blob: af7af2b16b9eb5a221f45e2fa38c6479e57dd04d [file] [log] [blame]
Benjamin Petersone711caf2008-06-11 16:44:04 +00001:mod:`multiprocessing` --- Process-based "threading" interface
2==============================================================
3
4.. module:: multiprocessing
5 :synopsis: Process-based "threading" interface.
6
Benjamin Petersone711caf2008-06-11 16:44:04 +00007
8Introduction
Georg Brandl49702152008-09-29 06:43:45 +00009------------
Benjamin Petersone711caf2008-06-11 16:44:04 +000010
Benjamin Peterson5289b2b2008-06-28 00:40:54 +000011:mod:`multiprocessing` is a package that supports spawning processes using an
12API similar to the :mod:`threading` module. The :mod:`multiprocessing` package
13offers both local and remote concurrency, effectively side-stepping the
14:term:`Global Interpreter Lock` by using subprocesses instead of threads. Due
15to this, the :mod:`multiprocessing` module allows the programmer to fully
16leverage multiple processors on a given machine. It runs on both Unix and
17Windows.
Benjamin Petersone711caf2008-06-11 16:44:04 +000018
Benjamin Petersone5384b02008-10-04 22:00:42 +000019.. warning::
20
21 Some of this package's functionality requires a functioning shared semaphore
Georg Brandl48310cd2009-01-03 21:18:54 +000022 implementation on the host operating system. Without one, the
23 :mod:`multiprocessing.synchronize` module will be disabled, and attempts to
24 import it will result in an :exc:`ImportError`. See
Benjamin Petersone5384b02008-10-04 22:00:42 +000025 :issue:`3770` for additional information.
Benjamin Petersone711caf2008-06-11 16:44:04 +000026
Jesse Noller45239682008-11-28 18:46:19 +000027.. note::
28
29 Functionality within this package requires that the ``__main__`` method be
30 importable by the children. This is covered in :ref:`multiprocessing-programming`
31 however it is worth pointing out here. This means that some examples, such
32 as the :class:`multiprocessing.Pool` examples will not work in the
33 interactive interpreter. For example::
34
35 >>> from multiprocessing import Pool
36 >>> p = Pool(5)
37 >>> def f(x):
Georg Brandla1c6a1c2009-01-03 21:26:05 +000038 ... return x*x
Georg Brandl48310cd2009-01-03 21:18:54 +000039 ...
Jesse Noller45239682008-11-28 18:46:19 +000040 >>> p.map(f, [1,2,3])
41 Process PoolWorker-1:
42 Process PoolWorker-2:
R. David Murray8e8099c2009-04-28 18:02:00 +000043 Process PoolWorker-3:
44 Traceback (most recent call last):
Jesse Noller45239682008-11-28 18:46:19 +000045 Traceback (most recent call last):
46 Traceback (most recent call last):
47 AttributeError: 'module' object has no attribute 'f'
48 AttributeError: 'module' object has no attribute 'f'
49 AttributeError: 'module' object has no attribute 'f'
50
R. David Murray8e8099c2009-04-28 18:02:00 +000051 (If you try this it will actually output three full tracebacks
52 interleaved in a semi-random fashion, and then you may have to
53 stop the master process somehow.)
54
Jesse Noller45239682008-11-28 18:46:19 +000055
Benjamin Petersone711caf2008-06-11 16:44:04 +000056The :class:`Process` class
57~~~~~~~~~~~~~~~~~~~~~~~~~~
58
59In :mod:`multiprocessing`, processes are spawned by creating a :class:`Process`
Benjamin Peterson5289b2b2008-06-28 00:40:54 +000060object and then calling its :meth:`~Process.start` method. :class:`Process`
Benjamin Petersone711caf2008-06-11 16:44:04 +000061follows the API of :class:`threading.Thread`. A trivial example of a
62multiprocess program is ::
63
Jesse Noller45239682008-11-28 18:46:19 +000064 from multiprocessing import Process
Benjamin Petersone711caf2008-06-11 16:44:04 +000065
66 def f(name):
Georg Brandl49702152008-09-29 06:43:45 +000067 print('hello', name)
Benjamin Petersone711caf2008-06-11 16:44:04 +000068
Jesse Noller45239682008-11-28 18:46:19 +000069 if __name__ == '__main__':
70 p = Process(target=f, args=('bob',))
71 p.start()
72 p.join()
Benjamin Petersone711caf2008-06-11 16:44:04 +000073
Jesse Noller45239682008-11-28 18:46:19 +000074To show the individual process IDs involved, here is an expanded example::
75
76 from multiprocessing import Process
77 import os
78
79 def info(title):
80 print title
81 print 'module name:', __name__
82 print 'parent process:', os.getppid()
83 print 'process id:', os.getpid()
Georg Brandl48310cd2009-01-03 21:18:54 +000084
Jesse Noller45239682008-11-28 18:46:19 +000085 def f(name):
86 info('function f')
87 print 'hello', name
Georg Brandl48310cd2009-01-03 21:18:54 +000088
Jesse Noller45239682008-11-28 18:46:19 +000089 if __name__ == '__main__':
90 info('main line')
91 p = Process(target=f, args=('bob',))
92 p.start()
93 p.join()
Benjamin Petersone711caf2008-06-11 16:44:04 +000094
95For an explanation of why (on Windows) the ``if __name__ == '__main__'`` part is
96necessary, see :ref:`multiprocessing-programming`.
97
98
99
100Exchanging objects between processes
101~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
102
103:mod:`multiprocessing` supports two types of communication channel between
104processes:
105
106**Queues**
107
Benjamin Peterson257060a2008-06-28 01:42:41 +0000108 The :class:`Queue` class is a near clone of :class:`queue.Queue`. For
Benjamin Petersone711caf2008-06-11 16:44:04 +0000109 example::
110
111 from multiprocessing import Process, Queue
112
113 def f(q):
114 q.put([42, None, 'hello'])
115
Georg Brandl1f01deb2009-01-03 22:47:39 +0000116 if __name__ == '__main__':
117 q = Queue()
118 p = Process(target=f, args=(q,))
119 p.start()
120 print(q.get()) # prints "[42, None, 'hello']"
121 p.join()
Benjamin Petersone711caf2008-06-11 16:44:04 +0000122
123 Queues are thread and process safe.
124
125**Pipes**
126
127 The :func:`Pipe` function returns a pair of connection objects connected by a
128 pipe which by default is duplex (two-way). For example::
129
130 from multiprocessing import Process, Pipe
131
132 def f(conn):
133 conn.send([42, None, 'hello'])
134 conn.close()
135
136 if __name__ == '__main__':
137 parent_conn, child_conn = Pipe()
138 p = Process(target=f, args=(child_conn,))
139 p.start()
Georg Brandl49702152008-09-29 06:43:45 +0000140 print(parent_conn.recv()) # prints "[42, None, 'hello']"
Benjamin Petersone711caf2008-06-11 16:44:04 +0000141 p.join()
142
143 The two connection objects returned by :func:`Pipe` represent the two ends of
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000144 the pipe. Each connection object has :meth:`~Connection.send` and
145 :meth:`~Connection.recv` methods (among others). Note that data in a pipe
146 may become corrupted if two processes (or threads) try to read from or write
147 to the *same* end of the pipe at the same time. Of course there is no risk
148 of corruption from processes using different ends of the pipe at the same
149 time.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000150
151
152Synchronization between processes
153~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
154
155:mod:`multiprocessing` contains equivalents of all the synchronization
156primitives from :mod:`threading`. For instance one can use a lock to ensure
157that only one process prints to standard output at a time::
158
159 from multiprocessing import Process, Lock
160
161 def f(l, i):
162 l.acquire()
Georg Brandl49702152008-09-29 06:43:45 +0000163 print('hello world', i)
Benjamin Petersone711caf2008-06-11 16:44:04 +0000164 l.release()
165
166 if __name__ == '__main__':
167 lock = Lock()
168
169 for num in range(10):
170 Process(target=f, args=(lock, num)).start()
171
172Without using the lock output from the different processes is liable to get all
173mixed up.
174
175
176Sharing state between processes
177~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
178
179As mentioned above, when doing concurrent programming it is usually best to
180avoid using shared state as far as possible. This is particularly true when
181using multiple processes.
182
183However, if you really do need to use some shared data then
184:mod:`multiprocessing` provides a couple of ways of doing so.
185
186**Shared memory**
187
188 Data can be stored in a shared memory map using :class:`Value` or
189 :class:`Array`. For example, the following code ::
190
191 from multiprocessing import Process, Value, Array
192
193 def f(n, a):
194 n.value = 3.1415927
195 for i in range(len(a)):
196 a[i] = -a[i]
197
198 if __name__ == '__main__':
199 num = Value('d', 0.0)
200 arr = Array('i', range(10))
201
202 p = Process(target=f, args=(num, arr))
203 p.start()
204 p.join()
205
Georg Brandl49702152008-09-29 06:43:45 +0000206 print(num.value)
207 print(arr[:])
Benjamin Petersone711caf2008-06-11 16:44:04 +0000208
209 will print ::
210
211 3.1415927
212 [0, -1, -2, -3, -4, -5, -6, -7, -8, -9]
213
214 The ``'d'`` and ``'i'`` arguments used when creating ``num`` and ``arr`` are
215 typecodes of the kind used by the :mod:`array` module: ``'d'`` indicates a
Georg Brandl2ee470f2008-07-16 12:55:28 +0000216 double precision float and ``'i'`` indicates a signed integer. These shared
Benjamin Petersone711caf2008-06-11 16:44:04 +0000217 objects will be process and thread safe.
218
219 For more flexibility in using shared memory one can use the
220 :mod:`multiprocessing.sharedctypes` module which supports the creation of
221 arbitrary ctypes objects allocated from shared memory.
222
223**Server process**
224
225 A manager object returned by :func:`Manager` controls a server process which
Georg Brandl2ee470f2008-07-16 12:55:28 +0000226 holds Python objects and allows other processes to manipulate them using
Benjamin Petersone711caf2008-06-11 16:44:04 +0000227 proxies.
228
229 A manager returned by :func:`Manager` will support types :class:`list`,
230 :class:`dict`, :class:`Namespace`, :class:`Lock`, :class:`RLock`,
231 :class:`Semaphore`, :class:`BoundedSemaphore`, :class:`Condition`,
232 :class:`Event`, :class:`Queue`, :class:`Value` and :class:`Array`. For
233 example, ::
234
235 from multiprocessing import Process, Manager
236
237 def f(d, l):
238 d[1] = '1'
239 d['2'] = 2
240 d[0.25] = None
241 l.reverse()
242
243 if __name__ == '__main__':
244 manager = Manager()
245
246 d = manager.dict()
247 l = manager.list(range(10))
248
249 p = Process(target=f, args=(d, l))
250 p.start()
251 p.join()
252
Georg Brandl49702152008-09-29 06:43:45 +0000253 print(d)
254 print(l)
Benjamin Petersone711caf2008-06-11 16:44:04 +0000255
256 will print ::
257
258 {0.25: None, 1: '1', '2': 2}
259 [9, 8, 7, 6, 5, 4, 3, 2, 1, 0]
260
261 Server process managers are more flexible than using shared memory objects
262 because they can be made to support arbitrary object types. Also, a single
263 manager can be shared by processes on different computers over a network.
264 They are, however, slower than using shared memory.
265
266
267Using a pool of workers
268~~~~~~~~~~~~~~~~~~~~~~~
269
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000270The :class:`~multiprocessing.pool.Pool` class represents a pool of worker
Benjamin Petersone711caf2008-06-11 16:44:04 +0000271processes. It has methods which allows tasks to be offloaded to the worker
272processes in a few different ways.
273
274For example::
275
276 from multiprocessing import Pool
277
278 def f(x):
279 return x*x
280
281 if __name__ == '__main__':
Jesse Noller45239682008-11-28 18:46:19 +0000282 pool = Pool(processes=4) # start 4 worker processes
283 result = pool.apply_async(f, [10]) # evaluate "f(10)" asynchronously
284 print result.get(timeout=1) # prints "100" unless your computer is *very* slow
285 print pool.map(f, range(10)) # prints "[0, 1, 4,..., 81]"
Benjamin Petersone711caf2008-06-11 16:44:04 +0000286
287
288Reference
289---------
290
291The :mod:`multiprocessing` package mostly replicates the API of the
292:mod:`threading` module.
293
294
295:class:`Process` and exceptions
296~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
297
298.. class:: Process([group[, target[, name[, args[, kwargs]]]]])
299
300 Process objects represent activity that is run in a separate process. The
301 :class:`Process` class has equivalents of all the methods of
302 :class:`threading.Thread`.
303
304 The constructor should always be called with keyword arguments. *group*
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000305 should always be ``None``; it exists solely for compatibility with
Benjamin Petersona786b022008-08-25 21:05:21 +0000306 :class:`threading.Thread`. *target* is the callable object to be invoked by
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000307 the :meth:`run()` method. It defaults to ``None``, meaning nothing is
Benjamin Petersone711caf2008-06-11 16:44:04 +0000308 called. *name* is the process name. By default, a unique name is constructed
309 of the form 'Process-N\ :sub:`1`:N\ :sub:`2`:...:N\ :sub:`k`' where N\
310 :sub:`1`,N\ :sub:`2`,...,N\ :sub:`k` is a sequence of integers whose length
311 is determined by the *generation* of the process. *args* is the argument
312 tuple for the target invocation. *kwargs* is a dictionary of keyword
313 arguments for the target invocation. By default, no arguments are passed to
314 *target*.
315
316 If a subclass overrides the constructor, it must make sure it invokes the
317 base class constructor (:meth:`Process.__init__`) before doing anything else
318 to the process.
319
320 .. method:: run()
321
322 Method representing the process's activity.
323
324 You may override this method in a subclass. The standard :meth:`run`
325 method invokes the callable object passed to the object's constructor as
326 the target argument, if any, with sequential and keyword arguments taken
327 from the *args* and *kwargs* arguments, respectively.
328
329 .. method:: start()
330
331 Start the process's activity.
332
333 This must be called at most once per process object. It arranges for the
334 object's :meth:`run` method to be invoked in a separate process.
335
336 .. method:: join([timeout])
337
338 Block the calling thread until the process whose :meth:`join` method is
339 called terminates or until the optional timeout occurs.
340
341 If *timeout* is ``None`` then there is no timeout.
342
343 A process can be joined many times.
344
345 A process cannot join itself because this would cause a deadlock. It is
346 an error to attempt to join a process before it has been started.
347
Benjamin Petersona786b022008-08-25 21:05:21 +0000348 .. attribute:: name
Benjamin Petersone711caf2008-06-11 16:44:04 +0000349
Benjamin Petersona786b022008-08-25 21:05:21 +0000350 The process's name.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000351
352 The name is a string used for identification purposes only. It has no
353 semantics. Multiple processes may be given the same name. The initial
354 name is set by the constructor.
355
Jesse Noller45239682008-11-28 18:46:19 +0000356 .. method:: is_alive
Benjamin Petersone711caf2008-06-11 16:44:04 +0000357
358 Return whether the process is alive.
359
360 Roughly, a process object is alive from the moment the :meth:`start`
361 method returns until the child process terminates.
362
Benjamin Petersona786b022008-08-25 21:05:21 +0000363 .. attribute:: daemon
Benjamin Petersone711caf2008-06-11 16:44:04 +0000364
Benjamin Petersonda10d3b2009-01-01 00:23:30 +0000365 The process's daemon flag, a Boolean value. This must be set before
Benjamin Petersona786b022008-08-25 21:05:21 +0000366 :meth:`start` is called.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000367
368 The initial value is inherited from the creating process.
369
370 When a process exits, it attempts to terminate all of its daemonic child
371 processes.
372
373 Note that a daemonic process is not allowed to create child processes.
374 Otherwise a daemonic process would leave its children orphaned if it gets
Alexandre Vassalotti260484d2009-07-17 11:43:26 +0000375 terminated when its parent process exits. Additionally, these are **not**
376 Unix daemons or services, they are normal processes that will be
377 terminated (and not joined) if non-dameonic processes have exited.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000378
Benjamin Petersona786b022008-08-25 21:05:21 +0000379 In addition to the :class:`Threading.Thread` API, :class:`Process` objects
380 also support the following attributes and methods:
Benjamin Petersone711caf2008-06-11 16:44:04 +0000381
Benjamin Petersona786b022008-08-25 21:05:21 +0000382 .. attribute:: pid
Benjamin Petersone711caf2008-06-11 16:44:04 +0000383
384 Return the process ID. Before the process is spawned, this will be
385 ``None``.
386
Benjamin Petersona786b022008-08-25 21:05:21 +0000387 .. attribute:: exitcode
Benjamin Petersone711caf2008-06-11 16:44:04 +0000388
Benjamin Petersona786b022008-08-25 21:05:21 +0000389 The child's exit code. This will be ``None`` if the process has not yet
390 terminated. A negative value *-N* indicates that the child was terminated
391 by signal *N*.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000392
Benjamin Petersona786b022008-08-25 21:05:21 +0000393 .. attribute:: authkey
Benjamin Petersone711caf2008-06-11 16:44:04 +0000394
Benjamin Petersona786b022008-08-25 21:05:21 +0000395 The process's authentication key (a byte string).
Benjamin Petersone711caf2008-06-11 16:44:04 +0000396
397 When :mod:`multiprocessing` is initialized the main process is assigned a
398 random string using :func:`os.random`.
399
400 When a :class:`Process` object is created, it will inherit the
Benjamin Petersona786b022008-08-25 21:05:21 +0000401 authentication key of its parent process, although this may be changed by
402 setting :attr:`authkey` to another byte string.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000403
404 See :ref:`multiprocessing-auth-keys`.
405
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000406 .. method:: terminate()
Benjamin Petersone711caf2008-06-11 16:44:04 +0000407
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000408 Terminate the process. On Unix this is done using the ``SIGTERM`` signal;
409 on Windows :cfunc:`TerminateProcess` is used. Note that exit handlers and
410 finally clauses, etc., will not be executed.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000411
412 Note that descendant processes of the process will *not* be terminated --
413 they will simply become orphaned.
414
415 .. warning::
416
417 If this method is used when the associated process is using a pipe or
418 queue then the pipe or queue is liable to become corrupted and may
419 become unusable by other process. Similarly, if the process has
420 acquired a lock or semaphore etc. then terminating it is liable to
421 cause other processes to deadlock.
422
423 Note that the :meth:`start`, :meth:`join`, :meth:`is_alive` and
Benjamin Petersona786b022008-08-25 21:05:21 +0000424 :attr:`exit_code` methods should only be called by the process that created
425 the process object.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000426
R. David Murray8e8099c2009-04-28 18:02:00 +0000427 Example usage of some of the methods of :class:`Process`:
428
429 .. doctest::
Benjamin Petersone711caf2008-06-11 16:44:04 +0000430
Benjamin Peterson206e3072008-10-19 14:07:49 +0000431 >>> import multiprocessing, time, signal
432 >>> p = multiprocessing.Process(target=time.sleep, args=(1000,))
Georg Brandl49702152008-09-29 06:43:45 +0000433 >>> print(p, p.is_alive())
Benjamin Petersone711caf2008-06-11 16:44:04 +0000434 <Process(Process-1, initial)> False
435 >>> p.start()
Georg Brandl49702152008-09-29 06:43:45 +0000436 >>> print(p, p.is_alive())
Benjamin Petersone711caf2008-06-11 16:44:04 +0000437 <Process(Process-1, started)> True
438 >>> p.terminate()
R. David Murray8e8099c2009-04-28 18:02:00 +0000439 >>> time.sleep(0.1)
Georg Brandl49702152008-09-29 06:43:45 +0000440 >>> print(p, p.is_alive())
Benjamin Petersone711caf2008-06-11 16:44:04 +0000441 <Process(Process-1, stopped[SIGTERM])> False
Benjamin Petersona786b022008-08-25 21:05:21 +0000442 >>> p.exitcode == -signal.SIGTERM
Benjamin Petersone711caf2008-06-11 16:44:04 +0000443 True
444
445
446.. exception:: BufferTooShort
447
448 Exception raised by :meth:`Connection.recv_bytes_into()` when the supplied
449 buffer object is too small for the message read.
450
451 If ``e`` is an instance of :exc:`BufferTooShort` then ``e.args[0]`` will give
452 the message as a byte string.
453
454
455Pipes and Queues
456~~~~~~~~~~~~~~~~
457
458When using multiple processes, one generally uses message passing for
459communication between processes and avoids having to use any synchronization
460primitives like locks.
461
462For passing messages one can use :func:`Pipe` (for a connection between two
463processes) or a queue (which allows multiple producers and consumers).
464
465The :class:`Queue` and :class:`JoinableQueue` types are multi-producer,
Benjamin Peterson257060a2008-06-28 01:42:41 +0000466multi-consumer FIFO queues modelled on the :class:`queue.Queue` class in the
Benjamin Petersone711caf2008-06-11 16:44:04 +0000467standard library. They differ in that :class:`Queue` lacks the
Benjamin Peterson257060a2008-06-28 01:42:41 +0000468:meth:`~queue.Queue.task_done` and :meth:`~queue.Queue.join` methods introduced
469into Python 2.5's :class:`queue.Queue` class.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000470
471If you use :class:`JoinableQueue` then you **must** call
472:meth:`JoinableQueue.task_done` for each task removed from the queue or else the
473semaphore used to count the number of unfinished tasks may eventually overflow
474raising an exception.
475
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000476Note that one can also create a shared queue by using a manager object -- see
477:ref:`multiprocessing-managers`.
478
Benjamin Petersone711caf2008-06-11 16:44:04 +0000479.. note::
480
Benjamin Peterson257060a2008-06-28 01:42:41 +0000481 :mod:`multiprocessing` uses the usual :exc:`queue.Empty` and
482 :exc:`queue.Full` exceptions to signal a timeout. They are not available in
Benjamin Petersone711caf2008-06-11 16:44:04 +0000483 the :mod:`multiprocessing` namespace so you need to import them from
Benjamin Peterson257060a2008-06-28 01:42:41 +0000484 :mod:`queue`.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000485
486
487.. warning::
488
489 If a process is killed using :meth:`Process.terminate` or :func:`os.kill`
490 while it is trying to use a :class:`Queue`, then the data in the queue is
491 likely to become corrupted. This may cause any other processes to get an
492 exception when it tries to use the queue later on.
493
494.. warning::
495
496 As mentioned above, if a child process has put items on a queue (and it has
497 not used :meth:`JoinableQueue.cancel_join_thread`), then that process will
498 not terminate until all buffered items have been flushed to the pipe.
499
500 This means that if you try joining that process you may get a deadlock unless
501 you are sure that all items which have been put on the queue have been
502 consumed. Similarly, if the child process is non-daemonic then the parent
Georg Brandl2ee470f2008-07-16 12:55:28 +0000503 process may hang on exit when it tries to join all its non-daemonic children.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000504
505 Note that a queue created using a manager does not have this issue. See
506 :ref:`multiprocessing-programming`.
507
Benjamin Petersone711caf2008-06-11 16:44:04 +0000508For an example of the usage of queues for interprocess communication see
509:ref:`multiprocessing-examples`.
510
511
512.. function:: Pipe([duplex])
513
514 Returns a pair ``(conn1, conn2)`` of :class:`Connection` objects representing
515 the ends of a pipe.
516
517 If *duplex* is ``True`` (the default) then the pipe is bidirectional. If
518 *duplex* is ``False`` then the pipe is unidirectional: ``conn1`` can only be
519 used for receiving messages and ``conn2`` can only be used for sending
520 messages.
521
522
523.. class:: Queue([maxsize])
524
525 Returns a process shared queue implemented using a pipe and a few
526 locks/semaphores. When a process first puts an item on the queue a feeder
527 thread is started which transfers objects from a buffer into the pipe.
528
Benjamin Peterson257060a2008-06-28 01:42:41 +0000529 The usual :exc:`queue.Empty` and :exc:`queue.Full` exceptions from the
Benjamin Petersone711caf2008-06-11 16:44:04 +0000530 standard library's :mod:`Queue` module are raised to signal timeouts.
531
Benjamin Peterson257060a2008-06-28 01:42:41 +0000532 :class:`Queue` implements all the methods of :class:`queue.Queue` except for
533 :meth:`~queue.Queue.task_done` and :meth:`~queue.Queue.join`.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000534
535 .. method:: qsize()
536
537 Return the approximate size of the queue. Because of
538 multithreading/multiprocessing semantics, this number is not reliable.
539
540 Note that this may raise :exc:`NotImplementedError` on Unix platforms like
Georg Brandlc575c902008-09-13 17:46:05 +0000541 Mac OS X where ``sem_getvalue()`` is not implemented.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000542
543 .. method:: empty()
544
545 Return ``True`` if the queue is empty, ``False`` otherwise. Because of
546 multithreading/multiprocessing semantics, this is not reliable.
