admin/purgeRoom requires server restart

matrix-org / dendrite

Dendrite is a second-generation Matrix homeserver written in Go!

https://matrix-org.github.io/dendrite/

Apache License 2.0

5.67k stars 664 forks source link

admin/purgeRoom requires server restart #3190

Open bones-was-here opened 1 year ago

bones-was-here commented 1 year ago

Background information

Dendrite version or git SHA: 0.13.2
SQLite3 or Postgres?: postgres
Running in Docker?: no
go version: 1.21
Client used (if applicable): element web

Description

What is the problem: After purging a room, dendrite must be restarted before the room can be rejoined
Who is affected: dendrite admins
How is this bug manifesting: When attempting to rejoin the room, a big spam of errors is logged and the join fails.
When did this first appear: unknown

Steps to reproduce

evacuate/leave and then purge a room with other servers participating
rejoin the room without first restarting dendrite

S7evinK commented 1 year ago

I'm not sure if/how we should solve this without slowing down the purge. The problem is most likely that the cache still has several events cached and uses them when trying to join a room. Given we're purging by room_id and roomNID, we don't have a way to also clear the cache for said events (we don't know which event_id/eventNID was deleted).

bones-was-here commented 11 months ago

If you can't invalidate the specific set of cached events, it might be best to invalidate the entire cache, for now. Losing some perf temporarily seems better than failing inscrutably and forcing the admin to restart the server.