The conversion of `device_lists_changes_in_room` rows into
`device_lists_outbound_pokes` is DB-bound on
`mark_redundant_device_lists_pokes`. The `UPDATE` uses the `(room_id,
stream_id)` index, so each call scans the unconverted backlog, getting
slower the further behind the conversion is.
Add a partial index matching the query via a background update, and skip
the (safe-to-skip) `UPDATE` until the index has been built. Also add a
metric reporting how far behind the conversion is, using the existing
`inserted_ts` column.
Fixes https://github.com/element-hq/backend-internal/issues/286
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Eric Eastwood <erice@element.io>
Follows: #20027
Noticed in https://github.com/element-hq/synapse/pull/20003
The problem was that `--no-root` prevents reinstalling the root project
(Synapse).
However, since we just did a `git checkout`, we need to reinstall the
root project in case the Rust code changed, as the
`poetry install` command is what causes the Rust to be recompiled.
Doing otherwise causes the 'Rust module outdated' error when importing
`synapse`.
---------
Signed-off-by: Olivier 'reivilibre <oliverw@matrix.org>
Issue spotted in: https://github.com/element-hq/synapse/pull/20098
Follows: #20027
We already use `faketime` for Postgres, but I forgot that the SQLite
schema delta would have the same problem
and somehow tuned it out of the preview diff on the original PR.
---------
Signed-off-by: Olivier 'reivilibre <oliverw@matrix.org>
We have an internal usage of `/scheduled_tasks` that would like to fetch
multiple actions at once (janitor).
We also make it so that invalid `status` values now return a 400 rather
than a 500.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
It seems a lot of time in our trial tests goes towards setting up the
database. (The same is probably true of Complement too)
We haven't done a full schema for about 20 schema versions, so no
surprise!
As a result, I want to produce a full schema soon.
In this PR I dust off `make_full_schema.sh` (which seems to have broken
after some SQLite changes)
and add a CI workflow that runs it (producing a diff) when someone
changes the schema.
The CI workflow also adds a sticky comment showing the diff on the
schema, so you can better appreciate the final effect of a change.
---
**Dead changes:**
I wanted to make it possible to generate a versioned full schema without
the manual work,
but you can't run the background updates without essentially starting up
a homeserver,
at which point it might fail because you haven't run all the deltas yet.
There's no actual good way to do this, short of deleting the latest
deltas (+ tweaking code to not crash without them)
or rolling back in the git history.
Backed out those changes, but they're preserved on the PR if interesting.
---------
Signed-off-by: Olivier 'reivilibre <oliverw@matrix.org>
*Spawning from
https://github.com/element-hq/synapse/pull/19057#discussion_r2427537811,*
Fix tests that use `homeserver_to_use=GenericWorkerServer` not being
able to be run standalone
Fix https://github.com/element-hq/synapse/issues/15671 (previously
https://github.com/matrix-org/synapse/issues/15671)
Before this change:
```shell
$ poetry run trial tests.storage.test_rollback_worker.WorkerSchemaTests.test_rolling_back
tests.storage.test_rollback_worker
WorkerSchemaTests
test_rolling_back ... [ERROR]
===============================================================================
[ERROR]
Traceback (most recent call last):
File "synapse/tests/unittest.py", line 129, in new
return code(orig, *args, **kwargs)
File "synapse/tests/unittest.py", line 223, in setUp
return orig()
File "synapse/tests/unittest.py", line 398, in setUp
self.hs = self.make_homeserver(self.reactor, self.clock)
File "synapse/tests/storage/test_rollback_worker.py", line 54, in make_homeserver
hs = self.setup_test_homeserver(homeserver_to_use=GenericWorkerServer)
File "synapse/tests/unittest.py", line 669, in setup_test_homeserver
hs = setup_test_homeserver(
File "synapse/tests/server.py", line 1260, in setup_test_homeserver
prepare_database(
File "synapse/synapse/storage/prepare_database.py", line 167, in prepare_database
raise UpgradeDatabaseException(EMPTY_DATABASE_ON_WORKER_ERROR)
synapse.storage.prepare_database.UpgradeDatabaseException: Uninitialised database: run the main synapse process to prepare the database schema before starting worker processes.
