It turns out that most of the time we were calling have_events, we were only
using half of the result. Replace have_events with have_seen_events and
get_rejection_reasons, so that we can see what's going on a bit more clearly.
This will allow us to measure how often we calculate state deltas in
event persistence that we would have been able to calculate at the same
time we calculated the state for the event.
The race happens when the user joins a room at the same time as doing a
sync. We fetch the current token and then get the rooms the user is in.
If the join happens after the current token, but before we get the rooms
we end up sending down a partial room entry in the sync.
This is fixed by looking at the stream ordering of the membership
returned by get_rooms_for_user, and handling the case when that stream
ordering is after the current token.
* Split state group persist into seperate storage func
* Add per database engine code for state group id gen
* Move store_state_group to StateReadStore
This allows other workers to use it, and so resolve state.
* Hook up store_state_group
* Fix tests
* Rename _store_mult_state_groups_txn
* Rename StateGroupReadStore
* Remove redundant _have_persisted_state_group_txn
* Update comments
* Comment compute_event_context
* Set start val for state_group_id_seq
... otherwise we try to recreate old state groups
* Update comments
* Don't store state for outliers
* Update comment
* Update docstring as state groups are ints
1. use `deferred.errback()` instead of `deferred.errback(e)`, which means that
a Failure object will be constructed using the current exception state,
*including* its stack trace - so the stack trace is saved in the Failure,
leading to better exception reports.
2. Set `consumeErrors=True` on the ObservableDeferred, because we know that
there will always be at least one observer - which avoids a spurious "CRITICAL:
unhandled exception in Deferred" error in the logs
ObserveableDeferred expects its callbacks to be called without any
logcontexts, whereas it turns out we were calling them with the logcontext of
the request which initiated the persistence loop.
It seems wrong that we are attributing work done in the persistence loop to the
request that happened to initiate it, so let's solve this by dropping the
logcontext for it.
(I'm not sure this actually causes any real problems other than messages in the
debug log, but let's clean it up anyway)
Richard van der Hoff