Recently I have observed that dendrite spends a lot of time (~390s) in
`selectRoomIDsWithAnyMembershipSQL` query
```
dendrite_syncapi=# select total_exec_time, left(query,100) from pg_stat_statements order by total_exec_time desc limit 5 ;
total_exec_time | left
--------------------+------------------------------------------------------------------------------------------------------
747826.5800519128 | SELECT event_id, id, headered_event_json, session_id, exclude_from_sync, transaction_id, history_vis
389130.5490339942 | SELECT DISTINCT room_id, membership FROM syncapi_current_room_state WHERE type = $2 AND state_key =
376104.17514700035 | SELECT psd.datname, xact_commit, xact_rollback, blks_read, blks_hit, tup_returned, tup_fetched, tup_
363644.164092031 | SELECT event_type_nid, event_state_key_nid, event_nid FROM roomserver_events WHERE event_nid = ANY($
58570.48104699995 | SELECT event_id, headered_event_json FROM syncapi_current_room_state WHERE room_id = $1 AND ( $2::te
(5 rows)
```
Explain analyze showed correct usage of `syncapi_room_state_unique`
index:
```
dendrite_syncapi=#
explain analyze SELECT distinct room_id, membership FROM syncapi_current_room_state WHERE type = 'm.room.member' AND state_key = '@qjfl:dendrite.stg.globekeeper.com';
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Unique (cost=2749.38..2749.56 rows=24 width=52) (actual time=2.933..2.956 rows=65 loops=1)
-> Sort (cost=2749.38..2749.44 rows=24 width=52) (actual time=2.932..2.937 rows=65 loops=1)
Sort Key: room_id, membership
Sort Method: quicksort Memory: 34kB
-> Index Scan using syncapi_room_state_unique on syncapi_current_room_state (cost=0.41..2748.83 rows=24 width=52) (actual time=0.030..2.890 rows=65 loops=1)
Index Cond: ((type = 'm.room.member'::text) AND (state_key = '@qjfl:dendrite.stg.globekeeper.com'::text))
Planning Time: 0.140 ms
Execution Time: 2.990 ms
(8 rows)
```
Multi-column indexes in Postgres shall perform well for leftmost
columns, but I gave it a try and created
`syncapi_current_room_state_type_state_key_idx` index. I could observe
significant performance improvement. Execution time dropped from 2.9 ms
to 0.24 ms:
```
explain analyze SELECT distinct room_id, membership FROM syncapi_current_room_state WHERE type = 'm.room.member' AND state_key = '@qjfl:dendrite.stg.globekeeper.com';
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------------------------------
Unique (cost=96.46..96.64 rows=24 width=52) (actual time=0.199..0.218 rows=65 loops=1)
-> Sort (cost=96.46..96.52 rows=24 width=52) (actual time=0.199..0.202 rows=65 loops=1)
Sort Key: room_id, membership
Sort Method: quicksort Memory: 34kB
-> Bitmap Heap Scan on syncapi_current_room_state (cost=4.53..95.91 rows=24 width=52) (actual time=0.048..0.139 rows=65 loops=1)
Recheck Cond: ((type = 'm.room.member'::text) AND (state_key = '@qjfl:dendrite.stg.globekeeper.com'::text))
Heap Blocks: exact=59
-> Bitmap Index Scan on syncapi_current_room_state_type_state_key_idx (cost=0.00..4.53 rows=24 width=0) (actual time=0.037..0.037 rows=65 loops=1)
Index Cond: ((type = 'm.room.member'::text) AND (state_key = '@qjfl:dendrite.stg.globekeeper.com'::text))
Planning Time: 0.236 ms
Execution Time: 0.242 ms
(11 rows)
```
Next improvement is skipping DISTINCT and rely on map assignment in
`SelectRoomIDsWithAnyMembership`. Execution time drops by almost half:
```
explain analyze SELECT room_id, membership FROM syncapi_current_room_state WHERE type = 'm.room.member' AND state_key = '@qjfl:dendrite.stg.globekeeper.com';
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on syncapi_current_room_state (cost=4.53..95.91 rows=24 width=52) (actual time=0.032..0.113 rows=65 loops=1)
Recheck Cond: ((type = 'm.room.member'::text) AND (state_key = '@qjfl:dendrite.stg.globekeeper.com'::text))
Heap Blocks: exact=59
-> Bitmap Index Scan on syncapi_current_room_state_type_state_key_idx (cost=0.00..4.53 rows=24 width=0) (actual time=0.021..0.021 rows=65 loops=1)
Index Cond: ((type = 'm.room.member'::text) AND (state_key = '@qjfl:dendrite.stg.globekeeper.com'::text))
Planning Time: 0.087 ms
Execution Time: 0.136 ms
(7 rows)
```
In our env we spend only 1s on inserting to table, so the write penalty
of creating an index should be small.
