Refactor how the `EventContext` class works, with the intention of reducing the amount of state we fetch from the DB during event processing.
The idea here is to get rid of the cached `current_state_ids` and `prev_state_ids` that live in the `EventContext`, and instead defer straight to the database (and its caching).
One change that may have a noticeable effect is that we now no longer prefill the `get_current_state_ids` cache on a state change. However, that query is relatively light, since its just a case of reading a table from the DB (unlike fetching state at an event which is more heavyweight). For deployments with workers this cache isn't even used.
Part of #12684
When we join a room via the faster-joins mechanism, we end up with "partial
state" at some points on the event DAG. Many parts of the codebase need to
wait for the full state to load. So, we implement a mechanism to keep track of
which events have partial state, and wait for them to be fully-populated.
We work through all the events with partial state, updating the state at each
of them. Once it's done, we recalculate the state for the whole room, and then
mark the room as having complete state.
Refactor and convert `Linearizer` to async. This makes a `Linearizer`
cancellation bug easier to fix.
Also refactor to use an async context manager, which eliminates an
unlikely footgun where code that doesn't immediately use the context
manager could forget to release the lock.
Signed-off-by: Sean Quah <seanq@element.io>
If we're missing most of the events in the room state, then we may as well call the /state endpoint, instead of individually requesting each and every event.
When we get a partial_state response from send_join, store information in the
database about it:
* store a record about the room as a whole having partial state, and stash the
list of member servers too.
* flag the join event itself as having partial state
* also, for any new events whose prev-events are partial-stated, note that
they will *also* be partial-stated.
We don't yet make any attempt to interpret this data, so API calls (and a bunch
of other things) are just going to get incorrect data.
msc3706 proposes changing the `/send_join` response:
> Any events returned within `state` can be omitted from `auth_chain`.
Currently, we rely on `m.room.create` being returned in `auth_chain`, but since
the `m.room.create` event must necessarily be part of the state, the above
change will break this.
In short, let's look for `m.room.create` in `state` rather than `auth_chain`.
I've never found this terribly useful. I think it was added in the early days
of Synapse, without much thought as to what would actually be useful to log,
and has just been cargo-culted ever since.
Rather, it tends to clutter up debug logs with useless information.
* `_auth_and_persist_outliers`: mark persisted events as outliers
Mark any events that get persisted via `_auth_and_persist_outliers` as, well,
outliers.
Currently this will be a no-op as everything will already be flagged as an
outlier, but I'm going to change that.
* `process_remote_join`: stop flagging as outlier
The events are now flagged as outliers later on, by `_auth_and_persist_outliers`.
* `send_join`: remove `outlier=True`
The events created here are returned in the result of `send_join` to
`FederationHandler.do_invite_join`. From there they are passed into
`FederationEventHandler.process_remote_join`, which passes them to
`_auth_and_persist_outliers`... which sets the `outlier` flag.
* `get_event_auth`: remove `outlier=True`
stop flagging the events returned by `get_event_auth` as outliers. This method
is only called by `_get_remote_auth_chain_for_event`, which passes the results
into `_auth_and_persist_outliers`, which will flag them as outliers.
* `_get_remote_auth_chain_for_event`: remove `outlier=True`
we pass all the events into `_auth_and_persist_outliers`, which will now flag
the events as outliers.
* `_check_sigs_and_hash_and_fetch`: remove unused `outlier` parameter
This param is now never set to True, so we can remove it.
* `_check_sigs_and_hash_and_fetch_one`: remove unused `outlier` param
This is no longer set anywhere, so we can remove it.
* `get_pdu`: remove unused `outlier` parameter
... and chase it down into `get_pdu_from_destination_raw`.
* `event_from_pdu_json`: remove redundant `outlier` param
This is never set to `True`, so can be removed.
* changelog
* update docstring
Events returned by `backfill` should not be flagged as outliers.
Fixes:
```
AssertionError: null
File "synapse/handlers/federation.py", line 313, in try_backfill
dom, room_id, limit=100, extremities=extremities
File "synapse/handlers/federation_event.py", line 517, in backfill
await self._process_pulled_events(dest, events, backfilled=True)
File "synapse/handlers/federation_event.py", line 642, in _process_pulled_events
await self._process_pulled_event(origin, ev, backfilled=backfilled)
File "synapse/handlers/federation_event.py", line 669, in _process_pulled_event
assert not event.internal_metadata.is_outlier()
```
See https://sentry.matrix.org/sentry/synapse-matrixorg/issues/231992Fixes#8894.
