Closed constantinpape closed 5 years ago
This does not seem to be a general issue, but rather some workflows which do not respect the thread limit set, or use more than one thread although they should be single threaded. E.g. the watershed workflow.
Turns out the most likely culprit is numpy: https://github.com/numpy/numpy/issues/11826 To fix it, either set the corresponding environment variables or use https://github.com/joblib/threadpoolctl.
This is indeed the issue. For now, the solution is to use this function and call it before the numpy import, like here.
Tasks tend to spawn way more threads / processes than specified. This might be related to the luigi scheduler or python
subprocess.call
?!