547
548 .. method:: full()
549
550 Return ``True`` if the queue is full, ``False`` otherwise. Because of
551 multithreading/multiprocessing semantics, this is not reliable.
552
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000553 .. method:: put(item[, block[, timeout]])
Benjamin Petersone711caf2008-06-11 16:44:04 +0000554
Georg Brandl48310cd2009-01-03 21:18:54 +0000555 Put item into the queue. If the optional argument *block* is ``True``
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000556 (the default) and *timeout* is ``None`` (the default), block if necessary until
Benjamin Petersone711caf2008-06-11 16:44:04 +0000557 a free slot is available. If *timeout* is a positive number, it blocks at
Benjamin Peterson257060a2008-06-28 01:42:41 +0000558 most *timeout* seconds and raises the :exc:`queue.Full` exception if no
Benjamin Petersone711caf2008-06-11 16:44:04 +0000559 free slot was available within that time. Otherwise (*block* is
560 ``False``), put an item on the queue if a free slot is immediately
Benjamin Peterson257060a2008-06-28 01:42:41 +0000561 available, else raise the :exc:`queue.Full` exception (*timeout* is
Benjamin Petersone711caf2008-06-11 16:44:04 +0000562 ignored in that case).
563
564 .. method:: put_nowait(item)
565
566 Equivalent to ``put(item, False)``.
567
568 .. method:: get([block[, timeout]])
569
570 Remove and return an item from the queue. If optional args *block* is
571 ``True`` (the default) and *timeout* is ``None`` (the default), block if
572 necessary until an item is available. If *timeout* is a positive number,
Benjamin Peterson257060a2008-06-28 01:42:41 +0000573 it blocks at most *timeout* seconds and raises the :exc:`queue.Empty`
Benjamin Petersone711caf2008-06-11 16:44:04 +0000574 exception if no item was available within that time. Otherwise (block is
575 ``False``), return an item if one is immediately available, else raise the
Benjamin Peterson257060a2008-06-28 01:42:41 +0000576 :exc:`queue.Empty` exception (*timeout* is ignored in that case).
Benjamin Petersone711caf2008-06-11 16:44:04 +0000577
578 .. method:: get_nowait()
579 get_no_wait()
580
581 Equivalent to ``get(False)``.
582
583 :class:`multiprocessing.Queue` has a few additional methods not found in
Georg Brandl2ee470f2008-07-16 12:55:28 +0000584 :class:`queue.Queue`. These methods are usually unnecessary for most
585 code:
Benjamin Petersone711caf2008-06-11 16:44:04 +0000586
587 .. method:: close()
588
589 Indicate that no more data will be put on this queue by the current
590 process. The background thread will quit once it has flushed all buffered
591 data to the pipe. This is called automatically when the queue is garbage
592 collected.
593
594 .. method:: join_thread()
595
596 Join the background thread. This can only be used after :meth:`close` has
597 been called. It blocks until the background thread exits, ensuring that
598 all data in the buffer has been flushed to the pipe.
599
600 By default if a process is not the creator of the queue then on exit it
601 will attempt to join the queue's background thread. The process can call
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000602 :meth:`cancel_join_thread` to make :meth:`join_thread` do nothing.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000603
604 .. method:: cancel_join_thread()
605
606 Prevent :meth:`join_thread` from blocking. In particular, this prevents
607 the background thread from being joined automatically when the process
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000608 exits -- see :meth:`join_thread`.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000609
610
611.. class:: JoinableQueue([maxsize])
612
613 :class:`JoinableQueue`, a :class:`Queue` subclass, is a queue which
614 additionally has :meth:`task_done` and :meth:`join` methods.
615
616 .. method:: task_done()
617
618 Indicate that a formerly enqueued task is complete. Used by queue consumer
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000619 threads. For each :meth:`~Queue.get` used to fetch a task, a subsequent
620 call to :meth:`task_done` tells the queue that the processing on the task
621 is complete.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000622
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000623 If a :meth:`~Queue.join` is currently blocking, it will resume when all
624 items have been processed (meaning that a :meth:`task_done` call was
625 received for every item that had been :meth:`~Queue.put` into the queue).
Benjamin Petersone711caf2008-06-11 16:44:04 +0000626
627 Raises a :exc:`ValueError` if called more times than there were items
628 placed in the queue.
629
630
631 .. method:: join()
632
633 Block until all items in the queue have been gotten and processed.
634
635 The count of unfinished tasks goes up whenever an item is added to the
636 queue. The count goes down whenever a consumer thread calls
637 :meth:`task_done` to indicate that the item was retrieved and all work on
638 it is complete. When the count of unfinished tasks drops to zero,
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000639 :meth:`~Queue.join` unblocks.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000640
641
642Miscellaneous
643~~~~~~~~~~~~~
644
645.. function:: active_children()
646
647 Return list of all live children of the current process.
648
649 Calling this has the side affect of "joining" any processes which have
650 already finished.
651
652.. function:: cpu_count()
653
654 Return the number of CPUs in the system. May raise
655 :exc:`NotImplementedError`.
656
657.. function:: current_process()
658
659 Return the :class:`Process` object corresponding to the current process.
660
661 An analogue of :func:`threading.current_thread`.
662
663.. function:: freeze_support()
664
665 Add support for when a program which uses :mod:`multiprocessing` has been
666 frozen to produce a Windows executable. (Has been tested with **py2exe**,
667 **PyInstaller** and **cx_Freeze**.)
668
669 One needs to call this function straight after the ``if __name__ ==
670 '__main__'`` line of the main module. For example::
671
672 from multiprocessing import Process, freeze_support
673
674 def f():
Georg Brandl49702152008-09-29 06:43:45 +0000675 print('hello world!')
Benjamin Petersone711caf2008-06-11 16:44:04 +0000676
677 if __name__ == '__main__':
678 freeze_support()
679 Process(target=f).start()
680
R. David Murray8e8099c2009-04-28 18:02:00 +0000681 If the ``freeze_support()`` line is omitted then trying to run the frozen
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000682 executable will raise :exc:`RuntimeError`.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000683
684 If the module is being run normally by the Python interpreter then
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000685 :func:`freeze_support` has no effect.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000686
687.. function:: set_executable()
688
689 Sets the path of the python interpreter to use when starting a child process.
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000690 (By default :data:`sys.executable` is used). Embedders will probably need to
691 do some thing like ::
Benjamin Petersone711caf2008-06-11 16:44:04 +0000692
693 setExecutable(os.path.join(sys.exec_prefix, 'pythonw.exe'))
694
R. David Murray8e8099c2009-04-28 18:02:00 +0000695 before they can create child processes. (Windows only)
Benjamin Petersone711caf2008-06-11 16:44:04 +0000696
697
698.. note::
699
700 :mod:`multiprocessing` contains no analogues of
701 :func:`threading.active_count`, :func:`threading.enumerate`,
702 :func:`threading.settrace`, :func:`threading.setprofile`,
703 :class:`threading.Timer`, or :class:`threading.local`.
704
705
706Connection Objects
707~~~~~~~~~~~~~~~~~~
708
709Connection objects allow the sending and receiving of picklable objects or
710strings. They can be thought of as message oriented connected sockets.
711
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000712Connection objects usually created using :func:`Pipe` -- see also
Benjamin Petersone711caf2008-06-11 16:44:04 +0000713:ref:`multiprocessing-listeners-clients`.
714
715.. class:: Connection
716
717 .. method:: send(obj)
718
719 Send an object to the other end of the connection which should be read
720 using :meth:`recv`.
721
Benjamin Peterson965ce872009-04-05 21:24:58 +0000722 The object must be picklable. Very large pickles (approximately 32 MB+,
723 though it depends on the OS) may raise a ValueError exception.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000724
725 .. method:: recv()
726
727 Return an object sent from the other end of the connection using
728 :meth:`send`. Raises :exc:`EOFError` if there is nothing left to receive
729 and the other end was closed.
730
731 .. method:: fileno()
732
733 Returns the file descriptor or handle used by the connection.
734
735 .. method:: close()
736
737 Close the connection.
738
739 This is called automatically when the connection is garbage collected.
740
741 .. method:: poll([timeout])
742
743 Return whether there is any data available to be read.
744
745 If *timeout* is not specified then it will return immediately. If
746 *timeout* is a number then this specifies the maximum time in seconds to
747 block. If *timeout* is ``None`` then an infinite timeout is used.
748
749 .. method:: send_bytes(buffer[, offset[, size]])
750
751 Send byte data from an object supporting the buffer interface as a
752 complete message.
753
754 If *offset* is given then data is read from that position in *buffer*. If
Benjamin Peterson965ce872009-04-05 21:24:58 +0000755 *size* is given then that many bytes will be read from buffer. Very large
756 buffers (approximately 32 MB+, though it depends on the OS) may raise a
757 ValueError exception
Benjamin Petersone711caf2008-06-11 16:44:04 +0000758
759 .. method:: recv_bytes([maxlength])
760
761 Return a complete message of byte data sent from the other end of the
762 connection as a string. Raises :exc:`EOFError` if there is nothing left
763 to receive and the other end has closed.
764
765 If *maxlength* is specified and the message is longer than *maxlength*
766 then :exc:`IOError` is raised and the connection will no longer be
767 readable.
768
769 .. method:: recv_bytes_into(buffer[, offset])
770
771 Read into *buffer* a complete message of byte data sent from the other end
772 of the connection and return the number of bytes in the message. Raises
773 :exc:`EOFError` if there is nothing left to receive and the other end was
774 closed.
775
776 *buffer* must be an object satisfying the writable buffer interface. If
777 *offset* is given then the message will be written into the buffer from
R. David Murray8e8099c2009-04-28 18:02:00 +0000778 that position. Offset must be a non-negative integer less than the
779 length of *buffer* (in bytes).
Benjamin Petersone711caf2008-06-11 16:44:04 +0000780
781 If the buffer is too short then a :exc:`BufferTooShort` exception is
782 raised and the complete message is available as ``e.args[0]`` where ``e``
783 is the exception instance.
784
785
786For example:
787
R. David Murray8e8099c2009-04-28 18:02:00 +0000788.. doctest::
789
Benjamin Petersone711caf2008-06-11 16:44:04 +0000790 >>> from multiprocessing import Pipe
791 >>> a, b = Pipe()
792 >>> a.send([1, 'hello', None])
793 >>> b.recv()
794 [1, 'hello', None]
795 >>> b.send_bytes('thank you')
796 >>> a.recv_bytes()
797 'thank you'
798 >>> import array
799 >>> arr1 = array.array('i', range(5))
800 >>> arr2 = array.array('i', [0] * 10)
801 >>> a.send_bytes(arr1)
802 >>> count = b.recv_bytes_into(arr2)
803 >>> assert count == len(arr1) * arr1.itemsize
804 >>> arr2
805 array('i', [0, 1, 2, 3, 4, 0, 0, 0, 0, 0])
806
807
808.. warning::
809
810 The :meth:`Connection.recv` method automatically unpickles the data it
811 receives, which can be a security risk unless you can trust the process
812 which sent the message.
813
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000814 Therefore, unless the connection object was produced using :func:`Pipe` you
815 should only use the :meth:`~Connection.recv` and :meth:`~Connection.send`
816 methods after performing some sort of authentication. See
817 :ref:`multiprocessing-auth-keys`.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000818
819.. warning::
820
821 If a process is killed while it is trying to read or write to a pipe then
822 the data in the pipe is likely to become corrupted, because it may become
823 impossible to be sure where the message boundaries lie.
824
825
826Synchronization primitives
827~~~~~~~~~~~~~~~~~~~~~~~~~~
828
829Generally synchronization primitives are not as necessary in a multiprocess
Georg Brandl2ee470f2008-07-16 12:55:28 +0000830program as they are in a multithreaded program. See the documentation for
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000831:mod:`threading` module.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000832
833Note that one can also create synchronization primitives by using a manager
834object -- see :ref:`multiprocessing-managers`.
835
836.. class:: BoundedSemaphore([value])
837
838 A bounded semaphore object: a clone of :class:`threading.BoundedSemaphore`.
839
Georg Brandlc575c902008-09-13 17:46:05 +0000840 (On Mac OS X this is indistinguishable from :class:`Semaphore` because
Benjamin Petersone711caf2008-06-11 16:44:04 +0000841 ``sem_getvalue()`` is not implemented on that platform).
