Should fix the following issues or make a lot less worse when using
Postgres:
The main issue behind #2911: The client gives up after a certain time,
causing a cascade of context errors, because the response couldn't be
built up fast enough. This mostly happens on accounts with many rooms,
due to the inefficient way we're getting recent events and current state
For #2777: The queries for getting the membership events for history
visibility were being executed for each room (I think 185?), resulting
in a whooping 2k queries for membership events. (Getting the
statesnapshot -> block nids -> actual wanted membership event)
Both should now be better by:
- Using a LATERAL join to get all recent events for all joined rooms in
one go (TODO: maybe do the same for room summary and current state etc)
- If we're lazy loading on initial syncs, we're now not getting the
whole current state, just to drop the majority of it because we're lazy
loading members - we add a filter to exclude membership events on the
first call to `CurrentState`.
- Using an optimized query to get the membership events needed to
calculate history visibility
---------
Co-authored-by: kegsay <kegan@matrix.org>
Since #2849 there is no limit for the current state we fetch to
calculate history visibility. In large rooms this can cause us to fetch
thousands of membership events we don't really care about.
This now only gets the state event types and senders in our timeline,
which should significantly reduce the amount of events we fetch from the
database.
Also removes `MaxTopologicalPosition`, as it is an unnecessary DB call,
given we use the result in `topological_position < $1` calls.
This is apparently some incorrect behaviour that we built as a result of
a spec bug (matrix-org/matrix-spec#1314) where we were applying a filter
to the `"state"` section of the `/sync` response incorrectly. The client
then has no way to know that the state was limited.
This PR removes the state limiting, which probably also helps #2842.
This should stop state events disappearing down a gap where we'd try to
separate out the sections *before* applying history visibility instead
of after.
This may be a better approach than #2843 but I hope @tak-hntlabs will
shout if it isn't.
If we're going backwards, we were selecting potentially thousands of
events, which in turn were fed to history visibility checks, resulting
in bad sync performance.
The problem was that we weren't getting enough recent events, as most of
them were removed by the history visibility filter. Now we're getting
all events between the given input range and re-slice the returned
values after applying history visibility.
Makes the tests
```
Can get rooms/{roomId}/members at a given point
Can filter rooms/{roomId}/members
```
pass, by moving `/members` and `/joined_members` to the SyncAPI.
This makes the following changes:
- get state deltas without the user supplied filter, so we can actually
"calculate" state transitions
- closes `stmt` when using SQLite
- Adds presence for users who newly joined a room, even if the syncing
user already knows about the presence status (should fix
https://github.com/matrix-org/complement/pull/516)
First attempt at removing empty fields from `/sync` responses. Needs
https://github.com/matrix-org/sytest/pull/1298 to keep Sytest happy.
Co-authored-by: Neil Alexander <neilalexander@users.noreply.github.com>
This now uses a transaction per stream, so that errors in one stream
don't propagate to another, and we therefore no longer need to do hacks
to reopen a new transaction after aborting a failed one.
This should transactional snapshot isolation for `/sync` etc requests.
For now we don't use repeatable read due to some odd test failures with
invites.
Previously `LoadMembershipAtEvent` would fail if the state before one of
the events was not known, i.e. because it was an outlier. This modifies
it so that it gracefully handles not knowing the state and returns no
memberships instead, so that history visibility doesn't freak out and
kill `/sync` requests dead.
This should fix an issue where we return less than the expected membership events, when doing an initial sync.
When doing an initial sync, the state limit is set to `math.MaxInt32`, while the default filter is set to 20.
- Reverts 9dc57122d9 as it was causing issues https://github.com/matrix-org/dendrite/issues/2660
- Updates the GMSL `DefaultStateFilter` to use a limit of 20 events
- Uses the timeline events to determine the new position instead of the state events
* Use existing current room state if we have it
* Don't dedupe before applying the history vis filter
* Revert "Don't dedupe before applying the history vis filter"
This reverts commit d27c4a0874.
* Revert "Use existing current room state if we have it"
This reverts commit 5819b4a7ce.
