Memory usage

[wwatu1337]

I'm interested in what is your memory usage, with supervisor and default config I see about 450mb memory usage without any traffic, seems a bit high.

[miggland]

I'm on docker, total of four containers (excluding Traefik proxy), using about 750 MB of memory.

[SchrodingersGat]

@wwatu1337 are you running the debug server or gunicorn?

[wwatu1337]

Gunicorn

[matmair]

@wwatu1337 that is somewhat normal for Django apps. In the normal setup it is running 1 webserver, 1 app-server, 4 worker + you propably have a database on there too. The good news is that it scales pretty well on kubernetes if you adjust a few things.

[matmair]

@wwatu1337 you can tune the number of worker processes in the supervisor setup or with the settings - that might help reduce the footprint.

[wwatu1337]

Default of numprocs is 1, as supervisorctl status returns only 1 cluster process is started, these are child processes spawned by manage.py, or is there any way in configurating those?

I have now further looked into this, I've set up a local docker env, while the server process has similar memory usage the worker process is very different, the memory usage is only around 150-200mb, this seems way more reasonable, it still spawns 8 processes as seen on the image, but the memory usage is very different, so something must definately be off, but out of ideas what that might be.

[matmair]

@wwatu1337 that number is way more in line with what I see. You can set the number of workers in the config.yaml.

[wwatu1337]

with only 1 worker, the memory usage drops about 200mb, but I would be more interested in finding out what causes the huge memory usage, any suggestions on what to try/look for?

[matmair]

@wwatu1337 tbh I find memory profiling with Django somewhat cumbersome and never bothered up to now as it wasn't a real problem. There is an extension out there that can be used with django-debug-toolbar but not sure if that will work with the worker.

You might also be able to hook into the python runtime with New Relic and analyze that way. I have not done that my self, colleagues have told me about the option.

[matmair]

@wwatu1337 could you please select an answer or let us know what you want further answered?

[wwatu1337]

sorry for late reply, just moved from bare metal to package manager installation, memory usage out of the box is around 2GB, setting gunicorn workers to 1 via
inventree config:set GUNICORN_CMD_ARGS='--workers=1 --access-logfile -'
brings the memory down to acceptable 1GB