```
dendrite_syncapi=# select total_exec_time, left(query,100) from pg_stat_statements where query like '%INSERT%syncapi_current_room_state%' order by total_exec_time desc;
total_exec_time | left
--------------------+------------------------------------------------------------------------------------------------------
1139.9057619999971 | INSERT INTO syncapi_current_room_state (room_id, event_id, type, sender, contains_url, state_key, he
(1 row)
```
This PR does not require test modifications.
### Pull Request Checklist
<!-- Please read docs/CONTRIBUTING.md before submitting your pull
request -->
* [x] I have added added tests for PR _or_ I have justified why this PR
doesn't need tests.
* [x] Pull request includes a [sign
off](https://github.com/matrix-org/dendrite/blob/main/docs/CONTRIBUTING.md#sign-off)
Signed-off-by: `Piotr Kozimor <p1996k@gmail.com>`
* Fix query issue, only add "changed" users if we actually share a room
* Avoid log spam if context is done
* Undo changes to filterSharedUsers
* Add logging again..
* Fix SQLite shared users query
* Change query to include invited users
* Add new db migration
* Update migrations
Remove goose
* Add possibility to test direct upgrades
* Try to fix WASM test
* Add checks for specific migrations
* Remove AddMigration
Use WithTransaction
Add Dendrite version to table
* Fix linter issues
* Update tests
* Update comments, outdent if
* Namespace migrations
* Add direct upgrade tests, skipping over one version
* Split migrations
* Update go version in CI
* Fix copy&paste mistake
* Use contexts in migrations
Co-authored-by: kegsay <kegan@matrix.org>
Co-authored-by: Neil Alexander <neilalexander@users.noreply.github.com>
* Add function to the sync API storage package for filtering shared users
* Use the database instead of asking the RS API
* Fix unit tests
* Fix map handling in `filterSharedUsers`
* Only load members of newly joined rooms
* Comment that the query is prepared at runtime
Co-authored-by: Neil Alexander <neilalexander@users.noreply.github.com>
* syncapi: add more tests; fix more bugs
bugfixes:
- The postgres impl of TopologyTable.SelectEventIDsInRange did not use the provided txn
- The postgres impl of EventsTable.SelectEvents did not preserve the ordering of the input event IDs in the output events slice
- The sqlite impl of EventsTable.SelectEvents did not use a bulk `IN ($1)` query.
