cpu usage skyrocketing for no apparent reason

[doums]

Background information

• Dendrite version or git SHA: 0.14.1+40bef6a
• SQLite3 or Postgres?: Postgres
• Running in Docker?: yes

Description

Hi, since few days ago my instance cpu usage is skyrocketing with one core at 100% constantly.
Thanks to docker I'm able to mitigate this by limiting the cpus amount I provide to the container.
The weird thing is, afaik, I didn't change the config recently. Also, presence feature is disabled since the beginning (both in/outbound set to false). I have no clue on what could be the root cause.
Restarting/recreating the container (but preserving any volume config and the db) does not help.
No error or warns in the logs.
All I can see is a decent pressure of http calls to

/_matrix/client/v3/sync?filter=1&timeout=30000&org.matrix.msc1234.use_state_after=true&set_presence=unavailable&since=blabla

Any ideas?

[Player3324]

Same here, it seems to be stuck in github.com/element-hq/dendrite/roomserver/state.(*authEventLoader).loadAuthEvents with most time in runtime.mapaccess2_faststr, which indicates that the queue isn't all that finite.

I used https://github.com/element-hq/dendrite/blob/main/docs/development/PROFILING.md