gh-132969: Fix error/hang when shutdown(wait=False) and task exited abnormally (GH-133222)
When shutdown is called with wait=False, the executor thread keeps running
even after the ProcessPoolExecutor's state is reset. The executor then tries
to replenish the worker processes pool resulting in an error and a potential hang
when it comes across a worker that has died. Fixed the issue by having
_adjust_process_count() return without doing anything if the ProcessPoolExecutor's
state has been reset.
Added unit tests to validate two scenarios:
max_workers < num_tasks (exception)
max_workers > num_tasks (exception + hang)
(cherry picked from commit 598aa7cc98)
Co-authored-by: Ajay Kamdar <140011370+ogbiggles@users.noreply.github.com>
gh-132063: ensure that `ProcessPoolExecutor` does not swallow falsey exceptions (GH-132129)
(cherry picked from commit 933c6653cb)
Co-authored-by: Duprat <yduprat@gmail.com>
There was a deadlock when `ProcessPoolExecutor` shuts down at the same
time that a queueing thread handles an error processing a task.
Don't use `_shutdown_lock` to protect the `_ThreadWakeup` pipes -- use
an internal lock instead. This fixes the ordering deadlock where the
`ExecutorManagerThread` holds the `_shutdown_lock` and joins the
queueing thread, while the queueing thread is attempting to acquire the
`_shutdown_lock` while closing the `_ThreadWakeup`.
(cherry picked from commit 760872efec)
Co-authored-by: Sam Gross <colesbury@gmail.com>
Deferred reference counting is not fully implemented yet. As a temporary
measure, we immortalize objects that would use deferred reference
counting to avoid multi-threaded scaling bottlenecks.
This is only performed in the free-threaded build once the first
non-main thread is started. Additionally, some tests, including refleak
tests, suppress this behavior.
concurrent.futures: The *executor manager thread* now catches
exceptions when adding an item to the *call queue*. During Python
finalization, creating a new thread can now raise RuntimeError. Catch
the exception and call terminate_broken() in this case.
Add test_python_finalization_error() to test_concurrent_futures.
concurrent.futures._ExecutorManagerThread changes:
* terminate_broken() no longer calls shutdown_workers() since the
call queue is no longer working anymore (read and write ends of
the queue pipe are closed).
* terminate_broken() now terminates child processes, not only
wait until they complete.
* _ExecutorManagerThread.terminate_broken() now holds shutdown_lock
to prevent race conditons with ProcessPoolExecutor.submit().
multiprocessing.Queue changes:
* Add _terminate_broken() method.
* _start_thread() sets _thread to None on exception to prevent
leaking "dangling threads" even if the thread was not started
yet.
Follow-up of gh-107219.
* Only close the connection writer on Windows.
* Also use existing constant _winapi.ERROR_OPERATION_ABORTED instead of
WSA_OPERATION_ABORTED.
This fixes issue #105829, https://github.com/python/cpython/issues/105829
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Antoine Pitrou <antoine@python.org>
Co-authored-by: Chris Withers <chris@withers.org>
Co-authored-by: Thomas Moreau <thomas.moreau.2010@gmail.com>
Fix a race condition in concurrent.futures. When a process in the
process pool was terminated abruptly (while the future was running or
pending), close the connection write end. If the call queue is
blocked on sending bytes to a worker process, closing the connection
write end interrupts the send, so the queue can be closed.
Changes:
* _ExecutorManagerThread.terminate_broken() now closes
call_queue._writer.
* multiprocessing PipeConnection.close() now interrupts
WaitForMultipleObjects() in _send_bytes() by cancelling the
overlapped operation.
Fix a race condition in _ExecutorManagerThread.terminate_broken():
ignore the InvalidStateError on future.set_exception(). It can happen
if the future is cancelled before the caller.
Moreover, test_crash_big_data() now waits explicitly until the
executor completes.
Fixes a hang in multiprocessing process pool executor when a child process crashes and code could otherwise block on writing to the pipe. See GH-94777 for more details.
This starts the process. Users who don't specify their own start method
and use the default on platforms where it is 'fork' will see a
DeprecationWarning upon multiprocessing.Pool() construction or upon
multiprocessing.Process.start() or concurrent.futures.ProcessPool use.
See the related issue and documentation within this change for details.
Do not spawn ProcessPool workers on demand when they spawn via fork.
