
-                    "user_name": user_id,
-                    "rule_id": rule_id,
-                    "priority_class": priority_class,
-                    "priority": priority,
-                    "conditions": conditions_json,
-                    "actions": actions_json,
-                },
-            )
-
-        if update_stream:
-            self._insert_push_rules_update_txn(
-                txn,
-                stream_id,
-                event_stream_ordering,
-                user_id,
-                rule_id,
-                op="ADD",
-                data={
-                    "priority_class": priority_class,
-                    "priority": priority,
-                    "conditions": conditions_json,
-                    "actions": actions_json,
-                },
-            )
-
-    @defer.inlineCallbacks
-    def delete_push_rule(self, user_id, rule_id):
-        """
-        Delete a push rule. Args specify the row to be deleted and can be
-        any of the columns in the push_rule table, but below are the
-        standard ones
-
-        Args:
-            user_id (str): The matrix ID of the push rule owner
-            rule_id (str): The rule_id of the rule to be deleted
-        """
-
-        def delete_push_rule_txn(txn, stream_id, event_stream_ordering):
-            self._simple_delete_one_txn(
-                txn, "push_rules", {"user_name": user_id, "rule_id": rule_id}
-            )
-
-            self._insert_push_rules_update_txn(
-                txn, stream_id, event_stream_ordering, user_id, rule_id, op="DELETE"
-            )
-
-        with self._push_rules_stream_id_gen.get_next() as ids:
-            stream_id, event_stream_ordering = ids
-            yield self.runInteraction(
-                "delete_push_rule",
-                delete_push_rule_txn,
-                stream_id,
-                event_stream_ordering,
-            )
-
-    @defer.inlineCallbacks
-    def set_push_rule_enabled(self, user_id, rule_id, enabled):
-        with self._push_rules_stream_id_gen.get_next() as ids:
-            stream_id, event_stream_ordering = ids
-            yield self.runInteraction(
-                "_set_push_rule_enabled_txn",
-                self._set_push_rule_enabled_txn,
-                stream_id,
-                event_stream_ordering,
-                user_id,
-                rule_id,
-                enabled,
-            )
-
-    def _set_push_rule_enabled_txn(
-        self, txn, stream_id, event_stream_ordering, user_id, rule_id, enabled
-    ):
-        new_id = self._push_rules_enable_id_gen.get_next()
-        self._simple_upsert_txn(
-            txn,
-            "push_rules_enable",
-            {"user_name": user_id, "rule_id": rule_id},
-            {"enabled": 1 if enabled else 0},
-            {"id": new_id},
-        )
-
-        self._insert_push_rules_update_txn(
-            txn,
-            stream_id,
-            event_stream_ordering,
-            user_id,
-            rule_id,
-            op="ENABLE" if enabled else "DISABLE",
-        )
-
-    @defer.inlineCallbacks
-    def set_push_rule_actions(self, user_id, rule_id, actions, is_default_rule):
-        actions_json = json.dumps(actions)
-
-        def set_push_rule_actions_txn(txn, stream_id, event_stream_ordering):
-            if is_default_rule:
-                # Add a dummy rule to the rules table with the user specified
-                # actions.
-                priority_class = -1
-                priority = 1
-                self._upsert_push_rule_txn(
-                    txn,
-                    stream_id,
-                    event_stream_ordering,
-                    user_id,
-                    rule_id,
-                    priority_class,
-                    priority,
-                    "[]",
-                    actions_json,
-                    update_stream=False,
-                )
-            else:
-                self._simple_update_one_txn(
-                    txn,
-                    "push_rules",
-                    {"user_name": user_id, "rule_id": rule_id},
-                    {"actions": actions_json},
-                )
-
-            self._insert_push_rules_update_txn(
-                txn,
-                stream_id,
-                event_stream_ordering,
-                user_id,
-                rule_id,
-                op="ACTIONS",
-                data={"actions": actions_json},
-            )
-
-        with self._push_rules_stream_id_gen.get_next() as ids:
-            stream_id, event_stream_ordering = ids
-            yield self.runInteraction(
-                "set_push_rule_actions",
-                set_push_rule_actions_txn,
-                stream_id,
-                event_stream_ordering,
-            )
-
-    def _insert_push_rules_update_txn(
-        self, txn, stream_id, event_stream_ordering, user_id, rule_id, op, data=None
-    ):
-        values = {
-            "stream_id": stream_id,
-            "event_stream_ordering": event_stream_ordering,
-            "user_id": user_id,
-            "rule_id": rule_id,
-            "op": op,
-        }
-        if data is not None:
-            values.update(data)
-
-        self._simple_insert_txn(txn, "push_rules_stream", values=values)
-
-        txn.call_after(self.get_push_rules_for_user.invalidate, (user_id,))
-        txn.call_after(self.get_push_rules_enabled_for_user.invalidate, (user_id,))
-        txn.call_after(
-            self.push_rules_stream_cache.entity_has_changed, user_id, stream_id
-        )
-
-    def get_all_push_rule_updates(self, last_id, current_id, limit):
-        """Get all the push rules changes that have happend on the server"""
-        if last_id == current_id:
-            return defer.succeed([])
-
-        def get_all_push_rule_updates_txn(txn):
-            sql = (
-                "SELECT stream_id, event_stream_ordering, user_id, rule_id,"
-                " op, priority_class, priority, conditions, actions"
-                " FROM push_rules_stream"
-                " WHERE ? < stream_id AND stream_id <= ?"
-                " ORDER BY stream_id ASC LIMIT ?"
-            )
-            txn.execute(sql, (last_id, current_id, limit))
-            return txn.fetchall()
-
-        return self.runInteraction(
-            "get_all_push_rule_updates", get_all_push_rule_updates_txn
-        )
-
-    def get_push_rules_stream_token(self):
-        """Get the position of the push rules stream.
-        Returns a pair of a stream id for the push_rules stream and the
-        room stream ordering it corresponds to."""
-        return self._push_rules_stream_id_gen.get_current_token()
-
-    def get_max_push_rules_stream_id(self):
-        return self.get_push_rules_stream_token()[0]
-
 
 class RuleNotFoundException(Exception):
     pass
diff --git a/synapse/storage/relations.py b/synapse/storage/relations.py
index fcb5f2f23a..d471ec9860 100644
--- a/synapse/storage/relations.py
+++ b/synapse/storage/relations.py
@@ -17,11 +17,7 @@ import logging
 
 import attr
 
-from synapse.api.constants import RelationTypes
 from synapse.api.errors import SynapseError
-from synapse.storage._base import SQLBaseStore
-from synapse.storage.stream import generate_pagination_where_clause
-from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
 
 logger = logging.getLogger(__name__)
 
@@ -113,358 +109,3 @@ class AggregationPaginationToken(object):
 
     def as_tuple(self):
         return attr.astuple(self)
-
-
-class RelationsWorkerStore(SQLBaseStore):
-    @cached(tree=True)
-    def get_relations_for_event(
-        self,
-        event_id,
-        relation_type=None,
-        event_type=None,
-        aggregation_key=None,
-        limit=5,
-        direction="b",
-        from_token=None,
-        to_token=None,
-    ):
-        """Get a list of relations for an event, ordered by topological ordering.
-
-        Args:
-            event_id (str): Fetch events that relate to this event ID.
-            relation_type (str|None): Only fetch events with this relation
-                type, if given.
-            event_type (str|None): Only fetch events with this event type, if
-                given.
-            aggregation_key (str|None): Only fetch events with this aggregation
-                key, if given.
-            limit (int): Only fetch the most recent `limit` events.
-            direction (str): Whether to fetch the most recent first (`"b"`) or
-                the oldest first (`"f"`).
-            from_token (RelationPaginationToken|None): Fetch rows from the given
-                token, or from the start if None.
-            to_token (RelationPaginationToken|None): Fetch rows up to the given
-                token, or up to the end if None.
-
-        Returns:
-            Deferred[PaginationChunk]: List of event IDs that match relations
-            requested. The rows are of the form `{"event_id": "..."}`.
-        """
-
-        where_clause = ["relates_to_id = ?"]
-        where_args = [event_id]
-
-        if relation_type is not None:
-            where_clause.append("relation_type = ?")
-            where_args.append(relation_type)
-
-        if event_type is not None:
-            where_clause.append("type = ?")
-            where_args.append(event_type)
-
-        if aggregation_key:
-            where_clause.append("aggregation_key = ?")
-            where_args.append(aggregation_key)
-
-        pagination_clause = generate_pagination_where_clause(
-            direction=direction,
-            column_names=("topological_ordering", "stream_ordering"),
-            from_token=attr.astuple(from_token) if from_token else None,
-            to_token=attr.astuple(to_token) if to_token else None,
-            engine=self.database_engine,
-        )
-
-        if pagination_clause:
-            where_clause.append(pagination_clause)
-
-        if direction == "b":
-            order = "DESC"
-        else:
-            order = "ASC"
-
-        sql = """
-            SELECT event_id, topological_ordering, stream_ordering
-            FROM event_relations
-            INNER JOIN events USING (event_id)
-            WHERE %s
-            ORDER BY topological_ordering %s, stream_ordering %s
-            LIMIT ?
-        """ % (
-            " AND ".join(where_clause),
-            order,
-            order,
-        )
-
-        def _get_recent_references_for_event_txn(txn):
-            txn.execute(sql, where_args + [limit + 1])
-
-            last_topo_id = None
-            last_stream_id = None
-            events = []
-            for row in txn:
-                events.append({"event_id": row[0]})
-                last_topo_id = row[1]
-                last_stream_id = row[2]
-
-            next_batch = None
-            if len(events) > limit and last_topo_id and last_stream_id:
-                next_batch = RelationPaginationToken(last_topo_id, last_stream_id)
-
-            return PaginationChunk(
-                chunk=list(events[:limit]), next_batch=next_batch, prev_batch=from_token
-            )
-
-        return self.runInteraction(
-            "get_recent_references_for_event", _get_recent_references_for_event_txn
-        )
-
-    @cached(tree=True)
-    def get_aggregation_groups_for_event(
-        self,
-        event_id,
-        event_type=None,
-        limit=5,
-        direction="b",
-        from_token=None,
-        to_token=None,
-    ):
-        """Get a list of annotations on the event, grouped by event type and
-        aggregation key, sorted by count.
-
-        This is used e.g. to get the what and how many reactions have happend
-        on an event.
-
-        Args:
-            event_id (str): Fetch events that relate to this event ID.
-            event_type (str|None): Only fetch events with this event type, if
-                given.
-            limit (int): Only fetch the `limit` groups.
-            direction (str): Whether to fetch the highest count first (`"b"`) or
-                the lowest count first (`"f"`).
-            from_token (AggregationPaginationToken|None): Fetch rows from the
-                given token, or from the start if None.
-            to_token (AggregationPaginationToken|None): Fetch rows up to the
-                given token, or up to the end if None.
-
-
-        Returns:
-            Deferred[PaginationChunk]: List of groups of annotations that
-            match. Each row is a dict with `type`, `key` and `count` fields.
-        """
-
-        where_clause = ["relates_to_id = ?", "relation_type = ?"]
-        where_args = [event_id, RelationTypes.ANNOTATION]
-
-        if event_type:
-            where_clause.append("type = ?")
-            where_args.append(event_type)
-
-        having_clause = generate_pagination_where_clause(
-            direction=direction,
-            column_names=("COUNT(*)", "MAX(stream_ordering)"),
-            from_token=attr.astuple(from_token) if from_token else None,
-            to_token=attr.astuple(to_token) if to_token else None,
-            engine=self.database_engine,
-        )
-
-        if direction == "b":
-            order = "DESC"
-        else:
-            order = "ASC"
-
-        if having_clause:
-            having_clause = "HAVING " + having_clause
-        else:
-            having_clause = ""
-
-        sql = """
-            SELECT type, aggregation_key, COUNT(DISTINCT sender), MAX(stream_ordering)
-            FROM event_relations
-            INNER JOIN events USING (event_id)
-            WHERE {where_clause}
-            GROUP BY relation_type, type, aggregation_key
-            {having_clause}
-            ORDER BY COUNT(*) {order}, MAX(stream_ordering) {order}
-            LIMIT ?
-        """.format(
-            where_clause=" AND ".join(where_clause),
-            order=order,
-            having_clause=having_clause,
-        )
-
-        def _get_aggregation_groups_for_event_txn(txn):
-            txn.execute(sql, where_args + [limit + 1])
-
-            next_batch = None
-            events = []
-            for row in txn:
-                events.append({"type": row[0], "key": row[1], "count": row[2]})
-                next_batch = AggregationPaginationToken(row[2], row[3])
-
-            if len(events) <= limit:
-                next_batch = None
-
-            return PaginationChunk(
-                chunk=list(events[:limit]), next_batch=next_batch, prev_batch=from_token
-            )
-
-        return self.runInteraction(
-            "get_aggregation_groups_for_event", _get_aggregation_groups_for_event_txn
-        )
-
-    @cachedInlineCallbacks()
-    def get_applicable_edit(self, event_id):
-        """Get the most recent edit (if any) that has happened for the given
-        event.
-
-        Correctly handles checking whether edits were allowed to happen.
-
-        Args:
-            event_id (str): The original event ID
-
-        Returns:
-            Deferred[EventBase|None]: Returns the most recent edit, if any.
-        """
-
-        # We only allow edits for `m.room.message` events that have the same sender
-        # and event type. We can't assert these things during regular event auth so
-        # we have to do the checks post hoc.
-
-        # Fetches latest edit that has the same type and sender as the
-        # original, and is an `m.room.message`.
-        sql = """
-            SELECT edit.event_id FROM events AS edit
-            INNER JOIN event_relations USING (event_id)
-            INNER JOIN events AS original ON
-                original.event_id = relates_to_id
-                AND edit.type = original.type
-                AND edit.sender = original.sender
-            WHERE
-                relates_to_id = ?
-                AND relation_type = ?
-                AND edit.type = 'm.room.message'
-            ORDER by edit.origin_server_ts DESC, edit.event_id DESC
-            LIMIT 1
-        """
-
-        def _get_applicable_edit_txn(txn):
-            txn.execute(sql, (event_id, RelationTypes.REPLACE))
-            row = txn.fetchone()
-            if row:
-                return row[0]
-
-        edit_id = yield self.runInteraction(
-            "get_applicable_edit", _get_applicable_edit_txn
-        )
-
-        if not edit_id:
-            return
-
-        edit_event = yield self.get_event(edit_id, allow_none=True)
-        return edit_event
-
-    def has_user_annotated_event(self, parent_id, event_type, aggregation_key, sender):
-        """Check if a user has already annotated an event with the same key
-        (e.g. already liked an event).
-
-        Args:
-            parent_id (str): The event being annotated
-            event_type (str): The event type of the annotation
-            aggregation_key (str): The aggregation key of the annotation
-            sender (str): The sender of the annotation
-
-        Returns:
-            Deferred[bool]
-        """
-
-        sql = """
-            SELECT 1 FROM event_relations
-            INNER JOIN events USING (event_id)
-            WHERE
-                relates_to_id = ?
-                AND relation_type = ?
-                AND type = ?
-                AND sender = ?
-                AND aggregation_key = ?
-            LIMIT 1;
-        """
-
-        def _get_if_user_has_annotated_event(txn):
-            txn.execute(
-                sql,
-                (
-                    parent_id,
-                    RelationTypes.ANNOTATION,
-                    event_type,
-                    sender,
-                    aggregation_key,
-                ),
-            )
-
-            return bool(txn.fetchone())
-
-        return self.runInteraction(
-            "get_if_user_has_annotated_event", _get_if_user_has_annotated_event
-        )
-
-
-class RelationsStore(RelationsWorkerStore):
-    def _handle_event_relations(self, txn, event):
-        """Handles inserting relation data during peristence of events
-
-        Args:
-            txn
-            event (EventBase)
-        """
-        relation = event.content.get("m.relates_to")
-        if not relation:
-            # No relations
-            return
-
-        rel_type = relation.get("rel_type")
-        if rel_type not in (
-            RelationTypes.ANNOTATION,
-            RelationTypes.REFERENCE,
-            RelationTypes.REPLACE,
-        ):
-            # Unknown relation type
-            return
-
-        parent_id = relation.get("event_id")
-        if not parent_id:
-            # Invalid relation
-            return
-
-        aggregation_key = relation.get("key")
-
-        self._simple_insert_txn(
-            txn,
-            table="event_relations",
-            values={
-                "event_id": event.event_id,
-                "relates_to_id": parent_id,
-                "relation_type": rel_type,
-                "aggregation_key": aggregation_key,
-            },
-        )
-
-        txn.call_after(self.get_relations_for_event.invalidate_many, (parent_id,))
-        txn.call_after(
-            self.get_aggregation_groups_for_event.invalidate_many, (parent_id,)
-        )
-
-        if rel_type == RelationTypes.REPLACE:
-            txn.call_after(self.get_applicable_edit.invalidate, (parent_id,))
-
-    def _handle_redaction(self, txn, redacted_event_id):
-        """Handles receiving a redaction and checking whether we need to remove
-        any redacted relations from the database.
-
-        Args:
-            txn
-            redacted_event_id (str): The event that was redacted.
-        """
-
-        self._simple_delete_txn(
-            txn, table="event_relations", keyvalues={"event_id": redacted_event_id}
-        )
diff --git a/synapse/storage/roommember.py b/synapse/storage/roommember.py
index eecb276465..8c4a83a840 100644
--- a/synapse/storage/roommember.py
+++ b/synapse/storage/roommember.py
@@ -17,22 +17,6 @@
 import logging
 from collections import namedtuple
 
-from six import iteritems, itervalues
-
-from canonicaljson import json
-
-from twisted.internet import defer
-
-from synapse.api.constants import EventTypes, Membership
-from synapse.metrics.background_process_metrics import run_as_background_process
-from synapse.storage._base import LoggingTransaction
-from synapse.storage.events_worker import EventsWorkerStore
-from synapse.types import get_domain_from_id
-from synapse.util.async_helpers import Linearizer
-from synapse.util.caches import intern_string
-from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
-from synapse.util.stringutils import to_ascii
-
 logger = logging.getLogger(__name__)
 
 
@@ -53,998 +37,3 @@ ProfileInfo = namedtuple("ProfileInfo", ("avatar_url", "display_name"))
 # a given membership type, suitable for use in calculating heroes for a room.
 # "count" points to the total numberr of users of a given membership type.
 MemberSummary = namedtuple("MemberSummary", ("members", "count"))
-
-_MEMBERSHIP_PROFILE_UPDATE_NAME = "room_membership_profile_update"
-_CURRENT_STATE_MEMBERSHIP_UPDATE_NAME = "current_state_events_membership"
-
-
-class RoomMemberWorkerStore(EventsWorkerStore):
-    def __init__(self, db_conn, hs):
-        super(RoomMemberWorkerStore, self).__init__(db_conn, hs)
-
-        # Is the current_state_events.membership up to date? Or is the
-        # background update still running?
-        self._current_state_events_membership_up_to_date = False
-
-        txn = LoggingTransaction(
-            db_conn.cursor(),
-            name="_check_safe_current_state_events_membership_updated",
-            database_engine=self.database_engine,
-        )
-        self._check_safe_current_state_events_membership_updated_txn(txn)
-        txn.close()
-
-    def _check_safe_current_state_events_membership_updated_txn(self, txn):
-        """Checks if it is safe to assume the new current_state_events
-        membership column is up to date
-        """
-
-        pending_update = self._simple_select_one_txn(
-            txn,
-            table="background_updates",
-            keyvalues={"update_name": _CURRENT_STATE_MEMBERSHIP_UPDATE_NAME},
-            retcols=["update_name"],
-            allow_none=True,
-        )
-
-        self._current_state_events_membership_up_to_date = not pending_update
-
-        # If the update is still running, reschedule to run.
-        if pending_update:
-            self._clock.call_later(
-                15.0,
-                run_as_background_process,
-                "_check_safe_current_state_events_membership_updated",
-                self.runInteraction,
-                "_check_safe_current_state_events_membership_updated",
-                self._check_safe_current_state_events_membership_updated_txn,
-            )
-
-    @cachedInlineCallbacks(max_entries=100000, iterable=True, cache_context=True)
-    def get_hosts_in_room(self, room_id, cache_context):
-        """Returns the set of all hosts currently in the room
-        """
-        user_ids = yield self.get_users_in_room(
-            room_id, on_invalidate=cache_context.invalidate
-        )
-        hosts = frozenset(get_domain_from_id(user_id) for user_id in user_ids)
-        return hosts
-
-    @cached(max_entries=100000, iterable=True)
-    def get_users_in_room(self, room_id):
-        def f(txn):
-            # If we can assume current_state_events.membership is up to date
-            # then we can avoid a join, which is a Very Good Thing given how
-            # frequently this function gets called.
-            if self._current_state_events_membership_up_to_date:
-                sql = """
-                    SELECT state_key FROM current_state_events
-                    WHERE type = 'm.room.member' AND room_id = ? AND membership = ?
-                """
-            else:
-                sql = """
-                    SELECT state_key FROM room_memberships as m
-                    INNER JOIN current_state_events as c
-                    ON m.event_id = c.event_id
-                    AND m.room_id = c.room_id
-                    AND m.user_id = c.state_key
-                    WHERE c.type = 'm.room.member' AND c.room_id = ? AND m.membership = ?
-                """
-
-            txn.execute(sql, (room_id, Membership.JOIN))
-            return [to_ascii(r[0]) for r in txn]
-
-        return self.runInteraction("get_users_in_room", f)
-
-    @cached(max_entries=100000)
-    def get_room_summary(self, room_id):
-        """ Get the details of a room roughly suitable for use by the room
-        summary extension to /sync. Useful when lazy loading room members.
-        Args:
-            room_id (str): The room ID to query
-        Returns:
-            Deferred[dict[str, MemberSummary]:
-                dict of membership states, pointing to a MemberSummary named tuple.
-        """
-
-        def _get_room_summary_txn(txn):
-            # first get counts.
-            # We do this all in one transaction to keep the cache small.
-            # FIXME: get rid of this when we have room_stats
-
-            # If we can assume current_state_events.membership is up to date
-            # then we can avoid a join, which is a Very Good Thing given how
-            # frequently this function gets called.
-            if self._current_state_events_membership_up_to_date:
-                # Note, rejected events will have a null membership field, so
-                # we we manually filter them out.
-                sql = """
-                    SELECT count(*), membership FROM current_state_events
-                    WHERE type = 'm.room.member' AND room_id = ?
-                        AND membership IS NOT NULL
-                    GROUP BY membership
-                """
-            else:
-                sql = """
-                    SELECT count(*), m.membership FROM room_memberships as m
-                    INNER JOIN current_state_events as c
-                    ON m.event_id = c.event_id
-                    AND m.room_id = c.room_id
-                    AND m.user_id = c.state_key
-                    WHERE c.type = 'm.room.member' AND c.room_id = ?
-                    GROUP BY m.membership
-                """
-
-            txn.execute(sql, (room_id,))
-            res = {}
-            for count, membership in txn:
-                summary = res.setdefault(to_ascii(membership), MemberSummary([], count))
-
-            # we order by membership and then fairly arbitrarily by event_id so
-            # heroes are consistent
-            if self._current_state_events_membership_up_to_date:
-                # Note, rejected events will have a null membership field, so
-                # we we manually filter them out.
-                sql = """
-                    SELECT state_key, membership, event_id
-                    FROM current_state_events
-                    WHERE type = 'm.room.member' AND room_id = ?
-                        AND membership IS NOT NULL
-                    ORDER BY
-                        CASE membership WHEN ? THEN 1 WHEN ? THEN 2 ELSE 3 END ASC,
-                        event_id ASC
-                    LIMIT ?
-                """
-            else:
-                sql = """
-                    SELECT c.state_key, m.membership, c.event_id
-                    FROM room_memberships as m
-                    INNER JOIN current_state_events as c USING (room_id, event_id)
-                    WHERE c.type = 'm.room.member' AND c.room_id = ?
-                    ORDER BY
-                        CASE m.membership WHEN ? THEN 1 WHEN ? THEN 2 ELSE 3 END ASC,
-                        c.event_id ASC
-                    LIMIT ?
-                """
-
-            # 6 is 5 (number of heroes) plus 1, in case one of them is the calling user.
-            txn.execute(sql, (room_id, Membership.JOIN, Membership.INVITE, 6))
-            for user_id, membership, event_id in txn:
-                summary = res[to_ascii(membership)]
-                # we will always have a summary for this membership type at this
-                # point given the summary currently contains the counts.
-                members = summary.members
-                members.append((to_ascii(user_id), to_ascii(event_id)))
-
-            return res
-
-        return self.runInteraction("get_room_summary", _get_room_summary_txn)
-
-    def _get_user_counts_in_room_txn(self, txn, room_id):
-        """
-        Get the user count in a room by membership.
-
-        Args:
-            room_id (str)
-            membership (Membership)
-
-        Returns:
-            Deferred[int]
-        """
-        sql = """
-        SELECT m.membership, count(*) FROM room_memberships as m
-            INNER JOIN current_state_events as c USING(event_id)
-            WHERE c.type = 'm.room.member' AND c.room_id = ?
-            GROUP BY m.membership
-        """
-
-        txn.execute(sql, (room_id,))
-        return {row[0]: row[1] for row in txn}
-
-    @cached()
-    def get_invited_rooms_for_user(self, user_id):
-        """ Get all the rooms the user is invited to
-        Args:
-            user_id (str): The user ID.
-        Returns:
-            A deferred list of RoomsForUser.
-        """
-
-        return self.get_rooms_for_user_where_membership_is(user_id, [Membership.INVITE])
-
-    @defer.inlineCallbacks
-    def get_invite_for_user_in_room(self, user_id, room_id):
-        """Gets the invite for the given user and room
-
-        Args:
-            user_id (str)
-            room_id (str)
-
-        Returns:
-            Deferred: Resolves to either a RoomsForUser or None if no invite was
-                found.
-        """
-        invites = yield self.get_invited_rooms_for_user(user_id)
-        for invite in invites:
-            if invite.room_id == room_id:
-                return invite
-        return None
-
-    @defer.inlineCallbacks
-    def get_rooms_for_user_where_membership_is(self, user_id, membership_list):
-        """ Get all the rooms for this user where the membership for this user
-        matches one in the membership list.
-
-        Filters out forgotten rooms.
-
-        Args:
-            user_id (str): The user ID.
-            membership_list (list): A list of synapse.api.constants.Membership
-            values which the user must be in.
-
-        Returns:
-            Deferred[list[RoomsForUser]]
-        """
-        if not membership_list:
-            return defer.succeed(None)
-
-        rooms = yield self.runInteraction(
-            "get_rooms_for_user_where_membership_is",
-            self._get_rooms_for_user_where_membership_is_txn,
-            user_id,
-            membership_list,
-        )
-
-        # Now we filter out forgotten rooms
-        forgotten_rooms = yield self.get_forgotten_rooms_for_user(user_id)
-        return [room for room in rooms if room.room_id not in forgotten_rooms]
-
-    def _get_rooms_for_user_where_membership_is_txn(
-        self, txn, user_id, membership_list
-    ):
-
-        do_invite = Membership.INVITE in membership_list
-        membership_list = [m for m in membership_list if m != Membership.INVITE]
-
-        results = []
-        if membership_list:
-            if self._current_state_events_membership_up_to_date:
-                sql = """
-                    SELECT room_id, e.sender, c.membership, event_id, e.stream_ordering
-                    FROM current_state_events AS c
-                    INNER JOIN events AS e USING (room_id, event_id)
-                    WHERE
-                        c.type = 'm.room.member'
-                        AND state_key = ?
-                        AND c.membership IN (%s)
-                """ % (
-                    ",".join("?" * len(membership_list))
-                )
-            else:
-                sql = """
-                    SELECT room_id, e.sender, m.membership, event_id, e.stream_ordering
-                    FROM current_state_events AS c
-                    INNER JOIN room_memberships AS m USING (room_id, event_id)
-                    INNER JOIN events AS e USING (room_id, event_id)
-                    WHERE
-                        c.type = 'm.room.member'
-                        AND state_key = ?
-                        AND m.membership IN (%s)
-                """ % (
-                    ",".join("?" * len(membership_list))
-                )
-
-            txn.execute(sql, (user_id, *membership_list))
-            results = [RoomsForUser(**r) for r in self.cursor_to_dict(txn)]
-
-        if do_invite:
-            sql = (
-                "SELECT i.room_id, inviter, i.event_id, e.stream_ordering"
-                " FROM local_invites as i"
-                " INNER JOIN events as e USING (event_id)"
-                " WHERE invitee = ? AND locally_rejected is NULL"
-                " AND replaced_by is NULL"
-            )
-
-            txn.execute(sql, (user_id,))
-            results.extend(
-                RoomsForUser(
-                    room_id=r["room_id"],
-                    sender=r["inviter"],
-                    event_id=r["event_id"],
-                    stream_ordering=r["stream_ordering"],
-                    membership=Membership.INVITE,
-                )
-                for r in self.cursor_to_dict(txn)
-            )
-
-        return results
-
-    @cachedInlineCallbacks(max_entries=500000, iterable=True)
-    def get_rooms_for_user_with_stream_ordering(self, user_id):
-        """Returns a set of room_ids the user is currently joined to
-
-        Args:
-            user_id (str)
-
-        Returns:
-            Deferred[frozenset[GetRoomsForUserWithStreamOrdering]]: Returns
-            the rooms the user is in currently, along with the stream ordering
-            of the most recent join for that user and room.
-        """
-        rooms = yield self.get_rooms_for_user_where_membership_is(
-            user_id, membership_list=[Membership.JOIN]
-        )
-        return frozenset(
-            GetRoomsForUserWithStreamOrdering(r.room_id, r.stream_ordering)
-            for r in rooms
-        )
-
-    @defer.inlineCallbacks
-    def get_rooms_for_user(self, user_id, on_invalidate=None):
-        """Returns a set of room_ids the user is currently joined to
-        """
-        rooms = yield self.get_rooms_for_user_with_stream_ordering(
-            user_id, on_invalidate=on_invalidate
-        )
-        return frozenset(r.room_id for r in rooms)
-
-    @cachedInlineCallbacks(max_entries=500000, cache_context=True, iterable=True)
-    def get_users_who_share_room_with_user(self, user_id, cache_context):
-        """Returns the set of users who share a room with `user_id`
-        """
-        room_ids = yield self.get_rooms_for_user(
-            user_id, on_invalidate=cache_context.invalidate
-        )
-
-        user_who_share_room = set()
-        for room_id in room_ids:
-            user_ids = yield self.get_users_in_room(
-                room_id, on_invalidate=cache_context.invalidate
-            )
-            user_who_share_room.update(user_ids)
-
-        return user_who_share_room
-
-    @defer.inlineCallbacks
-    def get_joined_users_from_context(self, event, context):
-        state_group = context.state_group
-        if not state_group:
-            # If state_group is None it means it has yet to be assigned a
-            # state group, i.e. we need to make sure that calls with a state_group
-            # of None don't hit previous cached calls with a None state_group.
-            # To do this we set the state_group to a new object as object() != object()
-            state_group = object()
-
-        current_state_ids = yield context.get_current_state_ids(self)
-        result = yield self._get_joined_users_from_context(
-            event.room_id, state_group, current_state_ids, event=event, context=context
-        )
-        return result
-
-    def get_joined_users_from_state(self, room_id, state_entry):
-        state_group = state_entry.state_group
-        if not state_group:
-            # If state_group is None it means it has yet to be assigned a
-            # state group, i.e. we need to make sure that calls with a state_group
-            # of None don't hit previous cached calls with a None state_group.
-            # To do this we set the state_group to a new object as object() != object()
-            state_group = object()
-
-        return self._get_joined_users_from_context(
-            room_id, state_group, state_entry.state, context=state_entry
-        )
-
-    @cachedInlineCallbacks(
-        num_args=2, cache_context=True, iterable=True, max_entries=100000
-    )
-    def _get_joined_users_from_context(
-        self,
-        room_id,
-        state_group,
-        current_state_ids,
-        cache_context,
-        event=None,
-        context=None,
-    ):
-        # We don't use `state_group`, it's there so that we can cache based
-        # on it. However, it's important that it's never None, since two current_states
-        # with a state_group of None are likely to be different.
-        # See bulk_get_push_rules_for_room for how we work around this.
-        assert state_group is not None
-
-        users_in_room = {}
-        member_event_ids = [
-            e_id
-            for key, e_id in iteritems(current_state_ids)
-            if key[0] == EventTypes.Member
-        ]
-
-        if context is not None:
-            # If we have a context with a delta from a previous state group,
-            # check if we also have the result from the previous group in cache.
-            # If we do then we can reuse that result and simply update it with
-            # any membership changes in `delta_ids`
-            if context.prev_group and context.delta_ids:
-                prev_res = self._get_joined_users_from_context.cache.get(
-                    (room_id, context.prev_group), None
-                )
-                if prev_res and isinstance(prev_res, dict):
-                    users_in_room = dict(prev_res)
-                    member_event_ids = [
-                        e_id
-                        for key, e_id in iteritems(context.delta_ids)
-                        if key[0] == EventTypes.Member
-                    ]
-                    for etype, state_key in context.delta_ids:
-                        users_in_room.pop(state_key, None)
-
-        # We check if we have any of the member event ids in the event cache
-        # before we ask the DB
-
-        # We don't update the event cache hit ratio as it completely throws off
-        # the hit ratio counts. After all, we don't populate the cache if we
-        # miss it here
-        event_map = self._get_events_from_cache(
-            member_event_ids, allow_rejected=False, update_metrics=False
-        )
-
-        missing_member_event_ids = []
-        for event_id in member_event_ids:
-            ev_entry = event_map.get(event_id)
-            if ev_entry:
-                if ev_entry.event.membership == Membership.JOIN:
-                    users_in_room[to_ascii(ev_entry.event.state_key)] = ProfileInfo(
-                        display_name=to_ascii(
-                            ev_entry.event.content.get("displayname", None)
-                        ),
-                        avatar_url=to_ascii(
-                            ev_entry.event.content.get("avatar_url", None)
-                        ),
-                    )
-            else:
-                missing_member_event_ids.append(event_id)
-
-        if missing_member_event_ids:
-            rows = yield self._simple_select_many_batch(
-                table="room_memberships",
-                column="event_id",
-                iterable=missing_member_event_ids,
-                retcols=("user_id", "display_name", "avatar_url"),
-                keyvalues={"membership": Membership.JOIN},
-                batch_size=500,
-                desc="_get_joined_users_from_context",
-            )
-
-            users_in_room.update(
-                {
-                    to_ascii(row["user_id"]): ProfileInfo(
-                        avatar_url=to_ascii(row["avatar_url"]),
-                        display_name=to_ascii(row["display_name"]),
-                    )
-                    for row in rows
-                }
-            )
-
-        if event is not None and event.type == EventTypes.Member:
-            if event.membership == Membership.JOIN:
-                if event.event_id in member_event_ids:
-                    users_in_room[to_ascii(event.state_key)] = ProfileInfo(
-                        display_name=to_ascii(event.content.get("displayname", None)),
-                        avatar_url=to_ascii(event.content.get("avatar_url", None)),
-                    )
-
-        return users_in_room
-
-    @cachedInlineCallbacks(max_entries=10000)
-    def is_host_joined(self, room_id, host):
-        if "%" in host or "_" in host:
-            raise Exception("Invalid host name")
-
-        sql = """
-            SELECT state_key FROM current_state_events AS c
-            INNER JOIN room_memberships AS m USING (event_id)
-            WHERE m.membership = 'join'
-                AND type = 'm.room.member'
-                AND c.room_id = ?
-                AND state_key LIKE ?
-            LIMIT 1
-        """
-
-        # We do need to be careful to ensure that host doesn't have any wild cards
-        # in it, but we checked above for known ones and we'll check below that
-        # the returned user actually has the correct domain.
-        like_clause = "%:" + host
-
-        rows = yield self._execute("is_host_joined", None, sql, room_id, like_clause)
-
-        if not rows:
-            return False
-
-        user_id = rows[0][0]
-        if get_domain_from_id(user_id) != host:
-            # This can only happen if the host name has something funky in it
-            raise Exception("Invalid host name")
-
-        return True
-
-    @cachedInlineCallbacks()
-    def was_host_joined(self, room_id, host):
-        """Check whether the server is or ever was in the room.
-
-        Args:
-            room_id (str)
-            host (str)
-
-        Returns:
-            Deferred: Resolves to True if the host is/was in the room, otherwise
-            False.
-        """
-        if "%" in host or "_" in host:
-            raise Exception("Invalid host name")
-
-        sql = """
-            SELECT user_id FROM room_memberships
-            WHERE room_id = ?
-                AND user_id LIKE ?
-                AND membership = 'join'
-            LIMIT 1
-        """
-
-        # We do need to be careful to ensure that host doesn't have any wild cards
-        # in it, but we checked above for known ones and we'll check below that
-        # the returned user actually has the correct domain.
-        like_clause = "%:" + host
-
-        rows = yield self._execute("was_host_joined", None, sql, room_id, like_clause)
-
-        if not rows:
-            return False
-
-        user_id = rows[0][0]
-        if get_domain_from_id(user_id) != host:
-            # This can only happen if the host name has something funky in it
-            raise Exception("Invalid host name")
-
-        return True
-
-    def get_joined_hosts(self, room_id, state_entry):
-        state_group = state_entry.state_group
-        if not state_group:
-            # If state_group is None it means it has yet to be assigned a
-            # state group, i.e. we need to make sure that calls with a state_group
-            # of None don't hit previous cached calls with a None state_group.
-            # To do this we set the state_group to a new object as object() != object()
-            state_group = object()
-
-        return self._get_joined_hosts(
-            room_id, state_group, state_entry.state, state_entry=state_entry
-        )
-
-    @cachedInlineCallbacks(num_args=2, max_entries=10000, iterable=True)
-    # @defer.inlineCallbacks
-    def _get_joined_hosts(self, room_id, state_group, current_state_ids, state_entry):
-        # We don't use `state_group`, its there so that we can cache based
-        # on it. However, its important that its never None, since two current_state's
-        # with a state_group of None are likely to be different.
-        # See bulk_get_push_rules_for_room for how we work around this.
-        assert state_group is not None
-
-        cache = self._get_joined_hosts_cache(room_id)
-        joined_hosts = yield cache.get_destinations(state_entry)
-
-        return joined_hosts
-
-    @cached(max_entries=10000)
-    def _get_joined_hosts_cache(self, room_id):
-        return _JoinedHostsCache(self, room_id)
-
-    @cachedInlineCallbacks(num_args=2)
-    def did_forget(self, user_id, room_id):
-        """Returns whether user_id has elected to discard history for room_id.
-
-        Returns False if they have since re-joined."""
-
-        def f(txn):
-            sql = (
-                "SELECT"
-                "  COUNT(*)"
-                " FROM"
-                "  room_memberships"
-                " WHERE"
-                "  user_id = ?"
-                " AND"
-                "  room_id = ?"
-                " AND"
-                "  forgotten = 0"
-            )
-            txn.execute(sql, (user_id, room_id))
-            rows = txn.fetchall()
-            return rows[0][0]
-
-        count = yield self.runInteraction("did_forget_membership", f)
-        return count == 0
-
-    @cached()
-    def get_forgotten_rooms_for_user(self, user_id):
-        """Gets all rooms the user has forgotten.
-
-        Args:
-            user_id (str)
-
-        Returns:
-            Deferred[set[str]]
-        """
-
-        def _get_forgotten_rooms_for_user_txn(txn):
-            # This is a slightly convoluted query that first looks up all rooms
-            # that the user has forgotten in the past, then rechecks that list
-            # to see if any have subsequently been updated. This is done so that
-            # we can use a partial index on `forgotten = 1` on the assumption
-            # that few users will actually forget many rooms.
-            #
-            # Note that a room is considered "forgotten" if *all* membership
-            # events for that user and room have the forgotten field set (as
-            # when a user forgets a room we update all rows for that user and
-            # room, not just the current one).
-            sql = """
-                SELECT room_id, (
-                    SELECT count(*) FROM room_memberships
-                    WHERE room_id = m.room_id AND user_id = m.user_id AND forgotten = 0
-                ) AS count
-                FROM room_memberships AS m
-                WHERE user_id = ? AND forgotten = 1
-                GROUP BY room_id, user_id;
-            """
-            txn.execute(sql, (user_id,))
-            return set(row[0] for row in txn if row[1] == 0)
-
-        return self.runInteraction(
-            "get_forgotten_rooms_for_user", _get_forgotten_rooms_for_user_txn
-        )
-
-    @defer.inlineCallbacks
-    def get_rooms_user_has_been_in(self, user_id):
-        """Get all rooms that the user has ever been in.
-
-        Args:
-            user_id (str)
-
-        Returns:
-            Deferred[set[str]]: Set of room IDs.
-        """
-
-        room_ids = yield self._simple_select_onecol(
-            table="room_memberships",
-            keyvalues={"membership": Membership.JOIN, "user_id": user_id},
-            retcol="room_id",
-            desc="get_rooms_user_has_been_in",
-        )
-
-        return set(room_ids)
-
-
-class RoomMemberStore(RoomMemberWorkerStore):
-    def __init__(self, db_conn, hs):
-        super(RoomMemberStore, self).__init__(db_conn, hs)
-        self.register_background_update_handler(
-            _MEMBERSHIP_PROFILE_UPDATE_NAME, self._background_add_membership_profile
-        )
-        self.register_background_update_handler(
-            _CURRENT_STATE_MEMBERSHIP_UPDATE_NAME,
-            self._background_current_state_membership,
-        )
-        self.register_background_index_update(
-            "room_membership_forgotten_idx",
-            index_name="room_memberships_user_room_forgotten",
-            table="room_memberships",
-            columns=["user_id", "room_id"],
-            where_clause="forgotten = 1",
-        )
-
-    def _store_room_members_txn(self, txn, events, backfilled):
-        """Store a room member in the database.
-        """
-        self._simple_insert_many_txn(
-            txn,
-            table="room_memberships",
-            values=[
-                {
-                    "event_id": event.event_id,
-                    "user_id": event.state_key,
-                    "sender": event.user_id,
-                    "room_id": event.room_id,
-                    "membership": event.membership,
-                    "display_name": event.content.get("displayname", None),
-                    "avatar_url": event.content.get("avatar_url", None),
-                }
-                for event in events
-            ],
-        )
-
-        for event in events:
-            txn.call_after(
-                self._membership_stream_cache.entity_has_changed,
-                event.state_key,
-                event.internal_metadata.stream_ordering,
-            )
-            txn.call_after(
-                self.get_invited_rooms_for_user.invalidate, (event.state_key,)
-            )
-
-            # We update the local_invites table only if the event is "current",
-            # i.e., its something that has just happened. If the event is an
-            # outlier it is only current if its an "out of band membership",
-            # like a remote invite or a rejection of a remote invite.
-            is_new_state = not backfilled and (
-                not event.internal_metadata.is_outlier()
-                or event.internal_metadata.is_out_of_band_membership()
-            )
-            is_mine = self.hs.is_mine_id(event.state_key)
-            if is_new_state and is_mine:
-                if event.membership == Membership.INVITE:
-                    self._simple_insert_txn(
-                        txn,
-                        table="local_invites",
-                        values={
-                            "event_id": event.event_id,
-                            "invitee": event.state_key,
-                            "inviter": event.sender,
-                            "room_id": event.room_id,
-                            "stream_id": event.internal_metadata.stream_ordering,
-                        },
-                    )
-                else:
-                    sql = (
-                        "UPDATE local_invites SET stream_id = ?, replaced_by = ? WHERE"
-                        " room_id = ? AND invitee = ? AND locally_rejected is NULL"
-                        " AND replaced_by is NULL"
-                    )
-
-                    txn.execute(
-                        sql,
-                        (
-                            event.internal_metadata.stream_ordering,
-                            event.event_id,
-                            event.room_id,
-                            event.state_key,
-                        ),
-                    )
-
-    @defer.inlineCallbacks
-    def locally_reject_invite(self, user_id, room_id):
-        sql = (
-            "UPDATE local_invites SET stream_id = ?, locally_rejected = ? WHERE"
-            " room_id = ? AND invitee = ? AND locally_rejected is NULL"
-            " AND replaced_by is NULL"
-        )
-
-        def f(txn, stream_ordering):
-            txn.execute(sql, (stream_ordering, True, room_id, user_id))
-
-        with self._stream_id_gen.get_next() as stream_ordering:
-            yield self.runInteraction("locally_reject_invite", f, stream_ordering)
-
-    def forget(self, user_id, room_id):
-        """Indicate that user_id wishes to discard history for room_id."""
-
-        def f(txn):
-            sql = (
-                "UPDATE"
-                "  room_memberships"
-                " SET"
-                "  forgotten = 1"
-                " WHERE"
-                "  user_id = ?"
-                " AND"
-                "  room_id = ?"
-            )
-            txn.execute(sql, (user_id, room_id))
-
-            self._invalidate_cache_and_stream(txn, self.did_forget, (user_id, room_id))
-            self._invalidate_cache_and_stream(
-                txn, self.get_forgotten_rooms_for_user, (user_id,)
-            )
-
-        return self.runInteraction("forget_membership", f)
-
-    @defer.inlineCallbacks
-    def _background_add_membership_profile(self, progress, batch_size):
-        target_min_stream_id = progress.get(
-            "target_min_stream_id_inclusive", self._min_stream_order_on_start
-        )
-        max_stream_id = progress.get(
-            "max_stream_id_exclusive", self._stream_order_on_start + 1
-        )
-
-        INSERT_CLUMP_SIZE = 1000
-
-        def add_membership_profile_txn(txn):
-            sql = """
-                SELECT stream_ordering, event_id, events.room_id, event_json.json
-                FROM events
-                INNER JOIN event_json USING (event_id)
-                INNER JOIN room_memberships USING (event_id)
-                WHERE ? <= stream_ordering AND stream_ordering < ?
-                AND type = 'm.room.member'
-                ORDER BY stream_ordering DESC
-                LIMIT ?
-            """
-
-            txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
-
-            rows = self.cursor_to_dict(txn)
-            if not rows:
-                return 0
-
-            min_stream_id = rows[-1]["stream_ordering"]
-
-            to_update = []
-            for row in rows:
-                event_id = row["event_id"]
-                room_id = row["room_id"]
-                try:
-                    event_json = json.loads(row["json"])
-                    content = event_json["content"]
-                except Exception:
-                    continue
-
-                display_name = content.get("displayname", None)
-                avatar_url = content.get("avatar_url", None)
-
-                if display_name or avatar_url:
-                    to_update.append((display_name, avatar_url, event_id, room_id))
-
-            to_update_sql = """
-                UPDATE room_memberships SET display_name = ?, avatar_url = ?
-                WHERE event_id = ? AND room_id = ?
-            """
-            for index in range(0, len(to_update), INSERT_CLUMP_SIZE):
-                clump = to_update[index : index + INSERT_CLUMP_SIZE]
-                txn.executemany(to_update_sql, clump)
-
-            progress = {
-                "target_min_stream_id_inclusive": target_min_stream_id,
-                "max_stream_id_exclusive": min_stream_id,
-            }
-
-            self._background_update_progress_txn(
-                txn, _MEMBERSHIP_PROFILE_UPDATE_NAME, progress
-            )
-
-            return len(rows)
-
-        result = yield self.runInteraction(
-            _MEMBERSHIP_PROFILE_UPDATE_NAME, add_membership_profile_txn
-        )
-
-        if not result:
-            yield self._end_background_update(_MEMBERSHIP_PROFILE_UPDATE_NAME)
-
-        return result
-
-    @defer.inlineCallbacks
-    def _background_current_state_membership(self, progress, batch_size):
-        """Update the new membership column on current_state_events.
-
-        This works by iterating over all rooms in alphebetical order.
-        """
-
-        def _background_current_state_membership_txn(txn, last_processed_room):
-            processed = 0
-            while processed < batch_size:
-                txn.execute(
-                    """
-                        SELECT MIN(room_id) FROM current_state_events WHERE room_id > ?
-                    """,
-                    (last_processed_room,),
-                )
-                row = txn.fetchone()
-                if not row or not row[0]:
-                    return processed, True
-
-                next_room, = row
-
-                sql = """
-                    UPDATE current_state_events
-                    SET membership = (
-                        SELECT membership FROM room_memberships
-                        WHERE event_id = current_state_events.event_id
-                    )
-                    WHERE room_id = ?
-                """
-                txn.execute(sql, (next_room,))
-                processed += txn.rowcount
-
-                last_processed_room = next_room
-
-            self._background_update_progress_txn(
-                txn,
-                _CURRENT_STATE_MEMBERSHIP_UPDATE_NAME,
-                {"last_processed_room": last_processed_room},
-            )
-
-            return processed, False
-
-        # If we haven't got a last processed room then just use the empty
-        # string, which will compare before all room IDs correctly.
-        last_processed_room = progress.get("last_processed_room", "")
-
-        row_count, finished = yield self.runInteraction(
-            "_background_current_state_membership_update",
-            _background_current_state_membership_txn,
-            last_processed_room,
-        )
-
-        if finished:
-            yield self._end_background_update(_CURRENT_STATE_MEMBERSHIP_UPDATE_NAME)
-
-        return row_count
-
-
-class _JoinedHostsCache(object):
-    """Cache for joined hosts in a room that is optimised to handle updates
-    via state deltas.
-    """
-
-    def __init__(self, store, room_id):
-        self.store = store
-        self.room_id = room_id
-
-        self.hosts_to_joined_users = {}
-
-        self.state_group = object()
-
-        self.linearizer = Linearizer("_JoinedHostsCache")
-
-        self._len = 0
-
-    @defer.inlineCallbacks
-    def get_destinations(self, state_entry):
-        """Get set of destinations for a state entry
-
-        Args:
-            state_entry(synapse.state._StateCacheEntry)
-        """
-        if state_entry.state_group == self.state_group:
-            return frozenset(self.hosts_to_joined_users)
-
-        with (yield self.linearizer.queue(())):
-            if state_entry.state_group == self.state_group:
-                pass
-            elif state_entry.prev_group == self.state_group:
-                for (typ, state_key), event_id in iteritems(state_entry.delta_ids):
-                    if typ != EventTypes.Member:
-                        continue
-
-                    host = intern_string(get_domain_from_id(state_key))
-                    user_id = state_key
-                    known_joins = self.hosts_to_joined_users.setdefault(host, set())
-
-                    event = yield self.store.get_event(event_id)
-                    if event.membership == Membership.JOIN:
-                        known_joins.add(user_id)
-                    else:
-                        known_joins.discard(user_id)
-
-                        if not known_joins:
-                            self.hosts_to_joined_users.pop(host, None)
-            else:
-                joined_users = yield self.store.get_joined_users_from_state(
-                    self.room_id, state_entry
-                )
-
-                self.hosts_to_joined_users = {}
-                for user_id in joined_users:
-                    host = intern_string(get_domain_from_id(user_id))
-                    self.hosts_to_joined_users.setdefault(host, set()).add(user_id)
-
-            if state_entry.state_group:
-                self.state_group = state_entry.state_group
-            else:
-                self.state_group = object()
-            self._len = sum(len(v) for v in itervalues(self.hosts_to_joined_users))
-        return frozenset(self.hosts_to_joined_users)
-
-    def __len__(self):
-        return self._len
diff --git a/synapse/storage/schema/delta/35/00background_updates_add_col.sql b/synapse/storage/schema/delta/35/00background_updates_add_col.sql
new file mode 100644
index 0000000000..c2d2a4f836
--- /dev/null
+++ b/synapse/storage/schema/delta/35/00background_updates_add_col.sql
@@ -0,0 +1,17 @@
+/* Copyright 2016 OpenMarket Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+ALTER TABLE background_updates ADD COLUMN depends_on TEXT;
diff --git a/synapse/storage/schema/full_schemas/54/full.sql b/synapse/storage/schema/full_schemas/54/full.sql
new file mode 100644
index 0000000000..1005880466
--- /dev/null
+++ b/synapse/storage/schema/full_schemas/54/full.sql
@@ -0,0 +1,8 @@
+
+
+CREATE TABLE background_updates (
+    update_name text NOT NULL,
+    progress_json text NOT NULL,
+    depends_on text,
+    CONSTRAINT background_updates_uniqueness UNIQUE (update_name)
+);
diff --git a/synapse/storage/state.py b/synapse/storage/state.py
index 1980a87108..3735846899 100644
--- a/synapse/storage/state.py
+++ b/synapse/storage/state.py
@@ -14,45 +14,18 @@
 # limitations under the License.
 
