respond to review comments

This commit is contained in:
Neil Johnson 2019-10-09 23:13:46 +01:00
parent 2ffa761c5e
commit df2833fadf
3 changed files with 53 additions and 41 deletions

View file

@ -605,13 +605,13 @@ def run(hs):
@defer.inlineCallbacks @defer.inlineCallbacks
def generate_monthly_active_users(): def generate_monthly_active_users():
current_mau_count = 0 current_mau_count = 0
reserved_count = 0 reserved_users = ()
store = hs.get_datastore() store = hs.get_datastore()
if hs.config.limit_usage_by_mau or hs.config.mau_stats_only: if hs.config.limit_usage_by_mau or hs.config.mau_stats_only:
current_mau_count = yield store.get_monthly_active_count() 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)) 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)) max_mau_gauge.set(float(hs.config.max_mau_value))
def start_generate_monthly_active_users(): def start_generate_monthly_active_users():

View file

@ -32,7 +32,6 @@ class MonthlyActiveUsersStore(SQLBaseStore):
super(MonthlyActiveUsersStore, self).__init__(None, hs) super(MonthlyActiveUsersStore, self).__init__(None, hs)
self._clock = hs.get_clock() self._clock = hs.get_clock()
self.hs = hs self.hs = hs
self.reserved_users = ()
# Do not add more reserved users than the total allowable number # Do not add more reserved users than the total allowable number
self._new_transaction( self._new_transaction(
dbconn, dbconn,
@ -51,7 +50,6 @@ class MonthlyActiveUsersStore(SQLBaseStore):
txn (cursor): txn (cursor):
threepids (list[dict]): List of threepid dicts to reserve threepids (list[dict]): List of threepid dicts to reserve
""" """
reserved_user_list = []
for tp in threepids: for tp in threepids:
user_id = self.get_user_id_by_threepid_txn(txn, tp["medium"], tp["address"]) user_id = self.get_user_id_by_threepid_txn(txn, tp["medium"], tp["address"])
@ -60,10 +58,8 @@ class MonthlyActiveUsersStore(SQLBaseStore):
is_support = self.is_support_user_txn(txn, user_id) is_support = self.is_support_user_txn(txn, user_id)
if not is_support: if not is_support:
self.upsert_monthly_active_user_txn(txn, user_id) self.upsert_monthly_active_user_txn(txn, user_id)
reserved_user_list.append(user_id)
else: else:
logger.warning("mau limit reserved threepid %s not found in db" % tp) logger.warning("mau limit reserved threepid %s not found in db" % tp)
self.reserved_users = tuple(reserved_user_list)
@defer.inlineCallbacks @defer.inlineCallbacks
def reap_monthly_active_users(self): def reap_monthly_active_users(self):
@ -74,8 +70,11 @@ class MonthlyActiveUsersStore(SQLBaseStore):
Deferred[] Deferred[]
""" """
def _reap_users(txn): def _reap_users(txn, reserved_users):
# Purge stale users """
Args:
reserved_users (tuple): reserved users to preserve
"""
thirty_days_ago = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24 * 30) thirty_days_ago = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24 * 30)
query_args = [thirty_days_ago] query_args = [thirty_days_ago]
@ -83,12 +82,12 @@ class MonthlyActiveUsersStore(SQLBaseStore):
# Need if/else since 'AND user_id NOT IN ({})' fails on Postgres # Need if/else since 'AND user_id NOT IN ({})' fails on Postgres
# when len(reserved_users) == 0. Works fine on sqlite. # 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 # questionmarks is a hack to overcome sqlite not supporting
# tuples in 'WHERE IN %s' # tuples in 'WHERE IN %s'
question_marks = ",".join("?" * len(self.reserved_users)) question_marks = ",".join("?" * len(reserved_users))
query_args.extend(self.reserved_users) query_args.extend(reserved_users)
sql = base_sql + """ AND user_id NOT IN ({})""".format(question_marks) sql = base_sql + """ AND user_id NOT IN ({})""".format(question_marks)
else: else:
sql = base_sql sql = base_sql
@ -105,7 +104,7 @@ class MonthlyActiveUsersStore(SQLBaseStore):
# While Postgres does not require 'LIMIT', but also does not support # 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 # negative LIMIT values. So there is no way to write it that both can
# support # support
if len(self.reserved_users) == 0: if len(reserved_users) == 0:
sql = """ sql = """
DELETE FROM monthly_active_users DELETE FROM monthly_active_users
WHERE user_id NOT IN ( WHERE user_id NOT IN (
@ -119,7 +118,9 @@ class MonthlyActiveUsersStore(SQLBaseStore):
# when len(reserved_users) == 0. Works fine on sqlite. # when len(reserved_users) == 0. Works fine on sqlite.
else: else:
# Must be >= 0 for postgres # Must be >= 0 for postgres
num_of_non_reserved_users_to_remove = max(max_mau_value - len(self.reserved_users), 0) num_of_non_reserved_users_to_remove = max(
max_mau_value - len(reserved_users), 0
)
# It is important to filter reserved users twice to guard # It is important to filter reserved users twice to guard
# against the case where the reserved user is present in the # against the case where the reserved user is present in the
@ -132,13 +133,22 @@ class MonthlyActiveUsersStore(SQLBaseStore):
ORDER BY timestamp DESC ORDER BY timestamp DESC
LIMIT ? LIMIT ?
) )
AND user_id NOT IN ({})""".format(question_marks, question_marks) AND user_id NOT IN ({})""".format(
question_marks, question_marks
)
query_args = [*self.reserved_users, num_of_non_reserved_users_to_remove, *self.reserved_users] query_args = [
*reserved_users,
num_of_non_reserved_users_to_remove,
*reserved_users,
]
txn.execute(sql, query_args) txn.execute(sql, query_args)
yield self.runInteraction("reap_monthly_active_users", _reap_users) 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 # It seems poor to invalidate the whole cache, Postgres supports
# 'Returning' which would allow me to invalidate only the # 'Returning' which would allow me to invalidate only the
# specific users, but sqlite has no way to do this and instead # specific users, but sqlite has no way to do this and instead
@ -167,27 +177,25 @@ class MonthlyActiveUsersStore(SQLBaseStore):
return self.runInteraction("count_users", _count_users) return self.runInteraction("count_users", _count_users)
@defer.inlineCallbacks @defer.inlineCallbacks
def get_registered_reserved_users_count(self): def get_registered_reserved_users(self):
"""Of the reserved threepids defined in config, how many are associated """Of the reserved threepids defined in config, which are associated
with registered users? with registered users?
Returns: Returns:
Defered[int]: Number of real reserved users Defered[tuple]: Real reserved users
""" """
count = 0
users = () users = ()
for tp in self.hs.config.mau_limits_reserved_threepids:
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( user_id = yield self.hs.get_datastore().get_user_id_by_threepid(
tp["medium"], tp["address"] tp["medium"], tp["address"]
) )
if user_id: if user_id:
count = count + 1
users = users + (user_id,) users = users + (user_id,)
# Update reserved_users to ensure it stays in sync, this is important return users
# for reaping.
self.reserved_users = users
return count
@defer.inlineCallbacks @defer.inlineCallbacks
def upsert_monthly_active_user(self, user_id): def upsert_monthly_active_user(self, user_id):

