Fix https://github.com/matrix-org/synapse/issues/13856
Fix https://github.com/matrix-org/synapse/issues/13865
> Discovered while trying to make Synapse fast enough for [this MSC2716 test for importing many batches](https://github.com/matrix-org/complement/pull/214#discussion_r741678240). As an example, disabling the `have_seen_event` cache saves 10 seconds for each `/messages` request in that MSC2716 Complement test because we're not making as many federation requests for `/state` (speeding up `have_seen_event` itself is related to https://github.com/matrix-org/synapse/issues/13625)
>
> But this will also make `/messages` faster in general so we can include it in the [faster `/messages` milestone](https://github.com/matrix-org/synapse/milestone/11).
>
> *-- https://github.com/matrix-org/synapse/issues/13856*
### The problem
`_invalidate_caches_for_event` doesn't run in monolith mode which means we never even tried to clear the `have_seen_event` and other caches. And even in worker mode, it only runs on the workers, not the master (AFAICT).
Additionally there was bug with the key being wrong so `_invalidate_caches_for_event` never invalidates the `have_seen_event` cache even when it does run.
Because we were using the `@cachedList` wrong, it was putting items in the cache under keys like `((room_id, event_id),)` with a `set` in a `set` (ex. `(('!TnCIJPKzdQdUlIyXdQ:test', '$Iu0eqEBN7qcyF1S9B3oNB3I91v2o5YOgRNPwi_78s-k'),)`) and we we're trying to invalidate with just `(room_id, event_id)` which did nothing.
Since MSC3715 has passed FCP, the stable parameter can be used.
This currently falls back to the unstable parameter if the stable
parameter is not provided (and MSC3715 support is enabled in
the configuration).
Since #11482, we're saving sessions IDs from upstream IdPs, but we've been losing them when the user goes through a user mapping session on account registration.
During a `lazy_load_members` `/sync`, we look through auth events in
rooms with partial state to find prior membership events. When such a
membership is not found, an error is logged.
Since the first join event for a user never has a prior membership event
to cite, the error would always be logged when one appeared in the room
timeline.
Avoid logging errors for such events.
Introduced in #13477.
Signed-off-by: Sean Quah <seanq@matrix.org>
This should mean that logs from worker processes are flushed before shutdown.
When a test completes, Complement stops the docker container, which means that
synapse will receive a SIGTERM. Currently, the `complement_fork_starter` exits
immediately (without notifying the worker processes), which means that the
workers never get a chance to flush their logs before the whole container is
vaped. We can fix this by propagating the SIGTERM to the children.
This is also using the partial state approximation if needed so we do
not block here during a fast join.
Signed-off-by: Mathieu Velten <mathieuv@matrix.org>
Co-authored-by: Sean Quah <8349537+squahtx@users.noreply.github.com>
This moves all the invalidations into a single place and de-duplicates
the code involved in invalidating caches for a given event by using
the base class method.
* Lockfile: update canonicaljson 1.6.0 -> 1.6.3
* Fix mypy errors with latest canonicaljson
The change to `_encode_json_bytes` definition wasn't sufficient:
```
synapse/http/server.py:751: error: Incompatible types in assignment (expression has type "Callable[[Arg(object, 'json_object')], bytes]", variable has type "Callable[[Arg(object, 'data')], bytes]") [assignment]
```
Which I think is mypy warning us that the two functions accept different
sets of kwargs. Fair enough!
* Changelog
Part of the work for #12993.
Once #12993 is fully resolved, we expect `/keys/changes` to behave
sensibly when joined to a room with partial state.
Signed-off-by: Sean Quah <seanq@matrix.org>
Use the provided list of servers in the room from the `/send_join`
response, since we will not know which users are in the room. This
isn't sufficient to ensure that all remote servers receive the right
device list updates, since the `/send_join` response may be inaccurate
or we may calculate the membership state of new users in the room
incorrectly.
Signed-off-by: Sean Quah <seanq@matrix.org>
This fixes a bug where the `/relations` API with `dir=f` would
skip the first item of each page (except the first page), causing
incomplete data to be returned to the client.
* Generate separate snapshots for sqlite, postgres and common
* Cleanup postgres dbs in the TRAP
* Say which logical DB we're applying updates to
* Run background updates on the state DB
* Add new option for accepting a SCHEMA_NUMBER
Adds a `thread_id` column to the `event_push_actions`, `event_push_actions_staging`,
and `event_push_summary` tables. This will notifications to be segmented by the thread
in a future pull request. The `thread_id` column stores the root event ID or the special
value `"main"`.
