/_matrix/client/v3/sync never finishes

[verymilan]

Description

Hi there, as of around january 9th, I have a botuser (mautrix-whatsapp) where the sync request seemingly never finishes. As of right now, it is taking about 8,5 hours and it's still going. Not even a goaway. If I reuse that since token with my auth token, it takes about 5s.

My workers don't seem to be under high load, but Redis curiously happily jumps to 500% cpu – which is not affected by that sync request as far as i can tell.

Steps to reproduce

unknown. since=s394247408_1207796618_5984102_309180540_44583100_66668_16267554_206909845_0_14158

Homeserver

https://matrix.tchncs.de

Synapse Version

1.99.0

Installation Method

pip (from PyPI)

Database

PostgreSQL

Workers

Multiple workers

Platform

Debian 12

Configuration

No response

Relevant log output

possibly

synapse.http.client - 928 - INFO - GET-545154 - Received response to POST synapse-replication://master/_synapse/replication/presence_set_state/%40whatsappbot%3Atchncs.de: 200
synapse.util.caches.response_cache - 274 - INFO - GET-545154 - [sync]: using incomplete cached result for [(@whatsappbot:tchncs.de, 30000, 's394247408_1207796618_5984102_309180540_44583100_66668_16267554_206909845_0_14158', '0', False, 'RDMCFJKLSJ', None)]

the testrequest from above is now on WARNING again, maybe should have left it INFO or something.

Anything else that would be useful to know?

No response