multiprocessing.util._afterfork_registry leak in threaded environment

[mzamazal]

BPO	26180
Nosy	@terryjreedy, @nirs, @jdemeyer, @applio
PRs	bpo-26180: Fix multiple registration of ForkAwareLocal atfork cleaner #13986
Files	bug.py bug3.py

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2016-01-22.15:24:46.852>
labels = ['library', '3.9', 'performance']
title = 'multiprocessing.util._afterfork_registry leak in threaded environment'
updated_at = <Date 2019-06-14.07:05:58.650>
user = 'https://bugs.python.org/mzamazal'

bugs.python.org fields:

activity = <Date 2019-06-14.07:05:58.650>
actor = 'mzamazal'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2016-01-22.15:24:46.852>
creator = 'mzamazal'
dependencies = []
files = ['41690', '41706']
hgrepos = []
issue_num = 26180
keywords = ['patch']
message_count = 6.0
messages = ['258816', '258850', '258896', '345473', '345524', '345561']
nosy_count = 5.0
nosy_names = ['terry.reedy', 'nirs', 'jdemeyer', 'davin', 'mzamazal']
pr_nums = ['13986']
priority = 'normal'
resolution = None
stage = 'patch review'
status = 'open'
superseder = None
type = 'resource usage'
url = 'https://bugs.python.org/issue26180'
versions = ['Python 3.9']

[mzamazal]

When calling multiprocessing.managers.BaseManager repeatedly from a client, each time from a new thread, new entry with (already existent) multiprocessing.util.ForkAwareLocal instance is inserted as a value into multiprocessing.util._afterfork_registry dictionary on each of the calls. So the dictionary grows on each client call and may grow so indefinitely, causing memory leak if nothing else.

The attached file demonstrates the problem (Python 2.7 version, it's reproducible on 3.4 as well after 2->3 adjustments). Just run it and look at the output. The printed dictionary contains 10 entries holding the same ForkAwareLocal instance although there should be probably just one such entry there.

[terryjreedy]

Please post your 3.x version. 3.4 only gets security patches now.

[mzamazal]

Python 3.x version attached.

[jdemeyer]

Why is it that __init__ is called multiple times? I'm just wondering if we could somehow ensure that __init__ is called only once.

[terryjreedy]

With master and Windows 10, the output looks like
(0, 61333912, <function ForkAwareThreadLock._reset at 0x03A783F0>): <multiprocessing.util.ForkAwareThreadLock object at 0x03A7E198>
(1, 63524512, <function ForkAwareLocal.__init__.<locals>.<lambda> at 0x03CA2940>): <multiprocessing.util.ForkAwareLocal object at 0x03C94EA0>
(2, 60787504, <function ProcessLocalSet.__init__.<locals>.<lambda> at 0x03CA2760>): ProcessLocalSet()
(5, 63524512, <function ForkAwareLocal.__init__.<locals>.<lambda> at 0x03CA2C60>): <multiprocessing.util.ForkAwareLocal object at 0x03C94EA0>
... <8 more lines like the two with ForkAwareLocal.__init__....>

I am not familiar with multiprocessing Managers and the doc is not completely clear to me. Does bug3.py only run in one process or does it result in multiple processes? If the latter, I expect it would need an "if __name__ == '__main__':" clause, at least on Windows. (Milan, I should have asked what OS you used, as mp has OS-specific behaviors.)
(https://docs.python.org/3/library/multiprocessing.html#the-spawn-and-forkserver-start-methods, Safe importing of main module)

[mzamazal]

I used Linux.