set env var in Python multiprocessing.Process

Question

In the subprocess Python 2 module, Popen can be given an env .

Seems that the equivalent way to do it with Process in multiprocessing module is to pass the env dictionnary in args or kwargs , and then use os.environ['FOO'] = value in the target .

Is it the right way?

Is it safe? I mean, no risk that the environment in the parent process or other child processes can be modified?

Here is an example (that works).

import multiprocessing
import time
import os

def target(someid):
    os.environ['FOO'] = "foo%i" % someid
    for i in range(10):
        print "Job %i: " % someid, os.environ['FOO']
        time.sleep(1)

if __name__ == '__main__':

    processes = []

    os.environ['FOO'] = 'foo'

    for someid in range(3):
        p = multiprocessing.Process(target=target, args=(someid,))
        p.start()
        processes.append(p)

    for i in range(10):
        print "Parent: ", os.environ['FOO']
        time.sleep(1)

    for p in processes:
        p.join()

Answer 1

Yes, that's the right way to do it. While the child will inherit its initial environment from the parent, subsequent changes to os.environ made in the child will not affect the parent, and vice-versa:

import os
import multiprocessing


def myfunc(q):
    print "child: " + os.environ['FOO']
    os.environ['FOO'] = "child_set"
    print "child new: " + os.environ['FOO']
    q.put(None)
    q.get()
    print "child new2: " + os.environ['FOO']


if __name__ == "__main__":
    os.environ['FOO'] = 'parent_set'
    q = multiprocessing.Queue()
    proc = multiprocessing.Process(target=myfunc, args=(q,))
    proc.start()
    q.get()
    print "parent: " + os.environ['FOO']
    os.environ['FOO'] = "parent_set_again"
    q.put(None)

Output:

child start: parent_set
child after changing: child_set
parent after child changing: parent_set
child after parent changing: child_set

If you need to pass an initial environment to the child, you would just pass it in the args or kwargs list:

def myfunc(env=None):
    time.sleep(3)
    if env is not None:
        os.environ = env
    print os.environ['FOO']


if __name__ == "__main__":
    child_env = os.environ.copy()

    for i in range(3):
        child_env['FOO'] = "foo%s" % (i,)
        proc = multiprocessing.Process(target=myfunc, kwargs ={'env' : child_env})
        proc.start()

Output:

foo0
foo1
foo2

Note that if you're using a multiprocessing.Pool , you can use the initializer / initargs keyword arguments to just set the correct environment once at the start of each process in the pool:

def init(env):
    os.environ = env

def myfunc():
    print os.environ['FOO']


if __name__ == "__main__":
    child_env = os.environ.copy()
    child_env['FOO'] = "foo"
    pool = multiprocessing.Pool(initializer=init, initargs=(child_env,))
    pool.apply(myfunc,()) 

Output:

foo