The `thread_id` column for `event_push_actions` and `event_push_summary` is
backfilled with `"main"` for all existing rows. New entries into `event_push_actions`
and `event_push_actions_staging` will get the proper thread ID.
`receipts_linearized` and `receipts_graph` also gain a `thread_id` column, which is similar,
except `NULL` is a special value meaning the receipt is "unthreaded".
See MSC3771 and MSC3773 for where this data will be useful.
Partial indices have been supported since SQLite 3.8, but Synapse
now requires >= 3.27, so we can enable support for them.
This requires rebuilding previous indices which were partial on
PostgreSQL, but not on SQLite.
* Remove incorrect migration file from `state` logical DB
The table `ex_outlier_stream` is part of the `main` logical DB; it
should not have been created in the `state` logical DB. We remove this
migration now as a tidy-up.
Note: we cannot `DROP TABLE IF EXISTS ex_outlier_stream` in a new
migration, because some (most) instances of Synapse host both of these
logical DBs on the same DB cluster.
* Changelog
When a remote user leaves the last room shared with the homeserver, we
have to mark their device list as unsubscribed, otherwise we would hold
on to a stale device list in our cache. Crucially, the device list would
remain cached even after the remote user rejoined the room, which could
lead to E2EE failures until the next change to the remote user's device
list.
Fixes#13651.
Signed-off-by: Sean Quah <seanq@matrix.org>
* Don't accept a trailing slash on the end of /get_missing_events
* Newsfile
Signed-off-by: Olivier Wilkinson (reivilibre) <oliverw@matrix.org>
Signed-off-by: Olivier Wilkinson (reivilibre) <oliverw@matrix.org>
* Remove checks for membership column in current_state_events
* Add schema script to force through the
`current_state_events_membership` background job
Contributed by Nick @ Beeper (@fizzadar).
Most of the time this function is heavily cached, but when that isn't
the case fetching the counts room by room slows down push delivery on
users with many (thousands) of rooms.
Signed off by Nick @ Beeper.
The problem with many services is that it makes it hard to find which service has the trace you want, see https://github.com/jaegertracing/jaeger-ui/issues/985
Previously, we split traces out into services based on their instance name like `matrix.org client_reader-1`, etc but there are many worker instances of the same `client_reader` so there is a lot to click through.
With this PR, all of the traces are just collected under the worker type like `client_reader`, `event_persister` 😇
Note: A Synapse worker instance name is an opaque string with the number convention only being our own thing for the `matrix.org` deployment. But seems pretty sensible to group things this way.
Update the docstrings for `get_users_in_room` and
`get_current_hosts_in_room` to explain the impact of partial state.
Signed-off-by: Sean Quah <seanq@matrix.org>
Handle malformed user IDs with no colons in `get_current_hosts_in_room`.
It's not currently possible for a malformed user ID to join a room, so
this error would never be hit.
Signed-off-by: Sean Quah <seanq@matrix.org>
Previously, `is_mine_id` would raise an exception when passed an ID with
no colons. Return `False` instead.
Fixes#13040.
Signed-off-by: Sean Quah <seanq@matrix.org>
When backfilling, `_get_state_ids_after_missing_prev_event` calls [`get_metadata_for_events`](26bc26586b/synapse/handlers/federation_event.py (L1133)). For `#matrix:matrix.org`, it's called with 77k `state_events` which means 77 calls to the database and takes 28 seconds.
This is a re-do of 57d334a13d (#13365),
which was backed out in 12abd72497 (#13501).
The `room_id` field represented the parent space for each room
and was made redundant by changes in the API shape where the
`children_state` is now nested underneath each `room`.
The room ID of each child is in the `state_key` field and is still
available.
This avoids doing work that will never be used (since the
resulting unread counts will never be sent in a /sync
response).
The negative of doing this is that unread counts will be
incorrect when the feature is initially enabled.
The method doesn't actually do any data fetching and the method that
does, `_get_joined_profile_from_event_id`, has its own cache.
Signed off by Nick @ Beeper (@Fizzadar).
We incorrectly didn't use the returned `Responder` if the client had
disconnected, which meant that the resource used by the Responder
wasn't correctly released.
In particular, this exhausted the thread pools so that *all* requests
timed out.
Media downloaded as part of a URL preview is normally deleted after two days.
