Multiprocessing Start Method Forced to Forkserver

[TheSkyentist]

#8093 Forces the multiprocessing start method globally:

jwst/jwst/ramp_fitting/ramp_fit_step.py

Line 26 in 613383f

multiprocessing.set_start_method('forkserver', force=True)

jwst/jwst/jump/jump_step.py

Line 11 in 613383f

multiprocessing.set_start_method('forkserver', force=True)

This can cause weird/unexpected behavior on Linux systems (default method fork) vs macOS (default method spawn). Therefore, even if the JWST pipeline is simply imported, it can change the behaviour of unrelated code that wraps the pipeline (this has been my experience). In general, the best practices recommended by Python are to only call the set_start_method within an if __name__ == '__main__' clause.

This may even effect other modules where multiprocessing is called (e.g. wfss_contam) and cause unexpected behaviour. I'm not sure what the best solution is moving forward, but it should be at least reflected in the documentation that this is required under the hood and may explain errors/problems.

[braingram]

Thanks for opening this issue. I'm taking a look at it now. I believe the issue will need to be addressed in a dependency (stcal) and then we should be able to remove the above mentioned lines and update wfss_contam.

Also, thanks for the detailed write-up!