From b33feff2f98bffacb8be877a50f7ed1afce4d890 Mon Sep 17 00:00:00 2001 From: Seemanshi2104 <143334335+Seemanshi2104@users.noreply.github.com> Date: Sat, 14 Nov 2026 02:32:20 +0530 Subject: [PATCH 1/2] gh-161931: Wake manager after spawning workers Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- Lib/concurrent/futures/process.py | 12 ++-- .../test_process_pool.py | 68 +++++++++++++++++++ 2 files changed, 75 insertions(+), 5 deletions(-) diff --git a/Lib/concurrent/futures/process.py b/Lib/concurrent/futures/process.py index 5d267481041e90..c3e62a69ca7fcc 100644 --- a/Lib/concurrent/futures/process.py +++ b/Lib/concurrent/futures/process.py @@ -929,11 +929,13 @@ def submit(self, fn, /, *args, **kwargs): self._pending_work_items[self._queue_count] = w self._work_ids.put(self._queue_count) self._queue_count += 1 - # Wake up queue management thread - self._executor_manager_thread_wakeup.wakeup() - - if self._safe_to_dynamically_spawn_children: - self._adjust_process_count() + try: + if self._safe_to_dynamically_spawn_children: + self._adjust_process_count() + finally: + # Wake up after spawning so the manager can wait on the new + # process's sentinel. + self._executor_manager_thread_wakeup.wakeup() self._start_executor_manager_thread() return f submit.__doc__ = _base.Executor.submit.__doc__ diff --git a/Lib/test/test_concurrent_futures/test_process_pool.py b/Lib/test/test_concurrent_futures/test_process_pool.py index dafbda862c51c2..3f930f4adf6b2d 100644 --- a/Lib/test/test_concurrent_futures/test_process_pool.py +++ b/Lib/test/test_concurrent_futures/test_process_pool.py @@ -71,6 +71,74 @@ def test_killed_child(self): # Submitting other jobs fails as well. self.assertRaises(BrokenProcessPool, self.executor.submit, pow, 2, 8) + def test_killed_child_during_submit(self): + context = self.get_context() + if context.get_start_method() == "fork": + self.skipTest("Incompatible with the fork start method.") + + started = self.manager.Queue() + gate = self.manager.Event() + second_submit = threading.Event() + wakeup_sent = threading.Event() + manager_waiting = threading.Event() + executor = self.executor + process_module = futures.process + original_wakeup = executor._executor_manager_thread_wakeup.wakeup + original_spawn_process = executor._spawn_process + original_connection_wait = process_module.mp.connection.wait + + # With the old submit ordering, let the manager rebuild its wait set + # after the wakeup but before the new process is registered. + def wakeup(): + if second_submit.is_set(): + wakeup_sent.set() + original_wakeup() + + def spawn_process(): + if second_submit.is_set() and wakeup_sent.is_set(): + self.assertTrue(manager_waiting.wait(support.SHORT_TIMEOUT)) + original_spawn_process() + + def connection_wait(object_list, timeout=None): + if (second_submit.is_set() and wakeup_sent.is_set() + and threading.current_thread() + is executor._executor_manager_thread + and len(executor._processes) == 1): + manager_waiting.set() + return original_connection_wait(object_list, timeout) + + worker_process = None + try: + with unittest.mock.patch.object( + executor._executor_manager_thread_wakeup, "wakeup", + wakeup), \ + unittest.mock.patch.object( + executor, "_spawn_process", spawn_process), \ + unittest.mock.patch.object( + process_module.mp.connection, "wait", + connection_wait): + future1 = executor.submit( + _report_wait_return, started, gate, 1) + self.assertEqual( + started.get(timeout=support.SHORT_TIMEOUT), 1) + + second_submit.set() + future2 = executor.submit( + _report_wait_return, started, gate, 2) + self.assertEqual( + started.get(timeout=support.SHORT_TIMEOUT), 2) + self.assertTrue(future1.running()) + + worker_process = list(executor._processes.values())[-1] + worker_process.terminate() + worker_process.join(support.SHORT_TIMEOUT) + with self.assertRaises(BrokenProcessPool): + future2.result(timeout=support.SHORT_TIMEOUT) + finally: + gate.set() + if worker_process is not None: + worker_process.join(support.SHORT_TIMEOUT) + @warnings_helper.ignore_fork_in_thread_deprecation_warnings() def test_broken_process_pool_traceback(self): # When a child process is abruptly terminated, the whole pool gets From 1cfab78a9cbab133100556b02e6033c6f9dca30f Mon Sep 17 00:00:00 2001 From: Seemanshi2104 <143334335+Seemanshi2104@users.noreply.github.com> Date: Sat, 14 Nov 2026 02:39:29 +0530 Subject: [PATCH 2/2] gh-161931: Add NEWS entry for ProcessPoolExecutor fix Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- .../Library/2026-11-14-02-38-26.gh-issue-161931.k7vQ2m.rst | 4 ++++ 1 file changed, 4 insertions(+) create mode 100644 Misc/NEWS.d/next/Library/2026-11-14-02-38-26.gh-issue-161931.k7vQ2m.rst diff --git a/Misc/NEWS.d/next/Library/2026-11-14-02-38-26.gh-issue-161931.k7vQ2m.rst b/Misc/NEWS.d/next/Library/2026-11-14-02-38-26.gh-issue-161931.k7vQ2m.rst new file mode 100644 index 00000000000000..88178946246628 --- /dev/null +++ b/Misc/NEWS.d/next/Library/2026-11-14-02-38-26.gh-issue-161931.k7vQ2m.rst @@ -0,0 +1,4 @@ +Fix a hang in :class:`concurrent.futures.ProcessPoolExecutor` when a worker +started with the ``spawn`` or ``forkserver`` method exits abruptly while other +workers are busy. The manager thread now waits on the new worker's process +sentinel.