 import logging
-from collections import namedtuple
 
 from six import iteritems, itervalues
-from six.moves import range
 
 import attr
 
 from twisted.internet import defer
 
 from synapse.api.constants import EventTypes
-from synapse.api.errors import NotFoundError
-from synapse.storage._base import SQLBaseStore
-from synapse.storage.background_updates import BackgroundUpdateStore
-from synapse.storage.engines import PostgresEngine
-from synapse.storage.events_worker import EventsWorkerStore
-from synapse.util.caches import get_cache_factor_for, intern_string
-from synapse.util.caches.descriptors import cached, cachedList
-from synapse.util.caches.dictionary_cache import DictionaryCache
-from synapse.util.stringutils import to_ascii
 
 logger = logging.getLogger(__name__)
 
 
-MAX_STATE_DELTA_HOPS = 100
-
-
-class _GetStateGroupDelta(
-    namedtuple("_GetStateGroupDelta", ("prev_group", "delta_ids"))
-):
-    """Return type of get_state_group_delta that implements __len__, which lets
-    us use the itrable flag when caching
-    """
-
-    __slots__ = []
-
-    def __len__(self):
-        return len(self.delta_ids) if self.delta_ids else 0
-
-
 @attr.s(slots=True)
 class StateFilter(object):
     """A filter used when querying for state.
@@ -353,252 +326,23 @@ class StateFilter(object):
         return member_filter, non_member_filter
 
 
-# this inherits from EventsWorkerStore because it calls self.get_events
-class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
-    """The parts of StateGroupStore that can be called from workers.
+class StateGroupStorage(object):
+    """High level interface to fetching state for event.
     """
 
-    STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
-    STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index"
-    CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx"
+    def __init__(self, hs, stores):
+        self.stores = stores
 
-    def __init__(self, db_conn, hs):
-        super(StateGroupWorkerStore, self).__init__(db_conn, hs)
-
-        # Originally the state store used a single DictionaryCache to cache the
-        # event IDs for the state types in a given state group to avoid hammering
-        # on the state_group* tables.
-        #
-        # The point of using a DictionaryCache is that it can cache a subset
-        # of the state events for a given state group (i.e. a subset of the keys for a
-        # given dict which is an entry in the cache for a given state group ID).
-        #
-        # However, this poses problems when performing complicated queries
-        # on the store - for instance: "give me all the state for this group, but
-        # limit members to this subset of users", as DictionaryCache's API isn't
-        # rich enough to say "please cache any of these fields, apart from this subset".
-        # This is problematic when lazy loading members, which requires this behaviour,
-        # as without it the cache has no choice but to speculatively load all
-        # state events for the group, which negates the efficiency being sought.
-        #
-        # Rather than overcomplicating DictionaryCache's API, we instead split the
-        # state_group_cache into two halves - one for tracking non-member events,
-        # and the other for tracking member_events.  This means that lazy loading
-        # queries can be made in a cache-friendly manner by querying both caches
-        # separately and then merging the result.  So for the example above, you
-        # would query the members cache for a specific subset of state keys
-        # (which DictionaryCache will handle efficiently and fine) and the non-members
-        # cache for all state (which DictionaryCache will similarly handle fine)
-        # and then just merge the results together.
-        #
-        # We size the non-members cache to be smaller than the members cache as the
-        # vast majority of state in Matrix (today) is member events.
-
-        self._state_group_cache = DictionaryCache(
-            "*stateGroupCache*",
-            # TODO: this hasn't been tuned yet
-            50000 * get_cache_factor_for("stateGroupCache"),
-        )
-        self._state_group_members_cache = DictionaryCache(
-            "*stateGroupMembersCache*",
-            500000 * get_cache_factor_for("stateGroupMembersCache"),
-        )
-
-    @defer.inlineCallbacks
-    def get_room_version(self, room_id):
-        """Get the room_version of a given room
-
-        Args:
-            room_id (str)
-
-        Returns:
-            Deferred[str]
-
-        Raises:
-            NotFoundError if the room is unknown
-        """
-        # for now we do this by looking at the create event. We may want to cache this
-        # more intelligently in future.
-
-        # Retrieve the room's create event
-        create_event = yield self.get_create_event_for_room(room_id)
-        return create_event.content.get("room_version", "1")
-
-    @defer.inlineCallbacks
-    def get_room_predecessor(self, room_id):
-        """Get the predecessor room of an upgraded room if one exists.
-        Otherwise return None.
-
-        Args:
-            room_id (str)
-
-        Returns:
-            Deferred[unicode|None]: predecessor room id
-
-        Raises:
-            NotFoundError if the room is unknown
-        """
-        # Retrieve the room's create event
-        create_event = yield self.get_create_event_for_room(room_id)
-
-        # Return predecessor if present
-        return create_event.content.get("predecessor", None)
-
-    @defer.inlineCallbacks
-    def get_create_event_for_room(self, room_id):
-        """Get the create state event for a room.
-
-        Args:
-            room_id (str)
-
-        Returns:
-            Deferred[EventBase]: The room creation event.
-
-        Raises:
-            NotFoundError if the room is unknown
-        """
-        state_ids = yield self.get_current_state_ids(room_id)
-        create_id = state_ids.get((EventTypes.Create, ""))
-
-        # If we can't find the create event, assume we've hit a dead end
-        if not create_id:
-            raise NotFoundError("Unknown room %s" % (room_id))
-
-        # Retrieve the room's create event and return
-        create_event = yield self.get_event(create_id)
-        return create_event
-
-    @cached(max_entries=100000, iterable=True)
-    def get_current_state_ids(self, room_id):
-        """Get the current state event ids for a room based on the
-        current_state_events table.
-
-        Args:
-            room_id (str)
-
-        Returns:
-            deferred: dict of (type, state_key) -> event_id
-        """
-
-        def _get_current_state_ids_txn(txn):
-            txn.execute(
-                """SELECT type, state_key, event_id FROM current_state_events
-                WHERE room_id = ?
-                """,
-                (room_id,),
-            )
-
-            return {
-                (intern_string(r[0]), intern_string(r[1])): to_ascii(r[2]) for r in txn
-            }
-
-        return self.runInteraction("get_current_state_ids", _get_current_state_ids_txn)
-
-    # FIXME: how should this be cached?
-    def get_filtered_current_state_ids(self, room_id, state_filter=StateFilter.all()):
-        """Get the current state event of a given type for a room based on the
-        current_state_events table.  This may not be as up-to-date as the result
-        of doing a fresh state resolution as per state_handler.get_current_state
-
-        Args:
-            room_id (str)
-            state_filter (StateFilter): The state filter used to fetch state
-                from the database.
-
-        Returns:
-            Deferred[dict[tuple[str, str], str]]: Map from type/state_key to
-            event ID.
-        """
-
-        where_clause, where_args = state_filter.make_sql_filter_clause()
-
-        if not where_clause:
-            # We delegate to the cached version
-            return self.get_current_state_ids(room_id)
-
-        def _get_filtered_current_state_ids_txn(txn):
-            results = {}
-            sql = """
-                SELECT type, state_key, event_id FROM current_state_events
-                WHERE room_id = ?
-            """
-
-            if where_clause:
-                sql += " AND (%s)" % (where_clause,)
-
-            args = [room_id]
-            args.extend(where_args)
-            txn.execute(sql, args)
-            for row in txn:
-                typ, state_key, event_id = row
-                key = (intern_string(typ), intern_string(state_key))
-                results[key] = event_id
-
-            return results
-
-        return self.runInteraction(
-            "get_filtered_current_state_ids", _get_filtered_current_state_ids_txn
-        )
-
-    @defer.inlineCallbacks
-    def get_canonical_alias_for_room(self, room_id):
-        """Get canonical alias for room, if any
-
-        Args:
-            room_id (str)
-
-        Returns:
-            Deferred[str|None]: The canonical alias, if any
-        """
-
-        state = yield self.get_filtered_current_state_ids(
-            room_id, StateFilter.from_types([(EventTypes.CanonicalAlias, "")])
-        )
-
-        event_id = state.get((EventTypes.CanonicalAlias, ""))
-        if not event_id:
-            return
-
-        event = yield self.get_event(event_id, allow_none=True)
-        if not event:
-            return
-
-        return event.content.get("canonical_alias")
-
-    @cached(max_entries=10000, iterable=True)
     def get_state_group_delta(self, state_group):
         """Given a state group try to return a previous group and a delta between
         the old and the new.
 
         Returns:
-            (prev_group, delta_ids), where both may be None.
+            Deferred[Tuple[Optional[int], Optional[list[dict[tuple[str, str], str]]]]]):
+                (prev_group, delta_ids)
         """
 
-        def _get_state_group_delta_txn(txn):
-            prev_group = self._simple_select_one_onecol_txn(
-                txn,
-                table="state_group_edges",
-                keyvalues={"state_group": state_group},
-                retcol="prev_state_group",
-                allow_none=True,
-            )
-
-            if not prev_group:
-                return _GetStateGroupDelta(None, None)
-
-            delta_ids = self._simple_select_list_txn(
-                txn,
-                table="state_groups_state",
-                keyvalues={"state_group": state_group},
-                retcols=("type", "state_key", "event_id"),
-            )
-
-            return _GetStateGroupDelta(
-                prev_group,
-                {(row["type"], row["state_key"]): row["event_id"] for row in delta_ids},
-            )
-
-        return self.runInteraction("get_state_group_delta", _get_state_group_delta_txn)
+        return self.stores.main.get_state_group_delta(state_group)
 
     @defer.inlineCallbacks
     def get_state_groups_ids(self, _room_id, event_ids):
@@ -615,10 +359,10 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
         if not event_ids:
             return {}
 
-        event_to_groups = yield self._get_state_group_for_events(event_ids)
+        event_to_groups = yield self.stores.main._get_state_group_for_events(event_ids)
 
         groups = set(itervalues(event_to_groups))
-        group_to_state = yield self._get_state_for_groups(groups)
+        group_to_state = yield self.stores.main._get_state_for_groups(groups)
 
         return group_to_state
 
@@ -639,7 +383,6 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
     @defer.inlineCallbacks
     def get_state_groups(self, room_id, event_ids):
         """ Get the state groups for the given list of event_ids
-
         Returns:
             Deferred[dict[int, list[EventBase]]]:
                 dict of state_group_id -> list of state events.
@@ -649,7 +392,7 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
 
         group_to_ids = yield self.get_state_groups_ids(room_id, event_ids)
 
-        state_event_map = yield self.get_events(
+        state_event_map = yield self.stores.main.get_events(
             [
                 ev_id
                 for group_ids in itervalues(group_to_ids)
@@ -667,7 +410,6 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
             for group, event_id_map in iteritems(group_to_ids)
         }
 
-    @defer.inlineCallbacks
     def _get_state_groups_from_groups(self, groups, state_filter):
         """Returns the state groups for a given set of groups, filtering on
         types of state events.
@@ -680,140 +422,28 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
             Deferred[dict[int, dict[tuple[str, str], str]]]:
                 dict of state_group_id -> (dict of (type, state_key) -> event id)
         """
-        results = {}
 
-        chunks = [groups[i : i + 100] for i in range(0, len(groups), 100)]
-        for chunk in chunks:
-            res = yield self.runInteraction(
-                "_get_state_groups_from_groups",
-                self._get_state_groups_from_groups_txn,
-                chunk,
-                state_filter,
-            )
-            results.update(res)
-
-        return results
-
-    def _get_state_groups_from_groups_txn(
-        self, txn, groups, state_filter=StateFilter.all()
-    ):
-        results = {group: {} for group in groups}
-
-        where_clause, where_args = state_filter.make_sql_filter_clause()
-
-        # Unless the filter clause is empty, we're going to append it after an
-        # existing where clause
-        if where_clause:
-            where_clause = " AND (%s)" % (where_clause,)
-
-        if isinstance(self.database_engine, PostgresEngine):
-            # Temporarily disable sequential scans in this transaction. This is
-            # a temporary hack until we can add the right indices in
-            txn.execute("SET LOCAL enable_seqscan=off")
-
-            # The below query walks the state_group tree so that the "state"
-            # table includes all state_groups in the tree. It then joins
-            # against `state_groups_state` to fetch the latest state.
-            # It assumes that previous state groups are always numerically
-            # lesser.
-            # The PARTITION is used to get the event_id in the greatest state
-            # group for the given type, state_key.
-            # This may return multiple rows per (type, state_key), but last_value
-            # should be the same.
-            sql = """
-                WITH RECURSIVE state(state_group) AS (
-                    VALUES(?::bigint)
-                    UNION ALL
-                    SELECT prev_state_group FROM state_group_edges e, state s
-                    WHERE s.state_group = e.state_group
-                )
-                SELECT DISTINCT type, state_key, last_value(event_id) OVER (
-                    PARTITION BY type, state_key ORDER BY state_group ASC
-                    ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
-                ) AS event_id FROM state_groups_state
-                WHERE state_group IN (
-                    SELECT state_group FROM state
-                )
-            """
-
-            for group in groups:
-                args = [group]
-                args.extend(where_args)
-
-                txn.execute(sql + where_clause, args)
-                for row in txn:
-                    typ, state_key, event_id = row
-                    key = (typ, state_key)
-                    results[group][key] = event_id
-        else:
-            max_entries_returned = state_filter.max_entries_returned()
-
-            # We don't use WITH RECURSIVE on sqlite3 as there are distributions
-            # that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
-            for group in groups:
-                next_group = group
-
-                while next_group:
-                    # We did this before by getting the list of group ids, and
-                    # then passing that list to sqlite to get latest event for
-                    # each (type, state_key). However, that was terribly slow
-                    # without the right indices (which we can't add until
-                    # after we finish deduping state, which requires this func)
-                    args = [next_group]
-                    args.extend(where_args)
-
-                    txn.execute(
-                        "SELECT type, state_key, event_id FROM state_groups_state"
-                        " WHERE state_group = ? " + where_clause,
-                        args,
-                    )
-                    results[group].update(
-                        ((typ, state_key), event_id)
-                        for typ, state_key, event_id in txn
-                        if (typ, state_key) not in results[group]
-                    )
-
-                    # If the number of entries in the (type,state_key)->event_id dict
-                    # matches the number of (type,state_keys) types we were searching
-                    # for, then we must have found them all, so no need to go walk
-                    # further down the tree... UNLESS our types filter contained
-                    # wildcards (i.e. Nones) in which case we have to do an exhaustive
-                    # search
-                    if (
-                        max_entries_returned is not None
-                        and len(results[group]) == max_entries_returned
-                    ):
-                        break
-
-                    next_group = self._simple_select_one_onecol_txn(
-                        txn,
-                        table="state_group_edges",
-                        keyvalues={"state_group": next_group},
-                        retcol="prev_state_group",
-                        allow_none=True,
-                    )
-
-        return results
+        return self.stores.main._get_state_groups_from_groups(groups, state_filter)
 
     @defer.inlineCallbacks
     def get_state_for_events(self, event_ids, state_filter=StateFilter.all()):
         """Given a list of event_ids and type tuples, return a list of state
         dicts for each event.
-
         Args:
             event_ids (list[string])
             state_filter (StateFilter): The state filter used to fetch state
                 from the database.
-
         Returns:
             deferred: A dict of (event_id) -> (type, state_key) -> [state_events]
         """
-        event_to_groups = yield self._get_state_group_for_events(event_ids)
+        event_to_groups = yield self.stores.main._get_state_group_for_events(event_ids)
 
         groups = set(itervalues(event_to_groups))
-        group_to_state = yield self._get_state_for_groups(groups, state_filter)
+        group_to_state = yield self.stores.main._get_state_for_groups(
+            groups, state_filter
+        )
 
-        state_event_map = yield self.get_events(
+        state_event_map = yield self.stores.main.get_events(
             [ev_id for sd in itervalues(group_to_state) for ev_id in itervalues(sd)],
             get_prev_content=False,
         )
@@ -843,10 +473,12 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
         Returns:
             A deferred dict from event_id -> (type, state_key) -> event_id
         """
-        event_to_groups = yield self._get_state_group_for_events(event_ids)
+        event_to_groups = yield self.stores.main._get_state_group_for_events(event_ids)
 
         groups = set(itervalues(event_to_groups))
-        group_to_state = yield self._get_state_for_groups(groups, state_filter)
+        group_to_state = yield self.stores.main._get_state_for_groups(
+            groups, state_filter
+        )
 
         event_to_state = {
             event_id: group_to_state[group]
@@ -887,76 +519,6 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
         state_map = yield self.get_state_ids_for_events([event_id], state_filter)
         return state_map[event_id]
 
-    @cached(max_entries=50000)
-    def _get_state_group_for_event(self, event_id):
-        return self._simple_select_one_onecol(
-            table="event_to_state_groups",
-            keyvalues={"event_id": event_id},
-            retcol="state_group",
-            allow_none=True,
-            desc="_get_state_group_for_event",
-        )
-
-    @cachedList(
-        cached_method_name="_get_state_group_for_event",
-        list_name="event_ids",
-        num_args=1,
-        inlineCallbacks=True,
-    )
-    def _get_state_group_for_events(self, event_ids):
-        """Returns mapping event_id -> state_group
-        """
-        rows = yield self._simple_select_many_batch(
-            table="event_to_state_groups",
-            column="event_id",
-            iterable=event_ids,
-            keyvalues={},
-            retcols=("event_id", "state_group"),
-            desc="_get_state_group_for_events",
-        )
-
-        return {row["event_id"]: row["state_group"] for row in rows}
-
-    def _get_state_for_group_using_cache(self, cache, group, state_filter):
-        """Checks if group is in cache. See `_get_state_for_groups`
-
-        Args:
-            cache(DictionaryCache): the state group cache to use
-            group(int): The state group to lookup
-            state_filter (StateFilter): The state filter used to fetch state
-                from the database.
-
-        Returns 2-tuple (`state_dict`, `got_all`).
-        `got_all` is a bool indicating if we successfully retrieved all
-        requests state from the cache, if False we need to query the DB for the
-        missing state.
-        """
-        is_all, known_absent, state_dict_ids = cache.get(group)
-
-        if is_all or state_filter.is_full():
-            # Either we have everything or want everything, either way
-            # `is_all` tells us whether we've gotten everything.
-            return state_filter.filter_state(state_dict_ids), is_all
-
-        # tracks whether any of our requested types are missing from the cache
-        missing_types = False
-
-        if state_filter.has_wildcards():
-            # We don't know if we fetched all the state keys for the types in
-            # the filter that are wildcards, so we have to assume that we may
-            # have missed some.
-            missing_types = True
-        else:
-            # There aren't any wild cards, so `concrete_types()` returns the
-            # complete list of event types we're wanting.
-            for key in state_filter.concrete_types():
-                if key not in state_dict_ids and key not in known_absent:
-                    missing_types = True
-                    break
-
-        return state_filter.filter_state(state_dict_ids), not missing_types
-
-    @defer.inlineCallbacks
     def _get_state_for_groups(self, groups, state_filter=StateFilter.all()):
         """Gets the state at each of a list of state groups, optionally
         filtering by type/state_key
@@ -970,154 +532,7 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
             Deferred[dict[int, dict[tuple[str, str], str]]]:
                 dict of state_group_id -> (dict of (type, state_key) -> event id)
         """
-
-        member_filter, non_member_filter = state_filter.get_member_split()
-
-        # Now we look them up in the member and non-member caches
-        non_member_state, incomplete_groups_nm, = (
-            yield self._get_state_for_groups_using_cache(
-                groups, self._state_group_cache, state_filter=non_member_filter
-            )
-        )
-
-        member_state, incomplete_groups_m, = (
-            yield self._get_state_for_groups_using_cache(
-                groups, self._state_group_members_cache, state_filter=member_filter
-            )
-        )
-
-        state = dict(non_member_state)
-        for group in groups:
-            state[group].update(member_state[group])
-
-        # Now fetch any missing groups from the database
-
-        incomplete_groups = incomplete_groups_m | incomplete_groups_nm
-
-        if not incomplete_groups:
-            return state
-
-        cache_sequence_nm = self._state_group_cache.sequence
-        cache_sequence_m = self._state_group_members_cache.sequence
-
-        # Help the cache hit ratio by expanding the filter a bit
-        db_state_filter = state_filter.return_expanded()
-
-        group_to_state_dict = yield self._get_state_groups_from_groups(
-            list(incomplete_groups), state_filter=db_state_filter
-        )
-
-        # Now lets update the caches
-        self._insert_into_cache(
-            group_to_state_dict,
-            db_state_filter,
-            cache_seq_num_members=cache_sequence_m,
-            cache_seq_num_non_members=cache_sequence_nm,
-        )
-
-        # And finally update the result dict, by filtering out any extra
-        # stuff we pulled out of the database.
-        for group, group_state_dict in iteritems(group_to_state_dict):
-            # We just replace any existing entries, as we will have loaded
-            # everything we need from the database anyway.
-            state[group] = state_filter.filter_state(group_state_dict)
-
-        return state
-
-    def _get_state_for_groups_using_cache(self, groups, cache, state_filter):
-        """Gets the state at each of a list of state groups, optionally
-        filtering by type/state_key, querying from a specific cache.
-
-        Args:
-            groups (iterable[int]): list of state groups for which we want
-                to get the state.
-            cache (DictionaryCache): the cache of group ids to state dicts which
-                we will pass through - either the normal state cache or the specific
-                members state cache.
-            state_filter (StateFilter): The state filter used to fetch state
-                from the database.
-
-        Returns:
-            tuple[dict[int, dict[tuple[str, str], str]], set[int]]: Tuple of
-            dict of state_group_id -> (dict of (type, state_key) -> event id)
-            of entries in the cache, and the state group ids either missing
-            from the cache or incomplete.
-        """
-        results = {}
-        incomplete_groups = set()
-        for group in set(groups):
-            state_dict_ids, got_all = self._get_state_for_group_using_cache(
-                cache, group, state_filter
-            )
-            results[group] = state_dict_ids
-
-            if not got_all:
-                incomplete_groups.add(group)
-
-        return results, incomplete_groups
-
-    def _insert_into_cache(
-        self,
-        group_to_state_dict,
-        state_filter,
-        cache_seq_num_members,
-        cache_seq_num_non_members,
-    ):
-        """Inserts results from querying the database into the relevant cache.
-
-        Args:
-            group_to_state_dict (dict): The new entries pulled from database.
-                Map from state group to state dict
-            state_filter (StateFilter): The state filter used to fetch state
-                from the database.
-            cache_seq_num_members (int): Sequence number of member cache since
-                last lookup in cache
-            cache_seq_num_non_members (int): Sequence number of member cache since
-                last lookup in cache
-        """
-
-        # We need to work out which types we've fetched from the DB for the
-        # member vs non-member caches. This should be as accurate as possible,
-        # but can be an underestimate (e.g. when we have wild cards)
-
-        member_filter, non_member_filter = state_filter.get_member_split()
-        if member_filter.is_full():
-            # We fetched all member events
-            member_types = None
-        else:
-            # `concrete_types()` will only return a subset when there are wild
-            # cards in the filter, but that's fine.
-            member_types = member_filter.concrete_types()
-
-        if non_member_filter.is_full():
-            # We fetched all non member events
-            non_member_types = None
-        else:
-            non_member_types = non_member_filter.concrete_types()
-
-        for group, group_state_dict in iteritems(group_to_state_dict):
-            state_dict_members = {}
-            state_dict_non_members = {}
-
-            for k, v in iteritems(group_state_dict):
-                if k[0] == EventTypes.Member:
-                    state_dict_members[k] = v
-                else:
-                    state_dict_non_members[k] = v
-
-            self._state_group_members_cache.update(
-                cache_seq_num_members,
-                key=group,
-                value=state_dict_members,
-                fetched_keys=member_types,
-            )
-
-            self._state_group_cache.update(
-                cache_seq_num_non_members,
-                key=group,
-                value=state_dict_non_members,
-                fetched_keys=non_member_types,
-            )
+        return self.stores.main._get_state_for_groups(groups, state_filter)
 
