Reduce the replication traffic of device lists, by not sending every
destination that needs to be sent the device list update over
replication. Instead a "hosts to send to have been calculated"
notification over replication, and then federation senders read the
destinations from the DB.
For non federation senders this should heavily reduce the impact of a
user in many large rooms changing a device.
Re-introduces #17191, and includes #17197 and #17214
The basic idea is to stop calling `get_rooms_for_user` everywhere, and
instead use the table `device_lists_changes_in_room`.
Commits reviewable one-by-one.
During the migration the automated script to update the copyright
headers accidentally got rid of some of the existing copyright lines.
Reinstate them.
If a worker reconnects to Redis we send out the current positions of all our streams. However, if we're also trying to send out a backlog of RDATA at the same time then we can end up sending a `POSITION` with the current token *before* we've sent all the RDATA before the current token.
This doesn't cause actual bugs as the receiving servers see the POSITION, fetch the relevant rows from the DB, and then ignore the old RDATA as they come in. However, this is inefficient so it'd be better if we didn't send out-of-order positions
* Fix bug where a new writer advances their token too quickly
When starting a new writer (for e.g. persisting events), the
`MultiWriterIdGenerator` doesn't have a minimum token for it as there
are no rows matching that new writer in the DB.
This results in the the first stream ID it acquired being announced as
persisted *before* it actually finishes persisting, if another writer
gets and persists a subsequent stream ID. This is due to the logic of
setting the minimum persisted position to the minimum known position of
across all writers, and the new writer starts off not being considered.
* Fix sending out POSITIONs when our token advances without update
Broke in #14820
* For replication HTTP requests, only wait for minimal position
This is because if a worker reaches ~100% CPU then everything starts
lagging and we hit the log line a lot. When at error we invoke sentry
and that has a lot of overhead, which then puts even more pressure on
the worker.
Refactoring to use both the user ID & the device ID when tracking
the currently syncing users in the presence handler.
This is done both locally and over replication. Note that the device
ID is discarded but will be used in a future change.
Refactoring to pass the device ID (in addition to the user ID) through
the presence handler (specifically the `user_syncing`, `set_state`,
and `bump_presence_active_time` methods and their replication
versions).
Simplify some of the presence code by reducing duplicated code between
worker & non-worker modes.
The main change is to push some of the logic from `user_syncing` into
`set_state`. This is done by passing whether the user is setting the presence
via a `/sync` with a new `is_sync` flag to `set_state`. If this is `true` some
additional logic is performed:
* Don't override `busy` presence.
* Update the `last_user_sync_ts`.
* Never update the status message.
All the information needed is already in the `instance_map`, so
use that instead of passing the hostname / IP & port manually
for each replication request.
This consolidates logic for future improvements of using e.g.
UNIX sockets for workers.
* Add SSL options to redis config
* fix lint issues
* Add documentation and changelog file
* add missing . at the end of the changelog
* Move client context factory to new file
* Rename ssl to tls and fix typo
* fix lint issues
* Added when redis attributes were added