* Tweaks
* Add possibility to set history_visibility and user AccountType
* Add new DB queries
* Add actual history_visibility changes for /messages
* Add passing tests
* Extract check function
* Cleanup
* Cleanup
* Fix build on 386
* Move ApplyHistoryVisibilityFilter to internal
* Move queries to topology table
* Add filtering to /sync and /context
Some cleanup
* Add passing tests; Remove failing tests :(
* Re-add passing tests
* Move filtering to own function to avoid duplication
* Re-add passing test
* Use newly added GMSL HistoryVisibility
* Update gomatrixserverlib
* Set the visibility when creating events
* Default to shared history visibility
* Remove unused query
* Update history visibility checks to use gmsl
Update tests
* Remove unused statement
* Update migrations to set "correct" history visibility
* Add method to fetch the membership at a given event
* Tweaks and logging
* Use actual internal rsAPI, default to shared visibility in tests
* Revert "Move queries to topology table"
This reverts commit 4f0d41be9c.
* Remove noise/unneeded code
* More cleanup
* Try to optimize database requests
* Fix imports
* PR peview fixes/changes
* Move setting history visibility to own migration, be more restrictive
* Fix unit tests
* Lint
* Fix missing entries
* Tweaks for incremental syncs
* Adapt generic changes
Co-authored-by: Neil Alexander <neilalexander@users.noreply.github.com>
Co-authored-by: kegsay <kegan@matrix.org>
* Bypass lazyLoadCache if we're doing an initial sync
* Make the linter happy again?
* Revert "Make the linter happy again?"
This reverts commit 52a5691ba3.
* Try that again
* Invalidate LazyLoadCache on initial syncs
* Remove unneeded check
* Add TODO
* Rename Invalite -> InvalidateLazyLoadedUser
* Thanks IDE
Issue: During conversation, under some conditions, sync cookie is not advanced, and, as a result, client loops on the same sync API call creating high traffic and CPU load.
Fix: pdu component of cookie was updated incorrectly.
* syncapi: use finer-grained interfaces when making the syncapi
* Use specific interfaces for syncapi-roomserver interactions
* Define query access token api for shared http auth code
* Initial work on lazyloading
* Partially implement lazy loading on /sync
* Rename methods
* Make missing tests pass
* Preallocate slice, even if it will end up with fewer values
* Let the cache handle the user mapping
* Linter
* Cap cache growth
* Add ignore users
* Ignore users in pushrules
Add passing tests
* Update sytest lists
* Store ignore knowledge in the sync API
* Fix copyrights
Co-authored-by: Neil Alexander <neilalexander@users.noreply.github.com>
* Include joined and invite member counts in room summary
This should fix#2314 and also fix the problem where some clients like Element Android, Fluffychat etc would display the wrong member count for a given room.
* Improve SQLite query precision
* Check existence of state key for membership events
* Use latest event position in response for advancing the stream position in an incremental sync
* Create some calm
* Use To in worst case
* Don't waste CPU cycles on an empty response after all
* Bug fixes
* Fix another bug
* Don't request state events if we already have the timeline events (Postgres only)
* Rename variable
* nocyclo
* Add SQLite
* Tweaks
* Revert query change
* Don't dedupe if asking for full state
* Update query
* Add some filtering (postgres only for now)
* Fix build error
* Try to use request filter
* Use default filter as a template when retrieving from the database
* Remove unused strut
* Update sytest-whitelist
* Add filtering to SelectEarlyEvents
* Fix Postgres selectEarlyEvents query
* Attempt filtering on SQLite
* Test limit, set field for limit/order in prepareWithFilters
* Remove debug logging, add comments
* Tweaks, debug logging
* Separate SQLite stream IDs
* Fix filtering in current state table
* Fix lock issues
* More tweaks
* Current state requires room ID
* Review comments
* It's half-alive
* Wakeups largely working
* Other tweaks, typing works
* Fix bugs, add receipt stream
* Delete notifier, other tweaks
* Dedupe a bit, add a template for the invite stream
* Clean up, add templates for other streams
* Don't leak channels
* Bring forward some more PDU logic, clean up other places
* Add some more wakeups
* Use addRoomDeltaToResponse
* Log tweaks, typing fixed?
* Fix timed out syncs
* Don't reset next batch position on timeout
* Add account data stream/position
* End of day
* Fix complete sync for receipt, typing
* Streams package
* Clean up a bit
* Complete sync send-to-device
* Don't drop errors
* More lightweight notifications
* Fix typing positions
* Don't advance position on remove again unless needed
* Device list updates
* Advance account data position
* Use limit for incremental sync
* Limit fixes, amongst other things
* Remove some fmt.Println
* Tweaks
* Re-add notifier
* Fix invite position
* Fixes
* Notify account data without advancing PDU position in notifier
* Apply account data position
* Get initial position for account data
* Fix position update
* Fix complete sync positions
* Review comments @Kegsay
* Room consumer parameters