842
843.. class:: Condition([lock])
844
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000845 A condition variable: a clone of :class:`threading.Condition`.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000846
847 If *lock* is specified then it should be a :class:`Lock` or :class:`RLock`
848 object from :mod:`multiprocessing`.
849
850.. class:: Event()
851
852 A clone of :class:`threading.Event`.
Benjamin Peterson965ce872009-04-05 21:24:58 +0000853 This method returns the state of the internal semaphore on exit, so it
854 will always return ``True`` except if a timeout is given and the operation
855 times out.
856
Raymond Hettinger35a88362009-04-09 00:08:24 +0000857 .. versionchanged:: 3.1
Benjamin Peterson965ce872009-04-05 21:24:58 +0000858 Previously, the method always returned ``None``.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000859
860.. class:: Lock()
861
862 A non-recursive lock object: a clone of :class:`threading.Lock`.
863
864.. class:: RLock()
865
866 A recursive lock object: a clone of :class:`threading.RLock`.
867
868.. class:: Semaphore([value])
869
870 A bounded semaphore object: a clone of :class:`threading.Semaphore`.
871
872.. note::
873
Benjamin Peterson5289b2b2008-06-28 00:40:54 +0000874 The :meth:`acquire` method of :class:`BoundedSemaphore`, :class:`Lock`,
Benjamin Petersone711caf2008-06-11 16:44:04 +0000875 :class:`RLock` and :class:`Semaphore` has a timeout parameter not supported
876 by the equivalents in :mod:`threading`. The signature is
877 ``acquire(block=True, timeout=None)`` with keyword parameters being
878 acceptable. If *block* is ``True`` and *timeout* is not ``None`` then it
879 specifies a timeout in seconds. If *block* is ``False`` then *timeout* is
880 ignored.
Georg Brandl48310cd2009-01-03 21:18:54 +0000881
R. David Murray8e8099c2009-04-28 18:02:00 +0000882.. note::
883 On OS/X ``sem_timedwait`` is unsupported, so timeout arguments for the
884 aforementioned :meth:`acquire` methods will be ignored on OS/X.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000885
886.. note::
887
888 If the SIGINT signal generated by Ctrl-C arrives while the main thread is
889 blocked by a call to :meth:`BoundedSemaphore.acquire`, :meth:`Lock.acquire`,
890 :meth:`RLock.acquire`, :meth:`Semaphore.acquire`, :meth:`Condition.acquire`
891 or :meth:`Condition.wait` then the call will be immediately interrupted and
892 :exc:`KeyboardInterrupt` will be raised.
893
894 This differs from the behaviour of :mod:`threading` where SIGINT will be
895 ignored while the equivalent blocking calls are in progress.
896
897
898Shared :mod:`ctypes` Objects
899~~~~~~~~~~~~~~~~~~~~~~~~~~~~
900
901It is possible to create shared objects using shared memory which can be
902inherited by child processes.
903
Jesse Nollerb0516a62009-01-18 03:11:38 +0000904.. function:: Value(typecode_or_type, *args[, lock])
Benjamin Petersone711caf2008-06-11 16:44:04 +0000905
906 Return a :mod:`ctypes` object allocated from shared memory. By default the
907 return value is actually a synchronized wrapper for the object.
908
909 *typecode_or_type* determines the type of the returned object: it is either a
910 ctypes type or a one character typecode of the kind used by the :mod:`array`
911 module. *\*args* is passed on to the constructor for the type.
912
913 If *lock* is ``True`` (the default) then a new lock object is created to
914 synchronize access to the value. If *lock* is a :class:`Lock` or
915 :class:`RLock` object then that will be used to synchronize access to the
916 value. If *lock* is ``False`` then access to the returned object will not be
917 automatically protected by a lock, so it will not necessarily be
918 "process-safe".
919
920 Note that *lock* is a keyword-only argument.
921
922.. function:: Array(typecode_or_type, size_or_initializer, *, lock=True)
923
924 Return a ctypes array allocated from shared memory. By default the return
925 value is actually a synchronized wrapper for the array.
926
927 *typecode_or_type* determines the type of the elements of the returned array:
928 it is either a ctypes type or a one character typecode of the kind used by
929 the :mod:`array` module. If *size_or_initializer* is an integer, then it
930 determines the length of the array, and the array will be initially zeroed.
931 Otherwise, *size_or_initializer* is a sequence which is used to initialize
932 the array and whose length determines the length of the array.
933
934 If *lock* is ``True`` (the default) then a new lock object is created to
935 synchronize access to the value. If *lock* is a :class:`Lock` or
936 :class:`RLock` object then that will be used to synchronize access to the
937 value. If *lock* is ``False`` then access to the returned object will not be
938 automatically protected by a lock, so it will not necessarily be
939 "process-safe".
940
941 Note that *lock* is a keyword only argument.
942
Amaury Forgeot d'Arcb0c29162008-11-22 22:18:04 +0000943 Note that an array of :data:`ctypes.c_char` has *value* and *raw*
Benjamin Petersone711caf2008-06-11 16:44:04 +0000944 attributes which allow one to use it to store and retrieve strings.
945
946
947The :mod:`multiprocessing.sharedctypes` module
948>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
949
950.. module:: multiprocessing.sharedctypes
951 :synopsis: Allocate ctypes objects from shared memory.
952
953The :mod:`multiprocessing.sharedctypes` module provides functions for allocating
954:mod:`ctypes` objects from shared memory which can be inherited by child
955processes.
956
957.. note::
958
Georg Brandl2ee470f2008-07-16 12:55:28 +0000959 Although it is possible to store a pointer in shared memory remember that
960 this will refer to a location in the address space of a specific process.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000961 However, the pointer is quite likely to be invalid in the context of a second
962 process and trying to dereference the pointer from the second process may
963 cause a crash.
964
965.. function:: RawArray(typecode_or_type, size_or_initializer)
966
967 Return a ctypes array allocated from shared memory.
968
969 *typecode_or_type* determines the type of the elements of the returned array:
970 it is either a ctypes type or a one character typecode of the kind used by
971 the :mod:`array` module. If *size_or_initializer* is an integer then it
972 determines the length of the array, and the array will be initially zeroed.
973 Otherwise *size_or_initializer* is a sequence which is used to initialize the
974 array and whose length determines the length of the array.
975
976 Note that setting and getting an element is potentially non-atomic -- use
977 :func:`Array` instead to make sure that access is automatically synchronized
978 using a lock.
979
980.. function:: RawValue(typecode_or_type, *args)
981
982 Return a ctypes object allocated from shared memory.
983
984 *typecode_or_type* determines the type of the returned object: it is either a
985 ctypes type or a one character typecode of the kind used by the :mod:`array`
Jesse Nollerb0516a62009-01-18 03:11:38 +0000986 module. *\*args* is passed on to the constructor for the type.
Benjamin Petersone711caf2008-06-11 16:44:04 +0000987
988 Note that setting and getting the value is potentially non-atomic -- use
989 :func:`Value` instead to make sure that access is automatically synchronized
990 using a lock.
991
Amaury Forgeot d'Arcb0c29162008-11-22 22:18:04 +0000992 Note that an array of :data:`ctypes.c_char` has ``value`` and ``raw``
Benjamin Petersone711caf2008-06-11 16:44:04 +0000993 attributes which allow one to use it to store and retrieve strings -- see
994 documentation for :mod:`ctypes`.
995
Jesse Nollerb0516a62009-01-18 03:11:38 +0000996.. function:: Array(typecode_or_type, size_or_initializer, *args[, lock])
Benjamin Petersone711caf2008-06-11 16:44:04 +0000997
998 The same as :func:`RawArray` except that depending on the value of *lock* a
999 process-safe synchronization wrapper may be returned instead of a raw ctypes
1000 array.
1001
1002 If *lock* is ``True`` (the default) then a new lock object is created to
1003 synchronize access to the value. If *lock* is a :class:`Lock` or
1004 :class:`RLock` object then that will be used to synchronize access to the
1005 value. If *lock* is ``False`` then access to the returned object will not be
1006 automatically protected by a lock, so it will not necessarily be
1007 "process-safe".
1008
1009 Note that *lock* is a keyword-only argument.
1010
1011.. function:: Value(typecode_or_type, *args[, lock])
1012
1013 The same as :func:`RawValue` except that depending on the value of *lock* a
1014 process-safe synchronization wrapper may be returned instead of a raw ctypes
1015 object.
1016
1017 If *lock* is ``True`` (the default) then a new lock object is created to
1018 synchronize access to the value. If *lock* is a :class:`Lock` or
1019 :class:`RLock` object then that will be used to synchronize access to the
1020 value. If *lock* is ``False`` then access to the returned object will not be
1021 automatically protected by a lock, so it will not necessarily be
1022 "process-safe".
1023
1024 Note that *lock* is a keyword-only argument.
1025
1026.. function:: copy(obj)
1027
1028 Return a ctypes object allocated from shared memory which is a copy of the
1029 ctypes object *obj*.
1030
1031.. function:: synchronized(obj[, lock])
1032
1033 Return a process-safe wrapper object for a ctypes object which uses *lock* to
1034 synchronize access. If *lock* is ``None`` (the default) then a
1035 :class:`multiprocessing.RLock` object is created automatically.
1036
1037 A synchronized wrapper will have two methods in addition to those of the
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001038 object it wraps: :meth:`get_obj` returns the wrapped object and
1039 :meth:`get_lock` returns the lock object used for synchronization.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001040
1041 Note that accessing the ctypes object through the wrapper can be a lot slower
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001042 than accessing the raw ctypes object.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001043
1044
1045The table below compares the syntax for creating shared ctypes objects from
1046shared memory with the normal ctypes syntax. (In the table ``MyStruct`` is some
1047subclass of :class:`ctypes.Structure`.)
1048
1049==================== ========================== ===========================
1050ctypes sharedctypes using type sharedctypes using typecode
1051==================== ========================== ===========================
1052c_double(2.4) RawValue(c_double, 2.4) RawValue('d', 2.4)
1053MyStruct(4, 6) RawValue(MyStruct, 4, 6)
1054(c_short * 7)() RawArray(c_short, 7) RawArray('h', 7)
1055(c_int * 3)(9, 2, 8) RawArray(c_int, (9, 2, 8)) RawArray('i', (9, 2, 8))
1056==================== ========================== ===========================
1057
1058
1059Below is an example where a number of ctypes objects are modified by a child
1060process::
1061
1062 from multiprocessing import Process, Lock
1063 from multiprocessing.sharedctypes import Value, Array
1064 from ctypes import Structure, c_double
1065
1066 class Point(Structure):
1067 _fields_ = [('x', c_double), ('y', c_double)]
1068
1069 def modify(n, x, s, A):
1070 n.value **= 2
1071 x.value **= 2
1072 s.value = s.value.upper()
1073 for a in A:
1074 a.x **= 2
1075 a.y **= 2
1076
1077 if __name__ == '__main__':
1078 lock = Lock()
1079
1080 n = Value('i', 7)
R. David Murray8e8099c2009-04-28 18:02:00 +00001081 x = Value(c_double, 1.0/3.0, lock=False)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001082 s = Array('c', 'hello world', lock=lock)
1083 A = Array(Point, [(1.875,-6.25), (-5.75,2.0), (2.375,9.5)], lock=lock)
1084
1085 p = Process(target=modify, args=(n, x, s, A))
1086 p.start()
1087 p.join()
1088
Georg Brandl49702152008-09-29 06:43:45 +00001089 print(n.value)
1090 print(x.value)
1091 print(s.value)
1092 print([(a.x, a.y) for a in A])
Benjamin Petersone711caf2008-06-11 16:44:04 +00001093
1094
Georg Brandl49702152008-09-29 06:43:45 +00001095.. highlight:: none
Benjamin Petersone711caf2008-06-11 16:44:04 +00001096
1097The results printed are ::
1098
1099 49
1100 0.1111111111111111
1101 HELLO WORLD
1102 [(3.515625, 39.0625), (33.0625, 4.0), (5.640625, 90.25)]
1103
Georg Brandl49702152008-09-29 06:43:45 +00001104.. highlight:: python
Benjamin Petersone711caf2008-06-11 16:44:04 +00001105
1106
1107.. _multiprocessing-managers:
1108
1109Managers
1110~~~~~~~~
1111
1112Managers provide a way to create data which can be shared between different
1113processes. A manager object controls a server process which manages *shared
1114objects*. Other processes can access the shared objects by using proxies.