     def store_state_group(
         self, event_id, room_id, prev_group, delta_ids, current_state_ids
@@ -1137,393 +552,6 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
         Returns:
             Deferred[int]: The state group ID
         """
-
-        def _store_state_group_txn(txn):
-            if current_state_ids is None:
-                # AFAIK, this can never happen
-                raise Exception("current_state_ids cannot be None")
-
-            state_group = self.database_engine.get_next_state_group_id(txn)
-
-            self._simple_insert_txn(
-                txn,
-                table="state_groups",
-                values={"id": state_group, "room_id": room_id, "event_id": event_id},
-            )
-
-            # We persist as a delta if we can, while also ensuring the chain
-            # of deltas isn't tooo long, as otherwise read performance degrades.
-            if prev_group:
-                is_in_db = self._simple_select_one_onecol_txn(
-                    txn,
-                    table="state_groups",
-                    keyvalues={"id": prev_group},
-                    retcol="id",
-                    allow_none=True,
-                )
-                if not is_in_db:
-                    raise Exception(
-                        "Trying to persist state with unpersisted prev_group: %r"
-                        % (prev_group,)
-                    )
-
-                potential_hops = self._count_state_group_hops_txn(txn, prev_group)
-            if prev_group and potential_hops < MAX_STATE_DELTA_HOPS:
-                self._simple_insert_txn(
-                    txn,
-                    table="state_group_edges",
-                    values={"state_group": state_group, "prev_state_group": prev_group},
-                )
-
-                self._simple_insert_many_txn(
-                    txn,
-                    table="state_groups_state",
-                    values=[
-                        {
-                            "state_group": state_group,
-                            "room_id": room_id,
-                            "type": key[0],
-                            "state_key": key[1],
-                            "event_id": state_id,
-                        }
-                        for key, state_id in iteritems(delta_ids)
-                    ],
-                )
-            else:
-                self._simple_insert_many_txn(
-                    txn,
-                    table="state_groups_state",
-                    values=[
-                        {
-                            "state_group": state_group,
-                            "room_id": room_id,
-                            "type": key[0],
-                            "state_key": key[1],
-                            "event_id": state_id,
-                        }
-                        for key, state_id in iteritems(current_state_ids)
-                    ],
-                )
-
-            # Prefill the state group caches with this group.
-            # It's fine to use the sequence like this as the state group map
-            # is immutable. (If the map wasn't immutable then this prefill could
-            # race with another update)
-
-            current_member_state_ids = {
-                s: ev
-                for (s, ev) in iteritems(current_state_ids)
-                if s[0] == EventTypes.Member
-            }
-            txn.call_after(
-                self._state_group_members_cache.update,
-                self._state_group_members_cache.sequence,
-                key=state_group,
-                value=dict(current_member_state_ids),
-            )
-
-            current_non_member_state_ids = {
-                s: ev
-                for (s, ev) in iteritems(current_state_ids)
-                if s[0] != EventTypes.Member
-            }
-            txn.call_after(
-                self._state_group_cache.update,
-                self._state_group_cache.sequence,
-                key=state_group,
-                value=dict(current_non_member_state_ids),
-            )
-
-            return state_group
-
-        return self.runInteraction("store_state_group", _store_state_group_txn)
-
-    def _count_state_group_hops_txn(self, txn, state_group):
-        """Given a state group, count how many hops there are in the tree.
-
-        This is used to ensure the delta chains don't get too long.
-        """
-        if isinstance(self.database_engine, PostgresEngine):
-            sql = """
-                WITH RECURSIVE state(state_group) AS (
-                    VALUES(?::bigint)
-                    UNION ALL
-                    SELECT prev_state_group FROM state_group_edges e, state s
-                    WHERE s.state_group = e.state_group
-                )
-                SELECT count(*) FROM state;
-            """
-
-            txn.execute(sql, (state_group,))
-            row = txn.fetchone()
-            if row and row[0]:
-                return row[0]
-            else:
-                return 0
-        else:
-            # We don't use WITH RECURSIVE on sqlite3 as there are distributions
-            # that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
-            next_group = state_group
-            count = 0
-
-            while next_group:
-                next_group = self._simple_select_one_onecol_txn(
-                    txn,
-                    table="state_group_edges",
-                    keyvalues={"state_group": next_group},
-                    retcol="prev_state_group",
-                    allow_none=True,
-                )
-                if next_group:
-                    count += 1
-
-            return count
-
-
-class StateStore(StateGroupWorkerStore, BackgroundUpdateStore):
-    """ Keeps track of the state at a given event.
-
-    This is done by the concept of `state groups`. Every event is a assigned
-    a state group (identified by an arbitrary string), which references a
-    collection of state events. The current state of an event is then the
-    collection of state events referenced by the event's state group.
-
-    Hence, every change in the current state causes a new state group to be
-    generated. However, if no change happens (e.g., if we get a message event
-    with only one parent it inherits the state group from its parent.)
-
-    There are three tables:
-      * `state_groups`: Stores group name, first event with in the group and
-        room id.
-      * `event_to_state_groups`: Maps events to state groups.
-      * `state_groups_state`: Maps state group to state events.
-    """
-
-    STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
-    STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index"
-    CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx"
-    EVENT_STATE_GROUP_INDEX_UPDATE_NAME = "event_to_state_groups_sg_index"
-
-    def __init__(self, db_conn, hs):
-        super(StateStore, self).__init__(db_conn, hs)
-        self.register_background_update_handler(
-            self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME,
-            self._background_deduplicate_state,
+        return self.stores.main.store_state_group(
+            event_id, room_id, prev_group, delta_ids, current_state_ids
         )
-        self.register_background_update_handler(
-            self.STATE_GROUP_INDEX_UPDATE_NAME, self._background_index_state
-        )
-        self.register_background_index_update(
-            self.CURRENT_STATE_INDEX_UPDATE_NAME,
-            index_name="current_state_events_member_index",
-            table="current_state_events",
-            columns=["state_key"],
-            where_clause="type='m.room.member'",
-        )
-        self.register_background_index_update(
-            self.EVENT_STATE_GROUP_INDEX_UPDATE_NAME,
-            index_name="event_to_state_groups_sg_index",
-            table="event_to_state_groups",
-            columns=["state_group"],
-        )
-
-    def _store_event_state_mappings_txn(self, txn, events_and_contexts):
-        state_groups = {}
-        for event, context in events_and_contexts:
-            if event.internal_metadata.is_outlier():
-                continue
-
-            # if the event was rejected, just give it the same state as its
-            # predecessor.
-            if context.rejected:
-                state_groups[event.event_id] = context.prev_group
-                continue
-
-            state_groups[event.event_id] = context.state_group
-
-        self._simple_insert_many_txn(
-            txn,
-            table="event_to_state_groups",
-            values=[
-                {"state_group": state_group_id, "event_id": event_id}
-                for event_id, state_group_id in iteritems(state_groups)
-            ],
-        )
-
-        for event_id, state_group_id in iteritems(state_groups):
-            txn.call_after(
-                self._get_state_group_for_event.prefill, (event_id,), state_group_id
-            )
-
-    @defer.inlineCallbacks
-    def _background_deduplicate_state(self, progress, batch_size):
-        """This background update will slowly deduplicate state by reencoding
-        them as deltas.
-        """
-        last_state_group = progress.get("last_state_group", 0)
-        rows_inserted = progress.get("rows_inserted", 0)
-        max_group = progress.get("max_group", None)
-
-        BATCH_SIZE_SCALE_FACTOR = 100
-
-        batch_size = max(1, int(batch_size / BATCH_SIZE_SCALE_FACTOR))
-
-        if max_group is None:
-            rows = yield self._execute(
-                "_background_deduplicate_state",
-                None,
-                "SELECT coalesce(max(id), 0) FROM state_groups",
-            )
-            max_group = rows[0][0]
-
-        def reindex_txn(txn):
-            new_last_state_group = last_state_group
-            for count in range(batch_size):
-                txn.execute(
-                    "SELECT id, room_id FROM state_groups"
-                    " WHERE ? < id AND id <= ?"
-                    " ORDER BY id ASC"
-                    " LIMIT 1",
-                    (new_last_state_group, max_group),
-                )
-                row = txn.fetchone()
-                if row:
-                    state_group, room_id = row
-
-                if not row or not state_group:
-                    return True, count
-
-                txn.execute(
-                    "SELECT state_group FROM state_group_edges"
-                    " WHERE state_group = ?",
-                    (state_group,),
-                )
-
-                # If we reach a point where we've already started inserting
-                # edges we should stop.
-                if txn.fetchall():
-                    return True, count
-
-                txn.execute(
-                    "SELECT coalesce(max(id), 0) FROM state_groups"
-                    " WHERE id < ? AND room_id = ?",
-                    (state_group, room_id),
-                )
-                prev_group, = txn.fetchone()
-                new_last_state_group = state_group
-
-                if prev_group:
-                    potential_hops = self._count_state_group_hops_txn(txn, prev_group)
-                    if potential_hops >= MAX_STATE_DELTA_HOPS:
-                        # We want to ensure chains are at most this long,#
-                        # otherwise read performance degrades.
-                        continue
-
-                    prev_state = self._get_state_groups_from_groups_txn(
-                        txn, [prev_group]
-                    )
-                    prev_state = prev_state[prev_group]
-
-                    curr_state = self._get_state_groups_from_groups_txn(
-                        txn, [state_group]
-                    )
-                    curr_state = curr_state[state_group]
-
-                    if not set(prev_state.keys()) - set(curr_state.keys()):
-                        # We can only do a delta if the current has a strict super set
-                        # of keys
-
-                        delta_state = {
-                            key: value
-                            for key, value in iteritems(curr_state)
-                            if prev_state.get(key, None) != value
-                        }
-
-                        self._simple_delete_txn(
-                            txn,
-                            table="state_group_edges",
-                            keyvalues={"state_group": state_group},
-                        )
-
-                        self._simple_insert_txn(
-                            txn,
-                            table="state_group_edges",
-                            values={
-                                "state_group": state_group,
-                                "prev_state_group": prev_group,
-                            },
-                        )
-
-                        self._simple_delete_txn(
-                            txn,
-                            table="state_groups_state",
-                            keyvalues={"state_group": state_group},
-                        )
-
-                        self._simple_insert_many_txn(
-                            txn,
-                            table="state_groups_state",
-                            values=[
-                                {
-                                    "state_group": state_group,
-                                    "room_id": room_id,
-                                    "type": key[0],
-                                    "state_key": key[1],
-                                    "event_id": state_id,
-                                }
-                                for key, state_id in iteritems(delta_state)
-                            ],
-                        )
-
-            progress = {
-                "last_state_group": state_group,
-                "rows_inserted": rows_inserted + batch_size,
-                "max_group": max_group,
-            }
-
-            self._background_update_progress_txn(
-                txn, self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, progress
-            )
-
-            return False, batch_size
-
-        finished, result = yield self.runInteraction(
-            self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, reindex_txn
-        )
-
-        if finished:
-            yield self._end_background_update(
-                self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME
-            )
-
-        return result * BATCH_SIZE_SCALE_FACTOR
-
-    @defer.inlineCallbacks
-    def _background_index_state(self, progress, batch_size):
-        def reindex_txn(conn):
-            conn.rollback()
-            if isinstance(self.database_engine, PostgresEngine):
-                # postgres insists on autocommit for the index
-                conn.set_session(autocommit=True)
-                try:
-                    txn = conn.cursor()
-                    txn.execute(
-                        "CREATE INDEX CONCURRENTLY state_groups_state_type_idx"
-                        " ON state_groups_state(state_group, type, state_key)"
-                    )
-                    txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
-                finally:
-                    conn.set_session(autocommit=False)
-            else:
-                txn = conn.cursor()
-                txn.execute(
-                    "CREATE INDEX state_groups_state_type_idx"
-                    " ON state_groups_state(state_group, type, state_key)"
-                )
-                txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
-
-        yield self.runWithConnection(reindex_txn)
-
-        yield self._end_background_update(self.STATE_GROUP_INDEX_UPDATE_NAME)
-
-        return 1
diff --git a/synapse/storage/stats.py b/synapse/storage/stats.py
deleted file mode 100644
index e13efed417..0000000000
--- a/synapse/storage/stats.py
+++ /dev/null
@@ -1,484 +0,0 @@
-# -*- coding: utf-8 -*-
-# Copyright 2018, 2019 New Vector Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import logging
-
-from twisted.internet import defer
-
-from synapse.api.constants import EventTypes, Membership
-from synapse.storage.prepare_database import get_statements
-from synapse.storage.state_deltas import StateDeltasStore
-from synapse.util.caches.descriptors import cached
-
-logger = logging.getLogger(__name__)
-
-# these fields track absolutes (e.g. total number of rooms on the server)
-ABSOLUTE_STATS_FIELDS = {
-    "room": (
-        "current_state_events",
-        "joined_members",
-        "invited_members",
-        "left_members",
-        "banned_members",
-        "state_events",
-    ),
-    "user": ("public_rooms", "private_rooms"),
-}
-
-TYPE_TO_ROOM = {"room": ("room_stats", "room_id"), "user": ("user_stats", "user_id")}
-
-TEMP_TABLE = "_temp_populate_stats"
-
-
-class StatsStore(StateDeltasStore):
-    def __init__(self, db_conn, hs):
-        super(StatsStore, self).__init__(db_conn, hs)
-
-        self.server_name = hs.hostname
-        self.clock = self.hs.get_clock()
-        self.stats_enabled = hs.config.stats_enabled
-        self.stats_bucket_size = hs.config.stats_bucket_size
-
-        self.register_background_update_handler(
-            "populate_stats_createtables", self._populate_stats_createtables
-        )
-        self.register_background_update_handler(
-            "populate_stats_process_rooms", self._populate_stats_process_rooms
-        )
-        self.register_background_update_handler(
-            "populate_stats_cleanup", self._populate_stats_cleanup
-        )
-
-    @defer.inlineCallbacks
-    def _populate_stats_createtables(self, progress, batch_size):
-
-        if not self.stats_enabled:
-            yield self._end_background_update("populate_stats_createtables")
-            return 1
-
-        # Get all the rooms that we want to process.
-        def _make_staging_area(txn):
-            # Create the temporary tables
-            stmts = get_statements(
-                """
-                -- We just recreate the table, we'll be reinserting the
-                -- correct entries again later anyway.
-                DROP TABLE IF EXISTS {temp}_rooms;
-
-                CREATE TABLE IF NOT EXISTS {temp}_rooms(
-                    room_id TEXT NOT NULL,
-                    events BIGINT NOT NULL
-                );
-
-                CREATE INDEX {temp}_rooms_events
-                    ON {temp}_rooms(events);
-                CREATE INDEX {temp}_rooms_id
-                    ON {temp}_rooms(room_id);
-            """.format(
-                    temp=TEMP_TABLE
-                ).splitlines()
-            )
-
-            for statement in stmts:
-                txn.execute(statement)
-
-            sql = (
-                "CREATE TABLE IF NOT EXISTS "
-                + TEMP_TABLE
-                + "_position(position TEXT NOT NULL)"
-            )
-            txn.execute(sql)
-
-            # Get rooms we want to process from the database, only adding
-            # those that we haven't (i.e. those not in room_stats_earliest_token)
-            sql = """
-                INSERT INTO %s_rooms (room_id, events)
-                SELECT c.room_id, count(*) FROM current_state_events AS c
-                LEFT JOIN room_stats_earliest_token AS t USING (room_id)
-                WHERE t.room_id IS NULL
-                GROUP BY c.room_id
-            """ % (
-                TEMP_TABLE,
-            )
-            txn.execute(sql)
-
-        new_pos = yield self.get_max_stream_id_in_current_state_deltas()
-        yield self.runInteraction("populate_stats_temp_build", _make_staging_area)
-        yield self._simple_insert(TEMP_TABLE + "_position", {"position": new_pos})
-        self.get_earliest_token_for_room_stats.invalidate_all()
-
-        yield self._end_background_update("populate_stats_createtables")
-        return 1
-
-    @defer.inlineCallbacks
-    def _populate_stats_cleanup(self, progress, batch_size):
-        """
-        Update the user directory stream position, then clean up the old tables.
-        """
-        if not self.stats_enabled:
-            yield self._end_background_update("populate_stats_cleanup")
-            return 1
-
-        position = yield self._simple_select_one_onecol(
-            TEMP_TABLE + "_position", None, "position"
-        )
-        yield self.update_stats_stream_pos(position)
-
-        def _delete_staging_area(txn):
-            txn.execute("DROP TABLE IF EXISTS " + TEMP_TABLE + "_rooms")
-            txn.execute("DROP TABLE IF EXISTS " + TEMP_TABLE + "_position")
-
-        yield self.runInteraction("populate_stats_cleanup", _delete_staging_area)
-
-        yield self._end_background_update("populate_stats_cleanup")
-        return 1
-
-    @defer.inlineCallbacks
-    def _populate_stats_process_rooms(self, progress, batch_size):
-
-        if not self.stats_enabled:
-            yield self._end_background_update("populate_stats_process_rooms")
-            return 1
-
-        # If we don't have progress filed, delete everything.
-        if not progress:
-            yield self.delete_all_stats()
-
-        def _get_next_batch(txn):
-            # Only fetch 250 rooms, so we don't fetch too many at once, even
-            # if those 250 rooms have less than batch_size state events.
-            sql = """
-                SELECT room_id, events FROM %s_rooms
-                ORDER BY events DESC
-                LIMIT 250
-            """ % (
-                TEMP_TABLE,
-            )
-            txn.execute(sql)
-            rooms_to_work_on = txn.fetchall()
-
-            if not rooms_to_work_on:
-                return None
-
-            # Get how many are left to process, so we can give status on how
-            # far we are in processing
-            txn.execute("SELECT COUNT(*) FROM " + TEMP_TABLE + "_rooms")
-            progress["remaining"] = txn.fetchone()[0]
-
-            return rooms_to_work_on
-
-        rooms_to_work_on = yield self.runInteraction(
-            "populate_stats_temp_read", _get_next_batch
-        )
-
-        # No more rooms -- complete the transaction.
-        if not rooms_to_work_on:
-            yield self._end_background_update("populate_stats_process_rooms")
-            return 1
-
-        logger.info(
-            "Processing the next %d rooms of %d remaining",
-            len(rooms_to_work_on),
-            progress["remaining"],
-        )
-
-        # Number of state events we've processed by going through each room
-        processed_event_count = 0
-
-        for room_id, event_count in rooms_to_work_on:
-
-            current_state_ids = yield self.get_current_state_ids(room_id)
-
-            join_rules_id = current_state_ids.get((EventTypes.JoinRules, ""))
-            history_visibility_id = current_state_ids.get(
-                (EventTypes.RoomHistoryVisibility, "")
-            )
-            encryption_id = current_state_ids.get((EventTypes.RoomEncryption, ""))
-            name_id = current_state_ids.get((EventTypes.Name, ""))
-            topic_id = current_state_ids.get((EventTypes.Topic, ""))
-            avatar_id = current_state_ids.get((EventTypes.RoomAvatar, ""))
-            canonical_alias_id = current_state_ids.get((EventTypes.CanonicalAlias, ""))
-
-            event_ids = [
-                join_rules_id,
-                history_visibility_id,
-                encryption_id,
-                name_id,
-                topic_id,
-                avatar_id,
-                canonical_alias_id,
-            ]
-
-            state_events = yield self.get_events(
-                [ev for ev in event_ids if ev is not None]
-            )
-
-            def _get_or_none(event_id, arg):
-                event = state_events.get(event_id)
-                if event:
-                    return event.content.get(arg)
-                return None
-
-            yield self.update_room_state(
-                room_id,
-                {
-                    "join_rules": _get_or_none(join_rules_id, "join_rule"),
-                    "history_visibility": _get_or_none(
-                        history_visibility_id, "history_visibility"
-                    ),
-                    "encryption": _get_or_none(encryption_id, "algorithm"),
-                    "name": _get_or_none(name_id, "name"),
-                    "topic": _get_or_none(topic_id, "topic"),
-                    "avatar": _get_or_none(avatar_id, "url"),
-                    "canonical_alias": _get_or_none(canonical_alias_id, "alias"),
-                },
-            )
-
-            now = self.hs.get_reactor().seconds()
-
-            # quantise time to the nearest bucket
-            now = (now // self.stats_bucket_size) * self.stats_bucket_size
-
-            def _fetch_data(txn):
-
-                # Get the current token of the room
-                current_token = self._get_max_stream_id_in_current_state_deltas_txn(txn)
-
-                current_state_events = len(current_state_ids)
-
-                membership_counts = self._get_user_counts_in_room_txn(txn, room_id)
-
-                total_state_events = self._get_total_state_event_counts_txn(
-                    txn, room_id
-                )
-
-                self._update_stats_txn(
-                    txn,
-                    "room",
-                    room_id,
-                    now,
-                    {
-                        "bucket_size": self.stats_bucket_size,
-                        "current_state_events": current_state_events,
-                        "joined_members": membership_counts.get(Membership.JOIN, 0),
-                        "invited_members": membership_counts.get(Membership.INVITE, 0),
-                        "left_members": membership_counts.get(Membership.LEAVE, 0),
-                        "banned_members": membership_counts.get(Membership.BAN, 0),
-                        "state_events": total_state_events,
-                    },
-                )
-                self._simple_insert_txn(
-                    txn,
-                    "room_stats_earliest_token",
-                    {"room_id": room_id, "token": current_token},
-                )
-
-                # We've finished a room. Delete it from the table.
-                self._simple_delete_one_txn(
-                    txn, TEMP_TABLE + "_rooms", {"room_id": room_id}
-                )
-
-            yield self.runInteraction("update_room_stats", _fetch_data)
-
-            # Update the remaining counter.
-            progress["remaining"] -= 1
-            yield self.runInteraction(
-                "populate_stats",
-                self._background_update_progress_txn,
-                "populate_stats_process_rooms",
-                progress,
-            )
-
-            processed_event_count += event_count
-
-            if processed_event_count > batch_size:
-                # Don't process any more rooms, we've hit our batch size.
-                return processed_event_count
-
-        return processed_event_count
-
-    def delete_all_stats(self):
-        """
-        Delete all statistics records.
-        """
-
-        def _delete_all_stats_txn(txn):
-            txn.execute("DELETE FROM room_state")
-            txn.execute("DELETE FROM room_stats")
-            txn.execute("DELETE FROM room_stats_earliest_token")
-            txn.execute("DELETE FROM user_stats")
-
-        return self.runInteraction("delete_all_stats", _delete_all_stats_txn)
-
-    def get_stats_stream_pos(self):
-        return self._simple_select_one_onecol(
-            table="stats_stream_pos",
-            keyvalues={},
-            retcol="stream_id",
-            desc="stats_stream_pos",
-        )
-
-    def update_stats_stream_pos(self, stream_id):
-        return self._simple_update_one(
-            table="stats_stream_pos",
-            keyvalues={},
-            updatevalues={"stream_id": stream_id},
-            desc="update_stats_stream_pos",
-        )
-
-    def update_room_state(self, room_id, fields):
-        """
-        Args:
-            room_id (str)
-            fields (dict[str:Any])
-        """
-
-        # For whatever reason some of the fields may contain null bytes, which
-        # postgres isn't a fan of, so we replace those fields with null.
-        for col in (
-            "join_rules",
-            "history_visibility",
-            "encryption",
-            "name",
-            "topic",
-            "avatar",
-            "canonical_alias",
-        ):
-            field = fields.get(col)
-            if field and "\0" in field:
-                fields[col] = None
-
-        return self._simple_upsert(
-            table="room_state",
-            keyvalues={"room_id": room_id},
-            values=fields,
-            desc="update_room_state",
-        )
-
-    def get_deltas_for_room(self, room_id, start, size=100):
-        """
-        Get statistics deltas for a given room.
-
-        Args:
-            room_id (str)
-            start (int): Pagination start. Number of entries, not timestamp.
-            size (int): How many entries to return.
-
-        Returns:
-            Deferred[list[dict]], where the dict has the keys of
-            ABSOLUTE_STATS_FIELDS["room"] and "ts".
-        """
-        return self._simple_select_list_paginate(
-            "room_stats",
-            {"room_id": room_id},
-            "ts",
-            start,
-            size,
-            retcols=(list(ABSOLUTE_STATS_FIELDS["room"]) + ["ts"]),
-            order_direction="DESC",
-        )
-
-    def get_all_room_state(self):
-        return self._simple_select_list(
-            "room_state", None, retcols=("name", "topic", "canonical_alias")
-        )
-
-    @cached()
-    def get_earliest_token_for_room_stats(self, room_id):
-        """
-        Fetch the "earliest token". This is used by the room stats delta
-        processor to ignore deltas that have been processed between the
-        start of the background task and any particular room's stats
-        being calculated.
-
-        Returns:
-            Deferred[int]
-        """
-        return self._simple_select_one_onecol(
-            "room_stats_earliest_token",
-            {"room_id": room_id},
-            retcol="token",
-            allow_none=True,
-        )
-
-    def update_stats(self, stats_type, stats_id, ts, fields):
-        table, id_col = TYPE_TO_ROOM[stats_type]
-        return self._simple_upsert(
-            table=table,
-            keyvalues={id_col: stats_id, "ts": ts},
-            values=fields,
-            desc="update_stats",
-        )
-
-    def _update_stats_txn(self, txn, stats_type, stats_id, ts, fields):
-        table, id_col = TYPE_TO_ROOM[stats_type]
-        return self._simple_upsert_txn(
-            txn, table=table, keyvalues={id_col: stats_id, "ts": ts}, values=fields
-        )
-
-    def update_stats_delta(self, ts, stats_type, stats_id, field, value):
-        def _update_stats_delta(txn):
-            table, id_col = TYPE_TO_ROOM[stats_type]
-
-            sql = (
-                "SELECT * FROM %s"
-                " WHERE %s=? and ts=("
-                "  SELECT MAX(ts) FROM %s"
-                "  WHERE %s=?"
-                ")"
-            ) % (table, id_col, table, id_col)
-            txn.execute(sql, (stats_id, stats_id))
-            rows = self.cursor_to_dict(txn)
-            if len(rows) == 0:
-                # silently skip as we don't have anything to apply a delta to yet.
-                # this tries to minimise any race between the initial sync and
-                # subsequent deltas arriving.
-                return
-
-            current_ts = ts
-            latest_ts = rows[0]["ts"]
-            if current_ts < latest_ts:
-                # This one is in the past, but we're just encountering it now.
-                # Mark it as part of the current bucket.
-                current_ts = latest_ts
-            elif ts != latest_ts:
-                # we have to copy our absolute counters over to the new entry.
-                values = {
-                    key: rows[0][key] for key in ABSOLUTE_STATS_FIELDS[stats_type]
-                }
-                values[id_col] = stats_id
-                values["ts"] = ts
-                values["bucket_size"] = self.stats_bucket_size
-
-                self._simple_insert_txn(txn, table=table, values=values)
-
-            # actually update the new value
-            if stats_type in ABSOLUTE_STATS_FIELDS[stats_type]:
-                self._simple_update_txn(
-                    txn,
-                    table=table,
-                    keyvalues={id_col: stats_id, "ts": current_ts},
-                    updatevalues={field: value},
-                )
-            else:
-                sql = ("UPDATE %s SET %s=%s+? WHERE %s=? AND ts=?") % (
-                    table,
-                    field,
-                    field,
-                    id_col,
-                )
-                txn.execute(sql, (value, stats_id, current_ts))
-
-        return self.runInteraction("update_stats_delta", _update_stats_delta)
diff --git a/synapse/storage/util/id_generators.py b/synapse/storage/util/id_generators.py
index cbb0a4810a..9d851beaa5 100644
--- a/synapse/storage/util/id_generators.py
+++ b/synapse/storage/util/id_generators.py
@@ -46,7 +46,7 @@ def _load_current_id(db_conn, table, column, step=1):
         cur.execute("SELECT MAX(%s) FROM %s" % (column, table))
     else:
         cur.execute("SELECT MIN(%s) FROM %s" % (column, table))
-    val, = cur.fetchone()
+    (val,) = cur.fetchone()
     cur.close()
     current_id = int(val) if val else step
     return (max if step > 0 else min)(current_id, step)
diff --git a/synapse/types.py b/synapse/types.py
index 51eadb6ad4..aafc3ffe74 100644
--- a/synapse/types.py
+++ b/synapse/types.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,6 +18,8 @@ import string
 from collections import namedtuple
 
 import attr
+from signedjson.key import decode_verify_key_bytes
+from unpaddedbase64 import decode_base64
 
 from synapse.api.errors import SynapseError
 
@@ -318,6 +321,7 @@ class StreamToken(
     )
 ):
     _SEPARATOR = "_"
+    START = None  # type: StreamToken
 
     @classmethod
     def from_string(cls, string):
@@ -402,7 +406,7 @@ class RoomStreamToken(namedtuple("_StreamToken", "topological stream")):
     followed by the "stream_ordering" id of the event it comes after.
     """
 
-    __slots__ = []
+    __slots__ = []  # type: list
 
     @classmethod
     def parse(cls, string):
@@ -475,3 +479,24 @@ class ReadReceipt(object):
     user_id = attr.ib()
     event_ids = attr.ib()
     data = attr.ib()
+
+
+def get_verify_key_from_cross_signing_key(key_info):
+    """Get the key ID and signedjson verify key from a cross-signing key dict
+
+    Args:
+        key_info (dict): a cross-signing key dict, which must have a "keys"
+            property that has exactly one item in it
+
+    Returns:
+        (str, VerifyKey): the key ID and verify key for the cross-signing key
+    """
+    # make sure that exactly one key is provided
+    if "keys" not in key_info:
+        raise ValueError("Invalid key")
+    keys = key_info["keys"]
+    if len(keys) != 1:
+        raise ValueError("Invalid key")
+    # and return that one key
+    for key_id, key_data in keys.items():
+        return (key_id, decode_verify_key_bytes(key_id, decode_base64(key_data)))
diff --git a/synapse/util/async_helpers.py b/synapse/util/async_helpers.py
index f1c46836b1..5c4de2e69f 100644
--- a/synapse/util/async_helpers.py
+++ b/synapse/util/async_helpers.py
@@ -13,12 +13,16 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 import collections
 import logging
 from contextlib import contextmanager
+from typing import Dict, Sequence, Set, Union
 
 from six.moves import range
 
+import attr
+
 from twisted.internet import defer
 from twisted.internet.defer import CancelledError
 from twisted.python import failure
@@ -82,11 +86,12 @@ class ObservableDeferred(object):
 
         deferred.addCallbacks(callback, errback)
 
-    def observe(self):
+    def observe(self) -> defer.Deferred:
         """Observe the underlying deferred.
 
-        Can return either a deferred if the underlying deferred is still pending
-        (or has failed), or the actual value. Callers may need to use maybeDeferred.
+        This returns a brand new deferred that is resolved when the underlying
+        deferred is resolved. Interacting with the returned deferred does not
+        effect the underdlying deferred.
         """
         if not self._result:
             d = defer.Deferred()
@@ -101,7 +106,7 @@ class ObservableDeferred(object):
             return d
         else:
             success, res = self._result
-            return res if success else defer.fail(res)
+            return defer.succeed(res) if success else defer.fail(res)
 
     def observers(self):
         return self._observers
@@ -134,7 +139,7 @@ def concurrently_execute(func, args, limit):
     the number of concurrent executions.
 
     Args:
-        func (func): Function to execute, should return a deferred.
+        func (func): Function to execute, should return a deferred or coroutine.
         args (list): List of arguments to pass to func, each invocation of func
             gets a signle argument.
         limit (int): Maximum number of conccurent executions.
@@ -144,11 +149,10 @@ def concurrently_execute(func, args, limit):
     """
     it = iter(args)
 
-    @defer.inlineCallbacks
-    def _concurrently_execute_inner():
+    async def _concurrently_execute_inner():
         try:
             while True:
-                yield func(next(it))
+                await maybe_awaitable(func(next(it)))
         except StopIteration:
             pass
 
@@ -213,7 +217,9 @@ class Linearizer(object):
         # the first element is the number of things executing, and
         # the second element is an OrderedDict, where the keys are deferreds for the
         # things blocked from executing.
-        self.key_to_defer = {}
+        self.key_to_defer = (
+            {}
+        )  # type: Dict[str, Sequence[Union[int, Dict[defer.Deferred, int]]]]
 
     def queue(self, key):
         # we avoid doing defer.inlineCallbacks here, so that cancellation works correctly.
@@ -303,7 +309,7 @@ class Linearizer(object):
                 )
 
             else:
-                logger.warn(
+                logger.warning(
                     "Unexpected exception waiting for linearizer lock %r for key %r",
                     self.name,
                     key,
@@ -340,10 +346,10 @@ class ReadWriteLock(object):
 
     def __init__(self):
         # Latest readers queued
-        self.key_to_current_readers = {}
+        self.key_to_current_readers = {}  # type: Dict[str, Set[defer.Deferred]]
 
         # Latest writer queued
-        self.key_to_current_writer = {}
+        self.key_to_current_writer = {}  # type: Dict[str, defer.Deferred]
 
     @defer.inlineCallbacks
     def read(self, key):
@@ -479,3 +485,30 @@ def timeout_deferred(deferred, timeout, reactor, on_timeout_cancel=None):
     deferred.addCallbacks(success_cb, failure_cb)
 
     return new_d
+
+
+@attr.s(slots=True, frozen=True)
+class DoneAwaitable(object):
+    """Simple awaitable that returns the provided value.
+    """
+
+    value = attr.ib()
+
+    def __await__(self):
+        return self
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        raise StopIteration(self.value)
+
+
+def maybe_awaitable(value):
+    """Convert a value to an awaitable if not already an awaitable.
+    """
+
+    if hasattr(value, "__await__"):
+        return value
+
+    return DoneAwaitable(value)
diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py
index b50e3503f0..da5077b471 100644
--- a/synapse/util/caches/__init__.py
+++ b/synapse/util/caches/__init__.py
@@ -16,6 +16,7 @@
 
 import logging
 import os
+from typing import Dict
 
 import six
 from six.moves import intern
@@ -37,7 +38,7 @@ def get_cache_factor_for(cache_name):
 
 
 caches_by_name = {}
-collectors_by_name = {}
+collectors_by_name = {}  # type: Dict
 
 cache_size = Gauge("synapse_util_caches_cache:size", "", ["name"])
 cache_hits = Gauge("synapse_util_caches_cache:hits", "", ["name"])
@@ -106,7 +107,7 @@ def register_cache(cache_type, cache_name, cache, collect_callback=None):
                 if collect_callback:
                     collect_callback()
             except Exception as e:
-                logger.warn("Error calculating metrics for %s: %s", cache_name, e)
+                logger.warning("Error calculating metrics for %s: %s", cache_name, e)
                 raise
 
             yield GaugeMetricFamily("__unused", "")
diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py
index 43f66ec4be..0e8da27f53 100644
--- a/synapse/util/caches/descriptors.py
+++ b/synapse/util/caches/descriptors.py
@@ -18,10 +18,12 @@ import inspect
 import logging
 import threading
 from collections import namedtuple
+from typing import Any, cast
 
 from six import itervalues
 
 from prometheus_client import Gauge
+from typing_extensions import Protocol
 
 from twisted.internet import defer
 
@@ -37,6 +39,18 @@ from . import register_cache
 logger = logging.getLogger(__name__)
 
 
+class _CachedFunction(Protocol):
+    invalidate = None  # type: Any
+    invalidate_all = None  # type: Any
+    invalidate_many = None  # type: Any
+    prefill = None  # type: Any
+    cache = None  # type: Any
+    num_args = None  # type: Any
+
+    def __name__(self):
+        ...
+
+
 cache_pending_metric = Gauge(
     "synapse_util_caches_cache_pending",
     "Number of lookups currently pending for this cache",
@@ -245,7 +259,9 @@ class Cache(object):
 
 
 class _CacheDescriptorBase(object):
-    def __init__(self, orig, num_args, inlineCallbacks, cache_context=False):
+    def __init__(
+        self, orig: _CachedFunction, num_args, inlineCallbacks, cache_context=False
+    ):
         self.orig = orig
 
         if inlineCallbacks:
@@ -404,7 +420,7 @@ class CacheDescriptor(_CacheDescriptorBase):
                 return tuple(get_cache_key_gen(args, kwargs))
 
         @functools.wraps(self.orig)
-        def wrapped(*args, **kwargs):
+        def _wrapped(*args, **kwargs):
             # If we're passed a cache_context then we'll want to call its invalidate()
             # whenever we are invalidated
             invalidate_callback = kwargs.pop("on_invalidate", None)
@@ -422,7 +438,7 @@ class CacheDescriptor(_CacheDescriptorBase):
                 if isinstance(cached_result_d, ObservableDeferred):
                     observer = cached_result_d.observe()
                 else:
-                    observer = cached_result_d
+                    observer = defer.succeed(cached_result_d)
 
             except KeyError:
                 ret = defer.maybeDeferred(
@@ -440,6 +456,8 @@ class CacheDescriptor(_CacheDescriptorBase):
 
             return make_deferred_yieldable(observer)
 
+        wrapped = cast(_CachedFunction, _wrapped)
+
         if self.num_args == 1:
             wrapped.invalidate = lambda key: cache.invalidate(key[0])
             wrapped.prefill = lambda key, val: cache.prefill(key[0], val)
@@ -464,9 +482,8 @@ class CacheListDescriptor(_CacheDescriptorBase):
     Given a list of keys it looks in the cache to find any hits, then passes
     the list of missing keys to the wrapped function.
 
-    Once wrapped, the function returns either a Deferred which resolves to
-    the list of results, or (if all results were cached), just the list of
-    results.
+    Once wrapped, the function returns a Deferred which resolves to the list
+    of results.
     """
 
     def __init__(
@@ -600,7 +617,7 @@ class CacheListDescriptor(_CacheDescriptorBase):
                 )
                 return make_deferred_yieldable(d)
             else:
-                return results
+                return defer.succeed(results)
 
         obj.__dict__[self.orig.__name__] = wrapped
 
diff --git a/synapse/util/caches/treecache.py b/synapse/util/caches/treecache.py
index 9a72218d85..2ea4e4e911 100644
--- a/synapse/util/caches/treecache.py
+++ b/synapse/util/caches/treecache.py
@@ -1,3 +1,5 @@
+from typing import Dict
+
 from six import itervalues
 
 SENTINEL = object()
@@ -12,7 +14,7 @@ class TreeCache(object):
 
     def __init__(self):
         self.size = 0
-        self.root = {}
+        self.root = {}  # type: Dict
 
     def __setitem__(self, key, value):
         return self.set(key, value)
diff --git a/synapse/util/hash.py b/synapse/util/hash.py
new file mode 100644
index 0000000000..359168704e
--- /dev/null
+++ b/synapse/util/hash.py
@@ -0,0 +1,33 @@
+# -*- coding: utf-8 -*-
+
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import hashlib
+
+import unpaddedbase64
+
+
+def sha256_and_url_safe_base64(input_text):
+    """SHA256 hash an input string, encode the digest as url-safe base64, and
+    return
+
+    :param input_text: string to hash
+    :type input_text: str
+
+    :returns a sha256 hashed and url-safe base64 encoded digest
+    :rtype: str
+    """
+    digest = hashlib.sha256(input_text.encode()).digest()
+    return unpaddedbase64.encode_base64(digest, urlsafe=True)
diff --git a/synapse/util/metrics.py b/synapse/util/metrics.py
index 0910930c21..3286804322 100644
--- a/synapse/util/metrics.py
+++ b/synapse/util/metrics.py
@@ -60,12 +60,14 @@ in_flight = InFlightGauge(
 )
 
 
-def measure_func(name):
+def measure_func(name=None):
     def wrapper(func):
+        block_name = func.__name__ if name is None else name
+
         @wraps(func)
         @defer.inlineCallbacks
         def measured_func(self, *args, **kwargs):
-            with Measure(self.clock, name):
+            with Measure(self.clock, block_name):
                 r = yield func(self, *args, **kwargs)
             return r
 
@@ -117,7 +119,7 @@ class Measure(object):
         context = LoggingContext.current_context()
 
         if context != self.start_context:
-            logger.warn(
+            logger.warning(
                 "Context has unexpectedly changed from '%s' to '%s'. (%r)",
                 self.start_context,
                 context,
@@ -126,7 +128,7 @@ class Measure(object):
             return
 
         if not context:
-            logger.warn("Expected context. (%r)", self.name)
+            logger.warning("Expected context. (%r)", self.name)
             return
 
         current = context.get_resource_usage()
@@ -138,7 +140,7 @@ class Measure(object):
             block_db_txn_duration.labels(self.name).inc(usage.db_txn_duration_sec)
             block_db_sched_duration.labels(self.name).inc(usage.db_sched_duration_sec)
         except ValueError:
-            logger.warn(
+            logger.warning(
                 "Failed to save metrics! OLD: %r, NEW: %r", self.start_usage, current
             )
 
diff --git a/synapse/util/module_loader.py b/synapse/util/module_loader.py
index 522acd5aa8..2705cbe5f8 100644
--- a/synapse/util/module_loader.py
+++ b/synapse/util/module_loader.py
@@ -14,12 +14,13 @@
 # limitations under the License.
 
 import importlib
+import importlib.util
 
 from synapse.config._base import ConfigError
 
 
 def load_module(provider):
-    """ Loads a module with its config
+    """ Loads a synapse module with its config
     Take a dict with keys 'module' (the module name) and 'config'
     (the config dict).
 
@@ -38,3 +39,20 @@ def load_module(provider):
         raise ConfigError("Failed to parse config for %r: %r" % (provider["module"], e))
 
     return provider_class, provider_config
+
+
+def load_python_module(location: str):
+    """Load a python module, and return a reference to its global namespace
+
+    Args:
+        location (str): path to the module
+
+    Returns:
+        python module object
+    """
+    spec = importlib.util.spec_from_file_location(location, location)
+    if spec is None:
+        raise Exception("Unable to load module at %s" % (location,))
+    mod = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)  # type: ignore
+    return mod
diff --git a/synapse/util/patch_inline_callbacks.py b/synapse/util/patch_inline_callbacks.py
new file mode 100644
index 0000000000..3925927f9f
--- /dev/null
+++ b/synapse/util/patch_inline_callbacks.py
@@ -0,0 +1,219 @@
+# -*- coding: utf-8 -*-
+# Copyright 2018 New Vector Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import functools
+import sys
+from typing import Any, Callable, List
+
+from twisted.internet import defer
+from twisted.internet.defer import Deferred
+from twisted.python.failure import Failure
+
+# Tracks if we've already patched inlineCallbacks
+_already_patched = False
+
+
+def do_patch():
+    """
+    Patch defer.inlineCallbacks so that it checks the state of the logcontext on exit
+    """
+
+    from synapse.logging.context import LoggingContext
+
+    global _already_patched
+
+    orig_inline_callbacks = defer.inlineCallbacks
+    if _already_patched:
+        return
+
+    def new_inline_callbacks(f):
+        @functools.wraps(f)
+        def wrapped(*args, **kwargs):
+            start_context = LoggingContext.current_context()
+            changes = []  # type: List[str]
+            orig = orig_inline_callbacks(_check_yield_points(f, changes))
+
+            try:
+                res = orig(*args, **kwargs)
+            except Exception:
+                if LoggingContext.current_context() != start_context:
+                    for err in changes:
+                        print(err, file=sys.stderr)
+
+                    err = "%s changed context from %s to %s on exception" % (
+                        f,
+                        start_context,
+                        LoggingContext.current_context(),
+                    )
+                    print(err, file=sys.stderr)
+                    raise Exception(err)
+                raise
+
+            if not isinstance(res, Deferred) or res.called:
+                if LoggingContext.current_context() != start_context:
+                    for err in changes:
+                        print(err, file=sys.stderr)
+
+                    err = "Completed %s changed context from %s to %s" % (
+                        f,
+                        start_context,
+                        LoggingContext.current_context(),
+                    )
+                    # print the error to stderr because otherwise all we
+                    # see in travis-ci is the 500 error
+                    print(err, file=sys.stderr)
+                    raise Exception(err)
+                return res
+
+            if LoggingContext.current_context() != LoggingContext.sentinel:
+                err = (
+                    "%s returned incomplete deferred in non-sentinel context "
+                    "%s (start was %s)"
+                ) % (f, LoggingContext.current_context(), start_context)
+                print(err, file=sys.stderr)
+                raise Exception(err)
+
+            def check_ctx(r):
+                if LoggingContext.current_context() != start_context:
+                    for err in changes:
+                        print(err, file=sys.stderr)
+                    err = "%s completion of %s changed context from %s to %s" % (
+                        "Failure" if isinstance(r, Failure) else "Success",
+                        f,
+                        start_context,
+                        LoggingContext.current_context(),
+                    )
+                    print(err, file=sys.stderr)
+                    raise Exception(err)
+                return r
+
+            res.addBoth(check_ctx)
+            return res
+
+        return wrapped
+
+    defer.inlineCallbacks = new_inline_callbacks
+    _already_patched = True
+
+
+def _check_yield_points(f: Callable, changes: List[str]):
+    """Wraps a generator that is about to be passed to defer.inlineCallbacks
+    checking that after every yield the log contexts are correct.
+
+    It's perfectly valid for log contexts to change within a function, e.g. due
+    to new Measure blocks, so such changes are added to the given `changes`
+    list instead of triggering an exception.
+
+    Args:
+        f: generator function to wrap
+        changes: A list of strings detailing how the contexts
+            changed within a function.
+
+    Returns:
+        function
+    """
+
+    from synapse.logging.context import LoggingContext
+
+    @functools.wraps(f)
+    def check_yield_points_inner(*args, **kwargs):
+        gen = f(*args, **kwargs)
+
+        last_yield_line_no = gen.gi_frame.f_lineno
+        result = None  # type: Any
+        while True:
+            expected_context = LoggingContext.current_context()
+
+            try:
+                isFailure = isinstance(result, Failure)
+                if isFailure:
+                    d = result.throwExceptionIntoGenerator(gen)
+                else:
+                    d = gen.send(result)
+            except (StopIteration, defer._DefGen_Return) as e:
+                if LoggingContext.current_context() != expected_context:
+                    # This happens when the context is lost sometime *after* the
+                    # final yield and returning. E.g. we forgot to yield on a
+                    # function that returns a deferred.
+                    #
+                    # We don't raise here as it's perfectly valid for contexts to
+                    # change in a function, as long as it sets the correct context
+                    # on resolving (which is checked separately).
+                    err = (
+                        "Function %r returned and changed context from %s to %s,"
+                        " in %s between %d and end of func"
+                        % (
+                            f.__qualname__,
+                            expected_context,
+                            LoggingContext.current_context(),
+                            f.__code__.co_filename,
+                            last_yield_line_no,
+                        )
+                    )
+                    changes.append(err)
+                return getattr(e, "value", None)
+
+            frame = gen.gi_frame
+
+            if isinstance(d, defer.Deferred) and not d.called:
+                # This happens if we yield on a deferred that doesn't follow
+                # the log context rules without wrapping in a `make_deferred_yieldable`.
+                # We raise here as this should never happen.
+                if LoggingContext.current_context() is not LoggingContext.sentinel:
+                    err = (
+                        "%s yielded with context %s rather than sentinel,"
+                        " yielded on line %d in %s"
+                        % (
+                            frame.f_code.co_name,
+                            LoggingContext.current_context(),
+                            frame.f_lineno,
+                            frame.f_code.co_filename,
+                        )
+                    )
+                    raise Exception(err)
+
+            try:
+                result = yield d
+            except Exception as e:
+                result = Failure(e)
+
+            if LoggingContext.current_context() != expected_context:
+
+                # This happens because the context is lost sometime *after* the
+                # previous yield and *after* the current yield. E.g. the
+                # deferred we waited on didn't follow the rules, or we forgot to
+                # yield on a function between the two yield points.
+                #
+                # We don't raise here as its perfectly valid for contexts to
+                # change in a function, as long as it sets the correct context
+                # on resolving (which is checked separately).
+                err = (
+                    "%s changed context from %s to %s, happened between lines %d and %d in %s"
+                    % (
+                        frame.f_code.co_name,
+                        expected_context,
+                        LoggingContext.current_context(),
+                        last_yield_line_no,
+                        frame.f_lineno,
+                        frame.f_code.co_filename,
+                    )
+                )
+                changes.append(err)
+
+            last_yield_line_no = frame.f_lineno
+
+    return check_yield_points_inner
diff --git a/synapse/util/retryutils.py b/synapse/util/retryutils.py
index 0862b5ca5a..af69587196 100644
--- a/synapse/util/retryutils.py
+++ b/synapse/util/retryutils.py
@@ -22,6 +22,15 @@ from synapse.api.errors import CodeMessageException
 
 logger = logging.getLogger(__name__)
 
+# the intial backoff, after the first transaction fails
+MIN_RETRY_INTERVAL = 10 * 60 * 1000
+
+# how much we multiply the backoff by after each subsequent fail
+RETRY_MULTIPLIER = 5
+
+# a cap on the backoff. (Essentially none)
+MAX_RETRY_INTERVAL = 2 ** 62
+
 
 class NotRetryingDestination(Exception):
     def __init__(self, retry_last_ts, retry_interval, destination):
@@ -71,11 +80,13 @@ def get_retry_limiter(destination, clock, store, ignore_backoff=False, **kwargs)
             # We aren't ready to retry that destination.
             raise
     """
+    failure_ts = None
     retry_last_ts, retry_interval = (0, 0)
 
     retry_timings = yield store.get_destination_retry_timings(destination)
 
     if retry_timings:
+        failure_ts = retry_timings["failure_ts"]
         retry_last_ts, retry_interval = (
             retry_timings["retry_last_ts"],
             retry_timings["retry_interval"],
@@ -99,6 +110,7 @@ def get_retry_limiter(destination, clock, store, ignore_backoff=False, **kwargs)
         destination,
         clock,
         store,
+        failure_ts,
         retry_interval,
         backoff_on_failure=backoff_on_failure,
         **kwargs
@@ -111,10 +123,8 @@ class RetryDestinationLimiter(object):
         destination,
         clock,
         store,
+        failure_ts,
         retry_interval,
-        min_retry_interval=10 * 60 * 1000,
-        max_retry_interval=24 * 60 * 60 * 1000,
-        multiplier_retry_interval=5,
         backoff_on_404=False,
         backoff_on_failure=True,
     ):
@@ -127,15 +137,11 @@ class RetryDestinationLimiter(object):
             destination (str)
             clock (Clock)
             store (DataStore)
+            failure_ts (int|None): when this destination started failing (in ms since
+                the epoch), or zero if the last request was successful
             retry_interval (int): The next retry interval taken from the
                 database in milliseconds, or zero if the last request was
                 successful.
-            min_retry_interval (int): The minimum retry interval to use after
-                a failed request, in milliseconds.
-            max_retry_interval (int): The maximum retry interval to use after
-                a failed request, in milliseconds.
-            multiplier_retry_interval (int): The multiplier to use to increase
-                the retry interval after a failed request.
             backoff_on_404 (bool): Back off if we get a 404
 
             backoff_on_failure (bool): set to False if we should not increase the
@@ -145,10 +151,8 @@ class RetryDestinationLimiter(object):
         self.store = store
         self.destination = destination
 