tests.storage.test_rollback_worker.WorkerSchemaTests.test_rolling_back
-------------------------------------------------------------------------------
Ran 1 tests in 0.034s
FAILED (errors=1)
```
### What was the problem before?
[`PREPPED_SQLITE_DB_CONN`](https://github.com/element-hq/synapse/blob/1a1af7b622f219ba0f2501709298caa230ad3912/tests/server.py#L1247-L1262)
is a process global and shared between all tests. Whichever test first
calls `setup_test_homeserver(...)` builds the template database for the
whole trial run.
`prepare_database(...)` has a built-in check to refuse upgrading the
database ["to avoid multiple workers doing it at
once."](https://github.com/element-hq/synapse/blob/1a1af7b622f219ba0f2501709298caa230ad3912/synapse/storage/prepare_database.py#L164-L167)
and throw `UpgradeDatabaseException`.
This means that if we happen to first run a test that uses a worker
(`homeserver_to_use=GenericWorkerServer`), `prepare_database(...)` will
just throw its `UpgradeDatabaseException`. And since
`PREPPED_SQLITE_DB_CONN` is assigned before `prepare_database(...)`, it
will never try to prepare again and the rest of the tests will fail.
The registration of `QuarantinedMediaStream` in
`ReplicationCommandHandler._streams_to_replicate` was missed when the
stream was added, so an instance configured as the
quarantined_media_changes stream writer never sent RDATA/POSITION for it
unless it was the main process.
Also add the stream to the `instance_map` config validation.
Stream was introduced in
https://github.com/element-hq/synapse/pull/19558
Fixes https://github.com/element-hq/synapse/issues/20080
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Set `idle_in_transaction_session_timeout` on new postgres connections
Idle transactions can block maintenance tasks server-side like vacuums,
which can lead to bloat and performance issues.
We should never hit this timeout in normal operation, as Synapse should
always be actively using the connection when in a transaction and so it
should only ever be briefly idle. If we do hit this timeout, it's likely
that no progress is being made and so aborting the session is safe.
In certain cases we have seen connections leak, particularly when using
a connection pooler like pgcat, and this timeout will help with that.
---------
Co-authored-by: Eric Eastwood <erice@element.io>
Broken out of #20027 because I'd like to have it land first.
> It seems a lot of time in our trial tests goes towards setting up the
database. (The same is probably true of Complement too)
>
> We haven't done a full schema for about 20 schema versions, so no
surprise!
>
> As a result, I want to produce a full schema soon.
When running `make_full_schema.sh`, these drop statements now cause the
error:
```
Parse error near line 2: table event_search_content may not be dropped
Parse error near line 3: table event_search_segments may not be dropped
Parse error near line 4: table event_search_segdir may not be dropped
Parse error near line 5: table event_search_docsize may not be dropped
Parse error near line 6: table event_search_stat may not be dropped
Parse error near line 7: table user_directory_search_content may not be dropped
Parse error near line 8: table user_directory_search_segments may not be dropped
Parse error near line 9: table user_directory_search_segdir may not be dropped
Parse error near line 10: table user_directory_search_docsize may not be dropped
Parse error near line 11: table user_directory_search_stat may not be dropped
```
It seems SQLite has cracked down on code that edits the internal tables.
Because SQLite dumps the schema with `CREATE TABLE IF NOT EXISTS` for
these virtual tables, it's harmless to leave them in the schema dump.
---------
Signed-off-by: Olivier 'reivilibre' <oliverw@matrix.org>
To make it easy to share to people and aid in debugging when they run into
performance issues.
Unfortunately, the Grafana UI doesn't make the export or import steps
easy so it involves some manual Grafana API calls.
As first explored in
https://github.com/element-hq/synapse-rust-apps/pull/397
Kegan Dougal