Semaphore leak when running CLI

[OReyesMatte]

Hi!

I tried running trackstra using the CLI tool on a custom segmented timelapse. However, halfway through the feature extraction, it stops and gives the following error message:

zsh: segmentation fault trackastra track -i trackastra_test/img/ -m trackastra_test/mask/ --output-ct (napari) reyesmatte@EVOCELL152 ~ % /Users/reyesmatte/miniforge3/envs/napari/lib/python3.10/multiprocessing/resource_tracker.py:224: UserWarning: resource_tracker: There appear to be 2 leaked semaphore objects to clean up at shutdown

Apparently it can be mitigated by changing the batch size, but I don't know if there's an easy way to do so on the CLI or on jupyter.
This only happens when using the CLI, as using the napari plugin works well. This was done on an M1 machine with trackastra 0.4.2.

Please let me know if you need anything else

Thanks!

[bentaculum]

Hi @OReyesMatte, thanks for reporting this. Would you mind sharing the full error message please?
In the mean time, I exposed the batch size to the CLI here #47.

[OReyesMatte]

Hi @bentaculum, thank you for the quick reply! Weirdly, that's the only message the CLI prints, as you can see here:

I also tried updating to the dev branch and it worked well! Setting the number of workers to 2 did the trick for me, independent of the batch size changes.

Thanks a lot!

[bentaculum]

That's an interesting patch! I'm curious now, could you try running with --n-workers 1 as well? I exposed it to CLI in the above PR.

[OReyesMatte]

I'll try that tomorrow on the computer that was giving me the issue. For what it's worth, a fresh environment (no napari, only the 0.4.3 dev version) did not require any additional tuning, and ran perfectly fine on cpu. So I'll retry a new fresh environment and start building and breaking again to reproduce the error. If you want, I can also upload tomorrow the .yaml file of the environment that had the issue

[OReyesMatte]

So I re-tried in the problem environment with --n-workers 1 and it gave the same error as before. Then again, a fresh install solved the problem, so something else in the environment I have (which I was trying to make a general napari one) is causing the issue. If you want to take a look, here's the .yml of the environment.

napari.yml

Still, I managed to solve the problem, and at least on the fresh install either the dev or stable version worked fine even without tuning the --n-workers or the --batch-size.

Thank you so much for the quick help!