Repository navigation
multiprocessing child segfault when passing lock as an argument #119413
Description
Crash report
What happened?
It seems that the below code will serialize the pointer to the lock handle over to the child process, and then rebuild the raw pointer in https://github.com/python/cpython/blob/main/Modules/_multiprocessing/clinic/semaphore.c.h#L331
this then segfaults the child process - which seems to cause new children to spin up.
import multiprocessing
import os
def _init(l):
print(os.getpid())
print(l)
def main():
ctx = multiprocessing.get_context(method="spawn")
lock = multiprocessing.Lock()
with ctx.Pool(1, initializer=_init, initargs=(lock,)) as pool:
ret = pool.map(print, [1])
if __name__ == "__main__":
main()CPython versions tested on:
3.10, 3.11
Operating systems tested on:
Linux
Output from running 'python -VV' on the command line:
No response
Activity
[-]multiprocessing segfault[/-][+]multiprocessing child segfault when passing lock as an argument[/+]Eclips4 commentedon May 22, 2024
Hello! I guess this has been fixed in #107275. I cannot reproduce this in any of the 3.12, 3.13, or 3.14 branches. Versions lower than 3.12 do not accept bug fixes.
DylanZA commentedon May 22, 2024
right - that makes sense. sorry for the noise in that case
DylanZA commentedon May 22, 2024
fwiw for future people - the new error message makes it obvious that this change fixes the problem: