### Background
As part of Element's plan to support a light form of vhosting (virtual
host) (multiple instances of Synapse in the same Python process), we're
currently diving into the details and implications of running multiple
instances of Synapse in the same Python process.
"Per-tenant logging" tracked internally by
https://github.com/element-hq/synapse-small-hosts/issues/48
### Prior art
Previously, we exposed `server_name` by providing a static logging
`MetadataFilter` that injected the values:
205d9e4fc4/synapse/config/logger.py (L216)
While this can work fine for the normal case of one Synapse instance per
Python process, this configures things globally and isn't compatible
when we try to start multiple Synapse instances because each subsequent
tenant will overwrite the previous tenant.
### What does this PR do?
We remove the `MetadataFilter` and replace it by tracking the
`server_name` in the `LoggingContext` and expose it with our existing
[`LoggingContextFilter`](205d9e4fc4/synapse/logging/context.py (L584-L622))
that we already use to expose information about the `request`.
This means that the `server_name` value follows wherever we log as
expected even when we have multiple Synapse instances running in the
same process.
### A note on logcontext
Anywhere, Synapse mistakenly uses the `sentinel` logcontext to log
something, we won't know which server sent the log. We've been fixing up
`sentinel` logcontext usage as tracked by
https://github.com/element-hq/synapse/issues/18905
Any further `sentinel` logcontext usage we find in the future can be
fixed piecemeal as normal.
d2a966f922/docs/log_contexts.md (L71-L81)
### Testing strategy
1. Adjust your logging config to include `%(server_name)s` in the format
```yaml
formatters:
precise:
format: '%(asctime)s - %(server_name)s - %(name)s - %(lineno)d -
%(levelname)s - %(request)s - %(message)s'
```
1. Start Synapse: `poetry run synapse_homeserver --config-path
homeserver.yaml`
1. Make some requests (`curl
http://localhost:8008/_matrix/client/versions`, etc)
1. Open the homeserver logs and notice the `server_name` in the logs as
expected. `unknown_server_from_sentinel_context` is expected for the
`sentinel` logcontext (things outside of Synapse).
Introduce `Clock.call_when_running(...)` to wrap startup code in a
logcontext, ensuring we can identify which server generated the logs.
Background:
> Ideally, nothing from the Synapse homeserver would be logged against the `sentinel`
> logcontext as we want to know which server the logs came from. In practice, this is not
> always the case yet especially outside of request handling.
>
> Global things outside of Synapse (e.g. Twisted reactor code) should run in the
> `sentinel` logcontext. It's only when it calls into application code that a logcontext
> gets activated. This means the reactor should be started in the `sentinel` logcontext,
> and any time an awaitable yields control back to the reactor, it should reset the
> logcontext to be the `sentinel` logcontext. This is important to avoid leaking the
> current logcontext to the reactor (which would then get picked up and associated with
> the next thing the reactor does).
>
> *-- `docs/log_contexts.md`
Also adds a lint to prefer `Clock.call_when_running(...)` over
`reactor.callWhenRunning(...)`
Part of https://github.com/element-hq/synapse/issues/18905
Spawning from @kegsay [pointing
out](https://matrix.to/#/!cnVVNLKqgUzNTOFQkz:matrix.org/$ExOO7J8uPUQSyH-9Uxc_QCa8jlXX9uK4VRtkSC0EI3o?via=element.io&via=matrix.org&via=jki.re)
that the Sliding Sync endpoint doesn't handle a large room with a lot of
state well on initial sync (requesting all state via `required_state: [
["*","*"] ]`) (it just takes forever).
After investigating further, the slow part is just
`get_events_as_list(...)` fetching all of the current state ID's out for
the room (which can be 100k+ events for rooms with a lot of membership).
This is just a slow thing in Synapse in general and the same thing
happens in Sync v2 or the `/state` endpoint.
---
The only idea I had to improve things was to use `batch_iter` to only
try fetching a fixed amount at a time instead of working with large
maps, lists, and sets. This doesn't seem to have much effect though.
There is already a `batch_iter(event_ids, 200)` in
`_fetch_event_rows(...)` for when we actually have to touch the database
and that's inside a queue to deduplicate work.
I did notice one slight optimization to use `get_events_as_list(...)`
directly instead of `get_events(...)`. `get_events(...)` just turns the
result from `get_events_as_list(...)` into a dict and since we're just
iterating over the events, we don't need the dict/map.
During the migration the automated script to update the copyright
headers accidentally got rid of some of the existing copyright lines.
Reinstate them.
There are a couple of things we need to be careful of here:
1. The current python code does no validation when loading from the DB,
so we need to be careful to ignore such errors (at least on jki.re there
are some old events with internal metadata fields of the wrong type).
2. We want to be memory efficient, as we often have many hundreds of
thousands of events in the cache at a time.
---------
Co-authored-by: Quentin Gliech <quenting@element.io>
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.
Some experimental prep work to enable external event caching based on #9379 & #12955. Doesn't actually move the cache at all, just lays the groundwork for async implemented caches.
Signed off by Nick @ Beeper (@Fizzadar)
I've seen a few errors which can only plausibly be explained by the calculated
event id for an event being different from the ID of the event in the
database. It should be cheap to check this, so let's do so and raise an
exception.
Multiple calls to `EventsWorkerStore._get_events_from_cache_or_db` can
reuse the same database fetch, which is initiated by the first call.
Ensure that cancelling the first call doesn't cancel the other calls
sharing the same database fetch.
Signed-off-by: Sean Quah <seanq@element.io>
The previous fix for the ongoing event fetches counter
(8eec25a1d9) was both insufficient and
incorrect.
When the database is unreachable, `_do_fetch` never gets run and so
`_event_fetch_ongoing` is never decremented.
The previous fix also moved the `_event_fetch_ongoing` decrement outside
of the `_event_fetch_lock` which allowed race conditions to corrupt the
counter.
Empirically, this helped my server considerably when handling gaps in Matrix HQ. The problem was that we would repeatedly call have_seen_events for the same set of (50K or so) auth_events, each of which would take many minutes to complete, even though it's only an index scan.
