Pusher instances not starting from v1.29.0

[roeltm]

Description

After updating to v1.29.0 the pusher pool no longer starts the worker.

Steps to reproduce

On v1.28.0 I had the following configuration:

start_pusher: False
instance_map:
  push_worker:
    host: <sygnal-host>
    port: <sygnal-port>

pusher_instances:
  - push_worker

which worked fine.

After updating to v1.29.0 it no longer starts a worker:

synapse.push.pusherpool - 92 - INFO - sentinel - Not starting pushers because they are disabled in the config

After changing the name of the worker to master it does however work.

I suspect it happened with the following PR: #9466

Version information

• Version: v1.29.0

• Install method: docker

[reivilibre]

Hi there, quite a few things confuse me about your issue.

To start with, v1.29.0 is fairly old (from March). It's difficult to support this issue, since it's possible that the way workers work may have changed — I'll assume you're using Redis-based workers.

I take it the configuration you've quoted is an excerpt from your overall homeserver.yaml?

instance_map:
  push_worker:
    host: <sygnal-host>
    port: <sygnal-port>

seems confused. This is not a place to put a Sygnal host and port (nowhere in Synapse's configuration needs to have Sygnal's host and port). At best, this would be the place to put a host and port of a Synapse pusher worker — but it looks like you don't need to do this for pusher workers because they don't have HTTP replication listeners.

After changing the name of the worker to master it does however work.

I don't really know what this means — where are you changing the name?

I don't know where you quoted the log line from — the master's log? If so, this is to be expected, since start_pusher: False disables the pushers from running on the master process (because you should be supplying a separate push worker to handle that).

[roeltm]

Ah, wait. I misunderstood some things. You are right, it doesn't make sense. Apologies.