<buttonid="sidebar-toggle"class="icon-button"type="button"title="Toggle Table of Contents"aria-label="Toggle Table of Contents"aria-controls="sidebar">
<inputtype="search"id="searchbar"name="searchbar"placeholder="Search this book ..."aria-controls="searchresults-outer"aria-describedby="searchresults-header">
<p>These should be named <code>*.sql</code>, or — for changes which should only be applied for a
given database engine — <code>*.sql.posgres</code> or <code>*.sql.sqlite</code>. For example, a delta which
adds a new column to the <code>foo</code> table might be called <code>01add_bar_to_foo.sql</code>.</p>
<p>Note that our SQL parser is a bit simple - it understands comments (<code>--</code> and <code>/*...*/</code>),
but complex statements which require a <code>;</code> in the middle of them (such as <code>CREATE TRIGGER</code>) are beyond it and you'll have to use a Python delta file.</p>
has no resolution yet (as of 2022-09-01). There are several places in Synapse
and even in the Matrix APIs like <ahref="https://spec.matrix.org/v1.1/server-server-api/#get_matrixfederationv1eventeventid"><code>GET /_matrix/federation/v1/event/{eventId}</code></a>
where we assume that event IDs are globally unique.</p>
<p>When scoping <code>event_id</code> in a database schema, it is often nice to accompany it
with <code>room_id</code> (<code>PRIMARY KEY (room_id, event_id)</code> and a <code>FOREIGN KEY(room_id) REFERENCES rooms(room_id)</code>) which makes flexible lookups easy. For example it
makes it very easy to find and clean up everything in a room when it needs to be
purged (no need to use sub-<code>select</code> query or join from the <code>events</code> table).</p>
<p>A note on collisions: In room versions <code>1</code> and <code>2</code> it's possible to end up with
two events with the same <code>event_id</code> (in the same or different rooms). After room
version <code>3</code>, that can only happen with a hash collision, which we basically hope
will never happen (SHA256 has a massive big key space).</p>
<h2id="worked-examples-of-gradual-migrations"><aclass="header"href="#worked-examples-of-gradual-migrations">Worked examples of gradual migrations</a></h2>
<p>Some migrations need to be performed gradually. A prime example of this is anything
which would need to do a large table scan — including adding columns, indices or
<code>NOT NULL</code> constraints to non-empty tables — such a migration should be done as a
background update where possible, at least on Postgres.
We can afford to be more relaxed about SQLite databases since they are usually
used on smaller deployments and SQLite does not support the same concurrent
DDL operations as Postgres.</p>
<p>We also typically insist on having at least one Synapse version's worth of
backwards compatibility, so that administrators can roll back Synapse if an upgrade
did not go smoothly.</p>
<p>This sometimes results in having to plan a migration across multiple versions
of Synapse.</p>
<p>This section includes an example and may include more in the future.</p>
<h3id="transforming-a-column-into-another-one-with-not-null-constraints"><aclass="header"href="#transforming-a-column-into-another-one-with-not-null-constraints">Transforming a column into another one, with <code>NOT NULL</code> constraints</a></h3>
<p>This example illustrates how you would introduce a new column, write data into it
based on data from an old column and then drop the old column.</p>
<li><code>old_column</code> is read by Synapse and written to by Synapse.</li>
<li><code>new_column</code> is written to by Synapse.</li>
</ol>
<p><strong>Notes:</strong></p>
<ol>
<li><code>new_column</code> can't have a <code>NOT NULL NOT VALID</code> constraint yet, because the previous Synapse version did not write to the new column (since we haven't bumped the <code>SCHEMA_COMPAT_VERSION</code> yet, we still need to be compatible with the previous version).</li>
</ol>
<h4id="synapse-version-n--2"><aclass="header"href="#synapse-version-n--2">Synapse version <code>N + 2</code></a></h4>
<pre><codeclass="language-python">SCHEMA_VERSION = S + 2
SCHEMA_COMPAT_VERSION = S + 1 # this signals that we can't roll back to a time before new_column existed
</code></pre>
<p><strong>Changes:</strong></p>
<ol>
<li>On Postgres, add a <code>NOT VALID</code> constraint to ensure new rows are compliant. <em>SQLite does not have such a construct, but it would be unnecessary anyway since there is no way to concurrently perform this migration on SQLite.</em>
<pre><codeclass="language-sql">ALTER TABLE mytable ADD CONSTRAINT CHECK new_column_not_null (new_column IS NOT NULL) NOT VALID;
</code></pre>
</li>
<li>Start a background update to perform migration: it should gradually run e.g.
<pre><codeclass="language-sql">UPDATE mytable SET new_column = old_column * 100 WHERE 0 < mytable_id AND mytable_id <= 5;
</code></pre>
This background update is technically pointless on SQLite, but you must schedule it anyway so that the <code>portdb</code> script to migrate to Postgres still works.</li>
<li>Upon completion of the background update, you should run <code>VALIDATE CONSTRAINT</code> on Postgres to turn the <code>NOT VALID</code> constraint into a valid one.
