1 # Copyright 2009 Brian Quinlan. All Rights Reserved.
2 # Licensed to PSF under a Contributor Agreement.
4 """Implements ProcessPoolExecutor.
6 The follow diagram and text describe the data-flow through the system:
8 |======================= In-process =====================|== Out-of-process ==|
10 +----------+ +----------+ +--------+ +-----------+ +---------+
11 | | => | Work Ids | => | | => | Call Q | => | |
12 | | +----------+ | | +-----------+ | |
13 | | | ... | | | | ... | | |
14 | | | 6 | | | | 5, call() | | |
15 | | | 7 | | | | ... | | |
16 | Process | | ... | | Local | +-----------+ | Process |
17 | Pool | +----------+ | Worker | | #1..n |
18 | Executor | | Thread | | |
19 | | +----------- + | | +-----------+ | |
20 | | <=> | Work Items | <=> | | <= | Result Q | <= | |
21 | | +------------+ | | +-----------+ | |
22 | | | 6: call() | | | | ... | | |
23 | | | future | | | | 4, result | | |
24 | | | ... | | | | 3, except | | |
25 +----------+ +------------+ +--------+ +-----------+ +---------+
27 Executor.submit() called:
28 - creates a uniquely numbered _WorkItem and adds it to the "Work Items" dict
29 - adds the id of the _WorkItem to the "Work Ids" queue
32 - reads work ids from the "Work Ids" queue and looks up the corresponding
33 WorkItem from the "Work Items" dict: if the work item has been cancelled then
34 it is simply removed from the dict, otherwise it is repackaged as a
35 _CallItem and put in the "Call Q". New _CallItems are put in the "Call Q"
36 until "Call Q" is full. NOTE: the size of the "Call Q" is kept small because
37 calls placed in the "Call Q" can no longer be cancelled with Future.cancel().
38 - reads _ResultItems from "Result Q", updates the future stored in the
39 "Work Items" dict and deletes the dict entry
42 - reads _CallItems from "Call Q", executes the calls, and puts the resulting
43 _ResultItems in "Request Q"
47 from concurrent.futures import _base
49 import multiprocessing
54 __author__ = 'Brian Quinlan (brian@sweetapp.com)'
56 # Workers are created as daemon threads and processes. This is done to allow the
57 # interpreter to exit when there are still idle processes in a
58 # ProcessPoolExecutor's process pool (i.e. shutdown() was not called). However,
59 # allowing workers to die with the interpreter has two undesirable properties:
60 # - The workers would still be running during interpretor shutdown,
61 # meaning that they would fail in unpredictable ways.
62 # - The workers could be killed while evaluating a work item, which could
63 # be bad if the callable being evaluated has external side-effects e.g.
66 # To work around this problem, an exit handler is installed which tells the
67 # workers to exit when their work queues are empty and then waits until the
68 # threads/processes finish.
70 _threads_queues = weakref.WeakKeyDictionary()
76 items = list(_threads_queues.items()) if _threads_queues else ()
82 # Controls how many more calls than processes will be queued in the call queue.
83 # A smaller number will mean that processes spend more time idle waiting for
84 # work while a larger number will make Future.cancel() succeed less frequently
85 # (Futures in the call queue cannot be cancelled).
86 EXTRA_QUEUED_CALLS = 1
88 class _WorkItem(object):
89 def __init__(self, future, fn, args, kwargs):
95 class _ResultItem(object):
96 def __init__(self, work_id, exception=None, result=None):
97 self.work_id = work_id
98 self.exception = exception
101 class _CallItem(object):
102 def __init__(self, work_id, fn, args, kwargs):
103 self.work_id = work_id
108 def _process_worker(call_queue, result_queue):
109 """Evaluates calls from call_queue and places the results in result_queue.
111 This worker is run in a separate process.
114 call_queue: A multiprocessing.Queue of _CallItems that will be read and
115 evaluated by the worker.
116 result_queue: A multiprocessing.Queue of _ResultItems that will written
118 shutdown: A multiprocessing.Event that will be set as a signal to the
119 worker that it should exit when call_queue is empty.
122 call_item = call_queue.get(block=True)
123 if call_item is None:
124 # Wake up queue management thread
125 result_queue.put(None)
128 r = call_item.fn(*call_item.args, **call_item.kwargs)
130 e = sys.exc_info()[1]
131 result_queue.put(_ResultItem(call_item.work_id,
134 result_queue.put(_ResultItem(call_item.work_id,
137 def _add_call_item_to_queue(pending_work_items,
140 """Fills call_queue with _WorkItems from pending_work_items.
142 This function never blocks.
145 pending_work_items: A dict mapping work ids to _WorkItems e.g.
146 {5: <_WorkItem...>, 6: <_WorkItem...>, ...}
147 work_ids: A queue.Queue of work ids e.g. Queue([5, 6, ...]). Work ids
148 are consumed and the corresponding _WorkItems from
149 pending_work_items are transformed into _CallItems and put in
151 call_queue: A multiprocessing.Queue that will be filled with _CallItems
152 derived from _WorkItems.
155 if call_queue.full():
158 work_id = work_ids.get(block=False)
162 work_item = pending_work_items[work_id]
164 if work_item.future.set_running_or_notify_cancel():
165 call_queue.put(_CallItem(work_id,
171 del pending_work_items[work_id]
174 def _queue_management_worker(executor_reference,
180 """Manages the communication between this process and the worker processes.
