It seems like test_socketserver leaks child processes on the "x86 Gentoo Refleaks 3.6" buildbot, but I'm unable to reproduce the bug.

http://buildbot.python.org/all/builders/x86%20Gentoo%20Refleaks%203.6/builds/49/steps/test/logs/stdio

0:48:11 load avg: 3.91 [154/405/2] test_socketserver failed -- running: test_decimal (630 sec)
Warning -- reap_children() reaped child process 1044
beginning 6 repetitions
123456
Warning -- reap_children() reaped child process 1104
Warning -- reap_children() reaped child process 1115
.Warning -- reap_children() reaped child process 1170
Warning -- reap_children() reaped child process 1175
Warning -- reap_children() reaped child process 1184
.Warning -- reap_children() reaped child process 1249
.Warning -- reap_children() reaped child process 1311
Warning -- reap_children() reaped child process 1316
...
test_socketserver leaked [1, 1, 1] memory blocks, sum=3

(...)

test_ForkingUnixStreamServer (test.test_socketserver.SocketServerTest) ... creating server
ADDR = /tmp/unix_socket.hqh5x95a
CLASS = <class 'test.test_socketserver.ForkingUnixStreamServer'>
server running
test client 0
test client 1
test client 2
waiting for server
done
Warning -- reap_children() reaped child process 17938
ok

test_ForkingUnixDatagramServer (test.test_socketserver.SocketServerTest) ... creating server
ADDR = /tmp/unix_socket.gry6ulhp
CLASS = <class 'test.test_socketserver.ForkingUnixDatagramServer'>
server running
test client 0
test client 1
test client 2
waiting for server
done
Warning -- reap_children() reaped child process 18212
ok

test_ForkingUDPServer (test.test_socketserver.SocketServerTest) ... creating server
ADDR = ('127.0.0.1', 43415)
CLASS = <class 'socketserver.ForkingUDPServer'>
server running
test client 0
test client 1
test client 2
waiting for server
done
Warning -- reap_children() reaped child process 18281
ok
test_ForkingUnixDatagramServer (test.test_socketserver.SocketServerTest)

Ah, the master branch is also affected. Example:

http://buildbot.python.org/all/builders/x86%20Gentoo%20Refleaks%203.x/builds/47/steps/test/logs/stdio

1:35:05 load avg: 4.64 [212/406] test_socketserver passed -- running: test_subprocess (370 sec)
beginning 6 repetitions
123456
Warning -- reap_children() reaped child process 4641
.Warning -- reap_children() reaped child process 4708
..Warning -- reap_children() reaped child process 4832
.Warning -- reap_children() reaped child process 4916
Warning -- reap_children() reaped child process 4921
..

The problem is that socketserver.ForkinMixin doesn't wait until all children completes. It's only calls os.waitpid() in non-blocking module (using os.WNOHANG) after each loop iteration. If a child process completes after the last call to ForkingMixIn.collect_children(), the server leaks zombie processes.

The server must wait until all children completes. Attached PR implements that.

The bug was be reproduced with the attached forkingmixin_sleep.patch.

haypo@selma$ ./python -m test -v -u all test_socketserver --fail-env-changed -m '*Fork*'
(...)
Warning -- reap_children() reaped child process 17093
Warning -- reap_children() reaped child process 17094
(...)

New changeset aa8ec34ad52bb3b274ce91169e1bc4a598655049 by Victor Stinner in branch 'master':
bpo-31151: Add socketserver.ForkingMixIn.server_close() (#3057)
https://github.com/python/cpython/commit/aa8ec34ad52bb3b274ce91169e1bc4a598655049

bpo-31010 has been marked as a duplicate of this issue.

I created a thread on the python-dev mailing list to discuss this issue:
https://mail.python.org/pipermail/python-dev/2017-August/148826.html

I tag this issue as release blocker as a remainder that we have to agree how to handle threads/processes before Python 3.7 feature freeze.

See also bpo-31233: socketserver.ThreadingMixIn leaks running threads after server_close().

I proposed the bpo-31593 to fix the issue differently in Python 3.6 (and maybe also Python 2.7).

What's the status of this?  Feature freeze is here!

> What's the status of this?  Feature freeze is here!

This issue is very to https://bugs.python.org/issue31233#msg311021 as it's status:
* The commit aa8ec34ad52bb3b274ce91169e1bc4a598655049 is backward incompatible and it's not documented
* There is no option to opt-in for the previous behaviour (exit without waiting for child processes)

For beta1, IMHO it's ok to keep the current status. But I would like to see this issue fixed before 3.7 final. Either revert my change, or do something else.

New changeset db8189bb8db609ca3993dec31fb95fdbe134469a by Victor Stinner in branch 'master':
bpo-31233, bpo-31151: Document socketserver changes (#5417)
https://github.com/python/cpython/commit/db8189bb8db609ca3993dec31fb95fdbe134469a

I created bpo-33540: "socketserver: Add an opt-in option to get Python 3.6 behaviour on server_close()". I set the release blocker priority on this new issue.

The bug has been fixed in Python 3.7.