Added tests:
- `TestGetEventsInRangeWithTopologyToken`
- `TestOutputRoomEventsTable`
- `TestTopologyTable`
* -p 1 for now
* 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
* Don't bail so quickly in fetchMissingStateEvents
* Don't recalculate event IDs so often in sync API
* Add comments
* Fix comments
* Update to matrix-org/gomatrixserverlib@eb6a890
* SendEventWithState events as new
* Use cumulative state IDs for final event
* Error wrapping in calculateAndSetState
* Handle overwriting same event type and state key
* Hacky way to spot historical events
* Don't exclude from sync
* Don't generate output events when rewriting forward extremities
* Update output event check
* Historical output events
* Define output room event type
* Notify key changes on state
* Don't send our membership event twice
* Deduplicate state entries
* Tweaks
* Remove unnecessary nolint
* Fix current state upsert in sync API
* Send auth events as outliers, state events as rewrite
* Sync API don't consume state events
* Process events actually
* Improve outlier check
* Fix local room check
* Remove extra room check, it seems to break the whole damn world
* Fix federated join check
* Fix nil pointer exception
* Better comments on DeduplicateStateEntries
* Reflow forced federated joins
* Don't force federated join for possibly even local invites
* Comment SendEventWithState better
* Rewrite room state in sync API storage
* Add TODO
* Clean up all room data when receiving create event
* Don't generate output events for rewrites, but instead notify that state is rewritten on the final new event
* Rename to PurgeRoom
* Exclude backfilled messages from /sync
* Split out rewriting state from updating state from state res
Co-authored-by: Kegan Dougal <kegan@matrix.org>
* Offset updates take place using TransactionWriter
* Refactor TransactionWriter in current state server
* Refactor TransactionWriter in federation sender
* Refactor TransactionWriter in key server
* Refactor TransactionWriter in media API
* Refactor TransactionWriter in server key API
* Refactor TransactionWriter in sync API
* Refactor TransactionWriter in user API
* Fix deadlocking Sync API tests
* Un-deadlock device database
* Fix appservice API
* Rename TransactionWriters to Writers
* Move writers up a layer in sync API
* Document sqlutil.Writer interface
* Add note to Writer documentation
* Updated TransactionWriters, moved locks in roomserver, various other tweaks
* Fix redaction deadlocks
* Fix lint issue
* Rename SQLiteTransactionWriter to ExclusiveTransactionWriter
* Fix us not sending transactions through in latest events updater
* Use TransactionWriter on other component SQLites
* Fix sync API tests
* Fix panic in media API
* Fix a couple of transactions
* Fix wrong query, add some logging output
* Add debug logging into StoreEvent
* Adjust InsertRoomNID
* Update logging
* Refactor all postgres tables; start work on sqlite
* wip sqlite merges; database is locked errors to investigate and failing tests
* Revert "wip sqlite merges; database is locked errors to investigate and failing tests"
This reverts commit 26cbfc5b75ae2dc4fb31a838b917aa39d758f162.
* convert current room state table
* port over sqlite topology table
* remove a few functions
* remove more functions
* Share more code
* factor out completesync and a bit more
* Remove remaining code
* Use HeaderedEvent in syncapi
* Update notifier test
* Fix persisting headered event
* Clean up unused API function
* Fix overshadowed err from linter
* Write headered JSON to invites table too
* Rename event_json to headered_event_json in syncapi database schemae
* Fix invites_table queries
* Update QueryRoomVersionCapabilitiesResponse comment
* Fix syncapi SQLite
* Use a fork of pq which supports userCurrent on wasm
* Use sqlite3_js driver when running in JS
* Add cmd/dendritejs to pull in sqlite3_js driver for wasm only
* Update to latest go-sqlite-js version
* Replace prometheus with a stub. sigh
* Hard-code a config and don't use opentracing
* Latest go-sqlite3-js version
* Generate a key for now
* Listen for fetch traffic rather than HTTP
* Latest hacks for js
* libp2p support
* More libp2p
* Fork gjson to allow us to enforce auth checks as before
Previously, all events would come down redacted because the hash
checks would fail. They would fail because sjson.DeleteBytes didn't
remove keys not used for hashing. This didn't work because of a build
tag which included a file which no-oped the index returned.
See https://github.com/tidwall/gjson/issues/157
When it's resolved, let's go back to mainline.
* Use gjson@1.6.0 as it fixes https://github.com/tidwall/gjson/issues/157
* Use latest gomatrixserverlib for sig checks
* Fix a bug which could cause exclude_from_sync to not be set
Caused when sending events over federation.
* Use query variadic to make lookups actually work!
* Latest gomatrixserverlib
* Add notes on getting p2p up and running
Partly so I don't forget myself!
* refactor: Move p2p specific stuff to cmd/dendritejs
This is important or else the normal build of dendrite will fail
because the p2p libraries depend on syscall/js which doesn't work
on normal builds.
Also, clean up main.go to read a bit better.
* Update ho-http-js-libp2p to return errors from RoundTrip
* Add an LRU cache around the key DB
We actually need this for P2P because otherwise we can *segfault*
with things like: "runtime: unexpected return pc for runtime.handleEvent"
where the event is a `syscall/js` event, caused by spamming sql.js
caused by "Checking event signatures for 14 events of room state" which
hammers the key DB repeatedly in quick succession.
