I'm getting sporadic fatal errors for my Github CI on windows-latest, python version v3.11. While pytest is running, I get this error:
Windows fatal exception: code 0x80000003
But, it doesn't always happen. I reran this failing job (no changes to the code in between) and it passed. Link to the failure point on a re-run.
I'm pretty sure this is due to improvements I made to parallelization with joblib. The tests that are running when the fatal error occur involve the code in this file in this PR.
Very roughly, the relevant code is:
from tqdm import tqdm
from joblib import Parallel, cpu_count, delayed
#...
n_jobs = cpu_count()
arg_list = [
## kwargs dicts
]
@delayed
def get_candidates(args_dict):
# stuff
return result
candidate_list = Parallel(n_jobs=n_jobs)(
get_candidates(args_dict=arg) for arg in tqdm(arg_list)
)
In case it's relevant, this is the test that's running when the fatal exception occurs.
Any insight would be appreciated!