reproduce.py

reproduce.py

'''
>>> dask.__version__
'1.0.0'

>>> umap.__version__
'0.3.7'
'''
from dask.distributed import Client

def do_run(arg1, arg2):
    # this import triggers the segfault
    import umap
    return 42

if __name__ == '__main__':
    client = Client(processes=True,
                    ip='127.0.0.1',
                    #n_workers=1,
                    memory_limit='40GB')

    config_path = ''
    name = ''
    lo_classes = [None, None, None]
    n_trial = 3
    pool = []

    # crashes
    for trial in range(n_trial):
        for lo_class in lo_classes:
            # the following line triggers a segfault in libffi.so
            #pool.append(client.submit(do_run, trial, lo_class))

            # this line alone works fine
            pool.append(client.submit(do_run, 1, lo_class))

            # following line crashes
            #pool.append(client.submit(do_run, trial, 1))

    result = client.gather(pool)
    print(result)