MSC4140: drop delayed state interference#19495
MSC4140: drop delayed state interference#19495AndrewFerr wants to merge 3 commits intoelement-hq:developfrom
Conversation
No longer cancel a user's delayed state events by a more recent state event sent from a different user. This behaviour has been delegated to an "alternative" in the delayed event MSC for some time now, and is not known to be used by any module other than experimental iterations of MatrixRTC.
reivilibre
left a comment
reivilibre
left a comment
There was a problem hiding this comment.
This makes sense, just need to go through the slightly tedious motions for the schema version, I think
| CREATE INDEX delayed_events_is_processed ON delayed_events (is_processed); | ||
| CREATE INDEX delayed_events_room_state_event_idx ON delayed_events (room_id, event_type, state_key) WHERE state_key IS NOT NULL; | ||
|
|
||
| -- TODO: delete this table in the next Synapse release that can handle a breaking DB change |
There was a problem hiding this comment.
This is possibly better tracked as a GitHub issue instead of a comment here.
More concretely, we need to study https://element-hq.github.io/synapse/latest/development/database_schema.html#synapse-schema-versions
There's a worked example with a table removal. Since I think it's fine to have degraded rollback support for this experimental feature (as in, I think it's probably OK to stop reading and writing to this table at the same time), I think for us this is something along the lines of:
-
SCHEMA_VERSIONneeds to be increased now, with a comment on the version saying that in the newSCHEMA_VERSIONwe no longer use this table. -
In a later version, we can make
SCHEMA_COMPAT_VERSIONmatch that as we drop the table, preventing rollback to the olderSCHEMA_VERSION.
There was a problem hiding this comment.
SCHEMA_VERSIONneeds to be increased now, with a comment on the version saying that in the newSCHEMA_VERSIONwe no longer use this table.
Since the schema isn't actually changed by this PR, would it be acceptable to put that comment on the latest SCHEMA_VERSION (which is a newer version than when the table was created) instead of creating an "empty" bump?
Co-authored-by: Olivier 'reivilibre' <olivier@librepush.net>
Don't test that a user's delayed state events are cancelled by a more recent state event from a different user. That behaviour [has been delegated to an "alternative"] (matrix-org/matrix-spec-proposals@3efecea) in the delayed event MSC4140 for some time now, and may be removed from Synapse ([Synapse PR](element-hq/synapse#19495))
2 participants
No longer cancel a user's delayed state events by a more recent state event sent from a different user. This behaviour has been delegated to an "alternative" in the delayed event MSC for some time now, and is not known to be used by any module other than experimental iterations of MatrixRTC.
See also: matrix-org/complement#845
Pull Request Checklist
EventStoretoEventWorkerStore.".code blocks.