However, while a background database migration is running, the process is
stopped. A long-running database migration can therefore cause the media
store to fill up with old preview files.
This logic was added in #2697 to make sure that we didn't try to run the expiry
without an index on `local_media_repository.created_ts`; the original logic that
needs that index was added in #2478 (in `get_url_cache_media_before`, as
amended by 93247a424a), and is still present.
Given that the background update was added before Synapse v1.0.0, just drop
this check and assume the index exists.
Optimize how we calculate `likely_domains` during backfill because I've seen this take 17s in production just to `get_current_state` which is used to `get_domains_from_state` (see case [*2. Loading tons of events* in the `/messages` investigation issue](https://github.com/matrix-org/synapse/issues/13356)).
There are 3 ways we currently calculate hosts that are in the room:
1. `get_current_state` -> `get_domains_from_state`
- Used in `backfill` to calculate `likely_domains` and `/timestamp_to_event` because it was cargo-culted from `backfill`
- This one is being eliminated in favor of `get_current_hosts_in_room` in this PR 🕳
1. `get_current_hosts_in_room`
- Used for other federation things like sending read receipts and typing indicators
1. `get_hosts_in_room_at_events`
- Used when pushing out events over federation to other servers in the `_process_event_queue_loop`
Fix https://github.com/matrix-org/synapse/issues/13626
Part of https://github.com/matrix-org/synapse/issues/13356
Mentioned in [internal doc](https://docs.google.com/document/d/1lvUoVfYUiy6UaHB6Rb4HicjaJAU40-APue9Q4vzuW3c/edit#bookmark=id.2tvwz3yhcafh)
### Query performance
#### Before
The query from `get_current_state` sucks just because we have to get all 80k events. And we see almost the exact same performance locally trying to get all of these events (16s vs 17s):
```
synapse=# SELECT type, state_key, event_id FROM current_state_events WHERE room_id = '!OGEhHVWSdvArJzumhm:matrix.org';
Time: 16035.612 ms (00:16.036)
synapse=# SELECT type, state_key, event_id FROM current_state_events WHERE room_id = '!OGEhHVWSdvArJzumhm:matrix.org';
Time: 4243.237 ms (00:04.243)
```
But what about `get_current_hosts_in_room`: When there is 8M rows in the `current_state_events` table, the previous query in `get_current_hosts_in_room` took 13s from complete freshness (when the events were first added). But takes 930ms after a Postgres restart or 390ms if running back to back to back.
```sh
$ psql synapse
synapse=# \timing on
synapse=# SELECT COUNT(DISTINCT substring(state_key FROM '@[^:]*:(.*)$'))
FROM current_state_events
WHERE
type = 'm.room.member'
AND membership = 'join'
AND room_id = '!OGEhHVWSdvArJzumhm:matrix.org';
count
-------
4130
(1 row)
Time: 13181.598 ms (00:13.182)
synapse=# SELECT COUNT(*) from current_state_events where room_id = '!OGEhHVWSdvArJzumhm:matrix.org';
count
-------
80814
synapse=# SELECT COUNT(*) from current_state_events;
count
---------
8162847
synapse=# SELECT pg_size_pretty( pg_total_relation_size('current_state_events') );
pg_size_pretty
----------------
4702 MB
```
#### After
I'm not sure how long it takes from complete freshness as I only really get that opportunity once (maybe restarting computer but that's cumbersome) and it's not really relevant to normal operating times. Maybe you get closer to the fresh times the more access variability there is so that Postgres caches aren't as exact. Update: The longest I've seen this run for is 6.4s and 4.5s after a computer restart.
After a Postgres restart, it takes 330ms and running back to back takes 260ms.
```sh
$ psql synapse
synapse=# \timing on
Timing is on.
synapse=# SELECT
substring(c.state_key FROM '@[^:]*:(.*)$') as host
FROM current_state_events c
/* Get the depth of the event from the events table */
INNER JOIN events AS e USING (event_id)
WHERE
c.type = 'm.room.member'
AND c.membership = 'join'
AND c.room_id = '!OGEhHVWSdvArJzumhm:matrix.org'
GROUP BY host
ORDER BY min(e.depth) ASC;
Time: 333.800 ms
```
#### Going further
To improve things further we could add a `limit` parameter to `get_current_hosts_in_room`. Realistically, we don't need 4k domains to choose from because there is no way we're going to query that many before we a) probably get an answer or b) we give up.