+        self.failure_ts = failure_ts
         self.retry_interval = retry_interval
-        self.min_retry_interval = min_retry_interval
-        self.max_retry_interval = max_retry_interval
-        self.multiplier_retry_interval = multiplier_retry_interval
         self.backoff_on_404 = backoff_on_404
         self.backoff_on_failure = backoff_on_failure
 
@@ -189,6 +193,7 @@ class RetryDestinationLimiter(object):
             logger.debug(
                 "Connection to %s was successful; clearing backoff", self.destination
             )
+            self.failure_ts = None
             retry_last_ts = 0
             self.retry_interval = 0
         elif not self.backoff_on_failure:
@@ -196,13 +201,14 @@ class RetryDestinationLimiter(object):
         else:
             # We couldn't connect.
             if self.retry_interval:
-                self.retry_interval *= self.multiplier_retry_interval
-                self.retry_interval *= int(random.uniform(0.8, 1.4))
+                self.retry_interval = int(
+                    self.retry_interval * RETRY_MULTIPLIER * random.uniform(0.8, 1.4)
+                )
 
-                if self.retry_interval >= self.max_retry_interval:
-                    self.retry_interval = self.max_retry_interval
+                if self.retry_interval >= MAX_RETRY_INTERVAL:
+                    self.retry_interval = MAX_RETRY_INTERVAL
             else:
-                self.retry_interval = self.min_retry_interval
+                self.retry_interval = MIN_RETRY_INTERVAL
 
             logger.info(
                 "Connection to %s was unsuccessful (%s(%s)); backoff now %i",
@@ -213,11 +219,17 @@ class RetryDestinationLimiter(object):
             )
             retry_last_ts = int(self.clock.time_msec())
 
+            if self.failure_ts is None:
+                self.failure_ts = retry_last_ts
+
         @defer.inlineCallbacks
         def store_retry_timings():
             try:
                 yield self.store.set_destination_retry_timings(
-                    self.destination, retry_last_ts, self.retry_interval
+                    self.destination,
+                    self.failure_ts,
+                    retry_last_ts,
+                    self.retry_interval,
                 )
             except Exception:
                 logger.exception("Failed to store destination_retry_timings")
diff --git a/synapse/util/rlimit.py b/synapse/util/rlimit.py
index 6c0f2bb0cf..207cd17c2a 100644
--- a/synapse/util/rlimit.py
+++ b/synapse/util/rlimit.py
@@ -33,4 +33,4 @@ def change_resource_limit(soft_file_no):
             resource.RLIMIT_CORE, (resource.RLIM_INFINITY, resource.RLIM_INFINITY)
         )
     except (ValueError, resource.error) as e:
-        logger.warn("Failed to set file or core limit: %s", e)
+        logger.warning("Failed to set file or core limit: %s", e)
diff --git a/synapse/util/versionstring.py b/synapse/util/versionstring.py
index fa404b9d75..ab7d03af3a 100644
--- a/synapse/util/versionstring.py
+++ b/synapse/util/versionstring.py
@@ -42,6 +42,7 @@ def get_version_string(module):
     try:
         null = open(os.devnull, "w")
         cwd = os.path.dirname(os.path.abspath(module.__file__))
+
         try:
             git_branch = (
                 subprocess.check_output(
@@ -51,7 +52,8 @@ def get_version_string(module):
                 .decode("ascii")
             )
             git_branch = "b=" + git_branch
-        except subprocess.CalledProcessError:
+        except (subprocess.CalledProcessError, FileNotFoundError):
+            # FileNotFoundError can arise when git is not installed
             git_branch = ""
 
         try:
@@ -63,7 +65,7 @@ def get_version_string(module):
                 .decode("ascii")
             )
             git_tag = "t=" + git_tag
-        except subprocess.CalledProcessError:
+        except (subprocess.CalledProcessError, FileNotFoundError):
             git_tag = ""
 
         try:
@@ -74,7 +76,7 @@ def get_version_string(module):
                 .strip()
                 .decode("ascii")
             )
-        except subprocess.CalledProcessError:
+        except (subprocess.CalledProcessError, FileNotFoundError):
             git_commit = ""
 
         try:
@@ -89,7 +91,7 @@ def get_version_string(module):
             )
 
             git_dirty = "dirty" if is_dirty else ""
-        except subprocess.CalledProcessError:
+        except (subprocess.CalledProcessError, FileNotFoundError):
             git_dirty = ""
 
         if git_branch or git_tag or git_commit or git_dirty:
diff --git a/synapse/visibility.py b/synapse/visibility.py
index bf0f1eebd8..8c843febd8 100644
--- a/synapse/visibility.py
+++ b/synapse/visibility.py
@@ -23,6 +23,7 @@ from twisted.internet import defer
 
 from synapse.api.constants import EventTypes, Membership
 from synapse.events.utils import prune_event
+from synapse.storage import Storage
 from synapse.storage.state import StateFilter
 from synapse.types import get_domain_from_id
 
@@ -43,14 +44,13 @@ MEMBERSHIP_PRIORITY = (
 
 @defer.inlineCallbacks
 def filter_events_for_client(
-    store, user_id, events, is_peeking=False, always_include_ids=frozenset()
+    storage: Storage, user_id, events, is_peeking=False, always_include_ids=frozenset()
 ):
     """
     Check which events a user is allowed to see
 
     Args:
-        store (synapse.storage.DataStore): our datastore (can also be a worker
-            store)
+        storage
         user_id(str): user id to be checked
         events(list[synapse.events.EventBase]): sequence of events to be checked
         is_peeking(bool): should be True if:
@@ -68,12 +68,12 @@ def filter_events_for_client(
     events = list(e for e in events if not e.internal_metadata.is_soft_failed())
 
     types = ((EventTypes.RoomHistoryVisibility, ""), (EventTypes.Member, user_id))
-    event_id_to_state = yield store.get_state_for_events(
+    event_id_to_state = yield storage.state.get_state_for_events(
         frozenset(e.event_id for e in events),
         state_filter=StateFilter.from_types(types),
     )
 
-    ignore_dict_content = yield store.get_global_account_data_by_type_for_user(
+    ignore_dict_content = yield storage.main.get_global_account_data_by_type_for_user(
         "m.ignored_user_list", user_id
     )
 
@@ -84,7 +84,7 @@ def filter_events_for_client(
         else []
     )
 
-    erased_senders = yield store.are_users_erased((e.sender for e in events))
+    erased_senders = yield storage.main.are_users_erased((e.sender for e in events))
 
     def allowed(event):
         """
@@ -213,13 +213,17 @@ def filter_events_for_client(
 
 @defer.inlineCallbacks
 def filter_events_for_server(
-    store, server_name, events, redact=True, check_history_visibility_only=False
+    storage: Storage,
+    server_name,
+    events,
+    redact=True,
+    check_history_visibility_only=False,
 ):
     """Filter a list of events based on whether given server is allowed to
     see them.
 
     Args:
-        store (DataStore)
+        storage
         server_name (str)
         events (iterable[FrozenEvent])
         redact (bool): Whether to return a redacted version of the event, or
@@ -274,7 +278,7 @@ def filter_events_for_server(
     # Lets check to see if all the events have a history visibility
     # of "shared" or "world_readable". If thats the case then we don't
     # need to check membership (as we know the server is in the room).
-    event_to_state_ids = yield store.get_state_ids_for_events(
+    event_to_state_ids = yield storage.state.get_state_ids_for_events(
         frozenset(e.event_id for e in events),
         state_filter=StateFilter.from_types(
             types=((EventTypes.RoomHistoryVisibility, ""),)
@@ -292,14 +296,14 @@ def filter_events_for_server(
     if not visibility_ids:
         all_open = True
     else:
-        event_map = yield store.get_events(visibility_ids)
+        event_map = yield storage.main.get_events(visibility_ids)
         all_open = all(
             e.content.get("history_visibility") in (None, "shared", "world_readable")
             for e in itervalues(event_map)
         )
 
     if not check_history_visibility_only:
-        erased_senders = yield store.are_users_erased((e.sender for e in events))
+        erased_senders = yield storage.main.are_users_erased((e.sender for e in events))
     else:
         # We don't want to check whether users are erased, which is equivalent
         # to no users having been erased.
@@ -328,7 +332,7 @@ def filter_events_for_server(
 
     # first, for each event we're wanting to return, get the event_ids
     # of the history vis and membership state at those events.
-    event_to_state_ids = yield store.get_state_ids_for_events(
+    event_to_state_ids = yield storage.state.get_state_ids_for_events(
         frozenset(e.event_id for e in events),
         state_filter=StateFilter.from_types(
             types=((EventTypes.RoomHistoryVisibility, ""), (EventTypes.Member, None))
@@ -358,7 +362,7 @@ def filter_events_for_server(
             return False
         return state_key[idx + 1 :] == server_name
 
-    event_map = yield store.get_events(
+    event_map = yield storage.main.get_events(
         [
             e_id
             for e_id, key in iteritems(event_id_to_state_key)
diff --git a/synctl b/synctl
index a9629cf0e8..45acece30b 100755
--- a/synctl
+++ b/synctl
@@ -71,7 +71,20 @@ def abort(message, colour=RED, stream=sys.stderr):
     sys.exit(1)
 
 
-def start(configfile, daemonize=True):
+def start(configfile: str, daemonize: bool = True) -> bool:
+    """Attempts to start synapse.
+    Args:
+        configfile: path to a yaml synapse config file
+        daemonize: whether to daemonize synapse or keep it attached to the current
+            session
+
+    Returns:
+        True if the process started successfully
+        False if there was an error starting the process
+
+        If deamonize is False it will only return once synapse exits.
+    """
+
     write("Starting ...")
     args = SYNAPSE
 
@@ -83,25 +96,40 @@ def start(configfile, daemonize=True):
     try:
         subprocess.check_call(args)
         write("started synapse.app.homeserver(%r)" % (configfile,), colour=GREEN)
+        return True
     except subprocess.CalledProcessError as e:
         write(
             "error starting (exit code: %d); see above for logs" % e.returncode,
             colour=RED,
         )
+        return False
 
 
-def start_worker(app, configfile, worker_configfile):
+def start_worker(app: str, configfile: str, worker_configfile: str) -> bool:
+    """Attempts to start a synapse worker.
+    Args:
+        app: name of the worker's appservice
+        configfile: path to a yaml synapse config file
+        worker_configfile: path to worker specific yaml synapse file
+
+    Returns:
+        True if the process started successfully
+        False if there was an error starting the process
+    """
+
     args = [sys.executable, "-B", "-m", app, "-c", configfile, "-c", worker_configfile]
 
     try:
         subprocess.check_call(args)
         write("started %s(%r)" % (app, worker_configfile), colour=GREEN)
+        return True
     except subprocess.CalledProcessError as e:
         write(
             "error starting %s(%r) (exit code: %d); see above for logs"
             % (app, worker_configfile, e.returncode),
             colour=RED,
         )
+        return False
 
 
 def stop(pidfile, app):
@@ -292,11 +320,14 @@ def main():
             write("All processes exited; now restarting...")
 
     if action == "start" or action == "restart":
+        error = False
         if start_stop_synapse:
             # Check if synapse is already running
             if os.path.exists(pidfile) and pid_running(int(open(pidfile).read())):
                 abort("synapse.app.homeserver already running")
-            start(configfile, bool(options.daemonize))
+
+            if not start(configfile, bool(options.daemonize)):
+                error = True
 
         for worker in workers:
             env = os.environ.copy()
@@ -307,12 +338,16 @@ def main():
             for cache_name, factor in iteritems(worker.cache_factors):
                 os.environ["SYNAPSE_CACHE_FACTOR_" + cache_name.upper()] = str(factor)
 
-            start_worker(worker.app, configfile, worker.configfile)
+            if not start_worker(worker.app, configfile, worker.configfile):
+                error = True
 
             # Reset env back to the original
             os.environ.clear()
             os.environ.update(env)
 
+        if error:
+            exit(1)
+
 
 if __name__ == "__main__":
     main()
diff --git a/tests/__init__.py b/tests/__init__.py
index f7fc502f01..ed805db1c2 100644
--- a/tests/__init__.py
+++ b/tests/__init__.py
@@ -16,9 +16,9 @@
 
 from twisted.trial import util
 
-import tests.patch_inline_callbacks
+from synapse.util.patch_inline_callbacks import do_patch
 
 # attempt to do the patch before we load any synapse code
-tests.patch_inline_callbacks.do_patch()
+do_patch()
 
 util.DEFAULT_TIMEOUT_DURATION = 20
diff --git a/tests/api/test_auth.py b/tests/api/test_auth.py
index c0cb8ef296..6121efcfa9 100644
--- a/tests/api/test_auth.py
+++ b/tests/api/test_auth.py
@@ -21,6 +21,7 @@ from twisted.internet import defer
 
 import synapse.handlers.auth
 from synapse.api.auth import Auth
+from synapse.api.constants import UserTypes
 from synapse.api.errors import (
     AuthError,
     Codes,
@@ -335,6 +336,23 @@ class AuthTestCase(unittest.TestCase):
         )
         yield self.auth.check_auth_blocking()
 
+    @defer.inlineCallbacks
+    def test_blocking_mau__depending_on_user_type(self):
+        self.hs.config.max_mau_value = 50
+        self.hs.config.limit_usage_by_mau = True
+
+        self.store.get_monthly_active_count = Mock(return_value=defer.succeed(100))
+        # Support users allowed
+        yield self.auth.check_auth_blocking(user_type=UserTypes.SUPPORT)
+        self.store.get_monthly_active_count = Mock(return_value=defer.succeed(100))
+        # Bots not allowed
+        with self.assertRaises(ResourceLimitError):
+            yield self.auth.check_auth_blocking(user_type=UserTypes.BOT)
+        self.store.get_monthly_active_count = Mock(return_value=defer.succeed(100))
+        # Real users not allowed
+        with self.assertRaises(ResourceLimitError):
+            yield self.auth.check_auth_blocking()
+
     @defer.inlineCallbacks
     def test_reserved_threepid(self):
         self.hs.config.limit_usage_by_mau = True
diff --git a/tests/api/test_filtering.py b/tests/api/test_filtering.py
index 6ba623de13..2dc5052249 100644
--- a/tests/api/test_filtering.py
+++ b/tests/api/test_filtering.py
@@ -19,6 +19,7 @@ import jsonschema
 
 from twisted.internet import defer
 
+from synapse.api.constants import EventContentFields
 from synapse.api.errors import SynapseError
 from synapse.api.filtering import Filter
 from synapse.events import FrozenEvent
@@ -95,6 +96,8 @@ class FilteringTestCase(unittest.TestCase):
                         "types": ["m.room.message"],
                         "not_rooms": ["!726s6s6q:example.com"],
                         "not_senders": ["@spam:example.com"],
+                        "org.matrix.labels": ["#fun"],
+                        "org.matrix.not_labels": ["#work"],
                     },
                     "ephemeral": {
                         "types": ["m.receipt", "m.typing"],
@@ -320,6 +323,46 @@ class FilteringTestCase(unittest.TestCase):
         )
         self.assertFalse(Filter(definition).check(event))
 
+    def test_filter_labels(self):
+        definition = {"org.matrix.labels": ["#fun"]}
+        event = MockEvent(
+            sender="@foo:bar",
+            type="m.room.message",
+            room_id="!secretbase:unknown",
+            content={EventContentFields.LABELS: ["#fun"]},
+        )
+
+        self.assertTrue(Filter(definition).check(event))
+
+        event = MockEvent(
+            sender="@foo:bar",
+            type="m.room.message",
+            room_id="!secretbase:unknown",
+            content={EventContentFields.LABELS: ["#notfun"]},
+        )
+
+        self.assertFalse(Filter(definition).check(event))
+
+    def test_filter_not_labels(self):
+        definition = {"org.matrix.not_labels": ["#fun"]}
+        event = MockEvent(
+            sender="@foo:bar",
+            type="m.room.message",
+            room_id="!secretbase:unknown",
+            content={EventContentFields.LABELS: ["#fun"]},
+        )
+
+        self.assertFalse(Filter(definition).check(event))
+
+        event = MockEvent(
+            sender="@foo:bar",
+            type="m.room.message",
+            room_id="!secretbase:unknown",
+            content={EventContentFields.LABELS: ["#notfun"]},
+        )
+
+        self.assertTrue(Filter(definition).check(event))
+
     @defer.inlineCallbacks
     def test_filter_presence_match(self):
         user_filter_json = {"presence": {"types": ["m.*"]}}
diff --git a/tests/config/test_generate.py b/tests/config/test_generate.py
index 5017cbce85..2684e662de 100644
--- a/tests/config/test_generate.py
+++ b/tests/config/test_generate.py
@@ -17,6 +17,8 @@ import os.path
 import re
 import shutil
 import tempfile
+from contextlib import redirect_stdout
+from io import StringIO
 
 from synapse.config.homeserver import HomeServerConfig
 
@@ -32,17 +34,18 @@ class ConfigGenerationTestCase(unittest.TestCase):
         shutil.rmtree(self.dir)
 
     def test_generate_config_generates_files(self):
-        HomeServerConfig.load_or_generate_config(
-            "",
-            [
-                "--generate-config",
-                "-c",
-                self.file,
-                "--report-stats=yes",
-                "-H",
-                "lemurs.win",
-            ],
-        )
+        with redirect_stdout(StringIO()):
+            HomeServerConfig.load_or_generate_config(
+                "",
+                [
+                    "--generate-config",
+                    "-c",
+                    self.file,
+                    "--report-stats=yes",
+                    "-H",
+                    "lemurs.win",
+                ],
+            )
 
         self.assertSetEqual(
             set(["homeserver.yaml", "lemurs.win.log.config", "lemurs.win.signing.key"]),
diff --git a/tests/config/test_load.py b/tests/config/test_load.py
index 6bfc1970ad..b3e557bd6a 100644
--- a/tests/config/test_load.py
+++ b/tests/config/test_load.py
@@ -15,6 +15,8 @@
 import os.path
 import shutil
 import tempfile
+from contextlib import redirect_stdout
+from io import StringIO
 
 import yaml
 
@@ -26,7 +28,6 @@ from tests import unittest
 class ConfigLoadingTestCase(unittest.TestCase):
     def setUp(self):
         self.dir = tempfile.mkdtemp()
-        print(self.dir)
         self.file = os.path.join(self.dir, "homeserver.yaml")
 
     def tearDown(self):
@@ -94,18 +95,27 @@ class ConfigLoadingTestCase(unittest.TestCase):
         )
         self.assertTrue(config.enable_registration)
 
+    def test_stats_enabled(self):
+        self.generate_config_and_remove_lines_containing("enable_metrics")
+        self.add_lines_to_config(["enable_metrics: true"])
+
+        # The default Metrics Flags are off by default.
+        config = HomeServerConfig.load_config("", ["-c", self.file])
+        self.assertFalse(config.metrics_flags.known_servers)
+
     def generate_config(self):
-        HomeServerConfig.load_or_generate_config(
-            "",
-            [
-                "--generate-config",
-                "-c",
-                self.file,
-                "--report-stats=yes",
-                "-H",
-                "lemurs.win",
-            ],
-        )
+        with redirect_stdout(StringIO()):
+            HomeServerConfig.load_or_generate_config(
+                "",
+                [
+                    "--generate-config",
+                    "-c",
+                    self.file,
+                    "--report-stats=yes",
+                    "-H",
+                    "lemurs.win",
+                ],
+            )
 
     def generate_config_and_remove_lines_containing(self, needle):
         self.generate_config()
diff --git a/tests/config/test_tls.py b/tests/config/test_tls.py
index 8e0c4b9533..1be6ff563b 100644
--- a/tests/config/test_tls.py
+++ b/tests/config/test_tls.py
@@ -16,21 +16,29 @@
 
 import os
 
+import idna
 import yaml
 
 from OpenSSL import SSL
 
+from synapse.config._base import Config, RootConfig
 from synapse.config.tls import ConfigError, TlsConfig
 from synapse.crypto.context_factory import ClientTLSOptionsFactory
 
 from tests.unittest import TestCase
 
 
-class TestConfig(TlsConfig):
+class FakeServer(Config):
+    section = "server"
+
     def has_tls_listener(self):
         return False
 
 
+class TestConfig(RootConfig):
+    config_classes = [FakeServer, TlsConfig]
+
+
 class TLSConfigTests(TestCase):
     def test_warn_self_signed(self):
         """
@@ -201,13 +209,13 @@ s4niecZKPBizL6aucT59CsunNmmb5Glq8rlAcU+1ZTZZzGYqVYhF6axB9Qg=
         conf = TestConfig()
         conf.read_config(
             yaml.safe_load(
-                TestConfig().generate_config_section(
+                TestConfig().generate_config(
                     "/config_dir_path",
                     "my_super_secure_server",
                     "/data_dir_path",
-                    "/tls_cert_path",
-                    "tls_private_key",
-                    None,  # This is the acme_domain
+                    tls_certificate_path="/tls_cert_path",
+                    tls_private_key_path="tls_private_key",
+                    acme_domain=None,  # This is the acme_domain
                 )
             ),
             "/config_dir_path",
@@ -222,16 +230,55 @@ s4niecZKPBizL6aucT59CsunNmmb5Glq8rlAcU+1ZTZZzGYqVYhF6axB9Qg=
         conf = TestConfig()
         conf.read_config(
             yaml.safe_load(
-                TestConfig().generate_config_section(
+                TestConfig().generate_config(
                     "/config_dir_path",
                     "my_super_secure_server",
                     "/data_dir_path",
-                    "/tls_cert_path",
-                    "tls_private_key",
-                    "my_supe_secure_server",  # This is the acme_domain
+                    tls_certificate_path="/tls_cert_path",
+                    tls_private_key_path="tls_private_key",
+                    acme_domain="my_supe_secure_server",  # This is the acme_domain
                 )
             ),
             "/config_dir_path",
         )
 
         self.assertTrue(conf.acme_enabled)
+
+    def test_whitelist_idna_failure(self):
+        """
+        The federation certificate whitelist will not allow IDNA domain names.
+        """
+        config = {
+            "federation_certificate_verification_whitelist": [
+                "example.com",
+                "*.ドメイン.テスト",
+            ]
+        }
+        t = TestConfig()
+        e = self.assertRaises(
+            ConfigError, t.read_config, config, config_dir_path="", data_dir_path=""
+        )
+        self.assertIn("IDNA domain names", str(e))
+
+    def test_whitelist_idna_result(self):
+        """
+        The federation certificate whitelist will match on IDNA encoded names.
+        """
+        config = {
+            "federation_certificate_verification_whitelist": [
+                "example.com",
+                "*.xn--eckwd4c7c.xn--zckzah",
+            ]
+        }
+        t = TestConfig()
+        t.read_config(config, config_dir_path="", data_dir_path="")
+
+        cf = ClientTLSOptionsFactory(t)
+
+        # Not in the whitelist
+        opts = cf.get_options(b"notexample.com")
+        self.assertTrue(opts._verifier._verify_certs)
+
+        # Caught by the wildcard
+        opts = cf.get_options(idna.encode("テスト.ドメイン.テスト"))
+        self.assertFalse(opts._verifier._verify_certs)
diff --git a/tests/crypto/test_keyring.py b/tests/crypto/test_keyring.py
index c4f0bbd3dd..8efd39c7f7 100644
--- a/tests/crypto/test_keyring.py
+++ b/tests/crypto/test_keyring.py
@@ -178,7 +178,7 @@ class KeyringTestCase(unittest.HomeserverTestCase):
         kr = keyring.Keyring(self.hs)
 
         key1 = signedjson.key.generate_signing_key(1)
-        r = self.hs.datastore.store_server_verify_keys(
+        r = self.hs.get_datastore().store_server_verify_keys(
             "server9",
             time.time() * 1000,
             [("server9", get_key_id(key1), FetchKeyResult(get_verify_key(key1), 1000))],
@@ -209,7 +209,7 @@ class KeyringTestCase(unittest.HomeserverTestCase):
         )
 
         key1 = signedjson.key.generate_signing_key(1)
-        r = self.hs.datastore.store_server_verify_keys(
+        r = self.hs.get_datastore().store_server_verify_keys(
             "server9",
             time.time() * 1000,
             [("server9", get_key_id(key1), FetchKeyResult(get_verify_key(key1), None))],
diff --git a/tests/handlers/test_e2e_keys.py b/tests/handlers/test_e2e_keys.py
index 8dccc6826e..854eb6c024 100644
--- a/tests/handlers/test_e2e_keys.py
+++ b/tests/handlers/test_e2e_keys.py
@@ -1,5 +1,7 @@
 # -*- coding: utf-8 -*-
 # Copyright 2016 OpenMarket Ltd
+# Copyright 2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -15,9 +17,11 @@
 
 import mock
 
+import signedjson.key as key
+import signedjson.sign as sign
+
 from twisted.internet import defer
 
-import synapse.api.errors
 import synapse.handlers.e2e_keys
 import synapse.storage
 from synapse.api import errors
@@ -145,3 +149,357 @@ class E2eKeysHandlerTestCase(unittest.TestCase):
                 "one_time_keys": {local_user: {device_id: {"alg1:k1": "key1"}}},
             },
         )
+
+    @defer.inlineCallbacks
+    def test_replace_master_key(self):
+        """uploading a new signing key should make the old signing key unavailable"""
+        local_user = "@boris:" + self.hs.hostname
+        keys1 = {
+            "master_key": {
+                # private key: 2lonYOM6xYKdEsO+6KrC766xBcHnYnim1x/4LFGF8B0
+                "user_id": local_user,
+                "usage": ["master"],
+                "keys": {
+                    "ed25519:nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk": "nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk"
+                },
+            }
+        }
+        yield self.handler.upload_signing_keys_for_user(local_user, keys1)
+
+        keys2 = {
+            "master_key": {
+                # private key: 4TL4AjRYwDVwD3pqQzcor+ez/euOB1/q78aTJ+czDNs
+                "user_id": local_user,
+                "usage": ["master"],
+                "keys": {
+                    "ed25519:Hq6gL+utB4ET+UvD5ci0kgAwsX6qP/zvf8v6OInU5iw": "Hq6gL+utB4ET+UvD5ci0kgAwsX6qP/zvf8v6OInU5iw"
+                },
+            }
+        }
+        yield self.handler.upload_signing_keys_for_user(local_user, keys2)
+
+        devices = yield self.handler.query_devices(
+            {"device_keys": {local_user: []}}, 0, local_user
+        )
+        self.assertDictEqual(devices["master_keys"], {local_user: keys2["master_key"]})
+
+    @defer.inlineCallbacks
+    def test_reupload_signatures(self):
+        """re-uploading a signature should not fail"""
+        local_user = "@boris:" + self.hs.hostname
+        keys1 = {
+            "master_key": {
+                # private key: HvQBbU+hc2Zr+JP1sE0XwBe1pfZZEYtJNPJLZJtS+F8
+                "user_id": local_user,
+                "usage": ["master"],
+                "keys": {
+                    "ed25519:EmkqvokUn8p+vQAGZitOk4PWjp7Ukp3txV2TbMPEiBQ": "EmkqvokUn8p+vQAGZitOk4PWjp7Ukp3txV2TbMPEiBQ"
+                },
+            },
+            "self_signing_key": {
+                # private key: 2lonYOM6xYKdEsO+6KrC766xBcHnYnim1x/4LFGF8B0
+                "user_id": local_user,
+                "usage": ["self_signing"],
+                "keys": {
+                    "ed25519:nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk": "nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk"
+                },
+            },
+        }
+        master_signing_key = key.decode_signing_key_base64(
+            "ed25519",
+            "EmkqvokUn8p+vQAGZitOk4PWjp7Ukp3txV2TbMPEiBQ",
+            "HvQBbU+hc2Zr+JP1sE0XwBe1pfZZEYtJNPJLZJtS+F8",
+        )
+        sign.sign_json(keys1["self_signing_key"], local_user, master_signing_key)
+        signing_key = key.decode_signing_key_base64(
+            "ed25519",
+            "nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk",
+            "2lonYOM6xYKdEsO+6KrC766xBcHnYnim1x/4LFGF8B0",
+        )
+        yield self.handler.upload_signing_keys_for_user(local_user, keys1)
+
+        # upload two device keys, which will be signed later by the self-signing key
+        device_key_1 = {
+            "user_id": local_user,
+            "device_id": "abc",
+            "algorithms": ["m.olm.curve25519-aes-sha256", "m.megolm.v1.aes-sha"],
+            "keys": {
+                "ed25519:abc": "base64+ed25519+key",
+                "curve25519:abc": "base64+curve25519+key",
+            },
+            "signatures": {local_user: {"ed25519:abc": "base64+signature"}},
+        }
+        device_key_2 = {
+            "user_id": local_user,
+            "device_id": "def",
+            "algorithms": ["m.olm.curve25519-aes-sha256", "m.megolm.v1.aes-sha"],
+            "keys": {
+                "ed25519:def": "base64+ed25519+key",
+                "curve25519:def": "base64+curve25519+key",
+            },
+            "signatures": {local_user: {"ed25519:def": "base64+signature"}},
+        }
+
+        yield self.handler.upload_keys_for_user(
+            local_user, "abc", {"device_keys": device_key_1}
+        )
+        yield self.handler.upload_keys_for_user(
+            local_user, "def", {"device_keys": device_key_2}
+        )
+
+        # sign the first device key and upload it
+        del device_key_1["signatures"]
+        sign.sign_json(device_key_1, local_user, signing_key)
+        yield self.handler.upload_signatures_for_device_keys(
+            local_user, {local_user: {"abc": device_key_1}}
+        )
+
+        # sign the second device key and upload both device keys.  The server
+        # should ignore the first device key since it already has a valid
+        # signature for it
+        del device_key_2["signatures"]
+        sign.sign_json(device_key_2, local_user, signing_key)
+        yield self.handler.upload_signatures_for_device_keys(
+            local_user, {local_user: {"abc": device_key_1, "def": device_key_2}}
+        )
+
+        device_key_1["signatures"][local_user]["ed25519:abc"] = "base64+signature"
+        device_key_2["signatures"][local_user]["ed25519:def"] = "base64+signature"
+        devices = yield self.handler.query_devices(
+            {"device_keys": {local_user: []}}, 0, local_user
+        )
+        del devices["device_keys"][local_user]["abc"]["unsigned"]
+        del devices["device_keys"][local_user]["def"]["unsigned"]
+        self.assertDictEqual(devices["device_keys"][local_user]["abc"], device_key_1)
+        self.assertDictEqual(devices["device_keys"][local_user]["def"], device_key_2)
+
+    @defer.inlineCallbacks
+    def test_self_signing_key_doesnt_show_up_as_device(self):
+        """signing keys should be hidden when fetching a user's devices"""
+        local_user = "@boris:" + self.hs.hostname
+        keys1 = {
+            "master_key": {
+                # private key: 2lonYOM6xYKdEsO+6KrC766xBcHnYnim1x/4LFGF8B0
+                "user_id": local_user,
+                "usage": ["master"],
+                "keys": {
+                    "ed25519:nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk": "nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk"
+                },
+            }
+        }
+        yield self.handler.upload_signing_keys_for_user(local_user, keys1)
+
+        res = None
+        try:
+            yield self.hs.get_device_handler().check_device_registered(
+                user_id=local_user,
+                device_id="nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk",
+                initial_device_display_name="new display name",
+            )
+        except errors.SynapseError as e:
+            res = e.code
+        self.assertEqual(res, 400)
+
+        res = yield self.handler.query_local_devices({local_user: None})
+        self.assertDictEqual(res, {local_user: {}})
+
+    @defer.inlineCallbacks
+    def test_upload_signatures(self):
+        """should check signatures that are uploaded"""
+        # set up a user with cross-signing keys and a device.  This user will
+        # try uploading signatures
+        local_user = "@boris:" + self.hs.hostname
+        device_id = "xyz"
+        # private key: OMkooTr76ega06xNvXIGPbgvvxAOzmQncN8VObS7aBA
+        device_pubkey = "NnHhnqiMFQkq969szYkooLaBAXW244ZOxgukCvm2ZeY"
+        device_key = {
+            "user_id": local_user,
+            "device_id": device_id,
+            "algorithms": ["m.olm.curve25519-aes-sha256", "m.megolm.v1.aes-sha"],
+            "keys": {"curve25519:xyz": "curve25519+key", "ed25519:xyz": device_pubkey},
+            "signatures": {local_user: {"ed25519:xyz": "something"}},
+        }
+        device_signing_key = key.decode_signing_key_base64(
+            "ed25519", "xyz", "OMkooTr76ega06xNvXIGPbgvvxAOzmQncN8VObS7aBA"
+        )
+
+        yield self.handler.upload_keys_for_user(
+            local_user, device_id, {"device_keys": device_key}
+        )
+
+        # private key: 2lonYOM6xYKdEsO+6KrC766xBcHnYnim1x/4LFGF8B0
+        master_pubkey = "nqOvzeuGWT/sRx3h7+MHoInYj3Uk2LD/unI9kDYcHwk"
+        master_key = {
+            "user_id": local_user,
+            "usage": ["master"],
+            "keys": {"ed25519:" + master_pubkey: master_pubkey},
+        }
+        master_signing_key = key.decode_signing_key_base64(
+            "ed25519", master_pubkey, "2lonYOM6xYKdEsO+6KrC766xBcHnYnim1x/4LFGF8B0"
+        )
+        usersigning_pubkey = "Hq6gL+utB4ET+UvD5ci0kgAwsX6qP/zvf8v6OInU5iw"
+        usersigning_key = {
+            # private key: 4TL4AjRYwDVwD3pqQzcor+ez/euOB1/q78aTJ+czDNs
+            "user_id": local_user,
+            "usage": ["user_signing"],
+            "keys": {"ed25519:" + usersigning_pubkey: usersigning_pubkey},
+        }
+        usersigning_signing_key = key.decode_signing_key_base64(
+            "ed25519", usersigning_pubkey, "4TL4AjRYwDVwD3pqQzcor+ez/euOB1/q78aTJ+czDNs"
+        )
+        sign.sign_json(usersigning_key, local_user, master_signing_key)
+        # private key: HvQBbU+hc2Zr+JP1sE0XwBe1pfZZEYtJNPJLZJtS+F8
+        selfsigning_pubkey = "EmkqvokUn8p+vQAGZitOk4PWjp7Ukp3txV2TbMPEiBQ"
+        selfsigning_key = {
+            "user_id": local_user,
+            "usage": ["self_signing"],
+            "keys": {"ed25519:" + selfsigning_pubkey: selfsigning_pubkey},
+        }
+        selfsigning_signing_key = key.decode_signing_key_base64(
+            "ed25519", selfsigning_pubkey, "HvQBbU+hc2Zr+JP1sE0XwBe1pfZZEYtJNPJLZJtS+F8"
+        )
+        sign.sign_json(selfsigning_key, local_user, master_signing_key)
+        cross_signing_keys = {
+            "master_key": master_key,
+            "user_signing_key": usersigning_key,
+            "self_signing_key": selfsigning_key,
+        }
+        yield self.handler.upload_signing_keys_for_user(local_user, cross_signing_keys)
+
+        # set up another user with a master key.  This user will be signed by
+        # the first user
+        other_user = "@otherboris:" + self.hs.hostname
+        other_master_pubkey = "fHZ3NPiKxoLQm5OoZbKa99SYxprOjNs4TwJUKP+twCM"
+        other_master_key = {
+            # private key: oyw2ZUx0O4GifbfFYM0nQvj9CL0b8B7cyN4FprtK8OI
+            "user_id": other_user,
+            "usage": ["master"],
+            "keys": {"ed25519:" + other_master_pubkey: other_master_pubkey},
+        }
+        yield self.handler.upload_signing_keys_for_user(
+            other_user, {"master_key": other_master_key}
+        )
+
+        # test various signature failures (see below)
+        ret = yield self.handler.upload_signatures_for_device_keys(
+            local_user,
+            {
+                local_user: {
+                    # fails because the signature is invalid
+                    # should fail with INVALID_SIGNATURE
+                    device_id: {
+                        "user_id": local_user,
+                        "device_id": device_id,
+                        "algorithms": [
+                            "m.olm.curve25519-aes-sha256",
+                            "m.megolm.v1.aes-sha",
+                        ],
+                        "keys": {
+                            "curve25519:xyz": "curve25519+key",
+                            # private key: OMkooTr76ega06xNvXIGPbgvvxAOzmQncN8VObS7aBA
+                            "ed25519:xyz": device_pubkey,
+                        },
+                        "signatures": {
+                            local_user: {"ed25519:" + selfsigning_pubkey: "something"}
+                        },
+                    },
+                    # fails because device is unknown
+                    # should fail with NOT_FOUND
+                    "unknown": {
+                        "user_id": local_user,
+                        "device_id": "unknown",
+                        "signatures": {
+                            local_user: {"ed25519:" + selfsigning_pubkey: "something"}
+                        },
+                    },
+                    # fails because the signature is invalid
+                    # should fail with INVALID_SIGNATURE
+                    master_pubkey: {
+                        "user_id": local_user,
+                        "usage": ["master"],
+                        "keys": {"ed25519:" + master_pubkey: master_pubkey},
+                        "signatures": {
+                            local_user: {"ed25519:" + device_pubkey: "something"}
+                        },
+                    },
+                },
+                other_user: {
+                    # fails because the device is not the user's master-signing key
+                    # should fail with NOT_FOUND
+                    "unknown": {
+                        "user_id": other_user,
+                        "device_id": "unknown",
+                        "signatures": {
+                            local_user: {"ed25519:" + usersigning_pubkey: "something"}
+                        },
+                    },
+                    other_master_pubkey: {
+                        # fails because the key doesn't match what the server has
+                        # should fail with UNKNOWN
+                        "user_id": other_user,
+                        "usage": ["master"],
+                        "keys": {"ed25519:" + other_master_pubkey: other_master_pubkey},
+                        "something": "random",
+                        "signatures": {
+                            local_user: {"ed25519:" + usersigning_pubkey: "something"}
+                        },
+                    },
+                },
+            },
+        )
+
+        user_failures = ret["failures"][local_user]
+        self.assertEqual(
+            user_failures[device_id]["errcode"], errors.Codes.INVALID_SIGNATURE
+        )
+        self.assertEqual(
+            user_failures[master_pubkey]["errcode"], errors.Codes.INVALID_SIGNATURE
+        )
+        self.assertEqual(user_failures["unknown"]["errcode"], errors.Codes.NOT_FOUND)
+
+        other_user_failures = ret["failures"][other_user]
+        self.assertEqual(
+            other_user_failures["unknown"]["errcode"], errors.Codes.NOT_FOUND
+        )
+        self.assertEqual(
+            other_user_failures[other_master_pubkey]["errcode"], errors.Codes.UNKNOWN
+        )
+
+        # test successful signatures
+        del device_key["signatures"]
+        sign.sign_json(device_key, local_user, selfsigning_signing_key)
+        sign.sign_json(master_key, local_user, device_signing_key)
+        sign.sign_json(other_master_key, local_user, usersigning_signing_key)
+        ret = yield self.handler.upload_signatures_for_device_keys(
+            local_user,
+            {
+                local_user: {device_id: device_key, master_pubkey: master_key},
+                other_user: {other_master_pubkey: other_master_key},
+            },
+        )
+
+        self.assertEqual(ret["failures"], {})
+
+        # fetch the signed keys/devices and make sure that the signatures are there
+        ret = yield self.handler.query_devices(
+            {"device_keys": {local_user: [], other_user: []}}, 0, local_user
+        )
+
+        self.assertEqual(
+            ret["device_keys"][local_user]["xyz"]["signatures"][local_user][
+                "ed25519:" + selfsigning_pubkey
+            ],
+            device_key["signatures"][local_user]["ed25519:" + selfsigning_pubkey],
+        )
+        self.assertEqual(
+            ret["master_keys"][local_user]["signatures"][local_user][
+                "ed25519:" + device_id
+            ],
+            master_key["signatures"][local_user]["ed25519:" + device_id],
+        )
+        self.assertEqual(
+            ret["master_keys"][other_user]["signatures"][local_user][
+                "ed25519:" + usersigning_pubkey
+            ],
+            other_master_key["signatures"][local_user]["ed25519:" + usersigning_pubkey],
+        )
diff --git a/tests/handlers/test_e2e_room_keys.py b/tests/handlers/test_e2e_room_keys.py
index 6d03342a4f..1d0007dda9 100644
--- a/tests/handlers/test_e2e_room_keys.py
+++ b/tests/handlers/test_e2e_room_keys.py
@@ -198,9 +198,8 @@ class E2eRoomKeysHandlerTestCase(unittest.TestCase):
         self.assertEqual(res, 404)
 
     @defer.inlineCallbacks
-    def test_update_bad_version(self):
-        """Check that we get a 400 if the version in the body is missing or
-        doesn't match
+    def test_update_omitted_version(self):
+        """Check that the update succeeds if the version is missing from the body
         """
         version = yield self.handler.create_version(
             self.local_user,
@@ -208,19 +207,35 @@ class E2eRoomKeysHandlerTestCase(unittest.TestCase):
         )
         self.assertEqual(version, "1")
 