This avoids potential deadlocks in the child processes due to forking from
a multithreaded process.
Prevent `max_tasks_per_child` use with a "fork" mp_context to avoid deadlocks.
Also defaults to "spawn" when no mp_context is supplied for safe convenience.
In the case of multiprocessing.synchronize() being missing, the
test_concurrent_futures test suite now skips only the tests that
require multiprocessing.synchronize().
Validate that multiprocessing.synchronize exists as part of
_check_system_limits(), allowing ProcessPoolExecutor to raise
NotImplementedError during __init__, rather than crashing with
ImportError during __init__ when creating a lock imported from
multiprocessing.synchronize.
Use _check_system_limits() to disable tests of
ProcessPoolExecutor on systems without multiprocessing.synchronize.
Running the test suite without multiprocessing.synchronize reveals
that Lib/compileall.py crashes when it uses a ProcessPoolExecutor.
Therefore, change Lib/compileall.py to call _check_system_limits()
before creating the ProcessPoolExecutor.
Note that both Lib/compileall.py and Lib/test/test_compileall.py
were attempting to sanity-check ProcessPoolExecutor by expecting
ImportError. In multiprocessing.resource_tracker, sem_unlink() is also absent
on platforms where POSIX semaphores aren't available. Avoid using
sem_unlink() if it, too, does not exist.
Co-authored-by: Pablo Galindo <Pablogsal@gmail.com>
Remove daemon threads from :mod:`concurrent.futures` by adding
an internal `threading._register_atexit()`, which calls registered functions
prior to joining all non-daemon threads. This allows for compatibility
with subinterpreters, which don't support daemon threads.
As reported initially by @rad-pat in #6084, the following script causes a deadlock.
```
from concurrent.futures import ProcessPoolExecutor
class ObjectWithPickleError():
"""Triggers a RuntimeError when sending job to the workers"""
def __reduce__(self):
raise RuntimeError()
if __name__ == "__main__":
e = ProcessPoolExecutor()
f = e.submit(id, ObjectWithPickleError())
e.shutdown(wait=False)
f.result() # Deadlock on get
```
This is caused by the fact that the main process is closing communication channels that might be necessary to the `queue_management_thread` later. To avoid this, this PR let the `queue_management_thread` manage all the closing.
https://bugs.python.org/issue39104
Automerge-Triggered-By: @pitrou
Deprecated passing the following arguments as keyword arguments:
- "func" in functools.partialmethod(), weakref.finalize(),
profile.Profile.runcall(), cProfile.Profile.runcall(),
bdb.Bdb.runcall(), trace.Trace.runfunc() and
curses.wrapper().
- "function" in unittest.addModuleCleanup() and
unittest.TestCase.addCleanup().
- "fn" in the submit() method of concurrent.futures.ThreadPoolExecutor
and concurrent.futures.ProcessPoolExecutor.
- "callback" in contextlib.ExitStack.callback(),
contextlib.AsyncExitStack.callback() and
contextlib.AsyncExitStack.push_async_callback().
- "c" and "typeid" in the create() method of multiprocessing.managers.Server
and multiprocessing.managers.SharedMemoryServer.
- "obj" in weakref.finalize().
Also allowed to pass arbitrary keyword arguments (even "self" and "func")
if the above arguments are passed as positional argument.
ProcessPoolExecutor workers will hold the return value of their last task in memory until the next task is received. Since the return value has already been propagated to the parent process's Future (or has been discarded by this point), the object can be safely released.
Executors in concurrent.futures accepted tasks after executor was shutdown by interpreter exit. Tasks were left in PENDING state forever. This fix changes submit to instead raise a RuntimeError.
Fix deadlocks in :class:`concurrent.futures.ProcessPoolExecutor` when task arguments or results cause pickling or unpickling errors.
This should make sure that calls to the :class:`ProcessPoolExecutor` API always eventually return.
* bpo-31326: ProcessPoolExecutor waits for the call queue thread
concurrent.futures.ProcessPoolExecutor.shutdown() now explicitly
closes the call queue. Moreover, shutdown(wait=True) now also join
the call queue thread, to prevent leaking a dangling thread.
* Fix for shutdown() being called twice.
* bpo-27144: concurrent.futures as_complie and map iterators do not keep
reference to returned object
* Some nits. Improve wordings in docstrings and comments, and avoid relying on
sys.getrefcount() in tests.