Another thing we can do is optimize the query to use a index skip scan:
- https://wiki.postgresql.org/wiki/Loose_indexscan
- Index Skip Scan, https://commitfest.postgresql.org/37/1741/
- https://www.timescale.com/blog/how-we-made-distinct-queries-up-to-8000x-faster-on-postgresql/
If things like the signing key file are missing, let's just try to generate
them on startup.
Again, this is useful for k8s-like deployments where we just want to generate
keys on the first run.
When loading current ids, sort by stream ID so that we don't want to overwrite the `current_position` of an instance to a lower stream ID than we're actually at ([discussion](https://github.com/matrix-org/synapse/pull/13585#discussion_r951795379)). Previously, it sorted alphabetically by instance name which can be `null` and throw errors but more importantly, accomplishes nothing.
Fixes the following startup error which is why I started looking into this area:
```
$ poetry run synapse_homeserver --config-path homeserver.yaml
****************************************************************
Error during initialisation:
'<' not supported between instances of 'NoneType' and 'str'
There may be more information in the logs.
****************************************************************
```
Somehow my database ended up looking like the following, notice the `instance_name` is `null` in the db, and we can't sort `NoneType` things. Another question is why do we see the `instance_name` as `null` sometimes instead of `master` in monolith mode?
```
$ psql synapse
synapse=# SELECT * FROM stream_positions;
stream_name | instance_name | stream_id
-----------------+---------------+-----------
account_data | master | 1242
events | master | 1787
to_device | master | 58
presence_stream | master | 485638
receipts | master | 341
backfill | master | -139106
(6 rows)
synapse=# SELECT instance_name, stream_id FROM receipts_linearized;
instance_name | stream_id
---------------+-----------
| 211
| 3
| 4
| 212
| 213
| 224
| 228
| 164
| 313
| 253
| 38
| 321
| 324
| 189
| 192
| 193
| 194
| 195
| 197
| 198
| 275
| 79
| 339
| 340
| 82
| 341
| 84
| 85
| 91
| 119
```
Use dedicated `get_local_users_in_room` to find local users when calculating `join_authorised_via_users_server` ("the authorising user for joining a restricted room") of a `/make_join` request.
Found while working on https://github.com/matrix-org/synapse/pull/13575#discussion_r953023755 but it's not related.
This speeds things up by ~2x.
The vast majority of the time is now spent in `LruCache` moving things around the linked lists.
We do this via two things:
1. Don't create a deferred per-key during bulk set operations in `DeferredCache`. Instead, only create them if a subsequent caller asks for the key.
2. Add a bulk lookup API to `DeferredCache` rather than use a loop.
Part of #13019
This changes all the permission-related methods to rely on the Requester instead of the UserID. This is a first step towards enabling scoped access tokens at some point, since I expect the Requester to have scope-related informations in it.
It also changes methods which figure out the user/device/appservice out of the access token to return a Requester instead of something else. This avoids having store-related objects in the methods signatures.
Use a state filter or accept partial state in a few places where we
request state, to avoid blocking.
To make lazy-loading `/sync`s work, we need to provide the memberships
of event senders, which are not guaranteed to be in the room state.
Instead we dig through auth events for memberships to present to
clients. The auth events of an event are guaranteed to contain a
passable membership event, otherwise the event would have been rejected.
Note that this only covers the common code paths encountered during
testing. There has been no exhaustive checking of all sync code paths.
Fixes#13146.
Signed-off-by: Sean Quah <seanq@matrix.org>
Broke by #13522
It looks like we have some rules in the DB with a priority class less
than 0 that don't override the base rules. Before these were just
dropped, but #13522 made that a hard error.
This improves load times for push rules:
| Version | Time per user | Time for 1k users |
| -------------------- | ------------- | ----------------- |
| Before | 138 µs | 138ms |
| Now (with custom) | 2.11 µs | 2.11ms |
| Now (without custom) | 49.7 ns | 0.05 ms |
This therefore has a large impact on send times for rooms
with large numbers of local users in the room.
This reverts commit f383b9b3ec. Other PRs
were seeing mypy failures that looked to be related to mypy-zope.
Confusingly, we didn't see this on #13521.
Revert this for now and investigate later.
* Clarifies comments.
* Fixes an erroneous comment (about return type) added in #13455
(ec24813220).
* Clarifies the name of a variable.
* Simplifies logic of pulling out the latest join for the requesting user.