-        res = None
-        try:
-            yield self.handler.update_version(
-                self.local_user,
-                version,
-                {
-                    "algorithm": "m.megolm_backup.v1",
-                    "auth_data": "revised_first_version_auth_data",
-                },
-            )
-        except errors.SynapseError as e:
-            res = e.code
-        self.assertEqual(res, 400)
+        yield self.handler.update_version(
+            self.local_user,
+            version,
+            {
+                "algorithm": "m.megolm_backup.v1",
+                "auth_data": "revised_first_version_auth_data",
+            },
+        )
+
+        # check we can retrieve it as the current version
+        res = yield self.handler.get_version_info(self.local_user)
+        self.assertDictEqual(
+            res,
+            {
+                "algorithm": "m.megolm_backup.v1",
+                "auth_data": "revised_first_version_auth_data",
+                "version": version,
+            },
+        )
+
+    @defer.inlineCallbacks
+    def test_update_bad_version(self):
+        """Check that we get a 400 if the version in the body doesn't match
+        """
+        version = yield self.handler.create_version(
+            self.local_user,
+            {"algorithm": "m.megolm_backup.v1", "auth_data": "first_version_auth_data"},
+        )
+        self.assertEqual(version, "1")
 
         res = None
         try:
diff --git a/tests/handlers/test_federation.py b/tests/handlers/test_federation.py
new file mode 100644
index 0000000000..b4d92cf732
--- /dev/null
+++ b/tests/handlers/test_federation.py
@@ -0,0 +1,207 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import logging
+
+from synapse.api.constants import EventTypes
+from synapse.api.errors import AuthError, Codes
+from synapse.federation.federation_base import event_from_pdu_json
+from synapse.logging.context import LoggingContext, run_in_background
+from synapse.rest import admin
+from synapse.rest.client.v1 import login, room
+
+from tests import unittest
+
+logger = logging.getLogger(__name__)
+
+
+class FederationTestCase(unittest.HomeserverTestCase):
+    servlets = [
+        admin.register_servlets,
+        login.register_servlets,
+        room.register_servlets,
+    ]
+
+    def make_homeserver(self, reactor, clock):
+        hs = self.setup_test_homeserver(http_client=None)
+        self.handler = hs.get_handlers().federation_handler
+        self.store = hs.get_datastore()
+        return hs
+
+    def test_exchange_revoked_invite(self):
+        user_id = self.register_user("kermit", "test")
+        tok = self.login("kermit", "test")
+
+        room_id = self.helper.create_room_as(room_creator=user_id, tok=tok)
+
+        # Send a 3PID invite event with an empty body so it's considered as a revoked one.
+        invite_token = "sometoken"
+        self.helper.send_state(
+            room_id=room_id,
+            event_type=EventTypes.ThirdPartyInvite,
+            state_key=invite_token,
+            body={},
+            tok=tok,
+        )
+
+        d = self.handler.on_exchange_third_party_invite_request(
+            room_id=room_id,
+            event_dict={
+                "type": EventTypes.Member,
+                "room_id": room_id,
+                "sender": user_id,
+                "state_key": "@someone:example.org",
+                "content": {
+                    "membership": "invite",
+                    "third_party_invite": {
+                        "display_name": "alice",
+                        "signed": {
+                            "mxid": "@alice:localhost",
+                            "token": invite_token,
+                            "signatures": {
+                                "magic.forest": {
+                                    "ed25519:3": "fQpGIW1Snz+pwLZu6sTy2aHy/DYWWTspTJRPyNp0PKkymfIsNffysMl6ObMMFdIJhk6g6pwlIqZ54rxo8SLmAg"
+                                }
+                            },
+                        },
+                    },
+                },
+            },
+        )
+
+        failure = self.get_failure(d, AuthError).value
+
+        self.assertEqual(failure.code, 403, failure)
+        self.assertEqual(failure.errcode, Codes.FORBIDDEN, failure)
+        self.assertEqual(failure.msg, "You are not invited to this room.")
+
+    def test_rejected_message_event_state(self):
+        """
+        Check that we store the state group correctly for rejected non-state events.
+
+        Regression test for #6289.
+        """
+        OTHER_SERVER = "otherserver"
+        OTHER_USER = "@otheruser:" + OTHER_SERVER
+
+        # create the room
+        user_id = self.register_user("kermit", "test")
+        tok = self.login("kermit", "test")
+        room_id = self.helper.create_room_as(room_creator=user_id, tok=tok)
+
+        # pretend that another server has joined
+        join_event = self._build_and_send_join_event(OTHER_SERVER, OTHER_USER, room_id)
+
+        # check the state group
+        sg = self.successResultOf(
+            self.store._get_state_group_for_event(join_event.event_id)
+        )
+
+        # build and send an event which will be rejected
+        ev = event_from_pdu_json(
+            {
+                "type": EventTypes.Message,
+                "content": {},
+                "room_id": room_id,
+                "sender": "@yetanotheruser:" + OTHER_SERVER,
+                "depth": join_event["depth"] + 1,
+                "prev_events": [join_event.event_id],
+                "auth_events": [],
+                "origin_server_ts": self.clock.time_msec(),
+            },
+            join_event.format_version,
+        )
+
+        with LoggingContext(request="send_rejected"):
+            d = run_in_background(self.handler.on_receive_pdu, OTHER_SERVER, ev)
+        self.get_success(d)
+
+        # that should have been rejected
+        e = self.get_success(self.store.get_event(ev.event_id, allow_rejected=True))
+        self.assertIsNotNone(e.rejected_reason)
+
+        # ... and the state group should be the same as before
+        sg2 = self.successResultOf(self.store._get_state_group_for_event(ev.event_id))
+
+        self.assertEqual(sg, sg2)
+
+    def test_rejected_state_event_state(self):
+        """
+        Check that we store the state group correctly for rejected state events.
+
+        Regression test for #6289.
+        """
+        OTHER_SERVER = "otherserver"
+        OTHER_USER = "@otheruser:" + OTHER_SERVER
+
+        # create the room
+        user_id = self.register_user("kermit", "test")
+        tok = self.login("kermit", "test")
+        room_id = self.helper.create_room_as(room_creator=user_id, tok=tok)
+
+        # pretend that another server has joined
+        join_event = self._build_and_send_join_event(OTHER_SERVER, OTHER_USER, room_id)
+
+        # check the state group
+        sg = self.successResultOf(
+            self.store._get_state_group_for_event(join_event.event_id)
+        )
+
+        # build and send an event which will be rejected
+        ev = event_from_pdu_json(
+            {
+                "type": "org.matrix.test",
+                "state_key": "test_key",
+                "content": {},
+                "room_id": room_id,
+                "sender": "@yetanotheruser:" + OTHER_SERVER,
+                "depth": join_event["depth"] + 1,
+                "prev_events": [join_event.event_id],
+                "auth_events": [],
+                "origin_server_ts": self.clock.time_msec(),
+            },
+            join_event.format_version,
+        )
+
+        with LoggingContext(request="send_rejected"):
+            d = run_in_background(self.handler.on_receive_pdu, OTHER_SERVER, ev)
+        self.get_success(d)
+
+        # that should have been rejected
+        e = self.get_success(self.store.get_event(ev.event_id, allow_rejected=True))
+        self.assertIsNotNone(e.rejected_reason)
+
+        # ... and the state group should be the same as before
+        sg2 = self.successResultOf(self.store._get_state_group_for_event(ev.event_id))
+
+        self.assertEqual(sg, sg2)
+
+    def _build_and_send_join_event(self, other_server, other_user, room_id):
+        join_event = self.get_success(
+            self.handler.on_make_join_request(other_server, room_id, other_user)
+        )
+        # the auth code requires that a signature exists, but doesn't check that
+        # signature... go figure.
+        join_event.signatures[other_server] = {"x": "y"}
+        with LoggingContext(request="send_join"):
+            d = run_in_background(
+                self.handler.on_send_join_request, other_server, join_event
+            )
+        self.get_success(d)
+
+        # sanity-check: the room should show that the new user is a member
+        r = self.get_success(self.store.get_current_state_ids(room_id))
+        self.assertEqual(r[(EventTypes.Member, other_user)], join_event.event_id)
+
+        return join_event
diff --git a/tests/handlers/test_presence.py b/tests/handlers/test_presence.py
index f70c6e7d65..d4293b4312 100644
--- a/tests/handlers/test_presence.py
+++ b/tests/handlers/test_presence.py
@@ -22,6 +22,7 @@ from synapse.api.constants import EventTypes, Membership, PresenceState
 from synapse.events import room_version_to_event_format
 from synapse.events.builder import EventBuilder
 from synapse.handlers.presence import (
+    EXTERNAL_PROCESS_EXPIRY,
     FEDERATION_PING_INTERVAL,
     FEDERATION_TIMEOUT,
     IDLE_TIMER,
@@ -413,6 +414,44 @@ class PresenceTimeoutTestCase(unittest.TestCase):
         self.assertEquals(state, new_state)
 
 
+class PresenceHandlerTestCase(unittest.HomeserverTestCase):
+    def prepare(self, reactor, clock, hs):
+        self.presence_handler = hs.get_presence_handler()
+        self.clock = hs.get_clock()
+
+    def test_external_process_timeout(self):
+        """Test that if an external process doesn't update the records for a while
+        we time out their syncing users presence.
+        """
+        process_id = 1
+        user_id = "@test:server"
+
+        # Notify handler that a user is now syncing.
+        self.get_success(
+            self.presence_handler.update_external_syncs_row(
+                process_id, user_id, True, self.clock.time_msec()
+            )
+        )
+
+        # Check that if we wait a while without telling the handler the user has
+        # stopped syncing that their presence state doesn't get timed out.
+        self.reactor.advance(EXTERNAL_PROCESS_EXPIRY / 2)
+
+        state = self.get_success(
+            self.presence_handler.get_state(UserID.from_string(user_id))
+        )
+        self.assertEqual(state.state, PresenceState.ONLINE)
+
+        # Check that if the external process timeout fires, then the syncing
+        # user gets timed out
+        self.reactor.advance(EXTERNAL_PROCESS_EXPIRY)
+
+        state = self.get_success(
+            self.presence_handler.get_state(UserID.from_string(user_id))
+        )
+        self.assertEqual(state.state, PresenceState.OFFLINE)
+
+
 class PresenceJoinTestCase(unittest.HomeserverTestCase):
     """Tests remote servers get told about presence of users in the room when
     they join and when new local users join.
diff --git a/tests/handlers/test_register.py b/tests/handlers/test_register.py
index e10296a5e4..1e9ba3a201 100644
--- a/tests/handlers/test_register.py
+++ b/tests/handlers/test_register.py
@@ -171,11 +171,11 @@ class RegistrationTestCase(unittest.HomeserverTestCase):
         rooms = self.get_success(self.store.get_rooms_for_user(user_id))
         self.assertEqual(len(rooms), 0)
 
-    def test_auto_create_auto_join_rooms_when_support_user_exists(self):
+    def test_auto_create_auto_join_rooms_when_user_is_not_a_real_user(self):
         room_alias_str = "#room:test"
         self.hs.config.auto_join_rooms = [room_alias_str]
 
-        self.store.is_support_user = Mock(return_value=True)
+        self.store.is_real_user = Mock(return_value=False)
         user_id = self.get_success(self.handler.register_user(localpart="support"))
         rooms = self.get_success(self.store.get_rooms_for_user(user_id))
         self.assertEqual(len(rooms), 0)
@@ -183,6 +183,31 @@ class RegistrationTestCase(unittest.HomeserverTestCase):
         room_alias = RoomAlias.from_string(room_alias_str)
         self.get_failure(directory_handler.get_association(room_alias), SynapseError)
 
+    def test_auto_create_auto_join_rooms_when_user_is_the_first_real_user(self):
+        room_alias_str = "#room:test"
+        self.hs.config.auto_join_rooms = [room_alias_str]
+
+        self.store.count_real_users = Mock(return_value=1)
+        self.store.is_real_user = Mock(return_value=True)
+        user_id = self.get_success(self.handler.register_user(localpart="real"))
+        rooms = self.get_success(self.store.get_rooms_for_user(user_id))
+        directory_handler = self.hs.get_handlers().directory_handler
+        room_alias = RoomAlias.from_string(room_alias_str)
+        room_id = self.get_success(directory_handler.get_association(room_alias))
+
+        self.assertTrue(room_id["room_id"] in rooms)
+        self.assertEqual(len(rooms), 1)
+
+    def test_auto_create_auto_join_rooms_when_user_is_not_the_first_real_user(self):
+        room_alias_str = "#room:test"
+        self.hs.config.auto_join_rooms = [room_alias_str]
+
+        self.store.count_real_users = Mock(return_value=2)
+        self.store.is_real_user = Mock(return_value=True)
+        user_id = self.get_success(self.handler.register_user(localpart="real"))
+        rooms = self.get_success(self.store.get_rooms_for_user(user_id))
+        self.assertEqual(len(rooms), 0)
+
     def test_auto_create_auto_join_where_no_consent(self):
         """Test to ensure that the first user is not auto-joined to a room if
         they have not given general consent.
diff --git a/tests/handlers/test_roomlist.py b/tests/handlers/test_roomlist.py
deleted file mode 100644
index 61eebb6985..0000000000
--- a/tests/handlers/test_roomlist.py
+++ /dev/null
@@ -1,39 +0,0 @@
-# -*- coding: utf-8 -*-
-# Copyright 2018 New Vector Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from synapse.handlers.room_list import RoomListNextBatch
-
-import tests.unittest
-import tests.utils
-
-
-class RoomListTestCase(tests.unittest.TestCase):
-    """ Tests RoomList's RoomListNextBatch. """
-
-    def setUp(self):
-        pass
-
-    def test_check_read_batch_tokens(self):
-        batch_token = RoomListNextBatch(
-            stream_ordering="abcdef",
-            public_room_stream_id="123",
-            current_limit=20,
-            direction_is_forward=True,
-        ).to_token()
-        next_batch = RoomListNextBatch.from_token(batch_token)
-        self.assertEquals(next_batch.stream_ordering, "abcdef")
-        self.assertEquals(next_batch.public_room_stream_id, "123")
-        self.assertEquals(next_batch.current_limit, 20)
-        self.assertEquals(next_batch.direction_is_forward, True)
diff --git a/tests/handlers/test_stats.py b/tests/handlers/test_stats.py
index a8b858eb4f..e0075ccd32 100644
--- a/tests/handlers/test_stats.py
+++ b/tests/handlers/test_stats.py
@@ -13,16 +13,17 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from mock import Mock
-
-from twisted.internet import defer
-
-from synapse.api.constants import EventTypes, Membership
 from synapse.rest import admin
 from synapse.rest.client.v1 import login, room
+from synapse.storage.data_stores.main import stats
 
 from tests import unittest
 
+# The expected number of state events in a fresh public room.
+EXPT_NUM_STATE_EVTS_IN_FRESH_PUBLIC_ROOM = 5
+# The expected number of state events in a fresh private room.
+EXPT_NUM_STATE_EVTS_IN_FRESH_PRIVATE_ROOM = 6
+
 
 class StatsRoomTests(unittest.HomeserverTestCase):
 
@@ -33,7 +34,6 @@ class StatsRoomTests(unittest.HomeserverTestCase):
     ]
 
     def prepare(self, reactor, clock, hs):
-
         self.store = hs.get_datastore()
         self.handler = self.hs.get_stats_handler()
 
@@ -47,7 +47,7 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         self.get_success(
             self.store._simple_insert(
                 "background_updates",
-                {"update_name": "populate_stats_createtables", "progress_json": "{}"},
+                {"update_name": "populate_stats_prepare", "progress_json": "{}"},
             )
         )
         self.get_success(
@@ -56,7 +56,17 @@ class StatsRoomTests(unittest.HomeserverTestCase):
                 {
                     "update_name": "populate_stats_process_rooms",
                     "progress_json": "{}",
-                    "depends_on": "populate_stats_createtables",
+                    "depends_on": "populate_stats_prepare",
+                },
+            )
+        )
+        self.get_success(
+            self.store._simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_stats_process_users",
+                    "progress_json": "{}",
+                    "depends_on": "populate_stats_process_rooms",
                 },
             )
         )
@@ -66,16 +76,46 @@ class StatsRoomTests(unittest.HomeserverTestCase):
                 {
                     "update_name": "populate_stats_cleanup",
                     "progress_json": "{}",
-                    "depends_on": "populate_stats_process_rooms",
+                    "depends_on": "populate_stats_process_users",
                 },
             )
         )
 
+    def get_all_room_state(self):
+        return self.store._simple_select_list(
+            "room_stats_state", None, retcols=("name", "topic", "canonical_alias")
+        )
+
+    def _get_current_stats(self, stats_type, stat_id):
+        table, id_col = stats.TYPE_TO_TABLE[stats_type]
+
+        cols = list(stats.ABSOLUTE_STATS_FIELDS[stats_type]) + list(
+            stats.PER_SLICE_FIELDS[stats_type]
+        )
+
+        end_ts = self.store.quantise_stats_time(self.reactor.seconds() * 1000)
+
+        return self.get_success(
+            self.store._simple_select_one(
+                table + "_historical",
+                {id_col: stat_id, end_ts: end_ts},
+                cols,
+                allow_none=True,
+            )
+        )
+
+    def _perform_background_initial_update(self):
+        # Do the initial population of the stats via the background update
+        self._add_background_updates()
+
+        while not self.get_success(self.store.has_completed_background_updates()):
+            self.get_success(self.store.do_next_background_update(100), by=0.1)
+
     def test_initial_room(self):
         """
         The background updates will build the table from scratch.
         """
-        r = self.get_success(self.store.get_all_room_state())
+        r = self.get_success(self.get_all_room_state())
         self.assertEqual(len(r), 0)
 
         # Disable stats
@@ -91,7 +131,7 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         )
 
         # Stats disabled, shouldn't have done anything
-        r = self.get_success(self.store.get_all_room_state())
+        r = self.get_success(self.get_all_room_state())
         self.assertEqual(len(r), 0)
 
         # Enable stats
@@ -104,7 +144,7 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         while not self.get_success(self.store.has_completed_background_updates()):
             self.get_success(self.store.do_next_background_update(100), by=0.1)
 
-        r = self.get_success(self.store.get_all_room_state())
+        r = self.get_success(self.get_all_room_state())
 
         self.assertEqual(len(r), 1)
         self.assertEqual(r[0]["topic"], "foo")
@@ -114,6 +154,7 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         Ingestion via notify_new_event will ignore tokens that the background
         update have already processed.
         """
+
         self.reactor.advance(86401)
 
         self.hs.config.stats_enabled = False
@@ -138,12 +179,18 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         self.hs.config.stats_enabled = True
         self.handler.stats_enabled = True
         self.store._all_done = False
-        self.get_success(self.store.update_stats_stream_pos(None))
+        self.get_success(
+            self.store._simple_update_one(
+                table="stats_incremental_position",
+                keyvalues={},
+                updatevalues={"stream_id": 0},
+            )
+        )
 
         self.get_success(
             self.store._simple_insert(
                 "background_updates",
-                {"update_name": "populate_stats_createtables", "progress_json": "{}"},
+                {"update_name": "populate_stats_prepare", "progress_json": "{}"},
             )
         )
 
@@ -154,6 +201,8 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         self.helper.invite(room=room_1, src=u1, targ=u2, tok=u1_token)
         self.helper.join(room=room_1, user=u2, tok=u2_token)
 
+        # orig_delta_processor = self.store.
+
         # Now do the initial ingestion.
         self.get_success(
             self.store._simple_insert(
@@ -185,8 +234,15 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         self.helper.invite(room=room_1, src=u1, targ=u3, tok=u1_token)
         self.helper.join(room=room_1, user=u3, tok=u3_token)
 
-        # Get the deltas! There should be two -- day 1, and day 2.
-        r = self.get_success(self.store.get_deltas_for_room(room_1, 0))
+        # self.handler.notify_new_event()
+
+        # We need to let the delta processor advance…
+        self.pump(10 * 60)
+
+        # Get the slices! There should be two -- day 1, and day 2.
+        r = self.get_success(self.store.get_statistics_for_subject("room", room_1, 0))
+
+        self.assertEqual(len(r), 2)
 
         # The oldest has 2 joined members
         self.assertEqual(r[-1]["joined_members"], 2)
@@ -194,111 +250,478 @@ class StatsRoomTests(unittest.HomeserverTestCase):
         # The newest has 3
         self.assertEqual(r[0]["joined_members"], 3)
 
-    def test_incorrect_state_transition(self):
+    def test_create_user(self):
         """
-        If the state transition is not one of (JOIN, INVITE, LEAVE, BAN) to
-        (JOIN, INVITE, LEAVE, BAN), an error is raised.
+        When we create a user, it should have statistics already ready.
         """
-        events = {
-            "a1": {"membership": Membership.LEAVE},
-            "a2": {"membership": "not a real thing"},
-        }
 
-        def get_event(event_id, allow_none=True):
-            m = Mock()
-            m.content = events[event_id]
-            d = defer.Deferred()
-            self.reactor.callLater(0.0, d.callback, m)
-            return d
-
-        def get_received_ts(event_id):
-            return defer.succeed(1)
-
-        self.store.get_received_ts = get_received_ts
-        self.store.get_event = get_event
-
-        deltas = [
-            {
-                "type": EventTypes.Member,
-                "state_key": "some_user",
-                "room_id": "room",
-                "event_id": "a1",
-                "prev_event_id": "a2",
-                "stream_id": 60,
-            }
-        ]
-
-        f = self.get_failure(self.handler._handle_deltas(deltas), ValueError)
-        self.assertEqual(
-            f.value.args[0], "'not a real thing' is not a valid prev_membership"
-        )
-
-        # And the other way...
-        deltas = [
-            {
-                "type": EventTypes.Member,
-                "state_key": "some_user",
-                "room_id": "room",
-                "event_id": "a2",
-                "prev_event_id": "a1",
-                "stream_id": 100,
-            }
-        ]
-
-        f = self.get_failure(self.handler._handle_deltas(deltas), ValueError)
-        self.assertEqual(
-            f.value.args[0], "'not a real thing' is not a valid membership"
-        )
-
-    def test_redacted_prev_event(self):
-        """
-        If the prev_event does not exist, then it is assumed to be a LEAVE.
-        """
         u1 = self.register_user("u1", "pass")
-        u1_token = self.login("u1", "pass")
 
-        room_1 = self.helper.create_room_as(u1, tok=u1_token)
+        u1stats = self._get_current_stats("user", u1)
 
-        # Do the initial population of the user directory via the background update
-        self._add_background_updates()
+        self.assertIsNotNone(u1stats)
+
+        # not in any rooms by default
+        self.assertEqual(u1stats["joined_rooms"], 0)
+
+    def test_create_room(self):
+        """
+        When we create a room, it should have statistics already ready.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+        r1stats = self._get_current_stats("room", r1)
+        r2 = self.helper.create_room_as(u1, tok=u1token, is_public=False)
+        r2stats = self._get_current_stats("room", r2)
+
+        self.assertIsNotNone(r1stats)
+        self.assertIsNotNone(r2stats)
+
+        # contains the default things you'd expect in a fresh room
+        self.assertEqual(
+            r1stats["total_events"],
+            EXPT_NUM_STATE_EVTS_IN_FRESH_PUBLIC_ROOM,
+            "Wrong number of total_events in new room's stats!"
+            " You may need to update this if more state events are added to"
+            " the room creation process.",
+        )
+        self.assertEqual(
+            r2stats["total_events"],
+            EXPT_NUM_STATE_EVTS_IN_FRESH_PRIVATE_ROOM,
+            "Wrong number of total_events in new room's stats!"
+            " You may need to update this if more state events are added to"
+            " the room creation process.",
+        )
+
+        self.assertEqual(
+            r1stats["current_state_events"], EXPT_NUM_STATE_EVTS_IN_FRESH_PUBLIC_ROOM
+        )
+        self.assertEqual(
+            r2stats["current_state_events"], EXPT_NUM_STATE_EVTS_IN_FRESH_PRIVATE_ROOM
+        )
+
+        self.assertEqual(r1stats["joined_members"], 1)
+        self.assertEqual(r1stats["invited_members"], 0)
+        self.assertEqual(r1stats["banned_members"], 0)
+
+        self.assertEqual(r2stats["joined_members"], 1)
+        self.assertEqual(r2stats["invited_members"], 0)
+        self.assertEqual(r2stats["banned_members"], 0)
+
+    def test_send_message_increments_total_events(self):
+        """
+        When we send a message, it increments total_events.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.send(r1, "hiss", tok=u1token)
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+
+    def test_send_state_event_nonoverwriting(self):
+        """
+        When we send a non-overwriting state event, it increments total_events AND current_state_events
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        self.helper.send_state(
+            r1, "cat.hissing", {"value": True}, tok=u1token, state_key="tabby"
+        )
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.send_state(
+            r1, "cat.hissing", {"value": False}, tok=u1token, state_key="moggy"
+        )
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            1,
+        )
+
+    def test_send_state_event_overwriting(self):
+        """
+        When we send an overwriting state event, it increments total_events ONLY
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        self.helper.send_state(
+            r1, "cat.hissing", {"value": True}, tok=u1token, state_key="tabby"
+        )
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.send_state(
+            r1, "cat.hissing", {"value": False}, tok=u1token, state_key="tabby"
+        )
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            0,
+        )
+
+    def test_join_first_time(self):
+        """
+        When a user joins a room for the first time, total_events, current_state_events and
+        joined_members should increase by exactly 1.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        u2 = self.register_user("u2", "pass")
+        u2token = self.login("u2", "pass")
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.join(r1, u2, tok=u2token)
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            1,
+        )
+        self.assertEqual(
+            r1stats_post["joined_members"] - r1stats_ante["joined_members"], 1
+        )
+
+    def test_join_after_leave(self):
+        """
+        When a user joins a room after being previously left, total_events and
+        joined_members should increase by exactly 1.
+        current_state_events should not increase.
+        left_members should decrease by exactly 1.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        u2 = self.register_user("u2", "pass")
+        u2token = self.login("u2", "pass")
+
+        self.helper.join(r1, u2, tok=u2token)
+        self.helper.leave(r1, u2, tok=u2token)
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.join(r1, u2, tok=u2token)
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            0,
+        )
+        self.assertEqual(
+            r1stats_post["joined_members"] - r1stats_ante["joined_members"], +1
+        )
+        self.assertEqual(
+            r1stats_post["left_members"] - r1stats_ante["left_members"], -1
+        )
+
+    def test_invited(self):
+        """
+        When a user invites another user, current_state_events, total_events and
+        invited_members should increase by exactly 1.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        u2 = self.register_user("u2", "pass")
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.invite(r1, u1, u2, tok=u1token)
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            1,
+        )
+        self.assertEqual(
+            r1stats_post["invited_members"] - r1stats_ante["invited_members"], +1
+        )
+
+    def test_join_after_invite(self):
+        """
+        When a user joins a room after being invited, total_events and
+        joined_members should increase by exactly 1.
+        current_state_events should not increase.
+        invited_members should decrease by exactly 1.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        u2 = self.register_user("u2", "pass")
+        u2token = self.login("u2", "pass")
+
+        self.helper.invite(r1, u1, u2, tok=u1token)
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.join(r1, u2, tok=u2token)
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            0,
+        )
+        self.assertEqual(
+            r1stats_post["joined_members"] - r1stats_ante["joined_members"], +1
+        )
+        self.assertEqual(
+            r1stats_post["invited_members"] - r1stats_ante["invited_members"], -1
+        )
+
+    def test_left(self):
+        """
+        When a user leaves a room after joining, total_events and
+        left_members should increase by exactly 1.
+        current_state_events should not increase.
+        joined_members should decrease by exactly 1.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        u2 = self.register_user("u2", "pass")
+        u2token = self.login("u2", "pass")
+
+        self.helper.join(r1, u2, tok=u2token)
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.leave(r1, u2, tok=u2token)
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            0,
+        )
+        self.assertEqual(
+            r1stats_post["left_members"] - r1stats_ante["left_members"], +1
+        )
+        self.assertEqual(
+            r1stats_post["joined_members"] - r1stats_ante["joined_members"], -1
+        )
+
+    def test_banned(self):
+        """
+        When a user is banned from a room after joining, total_events and
+        left_members should increase by exactly 1.
+        current_state_events should not increase.
+        banned_members should decrease by exactly 1.
+        """
+
+        self._perform_background_initial_update()
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        u2 = self.register_user("u2", "pass")
+        u2token = self.login("u2", "pass")
+
+        self.helper.join(r1, u2, tok=u2token)
+
+        r1stats_ante = self._get_current_stats("room", r1)
+
+        self.helper.change_membership(r1, u1, u2, "ban", tok=u1token)
+
+        r1stats_post = self._get_current_stats("room", r1)
+
+        self.assertEqual(r1stats_post["total_events"] - r1stats_ante["total_events"], 1)
+        self.assertEqual(
+            r1stats_post["current_state_events"] - r1stats_ante["current_state_events"],
+            0,
+        )
+        self.assertEqual(
+            r1stats_post["banned_members"] - r1stats_ante["banned_members"], +1
+        )
+        self.assertEqual(
+            r1stats_post["joined_members"] - r1stats_ante["joined_members"], -1
+        )
+
+    def test_initial_background_update(self):
+        """
+        Test that statistics can be generated by the initial background update
+        handler.
+
+        This test also tests that stats rows are not created for new subjects
+        when stats are disabled. However, it may be desirable to change this
+        behaviour eventually to still keep current rows.
+        """
+
+        self.hs.config.stats_enabled = False
+        self.handler.stats_enabled = False
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token)
+
+        # test that these subjects, which were created during a time of disabled
+        # stats, do not have stats.
+        self.assertIsNone(self._get_current_stats("room", r1))
+        self.assertIsNone(self._get_current_stats("user", u1))
+
+        self.hs.config.stats_enabled = True
+        self.handler.stats_enabled = True
+
+        self._perform_background_initial_update()
+
+        r1stats = self._get_current_stats("room", r1)
+        u1stats = self._get_current_stats("user", u1)
+
+        self.assertEqual(r1stats["joined_members"], 1)
+        self.assertEqual(
+            r1stats["current_state_events"], EXPT_NUM_STATE_EVTS_IN_FRESH_PUBLIC_ROOM
+        )
+
+        self.assertEqual(u1stats["joined_rooms"], 1)
+
+    def test_incomplete_stats(self):
+        """
+        This tests that we track incomplete statistics.
+
+        We first test that incomplete stats are incrementally generated,
+        following the preparation of a background regen.
+
+        We then test that these incomplete rows are completed by the background
+        regen.
+        """
+
+        u1 = self.register_user("u1", "pass")
+        u1token = self.login("u1", "pass")
+        u2 = self.register_user("u2", "pass")
+        u2token = self.login("u2", "pass")
+        u3 = self.register_user("u3", "pass")
+        r1 = self.helper.create_room_as(u1, tok=u1token, is_public=False)
+
+        # preparation stage of the initial background update
+        # Ugh, have to reset this flag
+        self.store._all_done = False
+
+        self.get_success(
+            self.store._simple_delete(
+                "room_stats_current", {"1": 1}, "test_delete_stats"
+            )
+        )
+        self.get_success(
+            self.store._simple_delete(
+                "user_stats_current", {"1": 1}, "test_delete_stats"
+            )
+        )
+
+        self.helper.invite(r1, u1, u2, tok=u1token)
+        self.helper.join(r1, u2, tok=u2token)
+        self.helper.invite(r1, u1, u3, tok=u1token)
+        self.helper.send(r1, "thou shalt yield", tok=u1token)
+
+        # now do the background updates
+
+        self.store._all_done = False
+        self.get_success(
+            self.store._simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_stats_process_rooms",
+                    "progress_json": "{}",
+                    "depends_on": "populate_stats_prepare",
+                },
+            )
+        )
+        self.get_success(
+            self.store._simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_stats_process_users",
+                    "progress_json": "{}",
+                    "depends_on": "populate_stats_process_rooms",
+                },
+            )
+        )
+        self.get_success(
+            self.store._simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_stats_cleanup",
+                    "progress_json": "{}",
+                    "depends_on": "populate_stats_process_users",
+                },
+            )
+        )
 
         while not self.get_success(self.store.has_completed_background_updates()):
             self.get_success(self.store.do_next_background_update(100), by=0.1)
 
-        events = {"a1": None, "a2": {"membership": Membership.JOIN}}
+        r1stats_complete = self._get_current_stats("room", r1)
+        u1stats_complete = self._get_current_stats("user", u1)
+        u2stats_complete = self._get_current_stats("user", u2)
 
-        def get_event(event_id, allow_none=True):
-            if events.get(event_id):
-                m = Mock()
-                m.content = events[event_id]
-            else:
-                m = None
-            d = defer.Deferred()
-            self.reactor.callLater(0.0, d.callback, m)
-            return d
+        # now we make our assertions
 
-        def get_received_ts(event_id):
-            return defer.succeed(1)
+        # check that _complete rows are complete and correct
+        self.assertEqual(r1stats_complete["joined_members"], 2)
+        self.assertEqual(r1stats_complete["invited_members"], 1)
 
-        self.store.get_received_ts = get_received_ts
-        self.store.get_event = get_event
+        self.assertEqual(
+            r1stats_complete["current_state_events"],
+            2 + EXPT_NUM_STATE_EVTS_IN_FRESH_PRIVATE_ROOM,
+        )
 
-        deltas = [
-            {
-                "type": EventTypes.Member,
-                "state_key": "some_user:test",
-                "room_id": room_1,
-                "event_id": "a2",
-                "prev_event_id": "a1",
-                "stream_id": 100,
-            }
-        ]
-
-        # Handle our fake deltas, which has a user going from LEAVE -> JOIN.
-        self.get_success(self.handler._handle_deltas(deltas))
-
-        # One delta, with two joined members -- the room creator, and our fake
-        # user.
-        r = self.get_success(self.store.get_deltas_for_room(room_1, 0))
-        self.assertEqual(len(r), 1)
-        self.assertEqual(r[0]["joined_members"], 2)
+        self.assertEqual(u1stats_complete["joined_rooms"], 1)
+        self.assertEqual(u2stats_complete["joined_rooms"], 1)
diff --git a/tests/handlers/test_typing.py b/tests/handlers/test_typing.py
index 5d5e324df2..5ec568f4e6 100644
--- a/tests/handlers/test_typing.py
+++ b/tests/handlers/test_typing.py
@@ -73,7 +73,7 @@ class TypingNotificationsTestCase(unittest.HomeserverTestCase):
                         "get_received_txn_response",
                         "set_received_txn_response",
                         "get_destination_retry_timings",
-                        "get_devices_by_remote",
+                        "get_device_updates_by_remote",
                         # Bits that user_directory needs
                         "get_user_directory_stream_pos",
                         "get_current_state_deltas",
@@ -99,12 +99,17 @@ class TypingNotificationsTestCase(unittest.HomeserverTestCase):
         self.event_source = hs.get_event_sources().sources["typing"]
 
         self.datastore = hs.get_datastore()
-        retry_timings_res = {"destination": "", "retry_last_ts": 0, "retry_interval": 0}
+        retry_timings_res = {
+            "destination": "",
+            "retry_last_ts": 0,
+            "retry_interval": 0,
+            "failure_ts": None,
+        }
         self.datastore.get_destination_retry_timings.return_value = defer.succeed(
             retry_timings_res
         )
 
-        self.datastore.get_devices_by_remote.return_value = (0, [])
+        self.datastore.get_device_updates_by_remote.return_value = (0, [])
 
         def get_received_txn_response(*args):
             return defer.succeed(None)
@@ -134,11 +139,14 @@ class TypingNotificationsTestCase(unittest.HomeserverTestCase):
             defer.succeed(1)
         )
 
-        self.datastore.get_current_state_deltas.return_value = None
+        self.datastore.get_current_state_deltas.return_value = (0, None)
 
         self.datastore.get_to_device_stream_token = lambda: 0
         self.datastore.get_new_device_msgs_for_remote = lambda *args, **kargs: ([], 0)
         self.datastore.delete_device_msgs_for_remote = lambda *args, **kargs: None
+        self.datastore.set_received_txn_response = lambda *args, **kwargs: defer.succeed(
+            None
+        )
 
     def test_started_typing_local(self):
         self.room_members = [U_APPLE, U_BANANA]
diff --git a/tests/http/__init__.py b/tests/http/__init__.py
index 2d5dba6464..2096ba3c91 100644
--- a/tests/http/__init__.py
+++ b/tests/http/__init__.py
@@ -20,6 +20,23 @@ from zope.interface import implementer
 from OpenSSL import SSL
 from OpenSSL.SSL import Connection
 from twisted.internet.interfaces import IOpenSSLServerConnectionCreator
+from twisted.internet.ssl import Certificate, trustRootFromCertificates
+from twisted.web.client import BrowserLikePolicyForHTTPS  # noqa: F401
+from twisted.web.iweb import IPolicyForHTTPS  # noqa: F401
+
+
+def get_test_https_policy():
+    """Get a test IPolicyForHTTPS which trusts the test CA cert
+
+    Returns:
+        IPolicyForHTTPS
+    """
+    ca_file = get_test_ca_cert_file()
+    with open(ca_file) as stream:
+        content = stream.read()
+    cert = Certificate.loadPEM(content)
+    trust_root = trustRootFromCertificates([cert])
+    return BrowserLikePolicyForHTTPS(trustRoot=trust_root)
 
 
 def get_test_ca_cert_file():
diff --git a/tests/http/federation/test_matrix_federation_agent.py b/tests/http/federation/test_matrix_federation_agent.py
index 71d7025264..cfcd98ff7d 100644
--- a/tests/http/federation/test_matrix_federation_agent.py
+++ b/tests/http/federation/test_matrix_federation_agent.py
@@ -124,19 +124,24 @@ class MatrixFederationAgentTests(unittest.TestCase):
             FakeTransport(client_protocol, self.reactor, server_tls_protocol)
         )
 
+        # grab a hold of the TLS connection, in case it gets torn down
+        server_tls_connection = server_tls_protocol._tlsConnection
+
+        # fish the test server back out of the server-side TLS protocol.
+        http_protocol = server_tls_protocol.wrappedProtocol
+
         # give the reactor a pump to get the TLS juices flowing.
         self.reactor.pump((0.1,))
 
         # check the SNI
-        server_name = server_tls_protocol._tlsConnection.get_servername()
+        server_name = server_tls_connection.get_servername()
         self.assertEqual(
             server_name,
             expected_sni,
             "Expected SNI %s but got %s" % (expected_sni, server_name),
         )
 
-        # fish the test server back out of the server-side TLS protocol.
-        return server_tls_protocol.wrappedProtocol
+        return http_protocol
 