Using a cache fixes this, though the underlying cause is probably a bug
in the version of Go I'm on (1.13.7)
* breaking: Add Tracing.Enabled to toggle whether we do opentracing
Defaults to false, which is why this is a breaking change. We need
this flag because WASM builds cannot do opentracing.
* Start adding conditional builds for wasm to handle lib/pq
The general idea here is to have the wasm build have a `NewXXXDatabase`
that doesn't import any postgres package and hence we never import
`lib/pq`, which doesn't work under WASM (undefined `userCurrent`).
* Remove lib/pq for wasm for syncapi
* Add conditional building to remaining storage APIs
* Update build script to set env vars correctly for dendritejs
* sqlite bug fixes
* Docs
* Add a no-op main for dendritejs when not building under wasm
* Use the real prometheus, even for WASM
Instead, the dendrite-sw.js must mock out `process.pid` and
`fs.stat` - which must invoke the callback with an error (e.g `EINVAL`)
in order for it to work:
```
global.process = {
pid: 1,
};
global.fs.stat = function(path, cb) {
cb({
code: "EINVAL",
});
}
```
* Linting
* bugfix: fix panic on new invite events from sytest
I'm unsure why the previous code didn't work, but it's
clearer, quicker and easier to read the `LastInsertID()` way.
Previously, the code would panic as the SELECT would fail
to find the last inserted row ID.
* sqlite: Fix UNIQUE violations and close more cursors
- Add missing `defer rows.Close()`
- Do not have the state block NID as a PRIMARY KEY else it breaks for blocks
with >1 state event in them. Instead, rejig the queries so we can still
have monotonically increasing integers without using AUTOINCREMENT (which
mandates PRIMARY KEY).
* sqlite: Add missing variadic function
* Use LastInsertId because empirically it works over the SELECT form (though I don't know why that is)
* sqlite: Fix invite table by using the global stream pos rather than one specific to invites
If we don't use the global, clients don't get notified about any invites
because the position is too low.
* linting: shadowing
* sqlite: do not use last rowid, we already know the stream pos!
* sqlite: Fix account data table in syncapi by commiting insert txns!
* sqlite: Fix failing federation invite
Was failing with 'database is locked' due to multiple write txns
being taken out.
* sqlite: Ensure we return exactly the number of events found in the database
Previously we would return exactly the number of *requested* events, which
meant that several zero-initialised events would bubble through the system,
failing at JSON serialisation time.
* sqlite: let's just ignore the problem for now....
* linting
* Move current work into single branch
* Initial massaging of clientapi etc (not working yet)
* Interfaces for accounts/devices databases
* Duplicate postgres package for sqlite3 (no changes made to it yet)
* Some keydb, accountdb, devicedb, common partition fixes, some more syncapi tweaking
* Fix accounts DB, device DB
* Update naffka dependency for SQLite
* Naffka SQLite
* Update naffka to latest master
* SQLite support for federationsender
* Mostly not-bad support for SQLite in syncapi (although there are problems where lots of events get classed incorrectly as backward extremities, probably because of IN/ANY clauses that are badly supported)
* Update Dockerfile -> Go 1.13.7, add build-base (as gcc and friends are needed for SQLite)
* Implement GET endpoints for account_data in clientapi
* Nuke filtering for now...
* Revert "Implement GET endpoints for account_data in clientapi"
This reverts commit 4d80dff4583d278620d9b3ed437e9fcd8d4674ee.
* Implement GET endpoints for account_data in clientapi (#861)
* Implement GET endpoints for account_data in clientapi
* Fix accountDB parameter
* Remove fmt.Println
* Fix insertAccountData SQLite query
* Fix accountDB storage interfaces
* Add empty push rules into account data on account creation (#862)
* Put SaveAccountData into the right function this time
* Not sure if roomserver is better or worse now
* sqlite work
* Allow empty last sent ID for the first event
* sqlite: room creation works
* Support sending messages
* Nuke fmt.println
* Move QueryVariadic etc into common, other device fixes
* Fix some linter issues
* Fix bugs
* Fix some linting errors
* Fix errcheck lint errors
* Make naffka use postgres as fallback, fix couple of compile errors
* What on earth happened to the /rooms/{roomID}/send/{eventType} routing
Co-authored-by: Neil Alexander <neilalexander@users.noreply.github.com>