* Push `get_room_{min,max_stream_ordering}` into StreamStore
Both implementations of this are identical, so we may as well push it down and
get rid of the abstract base class nonsense.
* Remove redundant `StreamStore` class
This is empty now
* Remove redundant `get_current_events_token`
This was an exact duplicate of `get_room_max_stream_ordering`, so let's get rid
of it.
* newsfile
This is the final piece of the jigsaw for #9595. As with other changes before this one (eg #10771), we need to make sure that we auth the auth events in the right order, and actually check that their predecessors haven't been rejected.
To do this I've reused the existing code we use when persisting outliers elsewhere.
I've removed the code for attempting to fetch missing auth_events - the events should have been present in the send_join response, so the likely reason they are missing is that we couldn't verify them, so requesting them again is unlikely to help. Instead, we simply drop any state which relies on those auth events, as we do at a backwards-extremity. See also matrix-org/complement#216 for a test for this.
Currently, when we receive an event whose auth_events differ from those we expect, we state-resolve between the two state sets, and check that the event passes auth based on the resolved state.
This means that it's possible for us to accept events which don't pass auth at their declared auth_events (or where the auth events themselves were rejected), leading to problems down the line like #10083.
This change means we will:
* ignore any events where we cannot find the auth events
* reject any events whose auth events were rejected
* reject any events which do not pass auth at their declared auth_events.
Together with a whole raft of previous work, this is a partial fix to #9595.
Fixes#6643.
Based on #11009.
This fixes a bug where we would accept an event whose `auth_events` include
rejected events, if the rejected event was shadowed by another `auth_event`
with same `(type, state_key)`.
The approach is to pass a list of auth events into
`check_auth_rules_for_event` instead of a dict, which of course means updating
the call sites.
This is an extension of #10956.
Found while working on the Gitter backfill script and noticed
it only happened after we sent 7 batches, https://gitlab.com/gitterHQ/webapp/-/merge_requests/2229#note_665906390
When there are more than 5 backward extremities for a given depth,
backfill will throw an error because we sliced the extremity list
to 5 but then try to iterate over the full list. This causes
us to look for state that we never fetched and we get a `KeyError`.
Before when calling `/messages` when there are more than 5 backward extremities:
```
Traceback (most recent call last):
File "/usr/local/lib/python3.8/site-packages/synapse/http/server.py", line 258, in _async_render_wrapper
callback_return = await self._async_render(request)
File "/usr/local/lib/python3.8/site-packages/synapse/http/server.py", line 446, in _async_render
callback_return = await raw_callback_return
File "/usr/local/lib/python3.8/site-packages/synapse/rest/client/room.py", line 580, in on_GET
msgs = await self.pagination_handler.get_messages(
File "/usr/local/lib/python3.8/site-packages/synapse/handlers/pagination.py", line 396, in get_messages
await self.hs.get_federation_handler().maybe_backfill(
File "/usr/local/lib/python3.8/site-packages/synapse/handlers/federation.py", line 133, in maybe_backfill
return await self._maybe_backfill_inner(room_id, current_depth, limit)
File "/usr/local/lib/python3.8/site-packages/synapse/handlers/federation.py", line 386, in _maybe_backfill_inner
likely_extremeties_domains = get_domains_from_state(states[e_id])
KeyError: '$zpFflMEBtZdgcMQWTakaVItTLMjLFdKcRWUPHbbSZJl'
```
We correctly allowed using the MSC2716 batch endpoint for
the room creator in existing room versions but accidentally didn't track
the events because of a logic flaw.
This prevented you from connecting subsequent chunks together because it would
throw the unknown batch ID error.
We only want to process MSC2716 events when:
- The room version supports MSC2716
- Any room where the homeserver has the `msc2716_enabled` experimental feature enabled and the event is from the room creator
`_check_event_auth` is only called in two places, and only one of those sets
`send_on_behalf_of`. Warming the cache isn't really part of auth anyway, so
moving it out makes a lot more sense.