     @defer.inlineCallbacks
     def _make_get_request(self, uri):
diff --git a/tests/http/test_proxyagent.py b/tests/http/test_proxyagent.py
new file mode 100644
index 0000000000..22abf76515
--- /dev/null
+++ b/tests/http/test_proxyagent.py
@@ -0,0 +1,334 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import logging
+
+import treq
+
+from twisted.internet import interfaces  # noqa: F401
+from twisted.internet.protocol import Factory
+from twisted.protocols.tls import TLSMemoryBIOFactory
+from twisted.web.http import HTTPChannel
+
+from synapse.http.proxyagent import ProxyAgent
+
+from tests.http import TestServerTLSConnectionFactory, get_test_https_policy
+from tests.server import FakeTransport, ThreadedMemoryReactorClock
+from tests.unittest import TestCase
+
+logger = logging.getLogger(__name__)
+
+HTTPFactory = Factory.forProtocol(HTTPChannel)
+
+
+class MatrixFederationAgentTests(TestCase):
+    def setUp(self):
+        self.reactor = ThreadedMemoryReactorClock()
+
+    def _make_connection(
+        self, client_factory, server_factory, ssl=False, expected_sni=None
+    ):
+        """Builds a test server, and completes the outgoing client connection
+
+        Args:
+            client_factory (interfaces.IProtocolFactory): the the factory that the
+                application is trying to use to make the outbound connection. We will
+                invoke it to build the client Protocol
+
+            server_factory (interfaces.IProtocolFactory): a factory to build the
+                server-side protocol
+
+            ssl (bool): If true, we will expect an ssl connection and wrap
+                server_factory with a TLSMemoryBIOFactory
+
+            expected_sni (bytes|None): the expected SNI value
+
+        Returns:
+            IProtocol: the server Protocol returned by server_factory
+        """
+        if ssl:
+            server_factory = _wrap_server_factory_for_tls(server_factory)
+
+        server_protocol = server_factory.buildProtocol(None)
+
+        # now, tell the client protocol factory to build the client protocol,
+        # and wire the output of said protocol up to the server via
+        # a FakeTransport.
+        #
+        # Normally this would be done by the TCP socket code in Twisted, but we are
+        # stubbing that out here.
+        client_protocol = client_factory.buildProtocol(None)
+        client_protocol.makeConnection(
+            FakeTransport(server_protocol, self.reactor, client_protocol)
+        )
+
+        # tell the server protocol to send its stuff back to the client, too
+        server_protocol.makeConnection(
+            FakeTransport(client_protocol, self.reactor, server_protocol)
+        )
+
+        if ssl:
+            http_protocol = server_protocol.wrappedProtocol
+            tls_connection = server_protocol._tlsConnection
+        else:
+            http_protocol = server_protocol
+            tls_connection = None
+
+        # give the reactor a pump to get the TLS juices flowing (if needed)
+        self.reactor.advance(0)
+
+        if expected_sni is not None:
+            server_name = tls_connection.get_servername()
+            self.assertEqual(
+                server_name,
+                expected_sni,
+                "Expected SNI %s but got %s" % (expected_sni, server_name),
+            )
+
+        return http_protocol
+
+    def test_http_request(self):
+        agent = ProxyAgent(self.reactor)
+
+        self.reactor.lookups["test.com"] = "1.2.3.4"
+        d = agent.request(b"GET", b"http://test.com")
+
+        # there should be a pending TCP connection
+        clients = self.reactor.tcpClients
+        self.assertEqual(len(clients), 1)
+        (host, port, client_factory, _timeout, _bindAddress) = clients[0]
+        self.assertEqual(host, "1.2.3.4")
+        self.assertEqual(port, 80)
+
+        # make a test server, and wire up the client
+        http_server = self._make_connection(
+            client_factory, _get_test_protocol_factory()
+        )
+
+        # the FakeTransport is async, so we need to pump the reactor
+        self.reactor.advance(0)
+
+        # now there should be a pending request
+        self.assertEqual(len(http_server.requests), 1)
+
+        request = http_server.requests[0]
+        self.assertEqual(request.method, b"GET")
+        self.assertEqual(request.path, b"/")
+        self.assertEqual(request.requestHeaders.getRawHeaders(b"host"), [b"test.com"])
+        request.write(b"result")
+        request.finish()
+
+        self.reactor.advance(0)
+
+        resp = self.successResultOf(d)
+        body = self.successResultOf(treq.content(resp))
+        self.assertEqual(body, b"result")
+
+    def test_https_request(self):
+        agent = ProxyAgent(self.reactor, contextFactory=get_test_https_policy())
+
+        self.reactor.lookups["test.com"] = "1.2.3.4"
+        d = agent.request(b"GET", b"https://test.com/abc")
+
+        # there should be a pending TCP connection
+        clients = self.reactor.tcpClients
+        self.assertEqual(len(clients), 1)
+        (host, port, client_factory, _timeout, _bindAddress) = clients[0]
+        self.assertEqual(host, "1.2.3.4")
+        self.assertEqual(port, 443)
+
+        # make a test server, and wire up the client
+        http_server = self._make_connection(
+            client_factory,
+            _get_test_protocol_factory(),
+            ssl=True,
+            expected_sni=b"test.com",
+        )
+
+        # the FakeTransport is async, so we need to pump the reactor
+        self.reactor.advance(0)
+
+        # now there should be a pending request
+        self.assertEqual(len(http_server.requests), 1)
+
+        request = http_server.requests[0]
+        self.assertEqual(request.method, b"GET")
+        self.assertEqual(request.path, b"/abc")
+        self.assertEqual(request.requestHeaders.getRawHeaders(b"host"), [b"test.com"])
+        request.write(b"result")
+        request.finish()
+
+        self.reactor.advance(0)
+
+        resp = self.successResultOf(d)
+        body = self.successResultOf(treq.content(resp))
+        self.assertEqual(body, b"result")
+
+    def test_http_request_via_proxy(self):
+        agent = ProxyAgent(self.reactor, http_proxy=b"proxy.com:8888")
+
+        self.reactor.lookups["proxy.com"] = "1.2.3.5"
+        d = agent.request(b"GET", b"http://test.com")
+
+        # there should be a pending TCP connection
+        clients = self.reactor.tcpClients
+        self.assertEqual(len(clients), 1)
+        (host, port, client_factory, _timeout, _bindAddress) = clients[0]
+        self.assertEqual(host, "1.2.3.5")
+        self.assertEqual(port, 8888)
+
+        # make a test server, and wire up the client
+        http_server = self._make_connection(
+            client_factory, _get_test_protocol_factory()
+        )
+
+        # the FakeTransport is async, so we need to pump the reactor
+        self.reactor.advance(0)
+
+        # now there should be a pending request
+        self.assertEqual(len(http_server.requests), 1)
+
+        request = http_server.requests[0]
+        self.assertEqual(request.method, b"GET")
+        self.assertEqual(request.path, b"http://test.com")
+        self.assertEqual(request.requestHeaders.getRawHeaders(b"host"), [b"test.com"])
+        request.write(b"result")
+        request.finish()
+
+        self.reactor.advance(0)
+
+        resp = self.successResultOf(d)
+        body = self.successResultOf(treq.content(resp))
+        self.assertEqual(body, b"result")
+
+    def test_https_request_via_proxy(self):
+        agent = ProxyAgent(
+            self.reactor,
+            contextFactory=get_test_https_policy(),
+            https_proxy=b"proxy.com",
+        )
+
+        self.reactor.lookups["proxy.com"] = "1.2.3.5"
+        d = agent.request(b"GET", b"https://test.com/abc")
+
+        # there should be a pending TCP connection
+        clients = self.reactor.tcpClients
+        self.assertEqual(len(clients), 1)
+        (host, port, client_factory, _timeout, _bindAddress) = clients[0]
+        self.assertEqual(host, "1.2.3.5")
+        self.assertEqual(port, 1080)
+
+        # make a test HTTP server, and wire up the client
+        proxy_server = self._make_connection(
+            client_factory, _get_test_protocol_factory()
+        )
+
+        # fish the transports back out so that we can do the old switcheroo
+        s2c_transport = proxy_server.transport
+        client_protocol = s2c_transport.other
+        c2s_transport = client_protocol.transport
+
+        # the FakeTransport is async, so we need to pump the reactor
+        self.reactor.advance(0)
+
+        # now there should be a pending CONNECT request
+        self.assertEqual(len(proxy_server.requests), 1)
+
+        request = proxy_server.requests[0]
+        self.assertEqual(request.method, b"CONNECT")
+        self.assertEqual(request.path, b"test.com:443")
+
+        # tell the proxy server not to close the connection
+        proxy_server.persistent = True
+
+        # this just stops the http Request trying to do a chunked response
+        # request.setHeader(b"Content-Length", b"0")
+        request.finish()
+
+        # now we can replace the proxy channel with a new, SSL-wrapped HTTP channel
+        ssl_factory = _wrap_server_factory_for_tls(_get_test_protocol_factory())
+        ssl_protocol = ssl_factory.buildProtocol(None)
+        http_server = ssl_protocol.wrappedProtocol
+
+        ssl_protocol.makeConnection(
+            FakeTransport(client_protocol, self.reactor, ssl_protocol)
+        )
+        c2s_transport.other = ssl_protocol
+
+        self.reactor.advance(0)
+
+        server_name = ssl_protocol._tlsConnection.get_servername()
+        expected_sni = b"test.com"
+        self.assertEqual(
+            server_name,
+            expected_sni,
+            "Expected SNI %s but got %s" % (expected_sni, server_name),
+        )
+
+        # now there should be a pending request
+        self.assertEqual(len(http_server.requests), 1)
+
+        request = http_server.requests[0]
+        self.assertEqual(request.method, b"GET")
+        self.assertEqual(request.path, b"/abc")
+        self.assertEqual(request.requestHeaders.getRawHeaders(b"host"), [b"test.com"])
+        request.write(b"result")
+        request.finish()
+
+        self.reactor.advance(0)
+
+        resp = self.successResultOf(d)
+        body = self.successResultOf(treq.content(resp))
+        self.assertEqual(body, b"result")
+
+
+def _wrap_server_factory_for_tls(factory, sanlist=None):
+    """Wrap an existing Protocol Factory with a test TLSMemoryBIOFactory
+
+    The resultant factory will create a TLS server which presents a certificate
+    signed by our test CA, valid for the domains in `sanlist`
+
+    Args:
+        factory (interfaces.IProtocolFactory): protocol factory to wrap
+        sanlist (iterable[bytes]): list of domains the cert should be valid for
+
+    Returns:
+        interfaces.IProtocolFactory
+    """
+    if sanlist is None:
+        sanlist = [b"DNS:test.com"]
+
+    connection_creator = TestServerTLSConnectionFactory(sanlist=sanlist)
+    return TLSMemoryBIOFactory(
+        connection_creator, isClient=False, wrappedFactory=factory
+    )
+
+
+def _get_test_protocol_factory():
+    """Get a protocol Factory which will build an HTTPChannel
+
+    Returns:
+        interfaces.IProtocolFactory
+    """
+    server_factory = Factory.forProtocol(HTTPChannel)
+
+    # Request.finish expects the factory to have a 'log' method.
+    server_factory.log = _log_request
+
+    return server_factory
+
+
+def _log_request(request):
+    """Implements Factory.log, which is expected by Request.finish"""
+    logger.info("Completed request %s", request)
diff --git a/tests/logging/test_structured.py b/tests/logging/test_structured.py
index a786de0233..451d05c0f0 100644
--- a/tests/logging/test_structured.py
+++ b/tests/logging/test_structured.py
@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
 import os
 import os.path
 import shutil
@@ -33,7 +34,20 @@ class FakeBeginner(object):
         self.observers = observers
 
 
-class StructuredLoggingTestCase(HomeserverTestCase):
+class StructuredLoggingTestBase(object):
+    """
+    Test base that registers a cleanup handler to reset the stdlib log handler
+    to 'unset'.
+    """
+
+    def prepare(self, reactor, clock, hs):
+        def _cleanup():
+            logging.getLogger("synapse").setLevel(logging.NOTSET)
+
+        self.addCleanup(_cleanup)
+
+
+class StructuredLoggingTestCase(StructuredLoggingTestBase, HomeserverTestCase):
     """
     Tests for Synapse's structured logging support.
     """
@@ -139,7 +153,9 @@ class StructuredLoggingTestCase(HomeserverTestCase):
         self.assertEqual(logs[0]["request"], "somereq")
 
 
-class StructuredLoggingConfigurationFileTestCase(HomeserverTestCase):
+class StructuredLoggingConfigurationFileTestCase(
+    StructuredLoggingTestBase, HomeserverTestCase
+):
     def make_homeserver(self, reactor, clock):
 
         tempdir = self.mktemp()
@@ -179,10 +195,11 @@ class StructuredLoggingConfigurationFileTestCase(HomeserverTestCase):
         """
         When a structured logging config is given, Synapse will use it.
         """
-        setup_logging(self.hs, self.hs.config)
+        beginner = FakeBeginner()
+        publisher = setup_logging(self.hs, self.hs.config, logBeginner=beginner)
 
         # Make a logger and send an event
-        logger = Logger(namespace="tests.logging.test_structured")
+        logger = Logger(namespace="tests.logging.test_structured", observer=publisher)
 
         with LoggingContext("testcontext", request="somereq"):
             logger.info("Hello there, {name}!", name="steve")
diff --git a/tests/logging/test_terse_json.py b/tests/logging/test_terse_json.py
index 514282591d..4cf81f7128 100644
--- a/tests/logging/test_terse_json.py
+++ b/tests/logging/test_terse_json.py
@@ -23,10 +23,10 @@ from synapse.logging._structured import setup_structured_logging
 from tests.server import connect_client
 from tests.unittest import HomeserverTestCase
 
-from .test_structured import FakeBeginner
+from .test_structured import FakeBeginner, StructuredLoggingTestBase
 
 
-class TerseJSONTCPTestCase(HomeserverTestCase):
+class TerseJSONTCPTestCase(StructuredLoggingTestBase, HomeserverTestCase):
     def test_log_output(self):
         """
         The Terse JSON outputter delivers simplified structured logs over TCP.
diff --git a/tests/patch_inline_callbacks.py b/tests/patch_inline_callbacks.py
deleted file mode 100644
index 220884311c..0000000000
--- a/tests/patch_inline_callbacks.py
+++ /dev/null
@@ -1,94 +0,0 @@
-# -*- coding: utf-8 -*-
-# Copyright 2018 New Vector Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from __future__ import print_function
-
-import functools
-import sys
-
-from twisted.internet import defer
-from twisted.internet.defer import Deferred
-from twisted.python.failure import Failure
-
-
-def do_patch():
-    """
-    Patch defer.inlineCallbacks so that it checks the state of the logcontext on exit
-    """
-
-    from synapse.logging.context import LoggingContext
-
-    orig_inline_callbacks = defer.inlineCallbacks
-
-    def new_inline_callbacks(f):
-
-        orig = orig_inline_callbacks(f)
-
-        @functools.wraps(f)
-        def wrapped(*args, **kwargs):
-            start_context = LoggingContext.current_context()
-
-            try:
-                res = orig(*args, **kwargs)
-            except Exception:
-                if LoggingContext.current_context() != start_context:
-                    err = "%s changed context from %s to %s on exception" % (
-                        f,
-                        start_context,
-                        LoggingContext.current_context(),
-                    )
-                    print(err, file=sys.stderr)
-                    raise Exception(err)
-                raise
-
-            if not isinstance(res, Deferred) or res.called:
-                if LoggingContext.current_context() != start_context:
-                    err = "%s changed context from %s to %s" % (
-                        f,
-                        start_context,
-                        LoggingContext.current_context(),
-                    )
-                    # print the error to stderr because otherwise all we
-                    # see in travis-ci is the 500 error
-                    print(err, file=sys.stderr)
-                    raise Exception(err)
-                return res
-
-            if LoggingContext.current_context() != LoggingContext.sentinel:
-                err = (
-                    "%s returned incomplete deferred in non-sentinel context "
-                    "%s (start was %s)"
-                ) % (f, LoggingContext.current_context(), start_context)
-                print(err, file=sys.stderr)
-                raise Exception(err)
-
-            def check_ctx(r):
-                if LoggingContext.current_context() != start_context:
-                    err = "%s completion of %s changed context from %s to %s" % (
-                        "Failure" if isinstance(r, Failure) else "Success",
-                        f,
-                        start_context,
-                        LoggingContext.current_context(),
-                    )
-                    print(err, file=sys.stderr)
-                    raise Exception(err)
-                return r
-
-            res.addBoth(check_ctx)
-            return res
-
-        return wrapped
-
-    defer.inlineCallbacks = new_inline_callbacks
diff --git a/tests/push/test_http.py b/tests/push/test_http.py
index 8ce6bb62da..af2327fb66 100644
--- a/tests/push/test_http.py
+++ b/tests/push/test_http.py
@@ -50,7 +50,7 @@ class HTTPPusherTests(HomeserverTestCase):
         config = self.default_config()
         config["start_pushers"] = True
 
-        hs = self.setup_test_homeserver(config=config, simple_http_client=m)
+        hs = self.setup_test_homeserver(config=config, proxied_http_client=m)
 
         return hs
 
diff --git a/tests/replication/slave/storage/_base.py b/tests/replication/slave/storage/_base.py
index 104349cdbd..4f924ce451 100644
--- a/tests/replication/slave/storage/_base.py
+++ b/tests/replication/slave/storage/_base.py
@@ -41,6 +41,7 @@ class BaseSlavedStoreTestCase(unittest.HomeserverTestCase):
     def prepare(self, reactor, clock, hs):
 
         self.master_store = self.hs.get_datastore()
+        self.storage = hs.get_storage()
         self.slaved_store = self.STORE_TYPE(self.hs.get_db_conn(), self.hs)
         self.event_id = 0
 
diff --git a/tests/replication/slave/storage/test_events.py b/tests/replication/slave/storage/test_events.py
index a368117b43..b68e9fe082 100644
--- a/tests/replication/slave/storage/test_events.py
+++ b/tests/replication/slave/storage/test_events.py
@@ -234,7 +234,9 @@ class SlavedEventStoreTestCase(BaseSlavedStoreTestCase):
             type="m.room.member", sender=USER_ID_2, key=USER_ID_2, membership="join"
         )
         msg, msgctx = self.build_event()
-        self.get_success(self.master_store.persist_events([(j2, j2ctx), (msg, msgctx)]))
+        self.get_success(
+            self.storage.persistence.persist_events([(j2, j2ctx), (msg, msgctx)])
+        )
         self.replicate()
 
         event_source = RoomEventSource(self.hs)
@@ -290,10 +292,12 @@ class SlavedEventStoreTestCase(BaseSlavedStoreTestCase):
 
         if backfill:
             self.get_success(
-                self.master_store.persist_events([(event, context)], backfilled=True)
+                self.storage.persistence.persist_events(
+                    [(event, context)], backfilled=True
+                )
             )
         else:
-            self.get_success(self.master_store.persist_event(event, context))
+            self.get_success(self.storage.persistence.persist_event(event, context))
 
         return event
 
diff --git a/tests/rest/admin/test_admin.py b/tests/rest/admin/test_admin.py
index 5877bb2133..8e1ca8b738 100644
--- a/tests/rest/admin/test_admin.py
+++ b/tests/rest/admin/test_admin.py
@@ -62,7 +62,7 @@ class UserRegisterTestCase(unittest.HomeserverTestCase):
         self.device_handler.check_device_registered = Mock(return_value="FAKE")
 
         self.datastore = Mock(return_value=Mock())
-        self.datastore.get_current_state_deltas = Mock(return_value=[])
+        self.datastore.get_current_state_deltas = Mock(return_value=(0, []))
 
         self.secrets = Mock()
 
@@ -561,3 +561,81 @@ class DeleteGroupTestCase(unittest.HomeserverTestCase):
         self.assertEqual(200, int(channel.result["code"]), msg=channel.result["body"])
 
         return channel.json_body["groups"]
+
+
+class PurgeRoomTestCase(unittest.HomeserverTestCase):
+    """Test /purge_room admin API.
+    """
+
+    servlets = [
+        synapse.rest.admin.register_servlets,
+        login.register_servlets,
+        room.register_servlets,
+    ]
+
+    def prepare(self, reactor, clock, hs):
+        self.store = hs.get_datastore()
+
+        self.admin_user = self.register_user("admin", "pass", admin=True)
+        self.admin_user_tok = self.login("admin", "pass")
+
+    def test_purge_room(self):
+        room_id = self.helper.create_room_as(self.admin_user, tok=self.admin_user_tok)
+
+        # All users have to have left the room.
+        self.helper.leave(room_id, user=self.admin_user, tok=self.admin_user_tok)
+
+        url = "/_synapse/admin/v1/purge_room"
+        request, channel = self.make_request(
+            "POST",
+            url.encode("ascii"),
+            {"room_id": room_id},
+            access_token=self.admin_user_tok,
+        )
+        self.render(request)
+
+        self.assertEqual(200, int(channel.result["code"]), msg=channel.result["body"])
+
+        # Test that the following tables have been purged of all rows related to the room.
+        for table in (
+            "current_state_events",
+            "event_backward_extremities",
+            "event_forward_extremities",
+            "event_json",
+            "event_push_actions",
+            "event_search",
+            "events",
+            "group_rooms",
+            "public_room_list_stream",
+            "receipts_graph",
+            "receipts_linearized",
+            "room_aliases",
+            "room_depth",
+            "room_memberships",
+            "room_stats_state",
+            "room_stats_current",
+            "room_stats_historical",
+            "room_stats_earliest_token",
+            "rooms",
+            "stream_ordering_to_exterm",
+            "users_in_public_rooms",
+            "users_who_share_private_rooms",
+            "appservice_room_list",
+            "e2e_room_keys",
+            "event_push_summary",
+            "pusher_throttle",
+            "group_summary_rooms",
+            "local_invites",
+            "room_account_data",
+            "room_tags",
+        ):
+            count = self.get_success(
+                self.store._simple_select_one_onecol(
+                    table="events",
+                    keyvalues={"room_id": room_id},
+                    retcol="COUNT(*)",
+                    desc="test_purge_room",
+                )
+            )
+
+            self.assertEqual(count, 0, msg="Rows not purged in {}".format(table))
diff --git a/tests/rest/client/test_redactions.py b/tests/rest/client/test_redactions.py
index fe66e397c4..d2bcf256fa 100644
--- a/tests/rest/client/test_redactions.py
+++ b/tests/rest/client/test_redactions.py
@@ -30,6 +30,14 @@ class RedactionsTestCase(HomeserverTestCase):
         sync.register_servlets,
     ]
 
+    def make_homeserver(self, reactor, clock):
+        config = self.default_config()
+
+        config["rc_message"] = {"per_second": 0.2, "burst_count": 10}
+        config["rc_admin_redaction"] = {"per_second": 1, "burst_count": 100}
+
+        return self.setup_test_homeserver(config=config)
+
     def prepare(self, reactor, clock, hs):
         # register a couple of users
         self.mod_user_id = self.register_user("user1", "pass")
@@ -177,3 +185,20 @@ class RedactionsTestCase(HomeserverTestCase):
         self._redact_event(
             self.other_access_token, self.room_id, create_event_id, expect_code=403
         )
+
+    def test_redact_event_as_moderator_ratelimit(self):
+        """Tests that the correct ratelimiting is applied to redactions
+        """
+
+        message_ids = []
+        # as a regular user, send messages to redact
+        for _ in range(20):
+            b = self.helper.send(room_id=self.room_id, tok=self.other_access_token)
+            message_ids.append(b["event_id"])
+            self.reactor.advance(10)  # To get around ratelimits
+
+        # as the moderator, send a bunch of redactions
+        for msg_id in message_ids:
+            # These should all succeed, even though this would be denied by
+            # the standard message ratelimiter
+            self._redact_event(self.mod_access_token, self.room_id, msg_id)
diff --git a/tests/rest/client/v1/test_rooms.py b/tests/rest/client/v1/test_rooms.py
index fe741637f5..5e38fd6ced 100644
--- a/tests/rest/client/v1/test_rooms.py
+++ b/tests/rest/client/v1/test_rooms.py
@@ -24,7 +24,7 @@ from six.moves.urllib import parse as urlparse
 from twisted.internet import defer
 
 import synapse.rest.admin
-from synapse.api.constants import Membership
+from synapse.api.constants import EventContentFields, EventTypes, Membership
 from synapse.rest.client.v1 import login, profile, room
 
 from tests import unittest
@@ -484,6 +484,15 @@ class RoomsCreateTestCase(RoomBase):
         self.render(request)
         self.assertEquals(400, channel.code)
 
+    def test_post_room_invitees_invalid_mxid(self):
+        # POST with invalid invitee, see https://github.com/matrix-org/synapse/issues/4088
+        # Note the trailing space in the MXID here!
+        request, channel = self.make_request(
+            "POST", "/createRoom", b'{"invite":["@alice:example.com "]}'
+        )
+        self.render(request)
+        self.assertEquals(400, channel.code)
+
 
 class RoomTopicTestCase(RoomBase):
     """ Tests /rooms/$room_id/topic REST events. """
@@ -802,6 +811,105 @@ class RoomMessageListTestCase(RoomBase):
         self.assertTrue("chunk" in channel.json_body)
         self.assertTrue("end" in channel.json_body)
 
+    def test_filter_labels(self):
+        """Test that we can filter by a label."""
+        message_filter = json.dumps(
+            {"types": [EventTypes.Message], "org.matrix.labels": ["#fun"]}
+        )
+
+        events = self._test_filter_labels(message_filter)
+
+        self.assertEqual(len(events), 2, [event["content"] for event in events])
+        self.assertEqual(events[0]["content"]["body"], "with right label", events[0])
+        self.assertEqual(events[1]["content"]["body"], "with right label", events[1])
+
+    def test_filter_not_labels(self):
+        """Test that we can filter by the absence of a label."""
+        message_filter = json.dumps(
+            {"types": [EventTypes.Message], "org.matrix.not_labels": ["#fun"]}
+        )
+
+        events = self._test_filter_labels(message_filter)
+
+        self.assertEqual(len(events), 3, [event["content"] for event in events])
+        self.assertEqual(events[0]["content"]["body"], "without label", events[0])
+        self.assertEqual(events[1]["content"]["body"], "with wrong label", events[1])
+        self.assertEqual(
+            events[2]["content"]["body"], "with two wrong labels", events[2]
+        )
+
+    def test_filter_labels_not_labels(self):
+        """Test that we can filter by both a label and the absence of another label."""
+        sync_filter = json.dumps(
+            {
+                "types": [EventTypes.Message],
+                "org.matrix.labels": ["#work"],
+                "org.matrix.not_labels": ["#notfun"],
+            }
+        )
+
+        events = self._test_filter_labels(sync_filter)
+
+        self.assertEqual(len(events), 1, [event["content"] for event in events])
+        self.assertEqual(events[0]["content"]["body"], "with wrong label", events[0])
+
+    def _test_filter_labels(self, message_filter):
+        self.helper.send_event(
+            room_id=self.room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with right label",
+                EventContentFields.LABELS: ["#fun"],
+            },
+        )
+
+        self.helper.send_event(
+            room_id=self.room_id,
+            type=EventTypes.Message,
+            content={"msgtype": "m.text", "body": "without label"},
+        )
+
+        self.helper.send_event(
+            room_id=self.room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with wrong label",
+                EventContentFields.LABELS: ["#work"],
+            },
+        )
+
+        self.helper.send_event(
+            room_id=self.room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with two wrong labels",
+                EventContentFields.LABELS: ["#work", "#notfun"],
+            },
+        )
+
+        self.helper.send_event(
+            room_id=self.room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with right label",
+                EventContentFields.LABELS: ["#fun"],
+            },
+        )
+
+        token = "s0_0_0_0_0_0_0_0_0"
+        request, channel = self.make_request(
+            "GET",
+            "/rooms/%s/messages?access_token=x&from=%s&filter=%s"
+            % (self.room_id, token, message_filter),
+        )
+        self.render(request)
+
+        return channel.json_body["chunk"]
+
 
 class RoomSearchTestCase(unittest.HomeserverTestCase):
     servlets = [
diff --git a/tests/rest/client/v1/utils.py b/tests/rest/client/v1/utils.py
index 9915367144..8ea0cb05ea 100644
--- a/tests/rest/client/v1/utils.py
+++ b/tests/rest/client/v1/utils.py
@@ -106,13 +106,22 @@ class RestHelper(object):
         self.auth_user_id = temp_id
 
     def send(self, room_id, body=None, txn_id=None, tok=None, expect_code=200):
-        if txn_id is None:
-            txn_id = "m%s" % (str(time.time()))
         if body is None:
             body = "body_text_here"
 
-        path = "/_matrix/client/r0/rooms/%s/send/m.room.message/%s" % (room_id, txn_id)
         content = {"msgtype": "m.text", "body": body}
+
+        return self.send_event(
+            room_id, "m.room.message", content, txn_id, tok, expect_code
+        )
+
+    def send_event(
+        self, room_id, type, content={}, txn_id=None, tok=None, expect_code=200
+    ):
+        if txn_id is None:
+            txn_id = "m%s" % (str(time.time()))
+
+        path = "/_matrix/client/r0/rooms/%s/send/%s/%s" % (room_id, type, txn_id)
         if tok:
             path = path + "?access_token=%s" % tok
 
@@ -128,8 +137,12 @@ class RestHelper(object):
 
         return channel.json_body
 
-    def send_state(self, room_id, event_type, body, tok, expect_code=200):
-        path = "/_matrix/client/r0/rooms/%s/state/%s" % (room_id, event_type)
+    def send_state(self, room_id, event_type, body, tok, expect_code=200, state_key=""):
+        path = "/_matrix/client/r0/rooms/%s/state/%s/%s" % (
+            room_id,
+            event_type,
+            state_key,
+        )
         if tok:
             path = path + "?access_token=%s" % tok
 
diff --git a/tests/rest/client/v2_alpha/test_account.py b/tests/rest/client/v2_alpha/test_account.py
index 920de41de4..0f51895b81 100644
--- a/tests/rest/client/v2_alpha/test_account.py
+++ b/tests/rest/client/v2_alpha/test_account.py
@@ -23,8 +23,8 @@ from email.parser import Parser
 import pkg_resources
 
 import synapse.rest.admin
-from synapse.api.constants import LoginType
-from synapse.rest.client.v1 import login
+from synapse.api.constants import LoginType, Membership
+from synapse.rest.client.v1 import login, room
 from synapse.rest.client.v2_alpha import account, register
 
 from tests import unittest
@@ -244,16 +244,66 @@ class DeactivateTestCase(unittest.HomeserverTestCase):
         synapse.rest.admin.register_servlets_for_client_rest_resource,
         login.register_servlets,
         account.register_servlets,
+        room.register_servlets,
     ]
 
     def make_homeserver(self, reactor, clock):
-        hs = self.setup_test_homeserver()
-        return hs
+        self.hs = self.setup_test_homeserver()
+        return self.hs
 
     def test_deactivate_account(self):
         user_id = self.register_user("kermit", "test")
         tok = self.login("kermit", "test")
 
+        self.deactivate(user_id, tok)
+
+        store = self.hs.get_datastore()
+
+        # Check that the user has been marked as deactivated.
+        self.assertTrue(self.get_success(store.get_user_deactivated_status(user_id)))
+
+        # Check that this access token has been invalidated.
+        request, channel = self.make_request("GET", "account/whoami")
+        self.render(request)
+        self.assertEqual(request.code, 401)
+
+    @unittest.INFO
+    def test_pending_invites(self):
+        """Tests that deactivating a user rejects every pending invite for them."""
+        store = self.hs.get_datastore()
+
+        inviter_id = self.register_user("inviter", "test")
+        inviter_tok = self.login("inviter", "test")
+
+        invitee_id = self.register_user("invitee", "test")
+        invitee_tok = self.login("invitee", "test")
+
+        # Make @inviter:test invite @invitee:test in a new room.
+        room_id = self.helper.create_room_as(inviter_id, tok=inviter_tok)
+        self.helper.invite(
+            room=room_id, src=inviter_id, targ=invitee_id, tok=inviter_tok
+        )
+
+        # Make sure the invite is here.
+        pending_invites = self.get_success(store.get_invited_rooms_for_user(invitee_id))
+        self.assertEqual(len(pending_invites), 1, pending_invites)
+        self.assertEqual(pending_invites[0].room_id, room_id, pending_invites)
+
+        # Deactivate @invitee:test.
+        self.deactivate(invitee_id, invitee_tok)
+
+        # Check that the invite isn't there anymore.
+        pending_invites = self.get_success(store.get_invited_rooms_for_user(invitee_id))
+        self.assertEqual(len(pending_invites), 0, pending_invites)
+
+        # Check that the membership of @invitee:test in the room is now "leave".
+        memberships = self.get_success(
+            store.get_rooms_for_user_where_membership_is(invitee_id, [Membership.LEAVE])
+        )
+        self.assertEqual(len(memberships), 1, memberships)
+        self.assertEqual(memberships[0].room_id, room_id, memberships)
+
+    def deactivate(self, user_id, tok):
         request_data = json.dumps(
             {
                 "auth": {
@@ -269,13 +319,3 @@ class DeactivateTestCase(unittest.HomeserverTestCase):
         )
         self.render(request)
         self.assertEqual(request.code, 200)
-
-        store = self.hs.get_datastore()
-
-        # Check that the user has been marked as deactivated.
-        self.assertTrue(self.get_success(store.get_user_deactivated_status(user_id)))
-
-        # Check that this access token has been invalidated.
-        request, channel = self.make_request("GET", "account/whoami")
-        self.render(request)
-        self.assertEqual(request.code, 401)
diff --git a/tests/rest/client/v2_alpha/test_auth.py b/tests/rest/client/v2_alpha/test_auth.py
index b9ef46e8fb..b6df1396ad 100644
--- a/tests/rest/client/v2_alpha/test_auth.py
+++ b/tests/rest/client/v2_alpha/test_auth.py
@@ -18,11 +18,22 @@ from twisted.internet.defer import succeed
 
 import synapse.rest.admin
 from synapse.api.constants import LoginType
+from synapse.handlers.ui_auth.checkers import UserInteractiveAuthChecker
 from synapse.rest.client.v2_alpha import auth, register
 
 from tests import unittest
 
 
+class DummyRecaptchaChecker(UserInteractiveAuthChecker):
+    def __init__(self, hs):
+        super().__init__(hs)
+        self.recaptcha_attempts = []
+
+    def check_auth(self, authdict, clientip):
+        self.recaptcha_attempts.append((authdict, clientip))
+        return succeed(True)
+
+
 class FallbackAuthTests(unittest.HomeserverTestCase):
 
     servlets = [
@@ -44,15 +55,9 @@ class FallbackAuthTests(unittest.HomeserverTestCase):
         return hs
 
     def prepare(self, reactor, clock, hs):
+        self.recaptcha_checker = DummyRecaptchaChecker(hs)
         auth_handler = hs.get_auth_handler()
-
-        self.recaptcha_attempts = []
-
-        def _recaptcha(authdict, clientip):
-            self.recaptcha_attempts.append((authdict, clientip))
-            return succeed(True)
-
-        auth_handler.checkers[LoginType.RECAPTCHA] = _recaptcha
+        auth_handler.checkers[LoginType.RECAPTCHA] = self.recaptcha_checker
 
     @unittest.INFO
     def test_fallback_captcha(self):
@@ -89,8 +94,9 @@ class FallbackAuthTests(unittest.HomeserverTestCase):
         self.assertEqual(request.code, 200)
 
         # The recaptcha handler is called with the response given
-        self.assertEqual(len(self.recaptcha_attempts), 1)
-        self.assertEqual(self.recaptcha_attempts[0][0]["response"], "a")
+        attempts = self.recaptcha_checker.recaptcha_attempts
+        self.assertEqual(len(attempts), 1)
+        self.assertEqual(attempts[0][0]["response"], "a")
 
         # also complete the dummy auth
         request, channel = self.make_request(
diff --git a/tests/rest/client/v2_alpha/test_filter.py b/tests/rest/client/v2_alpha/test_filter.py
index f42a8efbf4..e0e9e94fbf 100644
--- a/tests/rest/client/v2_alpha/test_filter.py
+++ b/tests/rest/client/v2_alpha/test_filter.py
@@ -92,7 +92,7 @@ class FilterTestCase(unittest.HomeserverTestCase):
         )
         self.render(request)
 
-        self.assertEqual(channel.result["code"], b"400")
+        self.assertEqual(channel.result["code"], b"404")
         self.assertEquals(channel.json_body["errcode"], Codes.NOT_FOUND)
 
     # Currently invalid params do not have an appropriate errcode
diff --git a/tests/rest/client/v2_alpha/test_register.py b/tests/rest/client/v2_alpha/test_register.py
index ab4d7d70d0..dab87e5edf 100644
--- a/tests/rest/client/v2_alpha/test_register.py
+++ b/tests/rest/client/v2_alpha/test_register.py
@@ -34,19 +34,12 @@ from tests import unittest
 class RegisterRestServletTestCase(unittest.HomeserverTestCase):
 
     servlets = [register.register_servlets]
+    url = b"/_matrix/client/r0/register"
 
-    def make_homeserver(self, reactor, clock):
-
-        self.url = b"/_matrix/client/r0/register"
-
-        self.hs = self.setup_test_homeserver()
-        self.hs.config.enable_registration = True
-        self.hs.config.registrations_require_3pid = []
-        self.hs.config.auto_join_rooms = []
-        self.hs.config.enable_registration_captcha = False
-        self.hs.config.allow_guest_access = True
-
-        return self.hs
+    def default_config(self, name="test"):
+        config = super().default_config(name)
+        config["allow_guest_access"] = True
+        return config
 
     def test_POST_appservice_registration_valid(self):
         user_id = "@as_user_kermit:test"
@@ -199,6 +192,73 @@ class RegisterRestServletTestCase(unittest.HomeserverTestCase):
 
         self.assertEquals(channel.result["code"], b"200", channel.result)
 
+    def test_advertised_flows(self):
+        request, channel = self.make_request(b"POST", self.url, b"{}")
+        self.render(request)
+        self.assertEquals(channel.result["code"], b"401", channel.result)
+        flows = channel.json_body["flows"]
+
+        # with the stock config, we only expect the dummy flow
+        self.assertCountEqual([["m.login.dummy"]], (f["stages"] for f in flows))
+
+    @unittest.override_config(
+        {
+            "enable_registration_captcha": True,
+            "user_consent": {
+                "version": "1",
+                "template_dir": "/",
+                "require_at_registration": True,
+            },
+            "account_threepid_delegates": {
+                "email": "https://id_server",
+                "msisdn": "https://id_server",
+            },
+        }
+    )
+    def test_advertised_flows_captcha_and_terms_and_3pids(self):
+        request, channel = self.make_request(b"POST", self.url, b"{}")
+        self.render(request)
+        self.assertEquals(channel.result["code"], b"401", channel.result)
+        flows = channel.json_body["flows"]
+
+        self.assertCountEqual(
+            [
+                ["m.login.recaptcha", "m.login.terms", "m.login.dummy"],
+                ["m.login.recaptcha", "m.login.terms", "m.login.email.identity"],
+                ["m.login.recaptcha", "m.login.terms", "m.login.msisdn"],
+                [
+                    "m.login.recaptcha",
+                    "m.login.terms",
+                    "m.login.msisdn",
+                    "m.login.email.identity",
+                ],
+            ],
+            (f["stages"] for f in flows),
+        )
+
+    @unittest.override_config(
+        {
+            "public_baseurl": "https://test_server",
+            "registrations_require_3pid": ["email"],
+            "disable_msisdn_registration": True,
+            "email": {
+                "smtp_host": "mail_server",
+                "smtp_port": 2525,
+                "notif_from": "sender@host",
+            },
+        }
+    )
+    def test_advertised_flows_no_msisdn_email_required(self):
+        request, channel = self.make_request(b"POST", self.url, b"{}")
+        self.render(request)
+        self.assertEquals(channel.result["code"], b"401", channel.result)
+        flows = channel.json_body["flows"]
+
+        # with the stock config, we expect all four combinations of 3pid
+        self.assertCountEqual(
+            [["m.login.email.identity"]], (f["stages"] for f in flows)
+        )
+
 
 class AccountValidityTestCase(unittest.HomeserverTestCase):
 
diff --git a/tests/rest/client/v2_alpha/test_sync.py b/tests/rest/client/v2_alpha/test_sync.py
index 71895094bd..3283c0e47b 100644
--- a/tests/rest/client/v2_alpha/test_sync.py
+++ b/tests/rest/client/v2_alpha/test_sync.py
@@ -12,10 +12,12 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import json
 
 from mock import Mock
 
 import synapse.rest.admin
+from synapse.api.constants import EventContentFields, EventTypes
 from synapse.rest.client.v1 import login, room
 from synapse.rest.client.v2_alpha import sync
 
@@ -26,7 +28,12 @@ from tests.server import TimedOutException
 class FilterTestCase(unittest.HomeserverTestCase):
 
     user_id = "@apple:test"
-    servlets = [sync.register_servlets]
+    servlets = [
+        synapse.rest.admin.register_servlets_for_client_rest_resource,
+        room.register_servlets,
+        login.register_servlets,
+        sync.register_servlets,
+    ]
 
     def make_homeserver(self, reactor, clock):
 
@@ -70,6 +77,140 @@ class FilterTestCase(unittest.HomeserverTestCase):
         )
 
 
+class SyncFilterTestCase(unittest.HomeserverTestCase):
+    servlets = [
+        synapse.rest.admin.register_servlets_for_client_rest_resource,
+        room.register_servlets,
+        login.register_servlets,
+        sync.register_servlets,
+    ]
+
+    def test_sync_filter_labels(self):
+        """Test that we can filter by a label."""
+        sync_filter = json.dumps(
+            {
+                "room": {
+                    "timeline": {
+                        "types": [EventTypes.Message],
+                        "org.matrix.labels": ["#fun"],
+                    }
+                }
+            }
+        )
+
+        events = self._test_sync_filter_labels(sync_filter)
+
+        self.assertEqual(len(events), 2, [event["content"] for event in events])
+        self.assertEqual(events[0]["content"]["body"], "with right label", events[0])
+        self.assertEqual(events[1]["content"]["body"], "with right label", events[1])
+
+    def test_sync_filter_not_labels(self):
+        """Test that we can filter by the absence of a label."""
+        sync_filter = json.dumps(
+            {
+                "room": {
+                    "timeline": {
+                        "types": [EventTypes.Message],
+                        "org.matrix.not_labels": ["#fun"],
+                    }
+                }
+            }
+        )
+
+        events = self._test_sync_filter_labels(sync_filter)
+
+        self.assertEqual(len(events), 3, [event["content"] for event in events])
+        self.assertEqual(events[0]["content"]["body"], "without label", events[0])
+        self.assertEqual(events[1]["content"]["body"], "with wrong label", events[1])
+        self.assertEqual(
+            events[2]["content"]["body"], "with two wrong labels", events[2]
+        )
+
+    def test_sync_filter_labels_not_labels(self):
+        """Test that we can filter by both a label and the absence of another label."""
+        sync_filter = json.dumps(
+            {
+                "room": {
+                    "timeline": {
+                        "types": [EventTypes.Message],
+                        "org.matrix.labels": ["#work"],
+                        "org.matrix.not_labels": ["#notfun"],
+                    }
+                }
+            }
+        )
+
+        events = self._test_sync_filter_labels(sync_filter)
+
+        self.assertEqual(len(events), 1, [event["content"] for event in events])
+        self.assertEqual(events[0]["content"]["body"], "with wrong label", events[0])
+
+    def _test_sync_filter_labels(self, sync_filter):
+        user_id = self.register_user("kermit", "test")
+        tok = self.login("kermit", "test")
+
+        room_id = self.helper.create_room_as(user_id, tok=tok)
+
+        self.helper.send_event(
+            room_id=room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with right label",
+                EventContentFields.LABELS: ["#fun"],
+            },
+            tok=tok,
+        )
+
+        self.helper.send_event(
+            room_id=room_id,
+            type=EventTypes.Message,
+            content={"msgtype": "m.text", "body": "without label"},
+            tok=tok,
+        )
+
+        self.helper.send_event(
+            room_id=room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with wrong label",
+                EventContentFields.LABELS: ["#work"],
+            },
+            tok=tok,
+        )
+
+        self.helper.send_event(
+            room_id=room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with two wrong labels",
+                EventContentFields.LABELS: ["#work", "#notfun"],
+            },
+            tok=tok,
+        )
+
+        self.helper.send_event(
+            room_id=room_id,
+            type=EventTypes.Message,
+            content={
+                "msgtype": "m.text",
+                "body": "with right label",
+                EventContentFields.LABELS: ["#fun"],
+            },
+            tok=tok,
+        )
+
+        request, channel = self.make_request(
+            "GET", "/sync?filter=%s" % sync_filter, access_token=tok
+        )
+        self.render(request)
+        self.assertEqual(channel.code, 200, channel.result)
+
+        return channel.json_body["rooms"]["join"][room_id]["timeline"]["events"]
+
+
 class SyncTypingTests(unittest.HomeserverTestCase):
 
     servlets = [
diff --git a/tests/rest/media/v1/test_url_preview.py b/tests/rest/media/v1/test_url_preview.py
index 976652aee8..852b8ab11c 100644
--- a/tests/rest/media/v1/test_url_preview.py
+++ b/tests/rest/media/v1/test_url_preview.py
@@ -247,6 +247,41 @@ class URLPreviewTests(unittest.HomeserverTestCase):
         self.assertEqual(channel.code, 200)
         self.assertEqual(channel.json_body["og:title"], "\u0434\u043a\u0430")
 
