Docker: Variable for which worker(s) to run

[turt2live]

Use case: Automatically scaling the parts of synapse that can be scaled, such as the event creator and synchrotrons. It would be desirable in this situation to have 1 worker per container.

[jfrederickson]

What do you think about this?

• A toggle var like SYNAPSE_ENABLE_REPLICATION to enable/disable the replication interface
• SYNAPSE_REPLICATION_BIND_ADDRESS defaulting to 127.0.0.1 (Does this even make sense in the context of Docker? I'm not sure... in Kubernetes you might bind to localhost and run all workers in one pod, or use NetworkPolicies to restrict access.)
• SYNAPSE_TCP_REPLICATION_PORT defaulting to 9092
• SYNAPSE_HTTP_REPLICATION_PORT defaulting to 9093

And then for the actual worker configs... does it make sense to try and template them out, or would it be better to just expect them to live in a directory somewhere? The former sounds simpler for administrators to deal with, but I don't know how stable those configs are.

[turt2live]

Something like that makes sense, although I did find that with worker setups the config generally becomes more complicated anyways. In the end I actually worked around the issue: https://github.com/t2bot/docker-synapse-worker

[richvdh]

As per #5518, we're going to recommend people configure this in a proper config file, as maintaining environment variables for every option is getting unwieldy, hard to maintain, and harder to write documentation for.