There's little point in doing a fancy state reconciliation dance if the event
itself is invalid.
Likewise, there's no point checking it again in `_check_for_soft_fail`.
Broadly, the existing `event_auth.check` function has two parts:
* a validation section: checks that the event isn't too big, that it has the rught signatures, etc.
This bit is independent of the rest of the state in the room, and so need only be done once
for each event.
* an auth section: ensures that the event is allowed, given the rest of the state in the room.
This gets done multiple times, against various sets of room state, because it forms part of
the state res algorithm.
Currently, this is implemented with `do_sig_check` and `do_size_check` parameters, but I think
that makes everything hard to follow. Instead, we split the function in two and call each part
separately where it is needed.
* Inline `_check_event_auth` for outliers
When we are persisting an outlier, most of `_check_event_auth` is redundant:
* `_update_auth_events_and_context_for_auth` does nothing, because the
`input_auth_events` are (now) exactly the event's auth_events,
which means that `missing_auth` is empty.
* we don't care about soft-fail, kicking guest users or `send_on_behalf_of`
for outliers
... so the only thing that matters is the auth itself, so let's just do that.
* `_auth_and_persist_fetched_events_inner`: de-async `prep`
`prep` no longer calls any `async` methods, so let's make it synchronous.
* Simplify `_check_event_auth`
We no longer need to support outliers here, which makes things rather simpler.
* changelog
* lint
* Factor more stuff out of `_get_events_and_persist`
It turns out that the event-sorting algorithm in `_get_events_and_persist` is
also useful in other circumstances. Here we move the current
`_auth_and_persist_fetched_events` to `_auth_and_persist_fetched_events_inner`,
and then factor the sorting part out to `_auth_and_persist_fetched_events`.
* `_get_remote_auth_chain_for_event`: remove redundant `outlier` assignment
`get_event_auth` returns events with the outlier flag already set, so this is
redundant (though we need to update a test where `get_event_auth` is mocked).
* `_get_remote_auth_chain_for_event`: move existing-event tests earlier
Move a couple of tests outside the loop. This is a bit inefficient for now, but
a future commit will make it better. It should be functionally identical.
* `_get_remote_auth_chain_for_event`: use `_auth_and_persist_fetched_events`
We can use the same codepath for persisting the events fetched as part of an
auth chain as for those fetched individually by `_get_events_and_persist` for
building the state at a backwards extremity.
* `_get_remote_auth_chain_for_event`: use a dict for efficiency
`_auth_and_persist_fetched_events` sorts the events itself, so we no longer
need to care about maintaining the ordering from `get_event_auth` (and no
longer need to sort by depth in `get_event_auth`).
That means that we can use a map, making it easier to filter out events we
already have, etc.
* changelog
* `_auth_and_persist_fetched_events`: improve docstring
Combine the two loops over the list of events, and hence get rid of
`_NewEventInfo`. Also pass the event back alongside the context, so that it's
easier to process the result.
* Reload auth events from db after fetching and persisting
In `_update_auth_events_and_context_for_auth`, when we fetch the remote auth
tree and persist the returned events: load the missing events from the database
rather than using the copies we got from the remote server.
This is mostly in preparation for additional refactors, but does have an
advantage in that if we later get around to checking the rejected status, we'll
be able to make use of it.
* Factor out `_get_remote_auth_chain_for_event` from `_update_auth_events_and_context_for_auth`
* changelog
Constructing an EventContext for an outlier is actually really simple, and
there's no sense in going via an `async` method in the `StateHandler`.
This also means that we can resolve a bunch of FIXMEs.
Adds missing type hints to methods in the synapse.handlers
module and requires all methods to have type hints there.
This also removes the unused construct_auth_difference method
from the FederationHandler.
I think I have finally teased apart the codepaths which handle outliers, and those that handle non-outliers.
Let's add some assertions to demonstrate my newfound knowledge.
If we're persisting an event E which has auth_events A1, A2, then we ought to make sure that we correctly auth
and persist A1 and A2, before we blindly accept E.
This PR does part of that - it persists the auth events first - but it does not fully solve the problem, because we
still don't check that the auth events weren't rejected.
