Advice for Production Tuning

[keithhackbarth]

Looking for general advice on how to think about tuning for optimized production usage of uvicorn+gunicorn?

• What things I should be monitoring for?
• How to know if workers is configured correctly? (ie, what would be the sign that we have too many)
• Other recommendation settings such as backlog, timeout, etc.?
• If my server is occasionally throwing a 502 on 1 out of 10,000 queries how to debug?
• Ideal CPU usage (average, max)

[acdha]

I was also thinking it would be useful to have more discussion about gunicorn, too. In testing I've found it to be less stable in a containerized environment and slower (I suspect this needs to involve tuning the backlogs in both cases).