How to tell when flask server using waitress is overloaded

Question

I have a simple flask application that runs a machine learning model on data sent in a post request to an endpoint (say /predict). The flask app uses waitress in production with the default parameters. Since prediction can take a while I have a readiness endpoint in my application that I would like to reply with a not ready 50x status code when the waitress task queue is greater than some number (lets say 5).

I need to know how to get the size of the waitress's task queue. Waitress does log "Task queue depth is 94" to stdout but I can't find a way to access the value programmatically. I would then use that number to decide whether my server was ready to respond to more requests or if I need to spin up new instances.

@Gordy I ended up creating a custom metric for kubernetes and using that. Waitress doesn't have a good way to do this. — scowan, Feb 10 '21 at 12:31
Isn't the waitress backlog setting meant to do backpressure? — Simon Kohlmeyer, Mar 01 '22 at 15:26

score 0 · Answer 1 · answered Sep 01 '22 at 05:36

This is not probably what you are looking for, but may be a good starting point. Using waitress.serve() you can set some configurations, one of which is the length of backlog you accept to be waiting in the queue. YOu can set waitress.serve(backlog=10) which means 10 requests can stay in the queue. The default value is 1024. However, to answer how to monitor the current number of request in the queue, this will not probably give you the answer. Ref

How to tell when flask server using waitress is overloaded

1 Answers1