1115
1116.. function:: multiprocessing.Manager()
1117
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001118 Returns a started :class:`~multiprocessing.managers.SyncManager` object which
1119 can be used for sharing objects between processes. The returned manager
1120 object corresponds to a spawned child process and has methods which will
1121 create shared objects and return corresponding proxies.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001122
1123.. module:: multiprocessing.managers
1124 :synopsis: Share data between process with shared objects.
1125
1126Manager processes will be shutdown as soon as they are garbage collected or
1127their parent process exits. The manager classes are defined in the
1128:mod:`multiprocessing.managers` module:
1129
1130.. class:: BaseManager([address[, authkey]])
1131
1132 Create a BaseManager object.
1133
1134 Once created one should call :meth:`start` or :meth:`serve_forever` to ensure
1135 that the manager object refers to a started manager process.
1136
1137 *address* is the address on which the manager process listens for new
1138 connections. If *address* is ``None`` then an arbitrary one is chosen.
1139
1140 *authkey* is the authentication key which will be used to check the validity
1141 of incoming connections to the server process. If *authkey* is ``None`` then
Benjamin Petersona786b022008-08-25 21:05:21 +00001142 ``current_process().authkey``. Otherwise *authkey* is used and it
Benjamin Petersone711caf2008-06-11 16:44:04 +00001143 must be a string.
1144
Benjamin Petersonf47ed4a2009-04-11 20:45:40 +00001145 .. method:: start([initializer[, initargs]])
Benjamin Petersone711caf2008-06-11 16:44:04 +00001146
Benjamin Petersonf47ed4a2009-04-11 20:45:40 +00001147 Start a subprocess to start the manager. If *initializer* is not ``None``
1148 then the subprocess will call ``initializer(*initargs)`` when it starts.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001149
Georg Brandl2ee470f2008-07-16 12:55:28 +00001150 .. method:: serve_forever()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001151
1152 Run the server in the current process.
1153
1154 .. method:: from_address(address, authkey)
1155
1156 A class method which creates a manager object referring to a pre-existing
1157 server process which is using the given address and authentication key.
1158
Jesse Noller45239682008-11-28 18:46:19 +00001159 .. method:: get_server()
Georg Brandl48310cd2009-01-03 21:18:54 +00001160
Jesse Noller45239682008-11-28 18:46:19 +00001161 Returns a :class:`Server` object which represents the actual server under
Georg Brandl48310cd2009-01-03 21:18:54 +00001162 the control of the Manager. The :class:`Server` object supports the
R. David Murray8e8099c2009-04-28 18:02:00 +00001163 :meth:`serve_forever` method::
Georg Brandl48310cd2009-01-03 21:18:54 +00001164
Georg Brandl1f01deb2009-01-03 22:47:39 +00001165 >>> from multiprocessing.managers import BaseManager
R. David Murray8e8099c2009-04-28 18:02:00 +00001166 >>> manager = BaseManager(address=('', 50000), authkey='abc')
1167 >>> server = manager.get_server()
1168 >>> server.serve_forever()
Georg Brandl48310cd2009-01-03 21:18:54 +00001169
R. David Murray8e8099c2009-04-28 18:02:00 +00001170 :class:`Server` additionally has an :attr:`address` attribute.
Jesse Noller45239682008-11-28 18:46:19 +00001171
1172 .. method:: connect()
Georg Brandl48310cd2009-01-03 21:18:54 +00001173
R. David Murray8e8099c2009-04-28 18:02:00 +00001174 Connect a local manager object to a remote manager process::
Georg Brandl48310cd2009-01-03 21:18:54 +00001175
Jesse Noller45239682008-11-28 18:46:19 +00001176 >>> from multiprocessing.managers import BaseManager
R. David Murray8e8099c2009-04-28 18:02:00 +00001177 >>> m = BaseManager(address=('127.0.0.1', 5000), authkey='abc')
Jesse Noller45239682008-11-28 18:46:19 +00001178 >>> m.connect()
1179
Benjamin Petersone711caf2008-06-11 16:44:04 +00001180 .. method:: shutdown()
1181
1182 Stop the process used by the manager. This is only available if
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001183 :meth:`start` has been used to start the server process.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001184
1185 This can be called multiple times.
1186
1187 .. method:: register(typeid[, callable[, proxytype[, exposed[, method_to_typeid[, create_method]]]]])
1188
1189 A classmethod which can be used for registering a type or callable with
1190 the manager class.
1191
1192 *typeid* is a "type identifier" which is used to identify a particular
1193 type of shared object. This must be a string.
1194
1195 *callable* is a callable used for creating objects for this type
1196 identifier. If a manager instance will be created using the
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001197 :meth:`from_address` classmethod or if the *create_method* argument is
Benjamin Petersone711caf2008-06-11 16:44:04 +00001198 ``False`` then this can be left as ``None``.
1199
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001200 *proxytype* is a subclass of :class:`BaseProxy` which is used to create
1201 proxies for shared objects with this *typeid*. If ``None`` then a proxy
1202 class is created automatically.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001203
1204 *exposed* is used to specify a sequence of method names which proxies for
1205 this typeid should be allowed to access using
1206 :meth:`BaseProxy._callMethod`. (If *exposed* is ``None`` then
1207 :attr:`proxytype._exposed_` is used instead if it exists.) In the case
1208 where no exposed list is specified, all "public methods" of the shared
1209 object will be accessible. (Here a "public method" means any attribute
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001210 which has a :meth:`__call__` method and whose name does not begin with
Benjamin Petersone711caf2008-06-11 16:44:04 +00001211 ``'_'``.)
1212
1213 *method_to_typeid* is a mapping used to specify the return type of those
1214 exposed methods which should return a proxy. It maps method names to
1215 typeid strings. (If *method_to_typeid* is ``None`` then
1216 :attr:`proxytype._method_to_typeid_` is used instead if it exists.) If a
1217 method's name is not a key of this mapping or if the mapping is ``None``
1218 then the object returned by the method will be copied by value.
1219
1220 *create_method* determines whether a method should be created with name
1221 *typeid* which can be used to tell the server process to create a new
1222 shared object and return a proxy for it. By default it is ``True``.
1223
1224 :class:`BaseManager` instances also have one read-only property:
1225
1226 .. attribute:: address
1227
1228 The address used by the manager.
1229
1230
1231.. class:: SyncManager
1232
1233 A subclass of :class:`BaseManager` which can be used for the synchronization
1234 of processes. Objects of this type are returned by
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001235 :func:`multiprocessing.Manager`.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001236
1237 It also supports creation of shared lists and dictionaries.
1238
1239 .. method:: BoundedSemaphore([value])
1240
1241 Create a shared :class:`threading.BoundedSemaphore` object and return a
1242 proxy for it.
1243
1244 .. method:: Condition([lock])
1245
1246 Create a shared :class:`threading.Condition` object and return a proxy for
1247 it.
1248
1249 If *lock* is supplied then it should be a proxy for a
1250 :class:`threading.Lock` or :class:`threading.RLock` object.
1251
1252 .. method:: Event()
1253
1254 Create a shared :class:`threading.Event` object and return a proxy for it.
1255
1256 .. method:: Lock()
1257
1258 Create a shared :class:`threading.Lock` object and return a proxy for it.
1259
1260 .. method:: Namespace()
1261
1262 Create a shared :class:`Namespace` object and return a proxy for it.
1263
1264 .. method:: Queue([maxsize])
1265
Benjamin Peterson257060a2008-06-28 01:42:41 +00001266 Create a shared :class:`queue.Queue` object and return a proxy for it.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001267
1268 .. method:: RLock()
1269
1270 Create a shared :class:`threading.RLock` object and return a proxy for it.
1271
1272 .. method:: Semaphore([value])
1273
1274 Create a shared :class:`threading.Semaphore` object and return a proxy for
1275 it.
1276
1277 .. method:: Array(typecode, sequence)
1278
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001279 Create an array and return a proxy for it.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001280
1281 .. method:: Value(typecode, value)
1282
1283 Create an object with a writable ``value`` attribute and return a proxy
1284 for it.
1285
1286 .. method:: dict()
1287 dict(mapping)
1288 dict(sequence)
1289
1290 Create a shared ``dict`` object and return a proxy for it.
1291
1292 .. method:: list()
1293 list(sequence)
1294
1295 Create a shared ``list`` object and return a proxy for it.
1296
1297
1298Namespace objects
1299>>>>>>>>>>>>>>>>>
1300
1301A namespace object has no public methods, but does have writable attributes.
1302Its representation shows the values of its attributes.
1303
1304However, when using a proxy for a namespace object, an attribute beginning with
R. David Murray8e8099c2009-04-28 18:02:00 +00001305``'_'`` will be an attribute of the proxy and not an attribute of the referent:
1306
1307.. doctest::
Benjamin Petersone711caf2008-06-11 16:44:04 +00001308
1309 >>> manager = multiprocessing.Manager()
1310 >>> Global = manager.Namespace()
1311 >>> Global.x = 10
1312 >>> Global.y = 'hello'
1313 >>> Global._z = 12.3 # this is an attribute of the proxy
Georg Brandl49702152008-09-29 06:43:45 +00001314 >>> print(Global)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001315 Namespace(x=10, y='hello')
1316
1317
1318Customized managers
1319>>>>>>>>>>>>>>>>>>>
1320
1321To create one's own manager, one creates a subclass of :class:`BaseManager` and
Georg Brandl1f01deb2009-01-03 22:47:39 +00001322use the :meth:`~BaseManager.register` classmethod to register new types or
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001323callables with the manager class. For example::
Benjamin Petersone711caf2008-06-11 16:44:04 +00001324
1325 from multiprocessing.managers import BaseManager
1326
1327 class MathsClass(object):
1328 def add(self, x, y):
1329 return x + y
1330 def mul(self, x, y):
1331 return x * y
1332
1333 class MyManager(BaseManager):
1334 pass
1335
1336 MyManager.register('Maths', MathsClass)
1337
1338 if __name__ == '__main__':
1339 manager = MyManager()
1340 manager.start()
1341 maths = manager.Maths()
Georg Brandl49702152008-09-29 06:43:45 +00001342 print(maths.add(4, 3)) # prints 7
1343 print(maths.mul(7, 8)) # prints 56
Benjamin Petersone711caf2008-06-11 16:44:04 +00001344
1345
1346Using a remote manager
1347>>>>>>>>>>>>>>>>>>>>>>
1348
1349It is possible to run a manager server on one machine and have clients use it
1350from other machines (assuming that the firewalls involved allow it).
1351
1352Running the following commands creates a server for a single shared queue which
1353remote clients can access::
1354
1355 >>> from multiprocessing.managers import BaseManager
Benjamin Peterson257060a2008-06-28 01:42:41 +00001356 >>> import queue
1357 >>> queue = queue.Queue()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001358 >>> class QueueManager(BaseManager): pass
Jesse Noller45239682008-11-28 18:46:19 +00001359 >>> QueueManager.register('get_queue', callable=lambda:queue)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001360 >>> m = QueueManager(address=('', 50000), authkey='abracadabra')
Jesse Noller45239682008-11-28 18:46:19 +00001361 >>> s = m.get_server()
R. David Murray8e8099c2009-04-28 18:02:00 +00001362 >>> s.serve_forever()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001363
1364One client can access the server as follows::
1365
1366 >>> from multiprocessing.managers import BaseManager
1367 >>> class QueueManager(BaseManager): pass
Jesse Noller45239682008-11-28 18:46:19 +00001368 >>> QueueManager.register('get_queue')
1369 >>> m = QueueManager(address=('foo.bar.org', 50000), authkey='abracadabra')
1370 >>> m.connect()
1371 >>> queue = m.get_queue()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001372 >>> queue.put('hello')
1373
1374Another client can also use it::
1375
1376 >>> from multiprocessing.managers import BaseManager
1377 >>> class QueueManager(BaseManager): pass
R. David Murray8e8099c2009-04-28 18:02:00 +00001378 >>> QueueManager.register('get_queue')
1379 >>> m = QueueManager(address=('foo.bar.org', 50000), authkey='abracadabra')
1380 >>> m.connect()
1381 >>> queue = m.get_queue()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001382 >>> queue.get()
1383 'hello'
1384
Georg Brandl48310cd2009-01-03 21:18:54 +00001385Local processes can also access that queue, using the code from above on the
Jesse Noller45239682008-11-28 18:46:19 +00001386client to access it remotely::
1387
1388 >>> from multiprocessing import Process, Queue
1389 >>> from multiprocessing.managers import BaseManager
1390 >>> class Worker(Process):
1391 ... def __init__(self, q):
1392 ... self.q = q
1393 ... super(Worker, self).__init__()
1394 ... def run(self):
1395 ... self.q.put('local hello')
Georg Brandl48310cd2009-01-03 21:18:54 +00001396 ...
Jesse Noller45239682008-11-28 18:46:19 +00001397 >>> queue = Queue()
1398 >>> w = Worker(queue)
1399 >>> w.start()
1400 >>> class QueueManager(BaseManager): pass
Georg Brandl48310cd2009-01-03 21:18:54 +00001401 ...
Jesse Noller45239682008-11-28 18:46:19 +00001402 >>> QueueManager.register('get_queue', callable=lambda: queue)
1403 >>> m = QueueManager(address=('', 50000), authkey='abracadabra')
1404 >>> s = m.get_server()
1405 >>> s.serve_forever()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001406
1407Proxy Objects
1408~~~~~~~~~~~~~
1409
1410A proxy is an object which *refers* to a shared object which lives (presumably)
1411in a different process. The shared object is said to be the *referent* of the
1412proxy. Multiple proxy objects may have the same referent.
1413
1414A proxy object has methods which invoke corresponding methods of its referent
1415(although not every method of the referent will necessarily be available through
1416the proxy). A proxy can usually be used in most of the same ways that its
R. David Murray8e8099c2009-04-28 18:02:00 +00001417referent can:
1418
1419.. doctest::
Benjamin Petersone711caf2008-06-11 16:44:04 +00001420
1421 >>> from multiprocessing import Manager
1422 >>> manager = Manager()
1423 >>> l = manager.list([i*i for i in range(10)])
Georg Brandl49702152008-09-29 06:43:45 +00001424 >>> print(l)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001425 [0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
Georg Brandl49702152008-09-29 06:43:45 +00001426 >>> print(repr(l))
R. David Murray8e8099c2009-04-28 18:02:00 +00001427 <ListProxy object, typeid 'list' at 0x...>
Benjamin Petersone711caf2008-06-11 16:44:04 +00001428 >>> l[4]
1429 16
1430 >>> l[2:5]
1431 [4, 9, 16]
1432
1433Notice that applying :func:`str` to a proxy will return the representation of
1434the referent, whereas applying :func:`repr` will return the representation of
1435the proxy.
1436
1437An important feature of proxy objects is that they are picklable so they can be
1438passed between processes. Note, however, that if a proxy is sent to the
1439corresponding manager's process then unpickling it will produce the referent
R. David Murray8e8099c2009-04-28 18:02:00 +00001440itself. This means, for example, that one shared object can contain a second:
1441
1442.. doctest::
Benjamin Petersone711caf2008-06-11 16:44:04 +00001443
1444 >>> a = manager.list()
1445 >>> b = manager.list()
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001446 >>> a.append(b) # referent of a now contains referent of b
Georg Brandl49702152008-09-29 06:43:45 +00001447 >>> print(a, b)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001448 [[]] []
1449 >>> b.append('hello')
Georg Brandl49702152008-09-29 06:43:45 +00001450 >>> print(a, b)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001451 [['hello']] ['hello']
1452
1453.. note::
1454
1455 The proxy types in :mod:`multiprocessing` do nothing to support comparisons
R. David Murray8e8099c2009-04-28 18:02:00 +00001456 by value. So, for instance, we have:
Benjamin Petersone711caf2008-06-11 16:44:04 +00001457
R. David Murray8e8099c2009-04-28 18:02:00 +00001458 .. doctest::
Benjamin Petersone711caf2008-06-11 16:44:04 +00001459
R. David Murray8e8099c2009-04-28 18:02:00 +00001460 >>> manager.list([1,2,3]) == [1,2,3]
1461 False
1462
1463 One should just use a copy of the referent instead when making comparisons.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001464
1465.. class:: BaseProxy
1466
1467 Proxy objects are instances of subclasses of :class:`BaseProxy`.
1468
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001469 .. method:: _callmethod(methodname[, args[, kwds]])
Benjamin Petersone711caf2008-06-11 16:44:04 +00001470
1471 Call and return the result of a method of the proxy's referent.
1472
1473 If ``proxy`` is a proxy whose referent is ``obj`` then the expression ::
1474
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001475 proxy._callmethod(methodname, args, kwds)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001476
1477 will evaluate the expression ::
1478
1479 getattr(obj, methodname)(*args, **kwds)
1480
1481 in the manager's process.
1482
1483 The returned value will be a copy of the result of the call or a proxy to
1484 a new shared object -- see documentation for the *method_to_typeid*
1485 argument of :meth:`BaseManager.register`.
1486
1487 If an exception is raised by the call, then then is re-raised by
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001488 :meth:`_callmethod`. If some other exception is raised in the manager's
Benjamin Petersone711caf2008-06-11 16:44:04 +00001489 process then this is converted into a :exc:`RemoteError` exception and is
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001490 raised by :meth:`_callmethod`.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001491
1492 Note in particular that an exception will be raised if *methodname* has
1493 not been *exposed*
1494
R. David Murray8e8099c2009-04-28 18:02:00 +00001495 An example of the usage of :meth:`_callmethod`:
1496
1497 .. doctest::
Benjamin Petersone711caf2008-06-11 16:44:04 +00001498
1499 >>> l = manager.list(range(10))
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001500 >>> l._callmethod('__len__')
Benjamin Petersone711caf2008-06-11 16:44:04 +00001501 10
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001502 >>> l._callmethod('__getslice__', (2, 7)) # equiv to `l[2:7]`
Benjamin Petersone711caf2008-06-11 16:44:04 +00001503 [2, 3, 4, 5, 6]
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001504 >>> l._callmethod('__getitem__', (20,)) # equiv to `l[20]`
Benjamin Petersone711caf2008-06-11 16:44:04 +00001505 Traceback (most recent call last):
1506 ...
1507 IndexError: list index out of range
1508
Benjamin Peterson6ebe78f2008-12-21 00:06:59 +00001509 .. method:: _getvalue()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001510
1511 Return a copy of the referent.
1512
1513 If the referent is unpicklable then this will raise an exception.
1514
1515 .. method:: __repr__
1516
1517 Return a representation of the proxy object.
1518
1519 .. method:: __str__
1520
1521 Return the representation of the referent.
1522
1523
1524Cleanup
1525>>>>>>>
1526
1527A proxy object uses a weakref callback so that when it gets garbage collected it
1528deregisters itself from the manager which owns its referent.
1529
1530A shared object gets deleted from the manager process when there are no longer
1531any proxies referring to it.
1532
1533
1534Process Pools
1535~~~~~~~~~~~~~
1536
1537.. module:: multiprocessing.pool
1538 :synopsis: Create pools of processes.
1539
1540One can create a pool of processes which will carry out tasks submitted to it
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001541with the :class:`Pool` class.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001542
1543.. class:: multiprocessing.Pool([processes[, initializer[, initargs]]])
1544
1545 A process pool object which controls a pool of worker processes to which jobs
1546 can be submitted. It supports asynchronous results with timeouts and
1547 callbacks and has a parallel map implementation.
1548
1549 *processes* is the number of worker processes to use. If *processes* is
1550 ``None`` then the number returned by :func:`cpu_count` is used. If
1551 *initializer* is not ``None`` then each worker process will call
1552 ``initializer(*initargs)`` when it starts.
1553
1554 .. method:: apply(func[, args[, kwds]])
1555
Benjamin Peterson37d2fe02008-10-24 22:28:58 +00001556 Call *func* with arguments *args* and keyword arguments *kwds*. It blocks
Jesse Noller7b3c89d2009-01-22 21:56:13 +00001557 till the result is ready. Given this blocks - :meth:`apply_async` is better suited
1558 for performing work in parallel. Additionally, the passed
1559 in function is only executed in one of the workers of the pool.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001560
1561 .. method:: apply_async(func[, args[, kwds[, callback]]])
1562
1563 A variant of the :meth:`apply` method which returns a result object.
1564
1565 If *callback* is specified then it should be a callable which accepts a
1566 single argument. When the result becomes ready *callback* is applied to
1567 it (unless the call failed). *callback* should complete immediately since
1568 otherwise the thread which handles the results will get blocked.
1569
1570 .. method:: map(func, iterable[, chunksize])
1571
Benjamin Petersond23f8222009-04-05 19:13:16 +00001572 A parallel equivalent of the :func:`map` builtin function (it supports only
1573 one *iterable* argument though). It blocks till the result is ready.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001574
1575 This method chops the iterable into a number of chunks which it submits to
1576 the process pool as separate tasks. The (approximate) size of these
1577 chunks can be specified by setting *chunksize* to a positive integer.
1578
1579 .. method:: map_async(func, iterable[, chunksize[, callback]])
1580
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001581 A variant of the :meth:`map` method which returns a result object.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001582
1583 If *callback* is specified then it should be a callable which accepts a
1584 single argument. When the result becomes ready *callback* is applied to
1585 it (unless the call failed). *callback* should complete immediately since
1586 otherwise the thread which handles the results will get blocked.
1587
1588 .. method:: imap(func, iterable[, chunksize])
1589
Georg Brandl92905032008-11-22 08:51:39 +00001590 A lazier version of :meth:`map`.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001591
1592 The *chunksize* argument is the same as the one used by the :meth:`.map`
1593 method. For very long iterables using a large value for *chunksize* can
1594 make make the job complete **much** faster than using the default value of
1595 ``1``.
1596
1597 Also if *chunksize* is ``1`` then the :meth:`next` method of the iterator
1598 returned by the :meth:`imap` method has an optional *timeout* parameter:
1599 ``next(timeout)`` will raise :exc:`multiprocessing.TimeoutError` if the
1600 result cannot be returned within *timeout* seconds.
1601
1602 .. method:: imap_unordered(func, iterable[, chunksize])
1603
1604 The same as :meth:`imap` except that the ordering of the results from the
1605 returned iterator should be considered arbitrary. (Only when there is
1606 only one worker process is the order guaranteed to be "correct".)
1607
1608 .. method:: close()
1609
1610 Prevents any more tasks from being submitted to the pool. Once all the
1611 tasks have been completed the worker processes will exit.
1612
1613 .. method:: terminate()
1614
1615 Stops the worker processes immediately without completing outstanding
1616 work. When the pool object is garbage collected :meth:`terminate` will be
1617 called immediately.
1618
1619 .. method:: join()
1620
1621 Wait for the worker processes to exit. One must call :meth:`close` or
1622 :meth:`terminate` before using :meth:`join`.
1623
1624
1625.. class:: AsyncResult
1626
1627 The class of the result returned by :meth:`Pool.apply_async` and
1628 :meth:`Pool.map_async`.
1629
Georg Brandle3d70ae2008-11-22 08:54:21 +00001630 .. method:: get([timeout])
Benjamin Petersone711caf2008-06-11 16:44:04 +00001631
1632 Return the result when it arrives. If *timeout* is not ``None`` and the
1633 result does not arrive within *timeout* seconds then
1634 :exc:`multiprocessing.TimeoutError` is raised. If the remote call raised
1635 an exception then that exception will be reraised by :meth:`get`.
1636
1637 .. method:: wait([timeout])
1638
1639 Wait until the result is available or until *timeout* seconds pass.
1640
1641 .. method:: ready()
1642
1643 Return whether the call has completed.
1644
1645 .. method:: successful()
1646
1647 Return whether the call completed without raising an exception. Will
1648 raise :exc:`AssertionError` if the result is not ready.
1649
1650The following example demonstrates the use of a pool::
1651
1652 from multiprocessing import Pool
1653
1654 def f(x):
1655 return x*x
1656
1657 if __name__ == '__main__':
1658 pool = Pool(processes=4) # start 4 worker processes
1659
Georg Brandle3d70ae2008-11-22 08:54:21 +00001660 result = pool.apply_async(f, (10,)) # evaluate "f(10)" asynchronously
Georg Brandl49702152008-09-29 06:43:45 +00001661 print(result.get(timeout=1)) # prints "100" unless your computer is *very* slow
Benjamin Petersone711caf2008-06-11 16:44:04 +00001662
Georg Brandl49702152008-09-29 06:43:45 +00001663 print(pool.map(f, range(10))) # prints "[0, 1, 4,..., 81]"
Benjamin Petersone711caf2008-06-11 16:44:04 +00001664
1665 it = pool.imap(f, range(10))
Georg Brandl49702152008-09-29 06:43:45 +00001666 print(next(it)) # prints "0"
1667 print(next(it)) # prints "1"
1668 print(it.next(timeout=1)) # prints "4" unless your computer is *very* slow
Benjamin Petersone711caf2008-06-11 16:44:04 +00001669
1670 import time
Georg Brandle3d70ae2008-11-22 08:54:21 +00001671 result = pool.apply_async(time.sleep, (10,))
Georg Brandl49702152008-09-29 06:43:45 +00001672 print(result.get(timeout=1)) # raises TimeoutError
Benjamin Petersone711caf2008-06-11 16:44:04 +00001673
1674
1675.. _multiprocessing-listeners-clients:
1676
1677Listeners and Clients
1678~~~~~~~~~~~~~~~~~~~~~
1679
1680.. module:: multiprocessing.connection
1681 :synopsis: API for dealing with sockets.
1682
1683Usually message passing between processes is done using queues or by using
1684:class:`Connection` objects returned by :func:`Pipe`.
1685
1686However, the :mod:`multiprocessing.connection` module allows some extra
1687flexibility. It basically gives a high level message oriented API for dealing
1688with sockets or Windows named pipes, and also has support for *digest
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001689authentication* using the :mod:`hmac` module.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001690
1691
1692.. function:: deliver_challenge(connection, authkey)
1693
1694 Send a randomly generated message to the other end of the connection and wait
1695 for a reply.
1696
1697 If the reply matches the digest of the message using *authkey* as the key
1698 then a welcome message is sent to the other end of the connection. Otherwise
1699 :exc:`AuthenticationError` is raised.
1700
1701.. function:: answerChallenge(connection, authkey)
1702
1703 Receive a message, calculate the digest of the message using *authkey* as the
1704 key, and then send the digest back.
1705
1706 If a welcome message is not received, then :exc:`AuthenticationError` is
1707 raised.
1708
1709.. function:: Client(address[, family[, authenticate[, authkey]]])
1710
1711 Attempt to set up a connection to the listener which is using address
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001712 *address*, returning a :class:`~multiprocessing.Connection`.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001713
1714 The type of the connection is determined by *family* argument, but this can
1715 generally be omitted since it can usually be inferred from the format of
1716 *address*. (See :ref:`multiprocessing-address-formats`)
1717
Alexandre Vassalottic57a84f2009-07-17 12:07:01 +00001718 If *authenticate* is ``True`` or *authkey* is a string then digest
Benjamin Petersone711caf2008-06-11 16:44:04 +00001719 authentication is used. The key used for authentication will be either
Benjamin Petersona786b022008-08-25 21:05:21 +00001720 *authkey* or ``current_process().authkey)`` if *authkey* is ``None``.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001721 If authentication fails then :exc:`AuthenticationError` is raised. See
1722 :ref:`multiprocessing-auth-keys`.
1723
1724.. class:: Listener([address[, family[, backlog[, authenticate[, authkey]]]]])
1725
1726 A wrapper for a bound socket or Windows named pipe which is 'listening' for
1727 connections.
1728
1729 *address* is the address to be used by the bound socket or named pipe of the
1730 listener object.
1731
Benjamin Petersond23f8222009-04-05 19:13:16 +00001732 .. note::
1733
1734 If an address of '0.0.0.0' is used, the address will not be a connectable
1735 end point on Windows. If you require a connectable end-point,
1736 you should use '127.0.0.1'.
1737
Benjamin Petersone711caf2008-06-11 16:44:04 +00001738 *family* is the type of socket (or named pipe) to use. This can be one of
1739 the strings ``'AF_INET'`` (for a TCP socket), ``'AF_UNIX'`` (for a Unix
1740 domain socket) or ``'AF_PIPE'`` (for a Windows named pipe). Of these only
1741 the first is guaranteed to be available. If *family* is ``None`` then the
1742 family is inferred from the format of *address*. If *address* is also
1743 ``None`` then a default is chosen. This default is the family which is
1744 assumed to be the fastest available. See
1745 :ref:`multiprocessing-address-formats`. Note that if *family* is
1746 ``'AF_UNIX'`` and address is ``None`` then the socket will be created in a
1747 private temporary directory created using :func:`tempfile.mkstemp`.
1748
1749 If the listener object uses a socket then *backlog* (1 by default) is passed
1750 to the :meth:`listen` method of the socket once it has been bound.
1751
1752 If *authenticate* is ``True`` (``False`` by default) or *authkey* is not
1753 ``None`` then digest authentication is used.
1754
1755 If *authkey* is a string then it will be used as the authentication key;
1756 otherwise it must be *None*.
1757
1758 If *authkey* is ``None`` and *authenticate* is ``True`` then
Benjamin Petersona786b022008-08-25 21:05:21 +00001759 ``current_process().authkey`` is used as the authentication key. If
Alexandre Vassalottic57a84f2009-07-17 12:07:01 +00001760 *authkey* is ``None`` and *authenticate* is ``False`` then no
Benjamin Petersone711caf2008-06-11 16:44:04 +00001761 authentication is done. If authentication fails then
1762 :exc:`AuthenticationError` is raised. See :ref:`multiprocessing-auth-keys`.
1763
1764 .. method:: accept()
1765
1766 Accept a connection on the bound socket or named pipe of the listener
1767 object and return a :class:`Connection` object. If authentication is
1768 attempted and fails, then :exc:`AuthenticationError` is raised.
1769
1770 .. method:: close()
1771
1772 Close the bound socket or named pipe of the listener object. This is
1773 called automatically when the listener is garbage collected. However it
1774 is advisable to call it explicitly.
1775
1776 Listener objects have the following read-only properties:
1777
1778 .. attribute:: address
1779
1780 The address which is being used by the Listener object.
1781
1782 .. attribute:: last_accepted
1783
1784 The address from which the last accepted connection came. If this is
1785 unavailable then it is ``None``.
1786
1787
1788The module defines two exceptions:
1789
1790.. exception:: AuthenticationError
1791
1792 Exception raised when there is an authentication error.
1793
Benjamin Petersone711caf2008-06-11 16:44:04 +00001794
1795**Examples**
1796
1797The following server code creates a listener which uses ``'secret password'`` as
1798an authentication key. It then waits for a connection and sends some data to
1799the client::
1800
1801 from multiprocessing.connection import Listener
1802 from array import array
1803
1804 address = ('localhost', 6000) # family is deduced to be 'AF_INET'
1805 listener = Listener(address, authkey='secret password')
1806
1807 conn = listener.accept()
Georg Brandl49702152008-09-29 06:43:45 +00001808 print('connection accepted from', listener.last_accepted)
Benjamin Petersone711caf2008-06-11 16:44:04 +00001809
1810 conn.send([2.25, None, 'junk', float])
1811
1812 conn.send_bytes('hello')
1813
1814 conn.send_bytes(array('i', [42, 1729]))
1815
1816 conn.close()
1817 listener.close()
1818
1819The following code connects to the server and receives some data from the
1820server::
1821
1822 from multiprocessing.connection import Client
1823 from array import array
1824
1825 address = ('localhost', 6000)
1826 conn = Client(address, authkey='secret password')
1827
Georg Brandl49702152008-09-29 06:43:45 +00001828 print(conn.recv()) # => [2.25, None, 'junk', float]
Benjamin Petersone711caf2008-06-11 16:44:04 +00001829
Georg Brandl49702152008-09-29 06:43:45 +00001830 print(conn.recv_bytes()) # => 'hello'
Benjamin Petersone711caf2008-06-11 16:44:04 +00001831
1832 arr = array('i', [0, 0, 0, 0, 0])
Georg Brandl49702152008-09-29 06:43:45 +00001833 print(conn.recv_bytes_into(arr)) # => 8
1834 print(arr) # => array('i', [42, 1729, 0, 0, 0])
Benjamin Petersone711caf2008-06-11 16:44:04 +00001835
1836 conn.close()
1837
1838
1839.. _multiprocessing-address-formats:
1840
1841Address Formats
1842>>>>>>>>>>>>>>>
1843
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001844* An ``'AF_INET'`` address is a tuple of the form ``(hostname, port)`` where
Benjamin Petersone711caf2008-06-11 16:44:04 +00001845 *hostname* is a string and *port* is an integer.
1846
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001847* An ``'AF_UNIX'`` address is a string representing a filename on the
Benjamin Petersone711caf2008-06-11 16:44:04 +00001848 filesystem.
1849
1850* An ``'AF_PIPE'`` address is a string of the form
Benjamin Petersonda10d3b2009-01-01 00:23:30 +00001851 :samp:`r'\\\\.\\pipe\\{PipeName}'`. To use :func:`Client` to connect to a named
Georg Brandl1f01deb2009-01-03 22:47:39 +00001852 pipe on a remote computer called *ServerName* one should use an address of the
Benjamin Peterson28d88b42009-01-09 03:03:23 +00001853 form :samp:`r'\\\\{ServerName}\\pipe\\{PipeName}'` instead.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001854
1855Note that any string beginning with two backslashes is assumed by default to be
1856an ``'AF_PIPE'`` address rather than an ``'AF_UNIX'`` address.
1857
1858
1859.. _multiprocessing-auth-keys:
1860
1861Authentication keys
1862~~~~~~~~~~~~~~~~~~~
1863
1864When one uses :meth:`Connection.recv`, the data received is automatically
1865unpickled. Unfortunately unpickling data from an untrusted source is a security
1866risk. Therefore :class:`Listener` and :func:`Client` use the :mod:`hmac` module
1867to provide digest authentication.
1868
1869An authentication key is a string which can be thought of as a password: once a
1870connection is established both ends will demand proof that the other knows the
1871authentication key. (Demonstrating that both ends are using the same key does
1872**not** involve sending the key over the connection.)
1873
1874If authentication is requested but do authentication key is specified then the
Benjamin Petersona786b022008-08-25 21:05:21 +00001875return value of ``current_process().authkey`` is used (see
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001876:class:`~multiprocessing.Process`). This value will automatically inherited by
1877any :class:`~multiprocessing.Process` object that the current process creates.
1878This means that (by default) all processes of a multi-process program will share
1879a single authentication key which can be used when setting up connections
Benjamin Petersond23f8222009-04-05 19:13:16 +00001880between themselves.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001881
1882Suitable authentication keys can also be generated by using :func:`os.urandom`.
1883
1884
1885Logging
1886~~~~~~~
1887
1888Some support for logging is available. Note, however, that the :mod:`logging`
1889package does not use process shared locks so it is possible (depending on the
1890handler type) for messages from different processes to get mixed up.
1891
1892.. currentmodule:: multiprocessing
1893.. function:: get_logger()
1894
1895 Returns the logger used by :mod:`multiprocessing`. If necessary, a new one
1896 will be created.
1897
Jesse Noller41faa542009-01-25 03:45:53 +00001898 When first created the logger has level :data:`logging.NOTSET` and no
1899 default handler. Messages sent to this logger will not by default propagate
1900 to the root logger.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001901
1902 Note that on Windows child processes will only inherit the level of the
1903 parent process's logger -- any other customization of the logger will not be
1904 inherited.
1905
Jesse Noller41faa542009-01-25 03:45:53 +00001906.. currentmodule:: multiprocessing
1907.. function:: log_to_stderr()
1908
1909 This function performs a call to :func:`get_logger` but in addition to
1910 returning the logger created by get_logger, it adds a handler which sends
1911 output to :data:`sys.stderr` using format
1912 ``'[%(levelname)s/%(processName)s] %(message)s'``.
1913
Benjamin Petersone711caf2008-06-11 16:44:04 +00001914Below is an example session with logging turned on::
1915
Benjamin Peterson206e3072008-10-19 14:07:49 +00001916 >>> import multiprocessing, logging
Jesse Noller41faa542009-01-25 03:45:53 +00001917 >>> logger = multiprocessing.log_to_stderr()
Benjamin Petersone711caf2008-06-11 16:44:04 +00001918 >>> logger.setLevel(logging.INFO)
1919 >>> logger.warning('doomed')
1920 [WARNING/MainProcess] doomed
Benjamin Peterson206e3072008-10-19 14:07:49 +00001921 >>> m = multiprocessing.Manager()
R. David Murray8e8099c2009-04-28 18:02:00 +00001922 [INFO/SyncManager-...] child process calling self.run()
1923 [INFO/SyncManager-...] created temp directory /.../pymp-...
1924 [INFO/SyncManager-...] manager serving at '/.../listener-...'
