Documentation: What does worker count entail?

[jiw-mh]

When extending the worker count I was wondering what kind of worker it is. My assumption is that the lifecycle hooks are run for each worker? Are they separate threads? Do they have memory/cpu limitations applied. These sort of questions may seem obvious to anyone working with this for a long time, but as a new user it might be nice to have an explanation on it.

[nggit]

Try print(os.getpid()). If they are different then worker means process.

[jiw-mh]

What I am trying to get is that yes: I can investigate how it works, but wouldn't it be good to also summarize how the variables are expected to work? Having the ramifications summarized seemed like a nice addition.

[nggit]

Got it. That's good. Now, would you like to volunteer to write the documentation?

[jiw-mh]

I do not have answers to a bunch of questions, which makes it hard to write the documentation: #2634 (comment)

[jiw-mh]

The questions regarding workers that have come to my mind (again: the answers may be straightforward. It seems like it would be good to have an answer for them in one place):

• Is the lifecycle script run once per worker?
• Are all workers expected to be start immediately or based on load?
• How are requests divided up to workers? Does it use round-robin? Is it based on cpu/memory load?
• How is CPU/memory divided between the workers? (Is there a way to limit memory/CPU per worker?)
• If one worker breaks for some reason: will it stop all other workers?
• Is there a way for server to know in which worker it runs?
• What factors do I need to consider when selecting workers? (Is it best to have one per CPU?)
• Is --limit-concurrency per worker or for the cluster?
• What happens if a worker dies?

Here is a stack-overflow link trying to cover the topic to some parts: https://stackoverflow.com/a/76240953

[jiw-mh]

Answers I think I have figured out but I am not sure if that is correct.

• Is the lifecycle script run once per worker? → Yes, started once per worker.
• Are all workers expected to be start immediately or based on load? → All are started at system start.
• Is there a way for server to know in which worker it runs? → Every worker is getting its own system thread. os.getpid() gives the process id of the thread but not if its the nth of m (Didn't find an environment variable or so)