gh-86354: Fix child processes not reusing forkserver created by parent

[moycat]

This fixes issue #86354 that a child process does not reuse the forkserver process started by its parent. Currently, each child process creates its own forkserver for its own child processes:

import multiprocessing as mp
import os

mp.set_start_method("forkserver", force=True)


def child(remaining):
    print(f"PID: {os.getpid()}, PPID: {os.getppid()}")
    remaining -= 1
    if remaining > 0:
        proc = mp.Process(target=child, args=(remaining,))
        proc.start()
        proc.join()


def main():
    proc = mp.Process(target=child, args=(5,))
    proc.start()
    proc.join()


if __name__ == '__main__':
    main()

PID: 60016, PPID: 60015
PID: 60018, PPID: 60017
PID: 60020, PPID: 60019
PID: 60022, PPID: 60021
PID: 60024, PPID: 60023

With this patch, all the descendants reuse the forkserver spawned by the main process, reducing the overhead when the process tree is deep:

PID: 60054, PPID: 60053
PID: 60055, PPID: 60053
PID: 60056, PPID: 60053
PID: 60057, PPID: 60053
PID: 60058, PPID: 60053

The new behavior matches the comment in forkserver.py:

cpython/Lib/multiprocessing/forkserver.py

Lines 123 to 129 in fd7dac0

	def ensure_running(self):
	'''Make sure that a fork server is running.
	This can be called from any process. Note that usually a child
	process will just reuse the forkserver started by its parent, so
	ensure_running() will do nothing.
	'''

• Issue: forkserver not reused in child processes #86354

[python-cla-bot]

All commit authors signed the Contributor License Agreement.