As a user, I can specify how long workers should wait during graceful shutdown

[bmbouter]

See this comment which discusses exposing the TASK_GRACE_INTERVAL as a setting. This is an RFE requesting that.

I don't know how long a heartbeat is so if the setting is expressed in heartbeats (which I think I'm ok with), then I need to know how long that is.

As a user, it would be nicer to me if it was expressable in seconds. For example we also have k8s hard timeout values and those are expressed in seconds, and we need to set these similarly and together, so having them all on a seconds timescale is nicer for me. It'll be ok though if not.

[mdellweg]

Well heartbeat duration is afaik a setting.
But also this grace time is not a precise time interval. It is measured by counting the ticking by of heartbeats. So the actual resulting time depends on where in the rhythm of the heart an event occured.

If we want more precise timing of stuff we need to rewrite the workers using async python.

[bmbouter]

I'm supportive to position the setting as a number of heartbeats and not "a specific number of seconds" if that makes it easier.

[mdellweg]

I think specifying seconds is fine. We just need to document that it's not a precise measure.

[mdellweg]

I just saw, that on sending SIGTERM or SIGHUP the current task should be awaited forever.
Only on SIGINT it will be killed after a grace time. (The second SIGINT will kill the task immediately.)