cc @pablogsal @stratakis

@stratakis: I addressed your review, thanks. Please review the updated PR.

Tests / Sanitizers / TSan (free-threading) (pull_request): Failing after 22m

Hum. test_socket failed with a strange error. I just re-ran the job.

ERROR: testSourceAddress (test.test_socket.NetworkConnectionAttributesTest.testSourceAddress)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/runner/work/cpython/cpython/Lib/test/test_socket.py", line 442, in raise_queued_exception
    raise self.queue.get()
  File "/home/runner/work/cpython/cpython/Lib/test/test_socket.py", line 479, in clientRun
    test_func()
    ~~~~~~~~~^^
  File "/home/runner/work/cpython/cpython/Lib/test/test_socket.py", line 6048, in _testSourceAddress
    self.assertEqual(self.cli.getsockname()[1], self.source_port)
                     ~~~~~~~~~~~~~~~~~~~~^^
OSError: [Errno 9] Bad file descriptor

Nitpick: I'd change the helper name to something other as kill has a very specific meaning and the helper also closes the pipes, it doesn't only kill the processes. Maybe cleanup_process_group() or something?

For testing the assert_usable() path (10s timeout):

Fake bpftrace script:
(sleep 300) & sleep 300

And adding it to the $PATH when running the tests:
PATH="fake bpftrace path:$PATH" ./python -m test test_dtrace -m 'BPFTraceNormalTests' -v

This hangs and also leaves leaked children:
$ pgrep -af 'sleep 300'
194272 sleep 300

For testing the run_case() path (60s timeout):

Fake bpftrace script:

case "$*" in
  *'probe: success'*)
  echo 'probe: success'
  exit 0
  ;;
esac
(sleep 300) & sleep 300

PATH="fake bpftrace path:$PATH" ./python -m unittest -v test.test_dtrace.BPFTraceNormalTests.test_gc

After 60 secs it should failed and children are leaked:
$ pgrep -af 'sleep 300'
250104 sleep 300
250105 sleep 300

With the PR the assert_usable() path doesn't hang anymore and doesn't leak and the run_case() path still times outs but doesn't leak.

Tests / Sanitizers / TSan (free-threading) (pull_request): Cancelled after 60m

Oh, test_abc of ./python -m test --tsan-parallel --parallel-threads=4 -j4 -W seems to be stuck, and the job was cancelled after 60 minutes :-(

Run ./python -m test --tsan-parallel --parallel-threads=4 -j4 -W
Using random seed: 2657327358
0:00:00 load avg: 2.63 mem: 274.1 MiB Run 2 tests in parallel using 2 worker processes
0:00:13 load avg: 2.96 mem: 588.5 MiB [1/2] test_hashlib passed
0:00:43 load avg: 1.79 mem: 588.6 MiB running (1): test_abc (42.9 sec)
0:01:13 load avg: 1.08 mem: 589.1 MiB running (1): test_abc (1 min 13 sec)
0:01:43 load avg: 0.66 mem: 590.1 MiB running (1): test_abc (1 min 43 sec)
...
0:36:00 load avg: 0.07 mem: 531.8 MiB running (1): test_abc (35 min 59 sec)
0:36:30 load avg: 0.04 mem: 531.8 MiB running (1): test_abc (36 min 29 sec)
Error: The operation was canceled.

With the PR the assert_usable() path doesn't hang anymore and doesn't leak and the run_case() path still times outs but doesn't leak.

Oh thanks for testing manually! It's good that you managed to confirm that my fix works as expected :-)

Nitpick: I'd change the helper name to something other as kill has a very specific meaning and the helper also closes the pipes, it doesn't only kill the processes. Maybe cleanup_process_group() or something?

Calling proc.communicate() # Clean up in kill_process_group() is just there to call waitpid(pid) to avoid zombie process. Since the process and its child processes are killed, it should no longer have to wait for pipes to be closed, they should be already closed.

I prefer to keep "kill" in the function name, since it's the main action of the function.

Thanks @vstinner for the PR 🌮🎉.. I'm working now to backport this PR to: 3.15.
🐍🍒⛏🤖

GH-151589 is a backport of this pull request to the 3.15 branch.