Python multiprocessing: abort map on first child error

Question

What's the proper way of aborting multiprocessing when one of the child aborts and/or throw an Exception?

I found various questions around that (generic multiprocessing error handling, how to close multiprocessing pool on exception but without answer, ...), but no clear answer on how to stop multiprocessing on child exception.

For instance, I expect the following code:

def f(x):
    sleep(x)
    print(f"f({x})")
    return 1.0 / (x - 2)


def main():
    with Pool(4) as p:
        try:
            r = p.map(f, range(7))
        except Exception as e:
            print(f"oops: {e}")
            p.close()
            p.terminate()
    print("end")


if __name__ == '__main__':
    main()

To output:

f(0)
f(1)
f(2)
oops: float division by zero
end

Instead, it applies f function on all items before detecting/handling the exception:

f(0)
f(1)
f(2)
f(4)
f(3)
f(5)
f(6)
oops: float division by zero
end

Isn't there any way to catch the exception directly?

Hairy

Note that the exception is only re-raised in the main process after map has finished. If you definitely want to use map there is no way to catch it any earlier, because the exception does not exist any earlier. So, the question is whether you want to stick to map or are looking for a custom solution? – Dekameter

Which python version? My Python:3.4.2 behave as you expected. – Vermeil

@Vermeil I've edited my post, I'm using Python 3.6.5 – Hairy

@MisterMiyagi I'm looking for the simplest multiprocessing with error handling - I'd like to avoid adding wrapper class or equivalent to handle it though. – Hairy

Answer 1

I think you're going to need apply_async for this, so you can act upon every single result instead of the cumulative result. pool.apply_async offers an error_callback parameter you can use to register your error-handler. apply_async is not blocking, so you'll need to join() the pool. I'm also using a flag terminated to know when results can be processed normally in case no exception occured.

from time import sleep
from multiprocessing import Pool

def f(x):
    sleep(x)
    print(f"f({x})")
    return 1.0 / (x - 2)

def on_error(e):
    global terminated
    terminated = True
    pool.terminate()
    print(f"oops:{e}")


def main():
    global pool
    global terminated

    terminated = False

    pool = Pool(4)
    results = [pool.apply_async(f, (x,), error_callback=on_error)
               for x in range(7)]
    pool.close()
    pool.join()

    if not terminated:
        for r in results:
            print(r.get())

    print("end")


if __name__ == '__main__':
    main()

Nuzzi