PageRenderTime 61ms CodeModel.GetById 10ms RepoModel.GetById 0ms app.codeStats 1ms

/Doc/library/multiprocessing.rst

http://unladen-swallow.googlecode.com/
ReStructuredText | 2225 lines | 1520 code | 705 blank | 0 comment | 0 complexity | 47505cbd9f34fec5d0574573b59de8ba MD5 | raw file
Possible License(s): 0BSD, BSD-3-Clause
  1. :mod:`multiprocessing` --- Process-based "threading" interface
  2. ==============================================================
  3. .. module:: multiprocessing
  4. :synopsis: Process-based "threading" interface.
  5. .. versionadded:: 2.6
  6. Introduction
  7. ----------------------
  8. :mod:`multiprocessing` is a package that supports spawning processes using an
  9. API similar to the :mod:`threading` module. The :mod:`multiprocessing` package
  10. offers both local and remote concurrency, effectively side-stepping the
  11. :term:`Global Interpreter Lock` by using subprocesses instead of threads. Due
  12. to this, the :mod:`multiprocessing` module allows the programmer to fully
  13. leverage multiple processors on a given machine. It runs on both Unix and
  14. Windows.
  15. .. warning::
  16. Some of this package's functionality requires a functioning shared semaphore
  17. implementation on the host operating system. Without one, the
  18. :mod:`multiprocessing.synchronize` module will be disabled, and attempts to
  19. import it will result in an :exc:`ImportError`. See
  20. :issue:`3770` for additional information.
  21. .. note::
  22. Functionality within this package requires that the ``__main__`` method be
  23. importable by the children. This is covered in :ref:`multiprocessing-programming`
  24. however it is worth pointing out here. This means that some examples, such
  25. as the :class:`multiprocessing.Pool` examples will not work in the
  26. interactive interpreter. For example::
  27. >>> from multiprocessing import Pool
  28. >>> p = Pool(5)
  29. >>> def f(x):
  30. ... return x*x
  31. ...
  32. >>> p.map(f, [1,2,3])
  33. Process PoolWorker-1:
  34. Process PoolWorker-2:
  35. Process PoolWorker-3:
  36. Traceback (most recent call last):
  37. Traceback (most recent call last):
  38. Traceback (most recent call last):
  39. AttributeError: 'module' object has no attribute 'f'
  40. AttributeError: 'module' object has no attribute 'f'
  41. AttributeError: 'module' object has no attribute 'f'
  42. (If you try this it will actually output three full tracebacks
  43. interleaved in a semi-random fashion, and then you may have to
  44. stop the master process somehow.)
  45. The :class:`Process` class
  46. ~~~~~~~~~~~~~~~~~~~~~~~~~~
  47. In :mod:`multiprocessing`, processes are spawned by creating a :class:`Process`
  48. object and then calling its :meth:`~Process.start` method. :class:`Process`
  49. follows the API of :class:`threading.Thread`. A trivial example of a
  50. multiprocess program is ::
  51. from multiprocessing import Process
  52. def f(name):
  53. print 'hello', name
  54. if __name__ == '__main__':
  55. p = Process(target=f, args=('bob',))
  56. p.start()
  57. p.join()
  58. To show the individual process IDs involved, here is an expanded example::
  59. from multiprocessing import Process
  60. import os
  61. def info(title):
  62. print title
  63. print 'module name:', __name__
  64. print 'parent process:', os.getppid()
  65. print 'process id:', os.getpid()
  66. def f(name):
  67. info('function f')
  68. print 'hello', name
  69. if __name__ == '__main__':
  70. info('main line')
  71. p = Process(target=f, args=('bob',))
  72. p.start()
  73. p.join()
  74. For an explanation of why (on Windows) the ``if __name__ == '__main__'`` part is
  75. necessary, see :ref:`multiprocessing-programming`.
  76. Exchanging objects between processes
  77. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  78. :mod:`multiprocessing` supports two types of communication channel between
  79. processes:
  80. **Queues**
  81. The :class:`Queue` class is a near clone of :class:`Queue.Queue`. For
  82. example::
  83. from multiprocessing import Process, Queue
  84. def f(q):
  85. q.put([42, None, 'hello'])
  86. if __name__ == '__main__':
  87. q = Queue()
  88. p = Process(target=f, args=(q,))
  89. p.start()
  90. print q.get() # prints "[42, None, 'hello']"
  91. p.join()
  92. Queues are thread and process safe.
  93. **Pipes**
  94. The :func:`Pipe` function returns a pair of connection objects connected by a
  95. pipe which by default is duplex (two-way). For example::
  96. from multiprocessing import Process, Pipe
  97. def f(conn):
  98. conn.send([42, None, 'hello'])
  99. conn.close()
  100. if __name__ == '__main__':
  101. parent_conn, child_conn = Pipe()
  102. p = Process(target=f, args=(child_conn,))
  103. p.start()
  104. print parent_conn.recv() # prints "[42, None, 'hello']"
  105. p.join()
  106. The two connection objects returned by :func:`Pipe` represent the two ends of
  107. the pipe. Each connection object has :meth:`~Connection.send` and
  108. :meth:`~Connection.recv` methods (among others). Note that data in a pipe
  109. may become corrupted if two processes (or threads) try to read from or write
  110. to the *same* end of the pipe at the same time. Of course there is no risk
  111. of corruption from processes using different ends of the pipe at the same
  112. time.
  113. Synchronization between processes
  114. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  115. :mod:`multiprocessing` contains equivalents of all the synchronization
  116. primitives from :mod:`threading`. For instance one can use a lock to ensure
  117. that only one process prints to standard output at a time::
  118. from multiprocessing import Process, Lock
  119. def f(l, i):
  120. l.acquire()
  121. print 'hello world', i
  122. l.release()
  123. if __name__ == '__main__':
  124. lock = Lock()
  125. for num in range(10):
  126. Process(target=f, args=(lock, num)).start()
  127. Without using the lock output from the different processes is liable to get all
  128. mixed up.