View file

@ -50,6 +50,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
{"medium": "email", "address": user2_email}, {"medium": "email", "address": user2_email},
{"medium": "email", "address": user3_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 # -1 because user3 is a support user and does not count
user_num = len(threepids) - 1 user_num = len(threepids) - 1
@ -84,6 +85,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
self.hs.config.max_mau_value = 0 self.hs.config.max_mau_value = 0
self.reactor.advance(FORTY_DAYS) self.reactor.advance(FORTY_DAYS)
self.hs.config.max_mau_value = 5
self.store.reap_monthly_active_users() self.store.reap_monthly_active_users()
self.pump() self.pump()
@ -173,23 +175,24 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
# monthly_active_users table are reserved # monthly_active_users table are reserved
now = int(self.hs.get_clock().time_msec()) now = int(self.hs.get_clock().time_msec())
if i != 0: if i != 0:
self.get_success(self.store.register_user(user_id=user, password_hash=None)) self.get_success(
#self.pump() self.store.register_user(user_id=user, password_hash=None)
self.get_success(self.store.user_add_threepid(user, "email", email, now, now)) )
self.get_success(
self.store.user_add_threepid(user, "email", email, now, now)
)
self.hs.config.mau_limits_reserved_threepids = threepids self.hs.config.mau_limits_reserved_threepids = threepids
self.store.runInteraction( self.store.runInteraction(
"initialise", self.store._initialise_reserved_users, threepids "initialise", self.store._initialise_reserved_users, threepids
) )
#self.pump()
count = self.store.get_monthly_active_count() count = self.store.get_monthly_active_count()
self.assertTrue(self.get_success(count), initial_users) self.assertTrue(self.get_success(count), initial_users)
count = self.store.get_registered_reserved_users_count() users = self.store.get_registered_reserved_users()
self.assertEquals(self.get_success(count), reserved_user_number) self.assertEquals(len(self.get_success(users)), reserved_user_number)
self.get_success(self.store.reap_monthly_active_users()) self.get_success(self.store.reap_monthly_active_users())
#self.pump()
count = self.store.get_monthly_active_count() count = self.store.get_monthly_active_count()
self.assertEquals(self.get_success(count), self.hs.config.max_mau_value) self.assertEquals(self.get_success(count), self.hs.config.max_mau_value)
@ -227,8 +230,8 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
def test_get_reserved_real_user_account(self): def test_get_reserved_real_user_account(self):
# Test no reserved users, or reserved threepids # Test no reserved users, or reserved threepids
count = self.store.get_registered_reserved_users_count() users = self.get_success(self.store.get_registered_reserved_users())
self.assertEquals(self.get_success(count), 0) self.assertEquals(len(users), 0)
# Test reserved users but no registered users # Test reserved users but no registered users
user1 = "@user1:example.com" user1 = "@user1:example.com"
@ -246,8 +249,8 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
) )
self.pump() self.pump()
count = self.store.get_registered_reserved_users_count() users = self.get_success(self.store.get_registered_reserved_users())
self.assertEquals(self.get_success(count), 0) self.assertEquals(len(users), 0)
# Test reserved registed users # Test reserved registed users
self.store.register_user(user_id=user1, password_hash=None) self.store.register_user(user_id=user1, password_hash=None)
@ -257,8 +260,9 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
now = int(self.hs.get_clock().time_msec()) now = int(self.hs.get_clock().time_msec())
self.store.user_add_threepid(user1, "email", user1_email, now, now) self.store.user_add_threepid(user1, "email", user1_email, now, now)
self.store.user_add_threepid(user2, "email", user2_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): def test_support_user_not_add_to_mau_limits(self):
support_user_id = "@support:test" support_user_id = "@support:test"