+    def test_overlong_title(self):
+        self.lookups["matrix.org"] = [(IPv4Address, "8.8.8.8")]
+
+        end_content = (
+            b""
+            b"" + b"x" * 2000 + b""
+            b''
+            b""
+        )
+
+        request, channel = self.make_request(
+            "GET", "url_preview?url=http://matrix.org", shorthand=False
+        )
+        request.render(self.preview_url)
+        self.pump()
+
+        client = self.reactor.tcpClients[0][2].buildProtocol(None)
+        server = AccumulatingProtocol()
+        server.makeConnection(FakeTransport(client, self.reactor))
+        client.makeConnection(FakeTransport(server, self.reactor))
+        client.dataReceived(
+            (
+                b"HTTP/1.0 200 OK\r\nContent-Length: %d\r\n"
+                b'Content-Type: text/html; charset="windows-1251"\r\n\r\n'
+            )
+            % (len(end_content),)
+            + end_content
+        )
+
+        self.pump()
+        self.assertEqual(channel.code, 200)
+        res = channel.json_body
+        # We should only see the `og:description` field, as `title` is too long and should be stripped out
+        self.assertCountEqual(["og:description"], res.keys())
+
     def test_ipaddr(self):
         """
         IP addresses can be previewed directly.
diff --git a/tests/server.py b/tests/server.py
index e397ebe8fa..f878aeaada 100644
--- a/tests/server.py
+++ b/tests/server.py
@@ -161,7 +161,11 @@ def make_request(
         path = path.encode("ascii")
 
     # Decorate it to be the full path, if we're using shorthand
-    if shorthand and not path.startswith(b"/_matrix"):
+    if (
+        shorthand
+        and not path.startswith(b"/_matrix")
+        and not path.startswith(b"/_synapse")
+    ):
         path = b"/_matrix/client/r0/" + path
         path = path.replace(b"//", b"/")
 
@@ -391,11 +395,24 @@ class FakeTransport(object):
             self.disconnecting = True
             if self._protocol:
                 self._protocol.connectionLost(reason)
-            self.disconnected = True
+
+            # if we still have data to write, delay until that is done
+            if self.buffer:
+                logger.info(
+                    "FakeTransport: Delaying disconnect until buffer is flushed"
+                )
+            else:
+                self.disconnected = True
 
     def abortConnection(self):
         logger.info("FakeTransport: abortConnection()")
-        self.loseConnection()
+
+        if not self.disconnecting:
+            self.disconnecting = True
+            if self._protocol:
+                self._protocol.connectionLost(None)
+
+        self.disconnected = True
 
     def pauseProducing(self):
         if not self.producer:
@@ -426,6 +443,9 @@ class FakeTransport(object):
             self._reactor.callLater(0.0, _produce)
 
     def write(self, byt):
+        if self.disconnecting:
+            raise Exception("Writing to disconnecting FakeTransport")
+
         self.buffer = self.buffer + byt
 
         # always actually do the write asynchronously. Some protocols (notably the
@@ -470,6 +490,10 @@ class FakeTransport(object):
         if self.buffer and self.autoflush:
             self._reactor.callLater(0.0, self.flush)
 
+        if not self.buffer and self.disconnecting:
+            logger.info("FakeTransport: Buffer now empty, completing disconnect")
+            self.disconnected = True
+
 
 def connect_client(reactor: IReactorTCP, client_id: int) -> AccumulatingProtocol:
     """
diff --git a/tests/server_notices/test_resource_limits_server_notices.py b/tests/server_notices/test_resource_limits_server_notices.py
index cdf89e3383..eb540e34f6 100644
--- a/tests/server_notices/test_resource_limits_server_notices.py
+++ b/tests/server_notices/test_resource_limits_server_notices.py
@@ -17,7 +17,7 @@ from mock import Mock
 
 from twisted.internet import defer
 
-from synapse.api.constants import EventTypes, ServerNoticeMsgType
+from synapse.api.constants import EventTypes, LimitBlockingTypes, ServerNoticeMsgType
 from synapse.api.errors import ResourceLimitError
 from synapse.server_notices.resource_limits_server_notices import (
     ResourceLimitsServerNotices,
@@ -133,7 +133,7 @@ class TestResourceLimitsServerNotices(unittest.HomeserverTestCase):
         self.get_success(self._rlsn.maybe_send_server_notice_to_user(self.user_id))
 
         # Would be better to check contents, but 2 calls == set blocking event
-        self.assertTrue(self._send_notice.call_count == 2)
+        self.assertEqual(self._send_notice.call_count, 2)
 
     def test_maybe_send_server_notice_to_user_add_blocked_notice_noop(self):
         """
@@ -158,6 +158,61 @@ class TestResourceLimitsServerNotices(unittest.HomeserverTestCase):
 
         self._send_notice.assert_not_called()
 
+    def test_maybe_send_server_notice_when_alerting_suppressed_room_unblocked(self):
+        """
+        Test that when server is over MAU limit and alerting is suppressed, then
+        an alert message is not sent into the room
+        """
+        self.hs.config.mau_limit_alerting = False
+        self._rlsn._auth.check_auth_blocking = Mock(
+            side_effect=ResourceLimitError(
+                403, "foo", limit_type=LimitBlockingTypes.MONTHLY_ACTIVE_USER
+            )
+        )
+        self.get_success(self._rlsn.maybe_send_server_notice_to_user(self.user_id))
+
+        self.assertTrue(self._send_notice.call_count == 0)
+
+    def test_check_hs_disabled_unaffected_by_mau_alert_suppression(self):
+        """
+        Test that when a server is disabled, that MAU limit alerting is ignored.
+        """
+        self.hs.config.mau_limit_alerting = False
+        self._rlsn._auth.check_auth_blocking = Mock(
+            side_effect=ResourceLimitError(
+                403, "foo", limit_type=LimitBlockingTypes.HS_DISABLED
+            )
+        )
+        self.get_success(self._rlsn.maybe_send_server_notice_to_user(self.user_id))
+
+        # Would be better to check contents, but 2 calls == set blocking event
+        self.assertEqual(self._send_notice.call_count, 2)
+
+    def test_maybe_send_server_notice_when_alerting_suppressed_room_blocked(self):
+        """
+        When the room is already in a blocked state, test that when alerting
+        is suppressed that the room is returned to an unblocked state.
+        """
+        self.hs.config.mau_limit_alerting = False
+        self._rlsn._auth.check_auth_blocking = Mock(
+            side_effect=ResourceLimitError(
+                403, "foo", limit_type=LimitBlockingTypes.MONTHLY_ACTIVE_USER
+            )
+        )
+        self._rlsn._server_notices_manager.__is_room_currently_blocked = Mock(
+            return_value=defer.succeed((True, []))
+        )
+
+        mock_event = Mock(
+            type=EventTypes.Message, content={"msgtype": ServerNoticeMsgType}
+        )
+        self._rlsn._store.get_events = Mock(
+            return_value=defer.succeed({"123": mock_event})
+        )
+        self.get_success(self._rlsn.maybe_send_server_notice_to_user(self.user_id))
+
+        self._send_notice.assert_called_once()
+
 
 class TestResourceLimitsServerNoticesWithRealRooms(unittest.HomeserverTestCase):
     def prepare(self, reactor, clock, hs):
diff --git a/tests/storage/test__base.py b/tests/storage/test__base.py
index dd49a14524..9b81b536f5 100644
--- a/tests/storage/test__base.py
+++ b/tests/storage/test__base.py
@@ -197,7 +197,7 @@ class CacheDecoratorTestCase(unittest.TestCase):
 
         a.func.prefill(("foo",), ObservableDeferred(d))
 
-        self.assertEquals(a.func("foo"), d.result)
+        self.assertEquals(a.func("foo").result, d.result)
         self.assertEquals(callcount[0], 0)
 
     @defer.inlineCallbacks
diff --git a/tests/storage/test_appservice.py b/tests/storage/test_appservice.py
index 622b16a071..dfeea24599 100644
--- a/tests/storage/test_appservice.py
+++ b/tests/storage/test_appservice.py
@@ -24,7 +24,7 @@ from twisted.internet import defer
 
 from synapse.appservice import ApplicationService, ApplicationServiceState
 from synapse.config._base import ConfigError
-from synapse.storage.appservice import (
+from synapse.storage.data_stores.main.appservice import (
     ApplicationServiceStore,
     ApplicationServiceTransactionStore,
 )
diff --git a/tests/storage/test_cleanup_extrems.py b/tests/storage/test_cleanup_extrems.py
index e9e2d5337c..69dcaa63d5 100644
--- a/tests/storage/test_cleanup_extrems.py
+++ b/tests/storage/test_cleanup_extrems.py
@@ -14,7 +14,13 @@
 # limitations under the License.
 
 import os.path
+from unittest.mock import patch
 
+from mock import Mock
+
+import synapse.rest.admin
+from synapse.api.constants import EventTypes
+from synapse.rest.client.v1 import login, room
 from synapse.storage import prepare_database
 from synapse.types import Requester, UserID
 
@@ -44,6 +50,8 @@ class CleanupExtremBackgroundUpdateStoreTestCase(HomeserverTestCase):
 
         schema_path = os.path.join(
             prepare_database.dir_path,
+            "data_stores",
+            "main",
             "schema",
             "delta",
             "54",
@@ -225,6 +233,14 @@ class CleanupExtremBackgroundUpdateStoreTestCase(HomeserverTestCase):
 
 
 class CleanupExtremDummyEventsTestCase(HomeserverTestCase):
+    CONSENT_VERSION = "1"
+    EXTREMITIES_COUNT = 50
+    servlets = [
+        synapse.rest.admin.register_servlets_for_client_rest_resource,
+        login.register_servlets,
+        room.register_servlets,
+    ]
+
     def make_homeserver(self, reactor, clock):
         config = self.default_config()
         config["cleanup_extremities_with_dummy_events"] = True
@@ -233,27 +249,19 @@ class CleanupExtremDummyEventsTestCase(HomeserverTestCase):
     def prepare(self, reactor, clock, homeserver):
         self.store = homeserver.get_datastore()
         self.room_creator = homeserver.get_room_creation_handler()
+        self.event_creator_handler = homeserver.get_event_creation_handler()
 
         # Create a test user and room
-        self.user = UserID("alice", "test")
+        self.user = UserID.from_string(self.register_user("user1", "password"))
+        self.token1 = self.login("user1", "password")
         self.requester = Requester(self.user, None, False, None, None)
         info = self.get_success(self.room_creator.create_room(self.requester, {}))
         self.room_id = info["room_id"]
+        self.event_creator = homeserver.get_event_creation_handler()
+        homeserver.config.user_consent_version = self.CONSENT_VERSION
 
     def test_send_dummy_event(self):
-        # Create a bushy graph with 50 extremities.
-
-        event_id_start = self.create_and_send_event(self.room_id, self.user)
-
-        for _ in range(50):
-            self.create_and_send_event(
-                self.room_id, self.user, prev_event_ids=[event_id_start]
-            )
-
-        latest_event_ids = self.get_success(
-            self.store.get_latest_event_ids_in_room(self.room_id)
-        )
-        self.assertEqual(len(latest_event_ids), 50)
+        self._create_extremity_rich_graph()
 
         # Pump the reactor repeatedly so that the background updates have a
         # chance to run.
@@ -263,3 +271,126 @@ class CleanupExtremDummyEventsTestCase(HomeserverTestCase):
             self.store.get_latest_event_ids_in_room(self.room_id)
         )
         self.assertTrue(len(latest_event_ids) < 10, len(latest_event_ids))
+
+    @patch("synapse.handlers.message._DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY", new=0)
+    def test_send_dummy_events_when_insufficient_power(self):
+        self._create_extremity_rich_graph()
+        # Criple power levels
+        self.helper.send_state(
+            self.room_id,
+            EventTypes.PowerLevels,
+            body={"users": {str(self.user): -1}},
+            tok=self.token1,
+        )
+        # Pump the reactor repeatedly so that the background updates have a
+        # chance to run.
+        self.pump(10 * 60)
+
+        latest_event_ids = self.get_success(
+            self.store.get_latest_event_ids_in_room(self.room_id)
+        )
+        # Check that the room has not been pruned
+        self.assertTrue(len(latest_event_ids) > 10)
+
+        # New user with regular levels
+        user2 = self.register_user("user2", "password")
+        token2 = self.login("user2", "password")
+        self.helper.join(self.room_id, user2, tok=token2)
+        self.pump(10 * 60)
+
+        latest_event_ids = self.get_success(
+            self.store.get_latest_event_ids_in_room(self.room_id)
+        )
+        self.assertTrue(len(latest_event_ids) < 10, len(latest_event_ids))
+
+    @patch("synapse.handlers.message._DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY", new=0)
+    def test_send_dummy_event_without_consent(self):
+        self._create_extremity_rich_graph()
+        self._enable_consent_checking()
+
+        # Pump the reactor repeatedly so that the background updates have a
+        # chance to run. Attempt to add dummy event with user that has not consented
+        # Check that dummy event send fails.
+        self.pump(10 * 60)
+        latest_event_ids = self.get_success(
+            self.store.get_latest_event_ids_in_room(self.room_id)
+        )
+        self.assertTrue(len(latest_event_ids) == self.EXTREMITIES_COUNT)
+
+        # Create new user, and add consent
+        user2 = self.register_user("user2", "password")
+        token2 = self.login("user2", "password")
+        self.get_success(
+            self.store.user_set_consent_version(user2, self.CONSENT_VERSION)
+        )
+        self.helper.join(self.room_id, user2, tok=token2)
+
+        # Background updates should now cause a dummy event to be added to the graph
+        self.pump(10 * 60)
+
+        latest_event_ids = self.get_success(
+            self.store.get_latest_event_ids_in_room(self.room_id)
+        )
+        self.assertTrue(len(latest_event_ids) < 10, len(latest_event_ids))
+
+    @patch("synapse.handlers.message._DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY", new=250)
+    def test_expiry_logic(self):
+        """Simple test to ensure that _expire_rooms_to_exclude_from_dummy_event_insertion()
+        expires old entries correctly.
+        """
+        self.event_creator_handler._rooms_to_exclude_from_dummy_event_insertion[
+            "1"
+        ] = 100000
+        self.event_creator_handler._rooms_to_exclude_from_dummy_event_insertion[
+            "2"
+        ] = 200000
+        self.event_creator_handler._rooms_to_exclude_from_dummy_event_insertion[
+            "3"
+        ] = 300000
+        self.event_creator_handler._expire_rooms_to_exclude_from_dummy_event_insertion()
+        # All entries within time frame
+        self.assertEqual(
+            len(
+                self.event_creator_handler._rooms_to_exclude_from_dummy_event_insertion
+            ),
+            3,
+        )
+        # Oldest room to expire
+        self.pump(1)
+        self.event_creator_handler._expire_rooms_to_exclude_from_dummy_event_insertion()
+        self.assertEqual(
+            len(
+                self.event_creator_handler._rooms_to_exclude_from_dummy_event_insertion
+            ),
+            2,
+        )
+        # All rooms to expire
+        self.pump(2)
+        self.assertEqual(
+            len(
+                self.event_creator_handler._rooms_to_exclude_from_dummy_event_insertion
+            ),
+            0,
+        )
+
+    def _create_extremity_rich_graph(self):
+        """Helper method to create bushy graph on demand"""
+
+        event_id_start = self.create_and_send_event(self.room_id, self.user)
+
+        for _ in range(self.EXTREMITIES_COUNT):
+            self.create_and_send_event(
+                self.room_id, self.user, prev_event_ids=[event_id_start]
+            )
+
+        latest_event_ids = self.get_success(
+            self.store.get_latest_event_ids_in_room(self.room_id)
+        )
+        self.assertEqual(len(latest_event_ids), 50)
+
+    def _enable_consent_checking(self):
+        """Helper method to enable consent checking"""
+        self.event_creator._block_events_without_consent_error = "No consent from user"
+        consent_uri_builder = Mock()
+        consent_uri_builder.build_user_consent_uri.return_value = "http://example.com"
+        self.event_creator._consent_uri_builder = consent_uri_builder
diff --git a/tests/storage/test_client_ips.py b/tests/storage/test_client_ips.py
index 09305c3bf1..afac5dec7f 100644
--- a/tests/storage/test_client_ips.py
+++ b/tests/storage/test_client_ips.py
@@ -55,7 +55,6 @@ class ClientIpStoreTestCase(unittest.HomeserverTestCase):
             {
                 "user_id": user_id,
                 "device_id": "device_id",
-                "access_token": "access_token",
                 "ip": "ip",
                 "user_agent": "user_agent",
                 "last_seen": 12345678000,
@@ -201,6 +200,156 @@ class ClientIpStoreTestCase(unittest.HomeserverTestCase):
         active = self.get_success(self.store.user_last_seen_monthly_active(user_id))
         self.assertTrue(active)
 
+    def test_devices_last_seen_bg_update(self):
+        # First make sure we have completed all updates.
+        while not self.get_success(self.store.has_completed_background_updates()):
+            self.get_success(self.store.do_next_background_update(100), by=0.1)
+
+        # Insert a user IP
+        user_id = "@user:id"
+        self.get_success(
+            self.store.insert_client_ip(
+                user_id, "access_token", "ip", "user_agent", "device_id"
+            )
+        )
+
+        # Force persisting to disk
+        self.reactor.advance(200)
+
+        # But clear the associated entry in devices table
+        self.get_success(
+            self.store._simple_update(
+                table="devices",
+                keyvalues={"user_id": user_id, "device_id": "device_id"},
+                updatevalues={"last_seen": None, "ip": None, "user_agent": None},
+                desc="test_devices_last_seen_bg_update",
+            )
+        )
+
+        # We should now get nulls when querying
+        result = self.get_success(
+            self.store.get_last_client_ip_by_device(user_id, "device_id")
+        )
+
+        r = result[(user_id, "device_id")]
+        self.assertDictContainsSubset(
+            {
+                "user_id": user_id,
+                "device_id": "device_id",
+                "ip": None,
+                "user_agent": None,
+                "last_seen": None,
+            },
+            r,
+        )
+
+        # Register the background update to run again.
+        self.get_success(
+            self.store._simple_insert(
+                table="background_updates",
+                values={
+                    "update_name": "devices_last_seen",
+                    "progress_json": "{}",
+                    "depends_on": None,
+                },
+            )
+        )
+
+        # ... and tell the DataStore that it hasn't finished all updates yet
+        self.store._all_done = False
+
+        # Now let's actually drive the updates to completion
+        while not self.get_success(self.store.has_completed_background_updates()):
+            self.get_success(self.store.do_next_background_update(100), by=0.1)
+
+        # We should now get the correct result again
+        result = self.get_success(
+            self.store.get_last_client_ip_by_device(user_id, "device_id")
+        )
+
+        r = result[(user_id, "device_id")]
+        self.assertDictContainsSubset(
+            {
+                "user_id": user_id,
+                "device_id": "device_id",
+                "ip": "ip",
+                "user_agent": "user_agent",
+                "last_seen": 0,
+            },
+            r,
+        )
+
+    def test_old_user_ips_pruned(self):
+        # First make sure we have completed all updates.
+        while not self.get_success(self.store.has_completed_background_updates()):
+            self.get_success(self.store.do_next_background_update(100), by=0.1)
+
+        # Insert a user IP
+        user_id = "@user:id"
+        self.get_success(
+            self.store.insert_client_ip(
+                user_id, "access_token", "ip", "user_agent", "device_id"
+            )
+        )
+
+        # Force persisting to disk
+        self.reactor.advance(200)
+
+        # We should see that in the DB
+        result = self.get_success(
+            self.store._simple_select_list(
+                table="user_ips",
+                keyvalues={"user_id": user_id},
+                retcols=["access_token", "ip", "user_agent", "device_id", "last_seen"],
+                desc="get_user_ip_and_agents",
+            )
+        )
+
+        self.assertEqual(
+            result,
+            [
+                {
+                    "access_token": "access_token",
+                    "ip": "ip",
+                    "user_agent": "user_agent",
+                    "device_id": "device_id",
+                    "last_seen": 0,
+                }
+            ],
+        )
+
+        # Now advance by a couple of months
+        self.reactor.advance(60 * 24 * 60 * 60)
+
+        # We should get no results.
+        result = self.get_success(
+            self.store._simple_select_list(
+                table="user_ips",
+                keyvalues={"user_id": user_id},
+                retcols=["access_token", "ip", "user_agent", "device_id", "last_seen"],
+                desc="get_user_ip_and_agents",
+            )
+        )
+
+        self.assertEqual(result, [])
+
+        # But we should still get the correct values for the device
+        result = self.get_success(
+            self.store.get_last_client_ip_by_device(user_id, "device_id")
+        )
+
+        r = result[(user_id, "device_id")]
+        self.assertDictContainsSubset(
+            {
+                "user_id": user_id,
+                "device_id": "device_id",
+                "ip": "ip",
+                "user_agent": "user_agent",
+                "last_seen": 0,
+            },
+            r,
+        )
+
 
 class ClientIpAuthTestCase(unittest.HomeserverTestCase):
 
diff --git a/tests/storage/test_devices.py b/tests/storage/test_devices.py
index 3cc18f9f1c..6f8d990959 100644
--- a/tests/storage/test_devices.py
+++ b/tests/storage/test_devices.py
@@ -72,7 +72,7 @@ class DeviceStoreTestCase(tests.unittest.TestCase):
         )
 
     @defer.inlineCallbacks
-    def test_get_devices_by_remote(self):
+    def test_get_device_updates_by_remote(self):
         device_ids = ["device_id1", "device_id2"]
 
         # Add two device updates with a single stream_id
@@ -81,7 +81,7 @@ class DeviceStoreTestCase(tests.unittest.TestCase):
         )
 
         # Get all device updates ever meant for this remote
-        now_stream_id, device_updates = yield self.store.get_devices_by_remote(
+        now_stream_id, device_updates = yield self.store.get_device_updates_by_remote(
             "somehost", -1, limit=100
         )
 
@@ -89,7 +89,7 @@ class DeviceStoreTestCase(tests.unittest.TestCase):
         self._check_devices_in_updates(device_ids, device_updates)
 
     @defer.inlineCallbacks
-    def test_get_devices_by_remote_limited(self):
+    def test_get_device_updates_by_remote_limited(self):
         # Test breaking the update limit in 1, 101, and 1 device_id segments
 
         # first add one device
@@ -115,20 +115,20 @@ class DeviceStoreTestCase(tests.unittest.TestCase):
         #
 
         # first we should get a single update
-        now_stream_id, device_updates = yield self.store.get_devices_by_remote(
+        now_stream_id, device_updates = yield self.store.get_device_updates_by_remote(
             "someotherhost", -1, limit=100
         )
         self._check_devices_in_updates(device_ids1, device_updates)
 
         # Then we should get an empty list back as the 101 devices broke the limit
-        now_stream_id, device_updates = yield self.store.get_devices_by_remote(
+        now_stream_id, device_updates = yield self.store.get_device_updates_by_remote(
             "someotherhost", now_stream_id, limit=100
         )
         self.assertEqual(len(device_updates), 0)
 
         # The 101 devices should've been cleared, so we should now just get one device
         # update
-        now_stream_id, device_updates = yield self.store.get_devices_by_remote(
+        now_stream_id, device_updates = yield self.store.get_device_updates_by_remote(
             "someotherhost", now_stream_id, limit=100
         )
         self._check_devices_in_updates(device_ids3, device_updates)
@@ -137,7 +137,9 @@ class DeviceStoreTestCase(tests.unittest.TestCase):
         """Check that an specific device ids exist in a list of device update EDUs"""
         self.assertEqual(len(device_updates), len(expected_device_ids))
 
-        received_device_ids = {update["device_id"] for update in device_updates}
+        received_device_ids = {
+            update["device_id"] for edu_type, update in device_updates
+        }
         self.assertEqual(received_device_ids, set(expected_device_ids))
 
     @defer.inlineCallbacks
diff --git a/tests/storage/test_e2e_room_keys.py b/tests/storage/test_e2e_room_keys.py
new file mode 100644
index 0000000000..d128fde441
--- /dev/null
+++ b/tests/storage/test_e2e_room_keys.py
@@ -0,0 +1,75 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from tests import unittest
+
+# sample room_key data for use in the tests
+room_key = {
+    "first_message_index": 1,
+    "forwarded_count": 1,
+    "is_verified": False,
+    "session_data": "SSBBTSBBIEZJU0gK",
+}
+
+
+class E2eRoomKeysHandlerTestCase(unittest.HomeserverTestCase):
+    def make_homeserver(self, reactor, clock):
+        hs = self.setup_test_homeserver("server", http_client=None)
+        self.store = hs.get_datastore()
+        return hs
+
+    def test_room_keys_version_delete(self):
+        # test that deleting a room key backup deletes the keys
+        version1 = self.get_success(
+            self.store.create_e2e_room_keys_version(
+                "user_id", {"algorithm": "rot13", "auth_data": {}}
+            )
+        )
+
+        self.get_success(
+            self.store.set_e2e_room_key(
+                "user_id", version1, "room", "session", room_key
+            )
+        )
+
+        version2 = self.get_success(
+            self.store.create_e2e_room_keys_version(
+                "user_id", {"algorithm": "rot13", "auth_data": {}}
+            )
+        )
+
+        self.get_success(
+            self.store.set_e2e_room_key(
+                "user_id", version2, "room", "session", room_key
+            )
+        )
+
+        # make sure the keys were stored properly
+        keys = self.get_success(self.store.get_e2e_room_keys("user_id", version1))
+        self.assertEqual(len(keys["rooms"]), 1)
+
+        keys = self.get_success(self.store.get_e2e_room_keys("user_id", version2))
+        self.assertEqual(len(keys["rooms"]), 1)
+
+        # delete version1
+        self.get_success(self.store.delete_e2e_room_keys_version("user_id", version1))
+
+        # make sure the key from version1 is gone, and the key from version2 is
+        # still there
+        keys = self.get_success(self.store.get_e2e_room_keys("user_id", version1))
+        self.assertEqual(len(keys["rooms"]), 0)
+
+        keys = self.get_success(self.store.get_e2e_room_keys("user_id", version2))
+        self.assertEqual(len(keys["rooms"]), 1)
diff --git a/tests/storage/test_end_to_end_keys.py b/tests/storage/test_end_to_end_keys.py
index c8ece15284..398d546280 100644
--- a/tests/storage/test_end_to_end_keys.py
+++ b/tests/storage/test_end_to_end_keys.py
@@ -38,7 +38,7 @@ class EndToEndKeyStoreTestCase(tests.unittest.TestCase):
         self.assertIn("user", res)
         self.assertIn("device", res["user"])
         dev = res["user"]["device"]
-        self.assertDictContainsSubset({"keys": json, "device_display_name": None}, dev)
+        self.assertDictContainsSubset(json, dev)
 
     @defer.inlineCallbacks
     def test_reupload_key(self):
@@ -68,7 +68,7 @@ class EndToEndKeyStoreTestCase(tests.unittest.TestCase):
         self.assertIn("device", res["user"])
         dev = res["user"]["device"]
         self.assertDictContainsSubset(
-            {"keys": json, "device_display_name": "display_name"}, dev
+            {"key": "value", "unsigned": {"device_display_name": "display_name"}}, dev
         )
 
     @defer.inlineCallbacks
@@ -80,10 +80,10 @@ class EndToEndKeyStoreTestCase(tests.unittest.TestCase):
         yield self.store.store_device("user2", "device1", None)
         yield self.store.store_device("user2", "device2", None)
 
-        yield self.store.set_e2e_device_keys("user1", "device1", now, "json11")
-        yield self.store.set_e2e_device_keys("user1", "device2", now, "json12")
-        yield self.store.set_e2e_device_keys("user2", "device1", now, "json21")
-        yield self.store.set_e2e_device_keys("user2", "device2", now, "json22")
+        yield self.store.set_e2e_device_keys("user1", "device1", now, {"key": "json11"})
+        yield self.store.set_e2e_device_keys("user1", "device2", now, {"key": "json12"})
+        yield self.store.set_e2e_device_keys("user2", "device1", now, {"key": "json21"})
+        yield self.store.set_e2e_device_keys("user2", "device2", now, {"key": "json22"})
 
         res = yield self.store.get_e2e_device_keys(
             (("user1", "device1"), ("user2", "device2"))
diff --git a/tests/storage/test_event_federation.py b/tests/storage/test_event_federation.py
index 86c7ac350d..2fe50377f8 100644
--- a/tests/storage/test_event_federation.py
+++ b/tests/storage/test_event_federation.py
@@ -57,7 +57,7 @@ class EventFederationWorkerStoreTestCase(tests.unittest.TestCase):
                     "(event_id, algorithm, hash) "
                     "VALUES (?, 'sha256', ?)"
                 ),
-                (event_id, b"ffff"),
+                (event_id, bytearray(b"ffff")),
             )
 
         for i in range(0, 11):
@@ -75,3 +75,43 @@ class EventFederationWorkerStoreTestCase(tests.unittest.TestCase):
             el = r[i]
             depth = el[2]
             self.assertLessEqual(5, depth)
+
+    @defer.inlineCallbacks
+    def test_get_rooms_with_many_extremities(self):
+        room1 = "#room1"
+        room2 = "#room2"
+        room3 = "#room3"
+
+        def insert_event(txn, i, room_id):
+            event_id = "$event_%i:local" % i
+            txn.execute(
+                (
+                    "INSERT INTO event_forward_extremities (room_id, event_id) "
+                    "VALUES (?, ?)"
+                ),
+                (room_id, event_id),
+            )
+
+        for i in range(0, 20):
+            yield self.store.runInteraction("insert", insert_event, i, room1)
+            yield self.store.runInteraction("insert", insert_event, i, room2)
+            yield self.store.runInteraction("insert", insert_event, i, room3)
+
+        # Test simple case
+        r = yield self.store.get_rooms_with_many_extremities(5, 5, [])
+        self.assertEqual(len(r), 3)
+
+        # Does filter work?
+
+        r = yield self.store.get_rooms_with_many_extremities(5, 5, [room1])
+        self.assertTrue(room2 in r)
+        self.assertTrue(room3 in r)
+        self.assertEqual(len(r), 2)
+
+        r = yield self.store.get_rooms_with_many_extremities(5, 5, [room1, room2])
+        self.assertEqual(r, [room3])
+
+        # Does filter and limit work?
+
+        r = yield self.store.get_rooms_with_many_extremities(5, 1, [room1])
+        self.assertTrue(r == [room2] or r == [room3])
diff --git a/tests/storage/test_monthly_active_users.py b/tests/storage/test_monthly_active_users.py
index 1494650d10..90a63dc477 100644
--- a/tests/storage/test_monthly_active_users.py
+++ b/tests/storage/test_monthly_active_users.py
@@ -50,6 +50,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
             {"medium": "email", "address": user2_email},
             {"medium": "email", "address": user3_email},
         ]
+        self.hs.config.mau_limits_reserved_threepids = threepids
         # -1 because user3 is a support user and does not count
         user_num = len(threepids) - 1
 
@@ -84,6 +85,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         self.hs.config.max_mau_value = 0
 
         self.reactor.advance(FORTY_DAYS)
+        self.hs.config.max_mau_value = 5
 
         self.store.reap_monthly_active_users()
         self.pump()
@@ -147,9 +149,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         self.store.reap_monthly_active_users()
         self.pump()
         count = self.store.get_monthly_active_count()
-        self.assertEquals(
-            self.get_success(count), initial_users - self.hs.config.max_mau_value
-        )
+        self.assertEquals(self.get_success(count), self.hs.config.max_mau_value)
 
         self.reactor.advance(FORTY_DAYS)
         self.store.reap_monthly_active_users()
@@ -158,6 +158,44 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         count = self.store.get_monthly_active_count()
         self.assertEquals(self.get_success(count), 0)
 
+    def test_reap_monthly_active_users_reserved_users(self):
+        """ Tests that reaping correctly handles reaping where reserved users are
+        present"""
+
+        self.hs.config.max_mau_value = 5
+        initial_users = 5
+        reserved_user_number = initial_users - 1
+        threepids = []
+        for i in range(initial_users):
+            user = "@user%d:server" % i
+            email = "user%d@example.com" % i
+            self.get_success(self.store.upsert_monthly_active_user(user))
+            threepids.append({"medium": "email", "address": email})
+            # Need to ensure that the most recent entries in the
+            # monthly_active_users table are reserved
+            now = int(self.hs.get_clock().time_msec())
+            if i != 0:
+                self.get_success(
+                    self.store.register_user(user_id=user, password_hash=None)
+                )
+                self.get_success(
+                    self.store.user_add_threepid(user, "email", email, now, now)
+                )
+
+        self.hs.config.mau_limits_reserved_threepids = threepids
+        self.store.runInteraction(
+            "initialise", self.store._initialise_reserved_users, threepids
+        )
+        count = self.store.get_monthly_active_count()
+        self.assertTrue(self.get_success(count), initial_users)
+
+        users = self.store.get_registered_reserved_users()
+        self.assertEquals(len(self.get_success(users)), reserved_user_number)
+
+        self.get_success(self.store.reap_monthly_active_users())
+        count = self.store.get_monthly_active_count()
+        self.assertEquals(self.get_success(count), self.hs.config.max_mau_value)
+
     def test_populate_monthly_users_is_guest(self):
         # Test that guest users are not added to mau list
         user_id = "@user_id:host"
@@ -192,12 +230,13 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
 
     def test_get_reserved_real_user_account(self):
         # Test no reserved users, or reserved threepids
-        count = self.store.get_registered_reserved_users_count()
-        self.assertEquals(self.get_success(count), 0)
+        users = self.get_success(self.store.get_registered_reserved_users())
+        self.assertEquals(len(users), 0)
         # Test reserved users but no registered users
 
         user1 = "@user1:example.com"
         user2 = "@user2:example.com"
+
         user1_email = "user1@example.com"
         user2_email = "user2@example.com"
         threepids = [
@@ -210,8 +249,8 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         )
 
         self.pump()
-        count = self.store.get_registered_reserved_users_count()
-        self.assertEquals(self.get_success(count), 0)
+        users = self.get_success(self.store.get_registered_reserved_users())
+        self.assertEquals(len(users), 0)
 
         # Test reserved registed users
         self.store.register_user(user_id=user1, password_hash=None)
@@ -221,8 +260,9 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         now = int(self.hs.get_clock().time_msec())
         self.store.user_add_threepid(user1, "email", user1_email, now, now)
         self.store.user_add_threepid(user2, "email", user2_email, now, now)
-        count = self.store.get_registered_reserved_users_count()
-        self.assertEquals(self.get_success(count), len(threepids))
+
+        users = self.get_success(self.store.get_registered_reserved_users())
+        self.assertEquals(len(users), len(threepids))
 
     def test_support_user_not_add_to_mau_limits(self):
         support_user_id = "@support:test"
diff --git a/tests/storage/test_profile.py b/tests/storage/test_profile.py
index 45824bd3b2..24c7fe16c3 100644
--- a/tests/storage/test_profile.py
+++ b/tests/storage/test_profile.py
@@ -16,7 +16,7 @@
 
 from twisted.internet import defer
 
-from synapse.storage.profile import ProfileStore
+from synapse.storage.data_stores.main.profile import ProfileStore
 from synapse.types import UserID
 
 from tests import unittest
diff --git a/tests/storage/test_redaction.py b/tests/storage/test_redaction.py
index d961b81d48..4561c3e383 100644
--- a/tests/storage/test_redaction.py
+++ b/tests/storage/test_redaction.py
@@ -17,6 +17,8 @@
 
 from mock import Mock
 
+from canonicaljson import json
+
 from twisted.internet import defer
 
 from synapse.api.constants import EventTypes, Membership
@@ -29,12 +31,15 @@ from tests.utils import create_room
 
 class RedactionTestCase(unittest.HomeserverTestCase):
     def make_homeserver(self, reactor, clock):
+        config = self.default_config()
+        config["redaction_retention_period"] = "30d"
         return self.setup_test_homeserver(
-            resource_for_federation=Mock(), http_client=None
+            resource_for_federation=Mock(), http_client=None, config=config
         )
 
     def prepare(self, reactor, clock, hs):
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
         self.event_builder_factory = hs.get_event_builder_factory()
         self.event_creation_handler = hs.get_event_creation_handler()
 
@@ -69,7 +74,7 @@ class RedactionTestCase(unittest.HomeserverTestCase):
             self.event_creation_handler.create_new_client_event(builder)
         )
 
-        self.get_success(self.store.persist_event(event, context))
+        self.get_success(self.storage.persistence.persist_event(event, context))
 
         return event
 
@@ -91,7 +96,7 @@ class RedactionTestCase(unittest.HomeserverTestCase):
             self.event_creation_handler.create_new_client_event(builder)
         )
 
-        self.get_success(self.store.persist_event(event, context))
+        self.get_success(self.storage.persistence.persist_event(event, context))
 
         return event
 
@@ -112,7 +117,9 @@ class RedactionTestCase(unittest.HomeserverTestCase):
             self.event_creation_handler.create_new_client_event(builder)
         )
 
-        self.get_success(self.store.persist_event(event, context))
+        self.get_success(self.storage.persistence.persist_event(event, context))
+
+        return event
 
     def test_redact(self):
         self.get_success(
@@ -257,7 +264,7 @@ class RedactionTestCase(unittest.HomeserverTestCase):
             )
         )
 
-        self.get_success(self.store.persist_event(event_1, context_1))
+        self.get_success(self.storage.persistence.persist_event(event_1, context_1))
 
         event_2, context_2 = self.get_success(
             self.event_creation_handler.create_new_client_event(
@@ -276,7 +283,7 @@ class RedactionTestCase(unittest.HomeserverTestCase):
                 )
             )
         )
-        self.get_success(self.store.persist_event(event_2, context_2))
+        self.get_success(self.storage.persistence.persist_event(event_2, context_2))
 
         # fetch one of the redactions
         fetched = self.get_success(self.store.get_event(redaction_event_id1))
@@ -286,3 +293,108 @@ class RedactionTestCase(unittest.HomeserverTestCase):
         self.assertEqual(
             fetched.unsigned["redacted_because"].event_id, redaction_event_id2
         )
+
+    def test_redact_censor(self):
+        """Test that a redacted event gets censored in the DB after a month
+        """
+
+        self.get_success(
+            self.inject_room_member(self.room1, self.u_alice, Membership.JOIN)
+        )
+
+        msg_event = self.get_success(self.inject_message(self.room1, self.u_alice, "t"))
+
+        # Check event has not been redacted:
+        event = self.get_success(self.store.get_event(msg_event.event_id))
+
+        self.assertObjectHasAttributes(
+            {
+                "type": EventTypes.Message,
+                "user_id": self.u_alice.to_string(),
+                "content": {"body": "t", "msgtype": "message"},
+            },
+            event,
+        )
+
+        self.assertFalse("redacted_because" in event.unsigned)
+
+        # Redact event
+        reason = "Because I said so"
+        self.get_success(
+            self.inject_redaction(self.room1, msg_event.event_id, self.u_alice, reason)
+        )
+
+        event = self.get_success(self.store.get_event(msg_event.event_id))
+
+        self.assertTrue("redacted_because" in event.unsigned)
+
+        self.assertObjectHasAttributes(
+            {
+                "type": EventTypes.Message,
+                "user_id": self.u_alice.to_string(),
+                "content": {},
+            },
+            event,
+        )
+
+        event_json = self.get_success(
+            self.store._simple_select_one_onecol(
+                table="event_json",
+                keyvalues={"event_id": msg_event.event_id},
+                retcol="json",
+            )
+        )
+
+        self.assert_dict(
+            {"content": {"body": "t", "msgtype": "message"}}, json.loads(event_json)
+        )
+
+        # Advance by 30 days, then advance again to ensure that the looping call
+        # for updating the stream position gets called and then the looping call
+        # for the censoring gets called.
+        self.reactor.advance(60 * 60 * 24 * 31)
+        self.reactor.advance(60 * 60 * 2)
+
+        event_json = self.get_success(
+            self.store._simple_select_one_onecol(
+                table="event_json",
+                keyvalues={"event_id": msg_event.event_id},
+                retcol="json",
+            )
+        )
+
+        self.assert_dict({"content": {}}, json.loads(event_json))
+
+    def test_redact_redaction(self):
+        """Tests that we can redact a redaction and can fetch it again.
+        """
+
+        self.get_success(
+            self.inject_room_member(self.room1, self.u_alice, Membership.JOIN)
+        )
+
+        msg_event = self.get_success(self.inject_message(self.room1, self.u_alice, "t"))
+
+        first_redact_event = self.get_success(
+            self.inject_redaction(
+                self.room1, msg_event.event_id, self.u_alice, "Redacting message"
+            )
+        )
+
+        self.get_success(
+            self.inject_redaction(
+                self.room1,
+                first_redact_event.event_id,
+                self.u_alice,
+                "Redacting redaction",
+            )
+        )
+
+        # Now lets jump to the future where we have censored the redaction event
+        # in the DB.
+        self.reactor.advance(60 * 60 * 24 * 31)
+
+        # We just want to check that fetching the event doesn't raise an exception.
+        self.get_success(
+            self.store.get_event(first_redact_event.event_id, allow_none=True)
+        )
diff --git a/tests/storage/test_room.py b/tests/storage/test_room.py
index 1bee45706f..3ddaa151fe 100644
--- a/tests/storage/test_room.py
+++ b/tests/storage/test_room.py
@@ -62,6 +62,7 @@ class RoomEventsStoreTestCase(unittest.TestCase):
         # Room events need the full datastore, for persist_event() and
         # get_room_state()
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
         self.event_factory = hs.get_event_factory()
 
         self.room = RoomID.from_string("!abcde:test")
@@ -72,7 +73,7 @@ class RoomEventsStoreTestCase(unittest.TestCase):
 