Benjamin Petersone711caf2008-06-11 16:44:04 +00001925 >>> del m
1926 [INFO/MainProcess] sending shutdown message to manager
R. David Murray8e8099c2009-04-28 18:02:00 +00001927 [INFO/SyncManager-...] manager exiting with exitcode 0
Benjamin Petersone711caf2008-06-11 16:44:04 +00001928
Jesse Noller41faa542009-01-25 03:45:53 +00001929In addition to having these two logging functions, the multiprocessing also
1930exposes two additional logging level attributes. These are :const:`SUBWARNING`
1931and :const:`SUBDEBUG`. The table below illustrates where theses fit in the
1932normal level hierarchy.
1933
1934+----------------+----------------+
1935| Level | Numeric value |
1936+================+================+
1937| ``SUBWARNING`` | 25 |
1938+----------------+----------------+
1939| ``SUBDEBUG`` | 5 |
1940+----------------+----------------+
1941
1942For a full table of logging levels, see the :mod:`logging` module.
1943
1944These additional logging levels are used primarily for certain debug messages
1945within the multiprocessing module. Below is the same example as above, except
1946with :const:`SUBDEBUG` enabled::
1947
1948 >>> import multiprocessing, logging
1949 >>> logger = multiprocessing.log_to_stderr()
1950 >>> logger.setLevel(multiprocessing.SUBDEBUG)
1951 >>> logger.warning('doomed')
1952 [WARNING/MainProcess] doomed
1953 >>> m = multiprocessing.Manager()
R. David Murray8e8099c2009-04-28 18:02:00 +00001954 [INFO/SyncManager-...] child process calling self.run()
1955 [INFO/SyncManager-...] created temp directory /.../pymp-...
1956 [INFO/SyncManager-...] manager serving at '/.../pymp-djGBXN/listener-...'
Jesse Noller41faa542009-01-25 03:45:53 +00001957 >>> del m
1958 [SUBDEBUG/MainProcess] finalizer calling ...
1959 [INFO/MainProcess] sending shutdown message to manager
R. David Murray8e8099c2009-04-28 18:02:00 +00001960 [DEBUG/SyncManager-...] manager received shutdown message
1961 [SUBDEBUG/SyncManager-...] calling <Finalize object, callback=unlink, ...
1962 [SUBDEBUG/SyncManager-...] finalizer calling <built-in function unlink> ...
1963 [SUBDEBUG/SyncManager-...] calling <Finalize object, dead>
1964 [SUBDEBUG/SyncManager-...] finalizer calling <function rmtree at 0x5aa730> ...
1965 [INFO/SyncManager-...] manager exiting with exitcode 0
Benjamin Petersone711caf2008-06-11 16:44:04 +00001966
1967The :mod:`multiprocessing.dummy` module
1968~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1969
1970.. module:: multiprocessing.dummy
1971 :synopsis: Dumb wrapper around threading.
1972
1973:mod:`multiprocessing.dummy` replicates the API of :mod:`multiprocessing` but is
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00001974no more than a wrapper around the :mod:`threading` module.
Benjamin Petersone711caf2008-06-11 16:44:04 +00001975
1976
1977.. _multiprocessing-programming:
1978
1979Programming guidelines
1980----------------------
1981
1982There are certain guidelines and idioms which should be adhered to when using
1983:mod:`multiprocessing`.
1984
1985
1986All platforms
1987~~~~~~~~~~~~~
1988
1989Avoid shared state
1990
1991 As far as possible one should try to avoid shifting large amounts of data
1992 between processes.
1993
1994 It is probably best to stick to using queues or pipes for communication
1995 between processes rather than using the lower level synchronization
1996 primitives from the :mod:`threading` module.
1997
1998Picklability
1999
2000 Ensure that the arguments to the methods of proxies are picklable.
2001
2002Thread safety of proxies
2003
2004 Do not use a proxy object from more than one thread unless you protect it
2005 with a lock.
2006
2007 (There is never a problem with different processes using the *same* proxy.)
2008
2009Joining zombie processes
2010
2011 On Unix when a process finishes but has not been joined it becomes a zombie.
2012 There should never be very many because each time a new process starts (or
2013 :func:`active_children` is called) all completed processes which have not
2014 yet been joined will be joined. Also calling a finished process's
2015 :meth:`Process.is_alive` will join the process. Even so it is probably good
2016 practice to explicitly join all the processes that you start.
2017
2018Better to inherit than pickle/unpickle
2019
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00002020 On Windows many types from :mod:`multiprocessing` need to be picklable so
Benjamin Petersone711caf2008-06-11 16:44:04 +00002021 that child processes can use them. However, one should generally avoid
2022 sending shared objects to other processes using pipes or queues. Instead
2023 you should arrange the program so that a process which need access to a
2024 shared resource created elsewhere can inherit it from an ancestor process.
2025
2026Avoid terminating processes
2027
2028 Using the :meth:`Process.terminate` method to stop a process is liable to
2029 cause any shared resources (such as locks, semaphores, pipes and queues)
2030 currently being used by the process to become broken or unavailable to other
2031 processes.
2032
2033 Therefore it is probably best to only consider using
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00002034 :meth:`Process.terminate` on processes which never use any shared resources.
Benjamin Petersone711caf2008-06-11 16:44:04 +00002035
2036Joining processes that use queues
2037
2038 Bear in mind that a process that has put items in a queue will wait before
2039 terminating until all the buffered items are fed by the "feeder" thread to
2040 the underlying pipe. (The child process can call the
Benjamin Petersonae5360b2008-09-08 23:05:23 +00002041 :meth:`Queue.cancel_join_thread` method of the queue to avoid this behaviour.)
Benjamin Petersone711caf2008-06-11 16:44:04 +00002042
2043 This means that whenever you use a queue you need to make sure that all
2044 items which have been put on the queue will eventually be removed before the
2045 process is joined. Otherwise you cannot be sure that processes which have
2046 put items on the queue will terminate. Remember also that non-daemonic
2047 processes will be automatically be joined.
2048
2049 An example which will deadlock is the following::
2050
2051 from multiprocessing import Process, Queue
2052
2053 def f(q):
2054 q.put('X' * 1000000)
2055
2056 if __name__ == '__main__':
2057 queue = Queue()
2058 p = Process(target=f, args=(queue,))
2059 p.start()
2060 p.join() # this deadlocks
2061 obj = queue.get()
2062
2063 A fix here would be to swap the last two lines round (or simply remove the
2064 ``p.join()`` line).
2065
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00002066Explicitly pass resources to child processes
Benjamin Petersone711caf2008-06-11 16:44:04 +00002067
2068 On Unix a child process can make use of a shared resource created in a
2069 parent process using a global resource. However, it is better to pass the
2070 object as an argument to the constructor for the child process.
2071
2072 Apart from making the code (potentially) compatible with Windows this also
2073 ensures that as long as the child process is still alive the object will not
2074 be garbage collected in the parent process. This might be important if some
2075 resource is freed when the object is garbage collected in the parent
2076 process.
2077
2078 So for instance ::
2079
2080 from multiprocessing import Process, Lock
2081
2082 def f():
2083 ... do something using "lock" ...
2084
2085 if __name__ == '__main__':
2086 lock = Lock()
2087 for i in range(10):
2088 Process(target=f).start()
2089
2090 should be rewritten as ::
2091
2092 from multiprocessing import Process, Lock
2093
2094 def f(l):
2095 ... do something using "l" ...
2096
2097 if __name__ == '__main__':
2098 lock = Lock()
2099 for i in range(10):
2100 Process(target=f, args=(lock,)).start()
2101
Alexandre Vassalottic57a84f2009-07-17 12:07:01 +00002102Beware replacing sys.stdin with a "file like object"
2103
2104 :mod:`multiprocessing` originally unconditionally called::
2105
2106 os.close(sys.stdin.fileno())
2107
2108 in the :meth:`multiprocessing.Process._bootstrap` method --- this resulted
2109 in issues with processes-in-processes. This has been changed to::
2110
2111 sys.stdin.close()
2112 sys.stdin = open(os.devnull)
2113
2114 Which solves the fundamental issue of processes colliding with each other
2115 resulting in a bad file descriptor error, but introduces a potential danger
2116 to applications which replace :func:`sys.stdin` with a "file-like object"
2117 with output buffering. This danger is that if multiple processes call
2118 :func:`close()` on this file-like object, it could result in the same
2119 data being flushed to the object multiple times, resulting in corruption.
2120
2121 If you write a file-like object and implement your own caching, you can
2122 make it fork-safe by storing the pid whenever you append to the cache,
2123 and discarding the cache when the pid changes. For example::
2124
2125 @property
2126 def cache(self):
2127 pid = os.getpid()
2128 if pid != self._pid:
2129 self._pid = pid
2130 self._cache = []
2131 return self._cache
2132
2133 For more information, see :issue:`5155`, :issue:`5313` and :issue:`5331`
Benjamin Petersone711caf2008-06-11 16:44:04 +00002134
2135Windows
2136~~~~~~~
2137
2138Since Windows lacks :func:`os.fork` it has a few extra restrictions:
2139
2140More picklability
2141
2142 Ensure that all arguments to :meth:`Process.__init__` are picklable. This
2143 means, in particular, that bound or unbound methods cannot be used directly
2144 as the ``target`` argument on Windows --- just define a function and use
2145 that instead.
2146
2147 Also, if you subclass :class:`Process` then make sure that instances will be
2148 picklable when the :meth:`Process.start` method is called.
2149
2150Global variables
2151
2152 Bear in mind that if code run in a child process tries to access a global
2153 variable, then the value it sees (if any) may not be the same as the value
2154 in the parent process at the time that :meth:`Process.start` was called.
2155
2156 However, global variables which are just module level constants cause no
2157 problems.
2158
2159Safe importing of main module
2160
2161 Make sure that the main module can be safely imported by a new Python
2162 interpreter without causing unintended side effects (such a starting a new
2163 process).
2164
2165 For example, under Windows running the following module would fail with a
2166 :exc:`RuntimeError`::
2167
2168 from multiprocessing import Process
2169
2170 def foo():
Georg Brandl49702152008-09-29 06:43:45 +00002171 print('hello')
Benjamin Petersone711caf2008-06-11 16:44:04 +00002172
2173 p = Process(target=foo)
2174 p.start()
2175
2176 Instead one should protect the "entry point" of the program by using ``if
2177 __name__ == '__main__':`` as follows::
2178
2179 from multiprocessing import Process, freeze_support
2180
2181 def foo():
Georg Brandl49702152008-09-29 06:43:45 +00002182 print('hello')
Benjamin Petersone711caf2008-06-11 16:44:04 +00002183
2184 if __name__ == '__main__':
2185 freeze_support()
2186 p = Process(target=foo)
2187 p.start()
2188
Benjamin Peterson5289b2b2008-06-28 00:40:54 +00002189 (The ``freeze_support()`` line can be omitted if the program will be run
Benjamin Petersone711caf2008-06-11 16:44:04 +00002190 normally instead of frozen.)
2191
2192 This allows the newly spawned Python interpreter to safely import the module
2193 and then run the module's ``foo()`` function.
2194
2195 Similar restrictions apply if a pool or manager is created in the main
2196 module.
2197
2198
2199.. _multiprocessing-examples:
2200
2201Examples
2202--------
2203
2204Demonstration of how to create and use customized managers and proxies:
2205
2206.. literalinclude:: ../includes/mp_newtype.py
2207
2208
2209Using :class:`Pool`:
2210
2211.. literalinclude:: ../includes/mp_pool.py
2212
2213
2214Synchronization types like locks, conditions and queues:
2215
2216.. literalinclude:: ../includes/mp_synchronize.py
2217
2218
2219An showing how to use queues to feed tasks to a collection of worker process and
2220collect the results:
2221
2222.. literalinclude:: ../includes/mp_workers.py
2223
2224
2225An example of how a pool of worker processes can each run a
2226:class:`SimpleHTTPServer.HttpServer` instance while sharing a single listening
2227socket.
2228
2229.. literalinclude:: ../includes/mp_webserver.py
2230
2231
2232Some simple benchmarks comparing :mod:`multiprocessing` with :mod:`threading`:
2233
2234.. literalinclude:: ../includes/mp_benchmarks.py
2235
2236An example/demo of how to use the :class:`managers.SyncManager`, :class:`Process`
Georg Brandl48310cd2009-01-03 21:18:54 +00002237and others to build a system which can distribute processes and work via a
Benjamin Petersone711caf2008-06-11 16:44:04 +00002238distributed queue to a "cluster" of machines on a network, accessible via SSH.
2239You will need to have private key authentication for all hosts configured for
2240this to work.
2241
Benjamin Peterson95a939c2008-06-14 02:23:29 +00002242.. literalinclude:: ../includes/mp_distributing.py