Gunicorn Workers Hang And Memory Does Not Get Deallocated

Hey,

I’m currently running a load test. I’m serving my Starlette app with gunicorn, with this line of code ;

gunicorn -w 48 -k uvicorn.workers.UvicornH11Worker -b 0.0.0.0:5000 app:app --timeout 10 --threads 2

So there is nothing wrong with the performance but the thing is memory consumption gets increased continuously. Even after the test is finished and app is not receiving any more requests, memory does not get deallocated. There was no request that app could not return a response not even an error. Could this be related to uvicorn workers?

[tomchristie]

Generally when I’ve seen any issue along these lines come up, the end result tends to come out as “this is just the Python runtime pre-emptively hanging onto memory that’s been allocated, rather than returning it to the O/S” rather than a memory leak. Eg. https://stackoverflow.com/questions/15455048/releasing-memory-in-python.

That may or may not be the case here, but it’s worth keeping in mind.

Having said that, your comment of “This continues until I run out of memory or restart the process.” does sound suspect.

The way to approach this would be:

• Try running under at least two of hypercorn/daphne/uvicorn to isolate if it replicates under different servers. If it does replicate under different servers, then it’s probably not a server issue.
• Reduce the issue to the simplest possible reproducing case you can - ideally just a starlette app with a single endpoint. Keep trying to remove code and simplify until you’ve got the simplest possible case you can come up with.

[Gavin-Evans]

Fortunately, I also encountered a similar problem, I don’t understand it, how did you solve it?