Support for configuring logic around worker exit to support Prometheus client multiprocessing

[Dominick-Peluso-Bose]

If you are using granian with multiple workers, I think to make use of Prometheus client we would need a way to hook into the lifecycle of when a worker exits to clean up Prometheus client multiprocessing.

In gunicorn, you add something like this to your gunicorn configuration file:

from prometheus_client import multiprocess

def child_exit(server, worker):
    multiprocess.mark_process_dead(worker.pid)

It would be great if granian added a way to hook into events like that. Another helpful one would be when the server starts.

[gi0baro]

That sounds smth you can already do overriding the spawning method thanks to the spawn_target parameter of Granian.serve, like:

import os

from granian import Granian
from prometheus_client import multiprocess

def custom_spawn(*args, **kwargs):
    Granian._spawn_wsgi_worker(*args, **kwargs)
    multiprocess.work_process_dead(os.getpid())

if __name__ == "__main__":
    Granian("mymodule:app", interface="wsgi").serve(spawn_target=custom_spawn)

of course you can switch wsgi with asgi and rsgi interfaces.

But my question would be: since granian doesn't re-spawn processes (at least right now) is this needed? I mean, if the worker is shutting down, it also implicitly means the entire server/app is shutting down, so you still won't produce any metrics in prometheus..