homepage

Maybe can not shutdown ThreadPoolExecutor when call the method shutdown.
Though the variable of _shutdown is set to true in the method of shutdown, it may also reads the variable of _shutdown from cpu cache in the method of _worker, and the worst case is that it could see an out-of-date value of _shutdown forever. so need to acquire lock before reading the variable of _shutdown to make sure see an up-to-date value.
the following is the new code:
def _worker(executor_reference, work_queue):
 try:
 while True:
 work_item = work_queue.get(block=True)
 if work_item is not None:
 work_item.run()
 continue
 executor = executor_reference()
 
 shutdown = False
 with executor._shutdown_lock.acquire():
 shutdown = executor._shutdown
 
 # Exit if:
 # - The interpreter is shutting down OR
 # - The executor that owns the worker has been collected OR
 # - The executor that owns the worker has been shutdown.
 if _shutdown or executor is None or shutdown:
 # Notice other workers
 work_queue.put(None)
 return
 del executor
 except BaseException:
 _base.LOGGER.critical('Exception in worker', exc_info=True)
 def shutdown(self, wait=True):
 with self._shutdown_lock:
 self._shutdown = True
 self._work_queue.put(None)
 if wait:
 for t in self._threads:
 t.join()

The attachment includes the new code

the attachment includes the new code

@Miles could you provide the code changes as a unified diff file, and if needed any changes to the test code as well, thanks.

The attachment includes the patch file

Correct me if I'm wrong, but this isn't actually an issue for CPython, right? The GIL ensures that when a thread writes to _shutdown, nothing else is reading it until the GIL is released and acquired by a new thread (which synchronizes _shutdown).
It might conceivably be a problem on non-CPython interpreters if they have no other form of inter-thread synchronization involved; that said, the locking involved in the self.work_queue.get(block=True) call likely synchronizes by side-effect even there.