     @defer.inlineCallbacks
     def inject_room_event(self, **kwargs):
-        yield self.store.persist_event(
+        yield self.storage.persistence.persist_event(
             self.event_factory.create_event(room_id=self.room.to_string(), **kwargs)
         )
 
diff --git a/tests/storage/test_roommember.py b/tests/storage/test_roommember.py
index 64cb294c37..9ddd17f73d 100644
--- a/tests/storage/test_roommember.py
+++ b/tests/storage/test_roommember.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,78 +14,130 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-
-from mock import Mock
-
-from twisted.internet import defer
+from unittest.mock import Mock
 
 from synapse.api.constants import EventTypes, Membership
 from synapse.api.room_versions import RoomVersions
-from synapse.types import Requester, RoomID, UserID
+from synapse.rest.admin import register_servlets_for_client_rest_resource
+from synapse.rest.client.v1 import login, room
+from synapse.types import Requester, UserID
 
 from tests import unittest
-from tests.utils import create_room, setup_test_homeserver
 
 
-class RoomMemberStoreTestCase(unittest.TestCase):
-    @defer.inlineCallbacks
-    def setUp(self):
-        hs = yield setup_test_homeserver(
-            self.addCleanup, resource_for_federation=Mock(), http_client=None
+class RoomMemberStoreTestCase(unittest.HomeserverTestCase):
+
+    servlets = [
+        login.register_servlets,
+        register_servlets_for_client_rest_resource,
+        room.register_servlets,
+    ]
+
+    def make_homeserver(self, reactor, clock):
+        hs = self.setup_test_homeserver(
+            resource_for_federation=Mock(), http_client=None
         )
+        return hs
+
+    def prepare(self, reactor, clock, hs):
+
         # We can't test the RoomMemberStore on its own without the other event
         # storage logic
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
         self.event_builder_factory = hs.get_event_builder_factory()
         self.event_creation_handler = hs.get_event_creation_handler()
 
-        self.u_alice = UserID.from_string("@alice:test")
-        self.u_bob = UserID.from_string("@bob:test")
+        self.u_alice = self.register_user("alice", "pass")
+        self.t_alice = self.login("alice", "pass")
+        self.u_bob = self.register_user("bob", "pass")
 
         # User elsewhere on another host
         self.u_charlie = UserID.from_string("@charlie:elsewhere")
 
-        self.room = RoomID.from_string("!abc123:test")
-
-        yield create_room(hs, self.room.to_string(), self.u_alice.to_string())
-
-    @defer.inlineCallbacks
     def inject_room_member(self, room, user, membership, replaces_state=None):
         builder = self.event_builder_factory.for_room_version(
             RoomVersions.V1,
             {
                 "type": EventTypes.Member,
-                "sender": user.to_string(),
-                "state_key": user.to_string(),
-                "room_id": room.to_string(),
+                "sender": user,
+                "state_key": user,
+                "room_id": room,
                 "content": {"membership": membership},
             },
         )
 
-        event, context = yield self.event_creation_handler.create_new_client_event(
-            builder
+        event, context = self.get_success(
+            self.event_creation_handler.create_new_client_event(builder)
         )
 
-        yield self.store.persist_event(event, context)
+        self.get_success(self.storage.persistence.persist_event(event, context))
 
         return event
 
-    @defer.inlineCallbacks
     def test_one_member(self):
-        yield self.inject_room_member(self.room, self.u_alice, Membership.JOIN)
 
-        self.assertEquals(
-            [self.room.to_string()],
-            [
-                m.room_id
-                for m in (
-                    yield self.store.get_rooms_for_user_where_membership_is(
-                        self.u_alice.to_string(), [Membership.JOIN]
-                    )
-                )
-            ],
+        # Alice creates the room, and is automatically joined
+        self.room = self.helper.create_room_as(self.u_alice, tok=self.t_alice)
+
+        rooms_for_user = self.get_success(
+            self.store.get_rooms_for_user_where_membership_is(
+                self.u_alice, [Membership.JOIN]
+            )
         )
 
+        self.assertEquals([self.room], [m.room_id for m in rooms_for_user])
+
+    def test_count_known_servers(self):
+        """
+        _count_known_servers will calculate how many servers are in a room.
+        """
+        self.room = self.helper.create_room_as(self.u_alice, tok=self.t_alice)
+        self.inject_room_member(self.room, self.u_bob, Membership.JOIN)
+        self.inject_room_member(self.room, self.u_charlie.to_string(), Membership.JOIN)
+
+        servers = self.get_success(self.store._count_known_servers())
+        self.assertEqual(servers, 2)
+
+    def test_count_known_servers_stat_counter_disabled(self):
+        """
+        If enabled, the metrics for how many servers are known will be counted.
+        """
+        self.assertTrue("_known_servers_count" not in self.store.__dict__.keys())
+
+        self.room = self.helper.create_room_as(self.u_alice, tok=self.t_alice)
+        self.inject_room_member(self.room, self.u_bob, Membership.JOIN)
+        self.inject_room_member(self.room, self.u_charlie.to_string(), Membership.JOIN)
+
+        self.pump(20)
+
+        self.assertTrue("_known_servers_count" not in self.store.__dict__.keys())
+
+    @unittest.override_config(
+        {"enable_metrics": True, "metrics_flags": {"known_servers": True}}
+    )
+    def test_count_known_servers_stat_counter_enabled(self):
+        """
+        If enabled, the metrics for how many servers are known will be counted.
+        """
+        # Initialises to 1 -- itself
+        self.assertEqual(self.store._known_servers_count, 1)
+
+        self.pump(20)
+
+        # No rooms have been joined, so technically the SQL returns 0, but it
+        # will still say it knows about itself.
+        self.assertEqual(self.store._known_servers_count, 1)
+
+        self.room = self.helper.create_room_as(self.u_alice, tok=self.t_alice)
+        self.inject_room_member(self.room, self.u_bob, Membership.JOIN)
+        self.inject_room_member(self.room, self.u_charlie.to_string(), Membership.JOIN)
+
+        self.pump(20)
+
+        # It now knows about Charlie's server.
+        self.assertEqual(self.store._known_servers_count, 2)
+
 
 class CurrentStateMembershipUpdateTestCase(unittest.HomeserverTestCase):
     def prepare(self, reactor, clock, homeserver):
diff --git a/tests/storage/test_state.py b/tests/storage/test_state.py
index 5c2cf3c2db..43200654f1 100644
--- a/tests/storage/test_state.py
+++ b/tests/storage/test_state.py
@@ -34,6 +34,8 @@ class StateStoreTestCase(tests.unittest.TestCase):
         hs = yield tests.utils.setup_test_homeserver(self.addCleanup)
 
         self.store = hs.get_datastore()
+        self.storage = hs.get_storage()
+        self.state_datastore = self.store
         self.event_builder_factory = hs.get_event_builder_factory()
         self.event_creation_handler = hs.get_event_creation_handler()
 
@@ -63,7 +65,7 @@ class StateStoreTestCase(tests.unittest.TestCase):
             builder
         )
 
-        yield self.store.persist_event(event, context)
+        yield self.storage.persistence.persist_event(event, context)
 
         return event
 
@@ -82,7 +84,7 @@ class StateStoreTestCase(tests.unittest.TestCase):
             self.room, self.u_alice, EventTypes.Name, "", {"name": "test room"}
         )
 
-        state_group_map = yield self.store.get_state_groups_ids(
+        state_group_map = yield self.storage.state.get_state_groups_ids(
             self.room, [e2.event_id]
         )
         self.assertEqual(len(state_group_map), 1)
@@ -101,7 +103,9 @@ class StateStoreTestCase(tests.unittest.TestCase):
             self.room, self.u_alice, EventTypes.Name, "", {"name": "test room"}
         )
 
-        state_group_map = yield self.store.get_state_groups(self.room, [e2.event_id])
+        state_group_map = yield self.storage.state.get_state_groups(
+            self.room, [e2.event_id]
+        )
         self.assertEqual(len(state_group_map), 1)
         state_list = list(state_group_map.values())[0]
 
@@ -141,7 +145,7 @@ class StateStoreTestCase(tests.unittest.TestCase):
         )
 
         # check we get the full state as of the final event
-        state = yield self.store.get_state_for_event(e5.event_id)
+        state = yield self.storage.state.get_state_for_event(e5.event_id)
 
         self.assertIsNotNone(e4)
 
@@ -157,21 +161,21 @@ class StateStoreTestCase(tests.unittest.TestCase):
         )
 
         # check we can filter to the m.room.name event (with a '' state key)
-        state = yield self.store.get_state_for_event(
+        state = yield self.storage.state.get_state_for_event(
             e5.event_id, StateFilter.from_types([(EventTypes.Name, "")])
         )
 
         self.assertStateMapEqual({(e2.type, e2.state_key): e2}, state)
 
         # check we can filter to the m.room.name event (with a wildcard None state key)
-        state = yield self.store.get_state_for_event(
+        state = yield self.storage.state.get_state_for_event(
             e5.event_id, StateFilter.from_types([(EventTypes.Name, None)])
         )
 
         self.assertStateMapEqual({(e2.type, e2.state_key): e2}, state)
 
         # check we can grab the m.room.member events (with a wildcard None state key)
-        state = yield self.store.get_state_for_event(
+        state = yield self.storage.state.get_state_for_event(
             e5.event_id, StateFilter.from_types([(EventTypes.Member, None)])
         )
 
@@ -181,7 +185,7 @@ class StateStoreTestCase(tests.unittest.TestCase):
 
         # check we can grab a specific room member without filtering out the
         # other event types
-        state = yield self.store.get_state_for_event(
+        state = yield self.storage.state.get_state_for_event(
             e5.event_id,
             state_filter=StateFilter(
                 types={EventTypes.Member: {self.u_alice.to_string()}},
@@ -199,7 +203,7 @@ class StateStoreTestCase(tests.unittest.TestCase):
         )
 
         # check that we can grab everything except members
-        state = yield self.store.get_state_for_event(
+        state = yield self.storage.state.get_state_for_event(
             e5.event_id,
             state_filter=StateFilter(
                 types={EventTypes.Member: set()}, include_others=True
@@ -215,13 +219,18 @@ class StateStoreTestCase(tests.unittest.TestCase):
         #######################################################
 
         room_id = self.room.to_string()
-        group_ids = yield self.store.get_state_groups_ids(room_id, [e5.event_id])
+        group_ids = yield self.storage.state.get_state_groups_ids(
+            room_id, [e5.event_id]
+        )
         group = list(group_ids.keys())[0]
 
         # test _get_state_for_group_using_cache correctly filters out members
         # with types=[]
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: set()}, include_others=True
@@ -237,8 +246,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
             state_dict,
         )
 
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: set()}, include_others=True
@@ -250,8 +262,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
 
         # test _get_state_for_group_using_cache correctly filters in members
         # with wildcard types
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: None}, include_others=True
@@ -267,8 +282,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
             state_dict,
         )
 
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: None}, include_others=True
@@ -287,8 +305,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
 
         # test _get_state_for_group_using_cache correctly filters in members
         # with specific types
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: {e5.state_key}}, include_others=True
@@ -304,8 +325,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
             state_dict,
         )
 
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: {e5.state_key}}, include_others=True
@@ -317,8 +341,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
 
         # test _get_state_for_group_using_cache correctly filters in members
         # with specific types
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: {e5.state_key}}, include_others=False
@@ -331,9 +358,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
         #######################################################
         # deliberately remove e2 (room name) from the _state_group_cache
 
-        (is_all, known_absent, state_dict_ids) = self.store._state_group_cache.get(
-            group
-        )
+        (
+            is_all,
+            known_absent,
+            state_dict_ids,
+        ) = self.state_datastore._state_group_cache.get(group)
 
         self.assertEqual(is_all, True)
         self.assertEqual(known_absent, set())
@@ -346,18 +375,20 @@ class StateStoreTestCase(tests.unittest.TestCase):
         )
 
         state_dict_ids.pop((e2.type, e2.state_key))
-        self.store._state_group_cache.invalidate(group)
-        self.store._state_group_cache.update(
-            sequence=self.store._state_group_cache.sequence,
+        self.state_datastore._state_group_cache.invalidate(group)
+        self.state_datastore._state_group_cache.update(
+            sequence=self.state_datastore._state_group_cache.sequence,
             key=group,
             value=state_dict_ids,
             # list fetched keys so it knows it's partial
             fetched_keys=((e1.type, e1.state_key),),
         )
 
-        (is_all, known_absent, state_dict_ids) = self.store._state_group_cache.get(
-            group
-        )
+        (
+            is_all,
+            known_absent,
+            state_dict_ids,
+        ) = self.state_datastore._state_group_cache.get(group)
 
         self.assertEqual(is_all, False)
         self.assertEqual(known_absent, set([(e1.type, e1.state_key)]))
@@ -369,8 +400,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
         # test _get_state_for_group_using_cache correctly filters out members
         # with types=[]
         room_id = self.room.to_string()
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: set()}, include_others=True
@@ -381,8 +415,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
         self.assertDictEqual({(e1.type, e1.state_key): e1.event_id}, state_dict)
 
         room_id = self.room.to_string()
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: set()}, include_others=True
@@ -394,8 +431,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
 
         # test _get_state_for_group_using_cache correctly filters in members
         # wildcard types
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: None}, include_others=True
@@ -405,8 +445,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
         self.assertEqual(is_all, False)
         self.assertDictEqual({(e1.type, e1.state_key): e1.event_id}, state_dict)
 
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: None}, include_others=True
@@ -424,8 +467,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
 
         # test _get_state_for_group_using_cache correctly filters in members
         # with specific types
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: {e5.state_key}}, include_others=True
@@ -435,8 +481,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
         self.assertEqual(is_all, False)
         self.assertDictEqual({(e1.type, e1.state_key): e1.event_id}, state_dict)
 
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: {e5.state_key}}, include_others=True
@@ -448,8 +497,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
 
         # test _get_state_for_group_using_cache correctly filters in members
         # with specific types
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: {e5.state_key}}, include_others=False
@@ -459,8 +511,11 @@ class StateStoreTestCase(tests.unittest.TestCase):
         self.assertEqual(is_all, False)
         self.assertDictEqual({}, state_dict)
 
-        (state_dict, is_all) = yield self.store._get_state_for_group_using_cache(
-            self.store._state_group_members_cache,
+        (
+            state_dict,
+            is_all,
+        ) = yield self.state_datastore._get_state_for_group_using_cache(
+            self.state_datastore._state_group_members_cache,
             group,
             state_filter=StateFilter(
                 types={EventTypes.Member: {e5.state_key}}, include_others=False
diff --git a/tests/storage/test_transactions.py b/tests/storage/test_transactions.py
index 14169afa96..8e817e2c7f 100644
--- a/tests/storage/test_transactions.py
+++ b/tests/storage/test_transactions.py
@@ -13,6 +13,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from synapse.util.retryutils import MAX_RETRY_INTERVAL
+
 from tests.unittest import HomeserverTestCase
 
 
@@ -29,17 +31,28 @@ class TransactionStoreTestCase(HomeserverTestCase):
         r = self.get_success(d)
         self.assertIsNone(r)
 
-        d = self.store.set_destination_retry_timings("example.com", 50, 100)
+        d = self.store.set_destination_retry_timings("example.com", 1000, 50, 100)
         self.get_success(d)
 
         d = self.store.get_destination_retry_timings("example.com")
         r = self.get_success(d)
 
-        self.assert_dict({"retry_last_ts": 50, "retry_interval": 100}, r)
+        self.assert_dict(
+            {"retry_last_ts": 50, "retry_interval": 100, "failure_ts": 1000}, r
+        )
 
     def test_initial_set_transactions(self):
         """Tests that we can successfully set the destination retries (there
         was a bug around invalidating the cache that broke this)
         """
-        d = self.store.set_destination_retry_timings("example.com", 50, 100)
+        d = self.store.set_destination_retry_timings("example.com", 1000, 50, 100)
+        self.get_success(d)
+
+    def test_large_destination_retry(self):
+        d = self.store.set_destination_retry_timings(
+            "example.com", MAX_RETRY_INTERVAL, MAX_RETRY_INTERVAL, MAX_RETRY_INTERVAL
+        )
+        self.get_success(d)
+
+        d = self.store.get_destination_retry_timings("example.com")
         self.get_success(d)
diff --git a/tests/storage/test_user_directory.py b/tests/storage/test_user_directory.py
index d7d244ce97..7eea57c0e2 100644
--- a/tests/storage/test_user_directory.py
+++ b/tests/storage/test_user_directory.py
@@ -15,7 +15,7 @@
 
 from twisted.internet import defer
 
-from synapse.storage import UserDirectoryStore
+from synapse.storage.data_stores.main.user_directory import UserDirectoryStore
 
 from tests import unittest
 from tests.utils import setup_test_homeserver
diff --git a/tests/test_federation.py b/tests/test_federation.py
index a73f18f88e..7d82b58466 100644
--- a/tests/test_federation.py
+++ b/tests/test_federation.py
@@ -36,7 +36,8 @@ class MessageAcceptTests(unittest.TestCase):
         # Figure out what the most recent event is
         most_recent = self.successResultOf(
             maybeDeferred(
-                self.homeserver.datastore.get_latest_event_ids_in_room, self.room_id
+                self.homeserver.get_datastore().get_latest_event_ids_in_room,
+                self.room_id,
             )
         )[0]
 
@@ -58,7 +59,9 @@ class MessageAcceptTests(unittest.TestCase):
         )
 
         self.handler = self.homeserver.get_handlers().federation_handler
-        self.handler.do_auth = lambda *a, **b: succeed(True)
+        self.handler.do_auth = lambda origin, event, context, auth_events: succeed(
+            context
+        )
         self.client = self.homeserver.get_federation_client()
         self.client._check_sigs_and_hash_and_fetch = lambda dest, pdus, **k: succeed(
             pdus
@@ -75,7 +78,8 @@ class MessageAcceptTests(unittest.TestCase):
         self.assertEqual(
             self.successResultOf(
                 maybeDeferred(
-                    self.homeserver.datastore.get_latest_event_ids_in_room, self.room_id
+                    self.homeserver.get_datastore().get_latest_event_ids_in_room,
+                    self.room_id,
                 )
             )[0],
             "$join:test.serv",
@@ -97,7 +101,8 @@ class MessageAcceptTests(unittest.TestCase):
         # Figure out what the most recent event is
         most_recent = self.successResultOf(
             maybeDeferred(
-                self.homeserver.datastore.get_latest_event_ids_in_room, self.room_id
+                self.homeserver.get_datastore().get_latest_event_ids_in_room,
+                self.room_id,
             )
         )[0]
 
@@ -137,6 +142,6 @@ class MessageAcceptTests(unittest.TestCase):
 
         # Make sure the invalid event isn't there
         extrem = maybeDeferred(
-            self.homeserver.datastore.get_latest_event_ids_in_room, self.room_id
+            self.homeserver.get_datastore().get_latest_event_ids_in_room, self.room_id
         )
         self.assertEqual(self.successResultOf(extrem)[0], "$join:test.serv")
diff --git a/tests/test_metrics.py b/tests/test_metrics.py
index 2edbae5c6d..270f853d60 100644
--- a/tests/test_metrics.py
+++ b/tests/test_metrics.py
@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2018 New Vector Ltd
+# Copyright 2019 Matrix.org Foundation C.I.C.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,8 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-
-from synapse.metrics import InFlightGauge
+from synapse.metrics import REGISTRY, InFlightGauge, generate_latest
 
 from tests import unittest
 
@@ -111,3 +111,21 @@ class TestMauLimit(unittest.TestCase):
             }
 
         return results
+
+
+class BuildInfoTests(unittest.TestCase):
+    def test_get_build(self):
+        """
+        The synapse_build_info metric reports the OS version, Python version,
+        and Synapse version.
+        """
+        items = list(
+            filter(
+                lambda x: b"synapse_build_info{" in x,
+                generate_latest(REGISTRY).split(b"\n"),
+            )
+        )
+        self.assertEqual(len(items), 1)
+        self.assertTrue(b"osversion=" in items[0])
+        self.assertTrue(b"pythonversion=" in items[0])
+        self.assertTrue(b"version=" in items[0])
diff --git a/tests/test_phone_home.py b/tests/test_phone_home.py
new file mode 100644
index 0000000000..7657bddea5
--- /dev/null
+++ b/tests/test_phone_home.py
@@ -0,0 +1,51 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import resource
+
+import mock
+
+from synapse.app.homeserver import phone_stats_home
+
+from tests.unittest import HomeserverTestCase
+
+
+class PhoneHomeStatsTestCase(HomeserverTestCase):
+    def test_performance_frozen_clock(self):
+        """
+        If time doesn't move, don't error out.
+        """
+        past_stats = [
+            (self.hs.get_clock().time(), resource.getrusage(resource.RUSAGE_SELF))
+        ]
+        stats = {}
+        self.get_success(phone_stats_home(self.hs, stats, past_stats))
+        self.assertEqual(stats["cpu_average"], 0)
+
+    def test_performance_100(self):
+        """
+        1 second of usage over 1 second is 100% CPU usage.
+        """
+        real_res = resource.getrusage(resource.RUSAGE_SELF)
+        old_resource = mock.Mock(spec=real_res)
+        old_resource.ru_utime = real_res.ru_utime - 1
+        old_resource.ru_stime = real_res.ru_stime
+        old_resource.ru_maxrss = real_res.ru_maxrss
+
+        past_stats = [(self.hs.get_clock().time(), old_resource)]
+        stats = {}
+        self.reactor.advance(1)
+        self.get_success(phone_stats_home(self.hs, stats, past_stats))
+        self.assertApproximates(stats["cpu_average"], 100, tolerance=2.5)
diff --git a/tests/test_state.py b/tests/test_state.py
index 610ec9fb46..176535947a 100644
--- a/tests/test_state.py
+++ b/tests/test_state.py
@@ -21,6 +21,7 @@ from synapse.api.auth import Auth
 from synapse.api.constants import EventTypes, Membership
 from synapse.api.room_versions import RoomVersions
 from synapse.events import FrozenEvent
+from synapse.events.snapshot import EventContext
 from synapse.state import StateHandler, StateResolutionHandler
 
 from tests import unittest
@@ -158,10 +159,12 @@ class Graph(object):
 class StateTestCase(unittest.TestCase):
     def setUp(self):
         self.store = StateGroupStore()
+        storage = Mock(main=self.store, state=self.store)
         hs = Mock(
             spec_set=[
                 "config",
                 "get_datastore",
+                "get_storage",
                 "get_auth",
                 "get_state_handler",
                 "get_clock",
@@ -174,6 +177,7 @@ class StateTestCase(unittest.TestCase):
         hs.get_clock.return_value = MockClock()
         hs.get_auth.return_value = Auth(hs)
         hs.get_state_resolution_handler = lambda: StateResolutionHandler(hs)
+        hs.get_storage.return_value = storage
 
         self.state = StateHandler(hs)
         self.event_id = 0
@@ -195,16 +199,22 @@ class StateTestCase(unittest.TestCase):
 
         self.store.register_events(graph.walk())
 
-        context_store = {}
+        context_store = {}  # type: dict[str, EventContext]
 
         for event in graph.walk():
             context = yield self.state.compute_event_context(event)
             self.store.register_event_context(event, context)
             context_store[event.event_id] = context
 
-        prev_state_ids = yield context_store["D"].get_prev_state_ids(self.store)
+        ctx_c = context_store["C"]
+        ctx_d = context_store["D"]
+
+        prev_state_ids = yield ctx_d.get_prev_state_ids(self.store)
         self.assertEqual(2, len(prev_state_ids))
 
+        self.assertEqual(ctx_c.state_group, ctx_d.state_group_before_event)
+        self.assertEqual(ctx_d.state_group_before_event, ctx_d.state_group)
+
     @defer.inlineCallbacks
     def test_branch_basic_conflict(self):
         graph = Graph(
@@ -238,12 +248,19 @@ class StateTestCase(unittest.TestCase):
             self.store.register_event_context(event, context)
             context_store[event.event_id] = context
 
-        prev_state_ids = yield context_store["D"].get_prev_state_ids(self.store)
+        # C ends up winning the resolution between B and C
 
+        ctx_c = context_store["C"]
+        ctx_d = context_store["D"]
+
+        prev_state_ids = yield ctx_d.get_prev_state_ids(self.store)
         self.assertSetEqual(
             {"START", "A", "C"}, {e_id for e_id in prev_state_ids.values()}
         )
 
+        self.assertEqual(ctx_c.state_group, ctx_d.state_group_before_event)
+        self.assertEqual(ctx_d.state_group_before_event, ctx_d.state_group)
+
     @defer.inlineCallbacks
     def test_branch_have_banned_conflict(self):
         graph = Graph(
@@ -289,11 +306,18 @@ class StateTestCase(unittest.TestCase):
             self.store.register_event_context(event, context)
             context_store[event.event_id] = context
 
-        prev_state_ids = yield context_store["E"].get_prev_state_ids(self.store)
+        # C ends up winning the resolution between C and D because bans win over other
+        # changes
 
+        ctx_c = context_store["C"]
+        ctx_e = context_store["E"]
+
+        prev_state_ids = yield ctx_e.get_prev_state_ids(self.store)
         self.assertSetEqual(
             {"START", "A", "B", "C"}, {e for e in prev_state_ids.values()}
         )
+        self.assertEqual(ctx_c.state_group, ctx_e.state_group_before_event)
+        self.assertEqual(ctx_e.state_group_before_event, ctx_e.state_group)
 
     @defer.inlineCallbacks
     def test_branch_have_perms_conflict(self):
@@ -357,12 +381,20 @@ class StateTestCase(unittest.TestCase):
             self.store.register_event_context(event, context)
             context_store[event.event_id] = context
 
-        prev_state_ids = yield context_store["D"].get_prev_state_ids(self.store)
+        # B ends up winning the resolution between B and C because power levels
+        # win over other changes.
 
+        ctx_b = context_store["B"]
+        ctx_d = context_store["D"]
+
+        prev_state_ids = yield ctx_d.get_prev_state_ids(self.store)
         self.assertSetEqual(
             {"A1", "A2", "A3", "A5", "B"}, {e for e in prev_state_ids.values()}
         )
 
+        self.assertEqual(ctx_b.state_group, ctx_d.state_group_before_event)
+        self.assertEqual(ctx_d.state_group_before_event, ctx_d.state_group)
+
     def _add_depths(self, nodes, edges):
         def _get_depth(ev):
             node = nodes[ev]
@@ -387,13 +419,16 @@ class StateTestCase(unittest.TestCase):
 
         context = yield self.state.compute_event_context(event, old_state=old_state)
 
-        current_state_ids = yield context.get_current_state_ids(self.store)
+        prev_state_ids = yield context.get_prev_state_ids(self.store)
+        self.assertCountEqual((e.event_id for e in old_state), prev_state_ids.values())
 
-        self.assertEqual(
-            set(e.event_id for e in old_state), set(current_state_ids.values())
+        current_state_ids = yield context.get_current_state_ids(self.store)
+        self.assertCountEqual(
+            (e.event_id for e in old_state), current_state_ids.values()
         )
 
-        self.assertIsNotNone(context.state_group)
+        self.assertIsNotNone(context.state_group_before_event)
+        self.assertEqual(context.state_group_before_event, context.state_group)
 
     @defer.inlineCallbacks
     def test_annotate_with_old_state(self):
@@ -408,11 +443,18 @@ class StateTestCase(unittest.TestCase):
         context = yield self.state.compute_event_context(event, old_state=old_state)
 
         prev_state_ids = yield context.get_prev_state_ids(self.store)
+        self.assertCountEqual((e.event_id for e in old_state), prev_state_ids.values())
 
-        self.assertEqual(
-            set(e.event_id for e in old_state), set(prev_state_ids.values())
+        current_state_ids = yield context.get_current_state_ids(self.store)
+        self.assertCountEqual(
+            (e.event_id for e in old_state + [event]), current_state_ids.values()
         )
 
+        self.assertIsNotNone(context.state_group_before_event)
+        self.assertNotEqual(context.state_group_before_event, context.state_group)
+        self.assertEqual(context.state_group_before_event, context.prev_group)
+        self.assertEqual({("state", ""): event.event_id}, context.delta_ids)
+
     @defer.inlineCallbacks
     def test_trivial_annotate_message(self):
         prev_event_id = "prev_event_id"
diff --git a/tests/test_terms_auth.py b/tests/test_terms_auth.py
index 52739fbabc..5ec5d2b358 100644
--- a/tests/test_terms_auth.py
+++ b/tests/test_terms_auth.py
@@ -28,6 +28,21 @@ from tests import unittest
 class TermsTestCase(unittest.HomeserverTestCase):
     servlets = [register_servlets]
 
+    def default_config(self, name="test"):
+        config = super().default_config(name)
+        config.update(
+            {
+                "public_baseurl": "https://example.org/",
+                "user_consent": {
+                    "version": "1.0",
+                    "policy_name": "My Cool Privacy Policy",
+                    "template_dir": "/",
+                    "require_at_registration": True,
+                },
+            }
+        )
+        return config
+
     def prepare(self, reactor, clock, hs):
         self.clock = MemoryReactorClock()
         self.hs_clock = Clock(self.clock)
@@ -35,17 +50,8 @@ class TermsTestCase(unittest.HomeserverTestCase):
         self.registration_handler = Mock()
         self.auth_handler = Mock()
         self.device_handler = Mock()
-        hs.config.enable_registration = True
-        hs.config.registrations_require_3pid = []
-        hs.config.auto_join_rooms = []
-        hs.config.enable_registration_captcha = False
 
     def test_ui_auth(self):
-        self.hs.config.user_consent_at_registration = True
-        self.hs.config.user_consent_policy_name = "My Cool Privacy Policy"
-        self.hs.config.public_baseurl = "https://example.org/"
-        self.hs.config.user_consent_version = "1.0"
-
         # Do a UI auth request
         request, channel = self.make_request(b"POST", self.url, b"{}")
         self.render(request)
diff --git a/tests/test_visibility.py b/tests/test_visibility.py
index 18f1a0035d..f7381b2885 100644
--- a/tests/test_visibility.py
+++ b/tests/test_visibility.py
@@ -14,6 +14,8 @@
 # limitations under the License.
 import logging
 
+from mock import Mock
+
 from twisted.internet import defer
 from twisted.internet.defer import succeed
 
@@ -36,6 +38,7 @@ class FilterEventsForServerTestCase(tests.unittest.TestCase):
         self.event_creation_handler = self.hs.get_event_creation_handler()
         self.event_builder_factory = self.hs.get_event_builder_factory()
         self.store = self.hs.get_datastore()
+        self.storage = self.hs.get_storage()
 
         yield create_room(self.hs, TEST_ROOM_ID, "@someone:ROOM")
 
@@ -62,7 +65,7 @@ class FilterEventsForServerTestCase(tests.unittest.TestCase):
             events_to_filter.append(evt)
 
         filtered = yield filter_events_for_server(
-            self.store, "test_server", events_to_filter
+            self.storage, "test_server", events_to_filter
         )
 
         # the result should be 5 redacted events, and 5 unredacted events.
@@ -100,7 +103,7 @@ class FilterEventsForServerTestCase(tests.unittest.TestCase):
 
         # ... and the filtering happens.
         filtered = yield filter_events_for_server(
-            self.store, "test_server", events_to_filter
+            self.storage, "test_server", events_to_filter
         )
 
         for i in range(0, len(events_to_filter)):
@@ -137,7 +140,7 @@ class FilterEventsForServerTestCase(tests.unittest.TestCase):
         event, context = yield self.event_creation_handler.create_new_client_event(
             builder
         )
-        yield self.hs.get_datastore().persist_event(event, context)
+        yield self.storage.persistence.persist_event(event, context)
         return event
 
     @defer.inlineCallbacks
@@ -159,7 +162,7 @@ class FilterEventsForServerTestCase(tests.unittest.TestCase):
             builder
         )
 
-        yield self.hs.get_datastore().persist_event(event, context)
+        yield self.storage.persistence.persist_event(event, context)
         return event
 
     @defer.inlineCallbacks
@@ -180,7 +183,7 @@ class FilterEventsForServerTestCase(tests.unittest.TestCase):
             builder
         )
 
-        yield self.hs.get_datastore().persist_event(event, context)
+        yield self.storage.persistence.persist_event(event, context)
         return event
 
     @defer.inlineCallbacks
@@ -257,6 +260,11 @@ class FilterEventsForServerTestCase(tests.unittest.TestCase):
 
         logger.info("Starting filtering")
         start = time.time()
+
+        storage = Mock()
+        storage.main = test_store
+        storage.state = test_store
+
         filtered = yield filter_events_for_server(
             test_store, "test_server", events_to_filter
         )
diff --git a/tests/util/caches/test_descriptors.py b/tests/util/caches/test_descriptors.py
index 5713870f48..39e360fe24 100644
--- a/tests/util/caches/test_descriptors.py
+++ b/tests/util/caches/test_descriptors.py
@@ -310,14 +310,14 @@ class DescriptorTestCase(unittest.TestCase):
 
         obj.mock.return_value = ["spam", "eggs"]
         r = obj.fn(1, 2)
-        self.assertEqual(r, ["spam", "eggs"])
+        self.assertEqual(r.result, ["spam", "eggs"])
         obj.mock.assert_called_once_with(1, 2)
         obj.mock.reset_mock()
 
         # a call with different params should call the mock again
         obj.mock.return_value = ["chips"]
         r = obj.fn(1, 3)
-        self.assertEqual(r, ["chips"])
+        self.assertEqual(r.result, ["chips"])
         obj.mock.assert_called_once_with(1, 3)
         obj.mock.reset_mock()
 
@@ -325,9 +325,9 @@ class DescriptorTestCase(unittest.TestCase):
         self.assertEqual(len(obj.fn.cache.cache), 3)
 
         r = obj.fn(1, 2)
-        self.assertEqual(r, ["spam", "eggs"])
+        self.assertEqual(r.result, ["spam", "eggs"])
         r = obj.fn(1, 3)
-        self.assertEqual(r, ["chips"])
+        self.assertEqual(r.result, ["chips"])
         obj.mock.assert_not_called()
 
     def test_cache_iterable_with_sync_exception(self):
diff --git a/tests/util/test_retryutils.py b/tests/util/test_retryutils.py
new file mode 100644
index 0000000000..9e348694ad
--- /dev/null
+++ b/tests/util/test_retryutils.py
@@ -0,0 +1,127 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from synapse.util.retryutils import (
+    MIN_RETRY_INTERVAL,
+    RETRY_MULTIPLIER,
+    NotRetryingDestination,
+    get_retry_limiter,
+)
+
+from tests.unittest import HomeserverTestCase
+
+
+class RetryLimiterTestCase(HomeserverTestCase):
+    def test_new_destination(self):
+        """A happy-path case with a new destination and a successful operation"""
+        store = self.hs.get_datastore()
+        d = get_retry_limiter("test_dest", self.clock, store)
+        self.pump()
+        limiter = self.successResultOf(d)
+
+        # advance the clock a bit before making the request
+        self.pump(1)
+
+        with limiter:
+            pass
+
+        d = store.get_destination_retry_timings("test_dest")
+        self.pump()
+        new_timings = self.successResultOf(d)
+        self.assertIsNone(new_timings)
+
+    def test_limiter(self):
+        """General test case which walks through the process of a failing request"""
+        store = self.hs.get_datastore()
+
+        d = get_retry_limiter("test_dest", self.clock, store)
+        self.pump()
+        limiter = self.successResultOf(d)
+
+        self.pump(1)
+        try:
+            with limiter:
+                self.pump(1)
+                failure_ts = self.clock.time_msec()
+                raise AssertionError("argh")
+        except AssertionError:
+            pass
+
+        # wait for the update to land
+        self.pump()
+
+        d = store.get_destination_retry_timings("test_dest")
+        self.pump()
+        new_timings = self.successResultOf(d)
+        self.assertEqual(new_timings["failure_ts"], failure_ts)
+        self.assertEqual(new_timings["retry_last_ts"], failure_ts)
+        self.assertEqual(new_timings["retry_interval"], MIN_RETRY_INTERVAL)
+
+        # now if we try again we should get a failure
+        d = get_retry_limiter("test_dest", self.clock, store)
+        self.pump()
+        self.failureResultOf(d, NotRetryingDestination)
+
+        #
+        # advance the clock and try again
+        #
+
+        self.pump(MIN_RETRY_INTERVAL)
+        d = get_retry_limiter("test_dest", self.clock, store)
+        self.pump()
+        limiter = self.successResultOf(d)
+
+        self.pump(1)
+        try:
+            with limiter:
+                self.pump(1)
+                retry_ts = self.clock.time_msec()
+                raise AssertionError("argh")
+        except AssertionError:
+            pass
+
+        # wait for the update to land
+        self.pump()
+
+        d = store.get_destination_retry_timings("test_dest")
+        self.pump()
+        new_timings = self.successResultOf(d)
+        self.assertEqual(new_timings["failure_ts"], failure_ts)
+        self.assertEqual(new_timings["retry_last_ts"], retry_ts)
+        self.assertGreaterEqual(
+            new_timings["retry_interval"], MIN_RETRY_INTERVAL * RETRY_MULTIPLIER * 0.5
+        )
+        self.assertLessEqual(
+            new_timings["retry_interval"], MIN_RETRY_INTERVAL * RETRY_MULTIPLIER * 2.0
+        )
+
+        #
+        # one more go, with success
+        #
+        self.pump(MIN_RETRY_INTERVAL * RETRY_MULTIPLIER * 2.0)
+        d = get_retry_limiter("test_dest", self.clock, store)
+        self.pump()
+        limiter = self.successResultOf(d)
+
+        self.pump(1)
+        with limiter:
+            self.pump(1)
+
+        # wait for the update to land
+        self.pump()
+
+        d = store.get_destination_retry_timings("test_dest")
+        self.pump()
+        new_timings = self.successResultOf(d)
+        self.assertIsNone(new_timings)
diff --git a/tests/utils.py b/tests/utils.py
index 46ef2959f2..7dc9bdc505 100644
--- a/tests/utils.py
+++ b/tests/utils.py
@@ -38,11 +38,7 @@ from synapse.logging.context import LoggingContext
 from synapse.server import HomeServer
 from synapse.storage import DataStore
 from synapse.storage.engines import PostgresEngine, create_engine
-from synapse.storage.prepare_database import (
-    _get_or_create_schema_state,
-    _setup_new_database,
-    prepare_database,
-)
+from synapse.storage.prepare_database import prepare_database
 from synapse.util.ratelimitutils import FederationRateLimiter
 
 # set this to True to run the tests against postgres instead of sqlite.
@@ -88,11 +84,7 @@ def setupdb():
             host=POSTGRES_HOST,
             password=POSTGRES_PASSWORD,
         )
-        cur = db_conn.cursor()
-        _get_or_create_schema_state(cur, db_engine)
-        _setup_new_database(cur, db_engine)
-        db_conn.commit()
-        cur.close()
+        prepare_database(db_conn, db_engine, None)
         db_conn.close()
 
         def _cleanup():
@@ -145,7 +137,6 @@ def default_config(name, parse=False):
         "limit_usage_by_mau": False,
         "hs_disabled": False,
         "hs_disabled_message": "",
-        "hs_disabled_limit_type": "",
         "max_mau_value": 50,
         "mau_trial_days": 0,
         "mau_stats_only": False,
@@ -334,10 +325,16 @@ def setup_test_homeserver(
         if homeserverToUse.__name__ == "TestHomeServer":
             hs.setup_master()
     else:
+        # If we have been given an explicit datastore we probably want to mock
+        # out the DataStores somehow too. This all feels a bit wrong, but then
+        # mocking the stores feels wrong too.
+        datastores = Mock(datastore=datastore)
+
         hs = homeserverToUse(
             name,
             db_pool=None,
             datastore=datastore,
+            datastores=datastores,
             config=config,
             version_string="Synapse/tests",
             database_engine=db_engine,
@@ -655,7 +652,7 @@ def create_room(hs, room_id, creator_id):
         creator_id (str)
     """
 
-    store = hs.get_datastore()
+    persistence_store = hs.get_storage().persistence
     event_builder_factory = hs.get_event_builder_factory()
     event_creation_handler = hs.get_event_creation_handler()
 
@@ -672,4 +669,4 @@ def create_room(hs, room_id, creator_id):
 
     event, context = yield event_creation_handler.create_new_client_event(builder)
 
-    yield store.persist_event(event, context)
+    yield persistence_store.persist_event(event, context)
diff --git a/tox.ini b/tox.ini
index 7cb40847b5..afe9bc909b 100644
--- a/tox.ini
+++ b/tox.ini
@@ -2,6 +2,7 @@
 envlist = packaging, py35, py36, py37, check_codestyle, check_isort
 
 [base]
+basepython = python3.7
 deps =
     mock
     python-subunit
@@ -113,15 +114,16 @@ skip_install = True
 basepython = python3.6
 deps =
     flake8
-    black
+    black==19.10b0  # We pin so that our tests don't start failing on new releases of black.
 commands =
     python -m black --check --diff .
-    /bin/sh -c "flake8 synapse tests scripts scripts-dev scripts/hash_password scripts/register_new_matrix_user scripts/synapse_port_db synctl {env:PEP8SUFFIX:}"
+    /bin/sh -c "flake8 synapse tests scripts scripts-dev synctl {env:PEP8SUFFIX:}"
+    {toxinidir}/scripts-dev/config-lint.sh
 
 [testenv:check_isort]
 skip_install = True
 deps = isort
-commands = /bin/sh -c "isort -c -df -sp setup.cfg -rc synapse tests"
+commands = /bin/sh -c "isort -c -df -sp setup.cfg -rc synapse tests scripts-dev scripts"
 
 [testenv:check-newsfragment]
 skip_install = True
@@ -137,18 +139,34 @@ commands = {toxinidir}/scripts-dev/generate_sample_config --check
 skip_install = True
 deps =
     coverage
-whitelist_externals =
-    bash
 commands=
     coverage combine
     coverage report
 
+[testenv:cov-erase]
+skip_install = True
+deps =
+    coverage
+commands=
+    coverage erase
+
+[testenv:cov-html]
+skip_install = True
+deps =
+    coverage
+commands=
+    coverage html
+
 [testenv:mypy]
-basepython = python3.5
+basepython = python3.7
+skip_install = True
 deps =
     {[base]deps}
-    mypy
+    mypy==0.730
+    mypy-zope
+env =
+    MYPYPATH = stubs/
 extras = all
-commands = mypy --ignore-missing-imports \
-            synapse/logging/_structured.py \
-            synapse/logging/_terse_json.py
+commands = mypy \
+            synapse/logging/ \
+            synapse/config/