This will take some time but does <strong>NOT</strong> hold an exclusive lock over the table.</li>
</ol>
<p><strong>Invariants:</strong></p>
<ol>
<li><code>old_column</code> is read by Synapse and written to by Synapse.</li>
<li><code>new_column</code> is written to by Synapse and new rows always have a non-<code>NULL</code> value in this field.</li>
</ol>
<p><strong>Notes:</strong></p>
<ol>
<li>If you wish, you can convert the <code>CHECK (new_column IS NOT NULL)</code> to a <code>NOT NULL</code> constraint free of charge in Postgres by adding the <code>NOT NULL</code> constraint and then dropping the <code>CHECK</code> constraint, because Postgres can statically verify that the <code>NOT NULL</code> constraint is implied by the <code>CHECK</code> constraint without performing a table scan.</li>
<li>It might be tempting to make version <code>N + 2</code> redundant by moving the background update to <code>N + 1</code> and delaying adding the <code>NOT NULL</code> constraint to <code>N + 3</code>, but that would mean the constraint would always be validated in the foreground in <code>N + 3</code>. Whereas if the <code>N + 2</code> step is kept, the migration in <code>N + 3</code> would be fast in the happy case.</li>
</ol>
<h4id="synapse-version-n--3"><aclass="header"href="#synapse-version-n--3">Synapse version <code>N + 3</code></a></h4>
<pre><codeclass="language-python">SCHEMA_VERSION = S + 3
SCHEMA_COMPAT_VERSION = S + 1 # we can't roll back to a time before new_column existed
</code></pre>
<p><strong>Changes:</strong></p>
<ol>
<li>(Postgres) Update the table to populate values of <code>new_column</code> in case the background update had not completed. Additionally, <code>VALIDATE CONSTRAINT</code> to make the check fully valid.
<pre><codeclass="language-sql">-- you ideally want an index on `new_column` or e.g. `(new_column) WHERE new_column IS NULL` first, or perhaps you can find a way to skip this if the `NOT NULL` constraint has already been validated.
UPDATE mytable SET new_column = old_column * 100 WHERE new_column IS NULL;
-- this is a no-op if it already ran as part of the background update
ALTER TABLE mytable VALIDATE CONSTRAINT new_column_not_null;
</code></pre>
</li>
<li>(SQLite) Recreate the table by precisely following <ahref="https://www.sqlite.org/lang_altertable.html#otheralter">the 12-step procedure for SQLite table schema changes</a>.
During this table rewrite, you should recreate <code>new_column</code> as <code>NOT NULL</code> and populate any outstanding <code>NULL</code> values at the same time.
Unfortunately, you can't drop <code>old_column</code> yet because it must be present for compatibility with the Postgres schema, as needed by <code>portdb</code>.
(Otherwise you could do this all in one go with SQLite!)</li>
</ol>
<p><strong>Invariants:</strong></p>
<ol>
<li><code>old_column</code> is written to by Synapse (but no longer read by Synapse!).</li>
<li><code>new_column</code> is read by Synapse and written to by Synapse. Moreover, all rows have a non-<code>NULL</code> value in this field, as guaranteed by a schema constraint.</li>
</ol>
<p><strong>Notes:</strong></p>
<ol>
<li>We can't drop <code>old_column</code> yet, or even stop writing to it, because that would break a rollback to the previous version of Synapse.</li>
<li>Application code can now rely on <code>new_column</code> being populated. The remaining steps are only motivated by the wish to clean-up old columns.</li>
</ol>
<h4id="synapse-version-n--4"><aclass="header"href="#synapse-version-n--4">Synapse version <code>N + 4</code></a></h4>
<pre><codeclass="language-python">SCHEMA_VERSION = S + 4
SCHEMA_COMPAT_VERSION = S + 3 # we can't roll back to a time before new_column was entirely non-NULL
</code></pre>
<p><strong>Invariants:</strong></p>
<ol>
<li><code>old_column</code> exists but is not written to or read from by Synapse.</li>
<li><code>new_column</code> is read by Synapse and written to by Synapse. Moreover, all rows have a non-<code>NULL</code> value in this field, as guaranteed by a schema constraint.</li>
</ol>
<p><strong>Notes:</strong></p>
<ol>
<li>We can't drop <code>old_column</code> yet because that would break a rollback to the previous version of Synapse. <br/>
<strong>TODO:</strong> It may be possible to relax this and drop the column straight away as long as the previous version of Synapse detected a rollback occurred and stopped attempting to write to the column. This could possibly be done by checking whether the database's schema compatibility version was <code>S + 3</code>.</li>
</ol>
<h4id="synapse-version-n--5"><aclass="header"href="#synapse-version-n--5">Synapse version <code>N + 5</code></a></h4>
<pre><codeclass="language-python">SCHEMA_VERSION = S + 5
SCHEMA_COMPAT_VERSION = S + 4 # we can't roll back to a time before old_column was no longer being touched
</code></pre>
<p><strong>Changes:</strong></p>
<ol>
<li>
<pre><codeclass="language-sql">ALTER TABLE mytable DROP COLUMN old_column;
