MirrorHub/synapse
0
0
Fork 1
mirror of https://mau.dev/maunium/synapse.git synced 2024-12-14 06:33:47 +01:00
Code Issues Projects Releases Wiki Activity

Remove _get_events_cache check optimisation from _have_seen_events_dict (#14161)

Browse source
This commit is contained in:
Andrew Morgan 2022-10-18 10:33:21 +01:00 committed by GitHub
parent 2c63cdcc3f
commit 828b5502cf
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 14 additions and 30 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
changelog.d/14161.bugfix Normal file
View file

@ -0,0 +1 @@
Fix a bug introduced in 1.30.0 where purging and rejoining a room without restarting in-between would result in a broken room.

31
synapse/storage/databases/main/events_worker.py
View file

@ -1502,21 +1502,15 @@ class EventsWorkerStore(SQLBaseStore):
Returns:
a dict {event_id -> bool}
"""
# if the event cache contains the event, obviously we've seen it.
# TODO: We used to query the _get_event_cache here as a fast-path before
# hitting the database. For if an event were in the cache, we've presumably
# seen it before.
#
# But this is currently an invalid assumption due to the _get_event_cache
# not being invalidated when purging events from a room. The optimisation can
# be re-added after https://github.com/matrix-org/synapse/issues/13476
cache_results = {
event_id
for event_id in event_ids
if await self._get_event_cache.contains((event_id,))
}
results = dict.fromkeys(cache_results, True)
remaining = [
event_id for event_id in event_ids if event_id not in cache_results
]
if not remaining:
return results
def have_seen_events_txn(txn: LoggingTransaction) -> None:
def have_seen_events_txn(txn: LoggingTransaction) -> Dict[str, bool]:
# we deliberately do *not* query the database for room_id, to make the
# query an index-only lookup on `events_event_id_key`.
#
@ -1524,16 +1518,17 @@ class EventsWorkerStore(SQLBaseStore):
sql = "SELECT event_id FROM events AS e WHERE "
clause, args = make_in_list_sql_clause(
txn.database_engine, "e.event_id", remaining
txn.database_engine, "e.event_id", event_ids
)
txn.execute(sql + clause, args)
found_events = {eid for eid, in txn}
# ... and then we can update the results for each key
results.update({eid: (eid in found_events) for eid in remaining})
return {eid: (eid in found_events) for eid in event_ids}
await self.db_pool.runInteraction("have_seen_events", have_seen_events_txn)
return results
return await self.db_pool.runInteraction(
"have_seen_events", have_seen_events_txn
)
@cached(max_entries=100000, tree=True)
async def have_seen_event(self, room_id: str, event_id: str) -> bool:

12
tests/storage/databases/main/test_events_worker.py
View file

@ -90,18 +90,6 @@ class HaveSeenEventsTestCase(unittest.HomeserverTestCase):
self.assertEqual(res, {self.event_ids[0]})
self.assertEqual(ctx.get_resource_usage().db_txn_count, 0)
def test_query_via_event_cache(self):
# fetch an event into the event cache
self.get_success(self.store.get_event(self.event_ids[0]))
# looking it up should now cause no db hits
with LoggingContext(name="test") as ctx:
res = self.get_success(
self.store.have_seen_events(self.room_id, [self.event_ids[0]])
)
self.assertEqual(res, {self.event_ids[0]})
self.assertEqual(ctx.get_resource_usage().db_txn_count, 0)
def test_persisting_event_invalidates_cache(self):
"""
Test to make sure that the `have_seen_event` cache