Currently whenever the current state changes in a room invalidate a lot
of caches, which cause *a lot* of traffic over replication. Instead,
lets batch up all those invalidations and send a single poke down
the replication streams.
Hopefully this will reduce load on the master process by substantially
reducing traffic.
Currently we only have the one event format version defined, but this
adds the necessary infrastructure to persist and fetch the format
versions alongside the events.
We specify the format version rather than the room version as:
1. We don't necessarily know the room version, existing events may be
either v1 or v2.
2. We'd need to be careful to prevent/handle correctly if different
events in the same room reported to be of different versions, which
sounds annoying.
* Fix race when persisting create event
When persisting a chunk of DAG it is sometimes requried to do a state
resolution, which requires knowledge of the room version. If this
happens while we're persisting the create event then we need to use that
event rather than attempting to look it up in the database.
Currently when fetching state groups from the data store we make two
hits two the database: once for members and once for non-members (unless
request is filtered to one or the other). This adds needless load to the
datbase, so this PR refactors the lookup to make only a single database
hit.
* speed up room summaries by pulling their data from room_memberships rather than room state
* disable LL for incr syncs, and log incr sync stats (#3840)
This field is no longer read from, so we should stop populating it. Once we're
happy that this doesn't break everything, and a rollback is unlikely, we can
think about dropping the column.
We've long passed the point where it's possible to have the same event_id in
different tables, so these join conditions are redundant: we can just join on
event_id.
event_edges is of non-trivial size, and the room_id column is wasteful, so
let's stop reading from it. In future, we can stop writing to it, and then drop
it.
(since it uses methods therein)
Turns out that we had a bunch of things which were incorrectly importing
EventWorkerStore from events.py rather than events_worker.py, which broke once
I removed the import into events.py.
This fixes a bug in _delete_existing_rows_txn which was introduced in #3435
(though it's been on matrix-org-hotfixes for *years*). This code is only called
when there is some sort of conflict the first time we try to persist an event,
so it only happens rarely. Still, the exceptions are annoying.
We need to run the errback in the sentinel context to avoid losing our own
context.
Also: add logging to runInteraction to help identify where "Starting db
connection from sentinel context" warnings are coming from
