Currently when using multiple event persisters we (in the worst case) don't tell clients about events until all event persisters have persisted new events after the original event. This is a suboptimal, especially if one of the event persisters goes down.
To handle this, we encode the position of each event persister in the room tokens so that we can send events to clients immediately. To reduce the size of the token we do two things:
1. We create a unique immutable persistent mapping between instance names and a generated small integer ID, which we can encode in the tokens instead of the instance name; and
2. We encode the "persisted upto position" of the room token and then only explicitly include instances that have positions strictly greater than that.
The new tokens look something like: `m3478~1.3488~2.3489`, where the first number is the min position, and the subsequent `-` separated pairs are the instance ID to positions map. (We use `.` and `~` as separators as they're URL safe and not already used by `StreamToken`).
This is so we can tell what is going on when things are taking a while to start up.
The main change here is to ensure that transactions that are created during startup get correctly logged like normal transactions.
On startup `MultiWriteIdGenerator` fetches the maximum stream ID for
each instance from the table and uses that as its initial "current
position" for each writer. This is problematic as a) it involves either
a scan of events table or an index (neither of which is ideal), and b)
if rows are being persisted out of order elsewhere while the process
restarts then using the maximum stream ID is not correct. This could
theoretically lead to race conditions where e.g. events that are
persisted out of order are not sent down sync streams.
We fix this by creating a new table that tracks the current positions of
each writer to the stream, and update it each time we finish persisting
a new entry. This is a relatively small overhead when persisting events.
However for the cache invalidation stream this is a much bigger relative
overhead, so instead we note that for invalidation we don't actually
care about reliability over restarts (as there's no caches to
invalidate) and simply don't bother reading and writing to the new table
in that particular case.
This is *not* ready for production yet. Caveats:
1. We should write some tests...
2. The stream token that we use for events can get stalled at the minimum position of all writers. This means that new events may not be processed and e.g. sent down sync streams if a writer isn't writing or is slow.
This fixes an issue where different methods (crop/scale) overwrite each other.
This first tries the new path. If that fails and we are looking for a
remote thumbnail, it tries the old path. If that still isn't found, it
continues as normal.
This should probably be removed in the future, after some of the newer
thumbnails were generated with the new path on most deployments. Then
the overhead should be minimal if the other thumbnails need to be
regenerated.
Signed-off-by: Nicolas Werner <nicolas.werner@hotmail.de>
Fixes https://github.com/matrix-org/synapse/issues/8238
Alongside the delta file, some changes were also necessary to the codebase to remove references to the now defunct `populate_stats_process_rooms_2` background job. Thankfully the latter doesn't seem to have made it into any documentation yet :)
* Fixup `ALTER TABLE` database queries
Make the new columns nullable, because doing otherwise can wedge a
server with a big database, as setting a default value rewrites the
table.
* Switch back to using the notifications count in the push badge
Clients are likely to be confused if we send a push but the badge count
is the unread messages one, and not the notifications one.
* Changelog
This is *not* ready for production yet. Caveats:
1. We should write some tests...
2. The stream token that we use for events can get stalled at the minimum position of all writers. This means that new events may not be processed and e.g. sent down sync streams if a writer isn't writing or is slow.
