Select a module using importlib and use in multipr

Question

I'd like to select in my main function which module to import based on an argument passed to the Python script. So, I'm using one of

blah = importlib.import_module("blah1")
blah = importlib.import_module("blah2")

where the 'blahX' are different implementations of the same interface.

I also want to pass the work off to different processes using the multiprocessing module.

blah = None

def f(a, b):
    print blah.f(a,b)

if __name__ == '__main__':

    # call importlib.import_module here...

    a = 1
    b = 2
    p = multiprocessing.Process(target=f, args=(a, b))
    p.start()
    p.join()

The problem is that the functions passed to multiprocessing.Process aren't aware of the module I imported in main. This is different than if I use import

import blah1 as blah
#import blah2 as blah

but then I loose the ability to choose a module at run-time.

How can I fix this design?

Answer 1

When you call mp.Process(...), the multiprocessing module forks a subprocess (on Unix) or starts a new Python process and imports the calling module (on Windows). On Unix your current code would work because all globals at the time of the fork are copied by the new subprocess.

On Windows, however, the calling module is imported. Since the definition of blah, e.g.,

blah = importlib.import_module("math")

is protected inside

if __name__ == '__main__':

the new definition of blah does not transfer to the subprocess.

So to make it work with Windows, you could pass the name of the module to the target function, and call importlib.import_module there:

import importlib
import multiprocessing as mp
blah = None

def f(a, b, name):
    blah = importlib.import_module(name)
    print blah.hypot(a,b)

if __name__ == '__main__':
    a = 1
    b = 2
    p = mp.Process(target=f, args=(a, b, "math"))
    p.start()
    p.join()