182 This function is run in a local thread.
185 executor_reference: A weakref.ref to the ProcessPoolExecutor that owns
186 this thread. Used to determine if the ProcessPoolExecutor has been
187 garbage collected and that this function can exit.
188 process: A list of the multiprocessing.Process instances used as
190 pending_work_items: A dict mapping work ids to _WorkItems e.g.
191 {5: <_WorkItem...>, 6: <_WorkItem...>, ...}
192 work_ids_queue: A queue.Queue of work ids e.g. Queue([5, 6, ...]).
193 call_queue: A multiprocessing.Queue that will be filled with _CallItems
194 derived from _WorkItems for processing by the process workers.
195 result_queue: A multiprocessing.Queue of _ResultItems generated by the
198 nb_shutdown_processes = [0]
199 def shutdown_one_process():
200 """Tell a worker to terminate, which will in turn wake us again"""
202 nb_shutdown_processes[0] += 1
204 _add_call_item_to_queue(pending_work_items,
208 result_item = result_queue.get(block=True)
209 if result_item is not None:
210 work_item = pending_work_items[result_item.work_id]
211 del pending_work_items[result_item.work_id]
213 if result_item.exception:
214 work_item.future.set_exception(result_item.exception)
216 work_item.future.set_result(result_item.result)
217 # Delete references to object. See issue16284
219 # Check whether we should start shutting down.
220 executor = executor_reference()
221 # No more work items can be added if:
222 # - The interpreter is shutting down OR
223 # - The executor that owns this worker has been collected OR
224 # - The executor that owns this worker has been shutdown.
225 if _shutdown or executor is None or executor._shutdown_thread:
226 # Since no new work items can be added, it is safe to shutdown
227 # this thread if there are no pending work items.
228 if not pending_work_items:
229 while nb_shutdown_processes[0] < len(processes):
230 shutdown_one_process()
231 # If .join() is not called on the created processes then
232 # some multiprocessing.Queue methods may deadlock on Mac OS
240 _system_limits_checked = False
241 _system_limited = None
242 def _check_system_limits():
243 global _system_limits_checked, _system_limited
244 if _system_limits_checked:
246 raise NotImplementedError(_system_limited)
247 _system_limits_checked = True
250 nsems_max = os.sysconf("SC_SEM_NSEMS_MAX")
251 except (AttributeError, ValueError):
252 # sysconf not available or setting not available
255 # indetermine limit, assume that limit is determined
256 # by available memory only
259 # minimum number of semaphores available
262 _system_limited = "system provides too few semaphores (%d available, 256 necessary)" % nsems_max
263 raise NotImplementedError(_system_limited)
266 class ProcessPoolExecutor(_base.Executor):
267 def __init__(self, max_workers=None):
268 """Initializes a new ProcessPoolExecutor instance.
271 max_workers: The maximum number of processes that can be used to
272 execute the given calls. If None or not given then as many
273 worker processes will be created as the machine has processors.
275 _check_system_limits()
277 if max_workers is None:
278 self._max_workers = multiprocessing.cpu_count()
281 raise ValueError("max_workers must be greater than 0")
283 self._max_workers = max_workers
285 # Make the call queue slightly larger than the number of processes to
286 # prevent the worker processes from idling. But don't make it too big
287 # because futures in the call queue cannot be cancelled.
288 self._call_queue = multiprocessing.Queue(self._max_workers +
290 self._result_queue = multiprocessing.Queue()
291 self._work_ids = queue.Queue()
292 self._queue_management_thread = None
293 self._processes = set()
295 # Shutdown is a two-step process.
296 self._shutdown_thread = False
297 self._shutdown_lock = threading.Lock()
298 self._queue_count = 0
299 self._pending_work_items = {}
301 def _start_queue_management_thread(self):
302 # When the executor gets lost, the weakref callback will wake up
303 # the queue management thread.
304 def weakref_cb(_, q=self._result_queue):
306 if self._queue_management_thread is None:
307 self._queue_management_thread = threading.Thread(
308 target=_queue_management_worker,
309 args=(weakref.ref(self, weakref_cb),
311 self._pending_work_items,
315 self._queue_management_thread.daemon = True
316 self._queue_management_thread.start()
317 _threads_queues[self._queue_management_thread] = self._result_queue
319 def _adjust_process_count(self):
320 for _ in range(len(self._processes), self._max_workers):
321 p = multiprocessing.Process(
322 target=_process_worker,
323 args=(self._call_queue,
326 self._processes.add(p)
328 def submit(self, fn, *args, **kwargs):
329 with self._shutdown_lock:
330 if self._shutdown_thread:
331 raise RuntimeError('cannot schedule new futures after shutdown')
334 w = _WorkItem(f, fn, args, kwargs)
336 self._pending_work_items[self._queue_count] = w
337 self._work_ids.put(self._queue_count)
338 self._queue_count += 1
339 # Wake up queue management thread
340 self._result_queue.put(None)
342 self._start_queue_management_thread()
343 self._adjust_process_count()
345 submit.__doc__ = _base.Executor.submit.__doc__
347 def shutdown(self, wait=True):
348 with self._shutdown_lock:
349 self._shutdown_thread = True
350 if self._queue_management_thread:
351 # Wake up queue management thread
352 self._result_queue.put(None)
354 self._queue_management_thread.join(sys.maxint)
355 # To reduce the risk of openning too many files, remove references to
356 # objects that use file descriptors.
357 self._queue_management_thread = None
358 self._call_queue = None
359 self._result_queue = None
360 self._processes = None
361 shutdown.__doc__ = _base.Executor.shutdown.__doc__
363 atexit.register(_python_exit)