Description

Prior issue over at element-web which has been closed as being a synapse issue rather than element-web's: element-hq/element-web#25874

Since element-web switched to doing presence using /sync vs using the presence API my synapse is using extremely high amounts of CPU due to presence requests happening. It also seems to kill both fed senders from the amount of work they have to do (I assume it is just the metrics endpoint timing out since kubernetes doesnt see them as broken).

The suspicion it being caused by the element-web change is because the peaks align perfectly with the time I start my element-web clients.

Steps to reproduce

• I believe just opening develop.element.io is enough.

Homeserver

matrix.midnightthoughts.space

Synapse Version

{"server_version":"1.87.0","python_version":"3.11.4"}

Installation Method

Docker (matrixdotorg/synapse)

Database

Postgresql 15 server with active + standby replication nodes

Workers

Single process

Platform

The server is running on a mixed arch (x86_64+arm64) kubernetes cluster.

Configuration

The config is at https://git.nordgedanken.dev/kubernetes/gitops/src/branch/main/apps/production/synapse-midnightthoughts.yaml#L30-L41 mainly there are 4 (3 non working) appservices

Relevant log output

Attached as a file as the log is too long for a comment

Anything else that would be useful to know?

There is an element-web rageshake available hooked to the other issue.

The pasted log is the last hour. Specifically interesting is the high amount of replication around presence_set_state.