http://buildbot.python.org/all/#/builders/1/builds/267

1:47:06 load avg: 4.77 [161/417/1] test_multiprocessing_forkserver failed (16 min 32 sec) -- running: test_support (56 sec 282 ms)
beginning 6 repetitions
123456
......
test_multiprocessing_forkserver leaked [1, 2, 1] memory blocks, sum=4
(...)
Re-running failed tests in verbose mode
Re-running test 'test_multiprocessing_forkserver' in verbose mode
(...)
OK (skipped=27)

> http://buildbot.python.org/all/#/builders/1/builds/267

Oh, this is: x86 Gentoo Refleaks 3.x.

--

x86 Gentoo Refleaks 3.7 also leaked:
http://buildbot.python.org/all/#/builders/114/builds/150

test_multiprocessing_forkserver leaked [1, 2, 1] memory blocks, sum=4
Re-running failed tests in verbose mode
Re-running test 'test_multiprocessing_forkserver' in verbose mode
test_multiprocessing_forkserver leaked [21, 2, 1] memory blocks, sum=24
test_multiprocessing_forkserver leaked [2, 0, 0] file descriptors, sum=2

Oh. I have a very good news: this issue is a duplicate of bpo-33735 and I just identified and fixed bpo-33735!

Hum. Sadly my commit 23401fb960bb94e6ea62d2999527968d53d3fc65 is not enough, the test still fails randomly:

pydev@stormageddon ~/cpython $ ./python -m test test_multiprocessing_forkserver -m test_imap_unordered -R 3:3 -F
Run tests sequentially
0:00:00 load avg: 5.79 [  1] test_multiprocessing_forkserver
beginning 6 repetitions
123456
......
0:00:28 load avg: 6.98 [  2] test_multiprocessing_forkserver
beginning 6 repetitions
123456
......
0:00:59 load avg: 7.69 [  3] test_multiprocessing_forkserver -- test_multiprocessing_forkserver passed in 30 sec 717 ms
beginning 6 repetitions
123456
......
0:01:28 load avg: 8.63 [  4] test_multiprocessing_forkserver
beginning 6 repetitions
123456
......
test_multiprocessing_forkserver leaked [2, 1, 1] memory blocks, sum=4
test_multiprocessing_forkserver failed

== Tests result: FAILURE ==

3 tests OK.

1 test failed:
    test_multiprocessing_forkserver

Total duration: 1 min 55 sec
Tests result: FAILURE


The test failed when the system load was high (8.63).

I hope that in pratice, my change will be enough to get less random failures. The "Re-run in verbose mode" on the "idle" buildbot should make the bug disappear.

superseder: test_multiprocessing_spawn leaked [1, 2, 1] memory blocks on AMD64 Windows8.1 Refleaks 3.7
stage: resolved