Monitoring the number of idle/busy workers?

[slava-pasedko]

We've got a big Django monolith that is deployed across 4 big vm's with each having 40+ workers running sync gunicorn.

We have a lot of integrations with different internal systems through HTTP.
We all know that sync worker is blocked until the integration's done. Sometimes our application's performance degrades because of slow responses from internal system.

I would like to know, is there any way to monitor the number of idle/busy workers in a moment? This kind of metric would indicate that we lack workers at the moment of related system's performance degradation.

If it is not available - please could you guide me how could I implement it? Is it possible to implement using gunicorn hooks f.e.?
Best regards, thank you for this project.

[benoitc]

there is no concept of idle / busy workers. The number of running sync woorkers gives you the number of simultaneous requests handled in guncorn. I think the right metric to scale is to check the duration of each requests and load. If requests time increase then it's a good moment to spawn a new machine or increase a worker. I would advise to use a finite number of workers per. vm/machine . Start low then increase until a max and r keep it once you get a good performance per machine.