  129. Sharing state between processes
  130. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  131. As mentioned above, when doing concurrent programming it is usually best to
  132. avoid using shared state as far as possible. This is particularly true when
  133. using multiple processes.
  134. However, if you really do need to use some shared data then
  135. :mod:`multiprocessing` provides a couple of ways of doing so.
  136. **Shared memory**
  137. Data can be stored in a shared memory map using :class:`Value` or
  138. :class:`Array`. For example, the following code ::
  139. from multiprocessing import Process, Value, Array
  140. def f(n, a):
  141. n.value = 3.1415927
  142. for i in range(len(a)):
  143. a[i] = -a[i]
  144. if __name__ == '__main__':
  145. num = Value('d', 0.0)
  146. arr = Array('i', range(10))
  147. p = Process(target=f, args=(num, arr))
  148. p.start()
  149. p.join()
  150. print num.value
  151. print arr[:]
  152. will print ::
  153. 3.1415927
  154. [0, -1, -2, -3, -4, -5, -6, -7, -8, -9]
  155. The ``'d'`` and ``'i'`` arguments used when creating ``num`` and ``arr`` are
  156. typecodes of the kind used by the :mod:`array` module: ``'d'`` indicates a
  157. double precision float and ``'i'`` indicates a signed integer. These shared
  158. objects will be process and thread safe.
  159. For more flexibility in using shared memory one can use the
  160. :mod:`multiprocessing.sharedctypes` module which supports the creation of
  161. arbitrary ctypes objects allocated from shared memory.
  162. **Server process**
  163. A manager object returned by :func:`Manager` controls a server process which
  164. holds Python objects and allows other processes to manipulate them using
  165. proxies.
  166. A manager returned by :func:`Manager` will support types :class:`list`,
  167. :class:`dict`, :class:`Namespace`, :class:`Lock`, :class:`RLock`,
  168. :class:`Semaphore`, :class:`BoundedSemaphore`, :class:`Condition`,
  169. :class:`Event`, :class:`Queue`, :class:`Value` and :class:`Array`. For
  170. example, ::
  171. from multiprocessing import Process, Manager
  172. def f(d, l):
  173. d[1] = '1'
  174. d['2'] = 2
  175. d[0.25] = None
  176. l.reverse()
  177. if __name__ == '__main__':
  178. manager = Manager()
  179. d = manager.dict()
  180. l = manager.list(range(10))
  181. p = Process(target=f, args=(d, l))
  182. p.start()
  183. p.join()
  184. print d
  185. print l
  186. will print ::
  187. {0.25: None, 1: '1', '2': 2}
  188. [9, 8, 7, 6, 5, 4, 3, 2, 1, 0]
  189. Server process managers are more flexible than using shared memory objects
  190. because they can be made to support arbitrary object types. Also, a single
  191. manager can be shared by processes on different computers over a network.
  192. They are, however, slower than using shared memory.
  193. Using a pool of workers
  194. ~~~~~~~~~~~~~~~~~~~~~~~
  195. The :class:`~multiprocessing.pool.Pool` class represents a pool of worker
  196. processes. It has methods which allows tasks to be offloaded to the worker
  197. processes in a few different ways.
  198. For example::
  199. from multiprocessing import Pool
  200. def f(x):
  201. return x*x
  202. if __name__ == '__main__':
  203. pool = Pool(processes=4) # start 4 worker processes
  204. result = pool.apply_async(f, [10]) # evaluate "f(10)" asynchronously
  205. print result.get(timeout=1) # prints "100" unless your computer is *very* slow
  206. print pool.map(f, range(10)) # prints "[0, 1, 4,..., 81]"
  207. Reference
  208. ---------
  209. The :mod:`multiprocessing` package mostly replicates the API of the
  210. :mod:`threading` module.
  211. :class:`Process` and exceptions
  212. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213. .. class:: Process([group[, target[, name[, args[, kwargs]]]]])
  214. Process objects represent activity that is run in a separate process. The
  215. :class:`Process` class has equivalents of all the methods of
  216. :class:`threading.Thread`.
  217. The constructor should always be called with keyword arguments. *group*
  218. should always be ``None``; it exists solely for compatibility with
  219. :class:`threading.Thread`. *target* is the callable object to be invoked by
  220. the :meth:`run()` method. It defaults to ``None``, meaning nothing is
  221. called. *name* is the process name. By default, a unique name is constructed
  222. of the form 'Process-N\ :sub:`1`:N\ :sub:`2`:...:N\ :sub:`k`' where N\
  223. :sub:`1`,N\ :sub:`2`,...,N\ :sub:`k` is a sequence of integers whose length
  224. is determined by the *generation* of the process. *args* is the argument
  225. tuple for the target invocation. *kwargs* is a dictionary of keyword
  226. arguments for the target invocation. By default, no arguments are passed to
  227. *target*.
  228. If a subclass overrides the constructor, it must make sure it invokes the
  229. base class constructor (:meth:`Process.__init__`) before doing anything else
  230. to the process.
  231. .. method:: run()
  232. Method representing the process's activity.
  233. You may override this method in a subclass. The standard :meth:`run`
  234. method invokes the callable object passed to the object's constructor as
  235. the target argument, if any, with sequential and keyword arguments taken
  236. from the *args* and *kwargs* arguments, respectively.
  237. .. method:: start()
  238. Start the process's activity.
  239. This must be called at most once per process object. It arranges for the
  240. object's :meth:`run` method to be invoked in a separate process.
  241. .. method:: join([timeout])
  242. Block the calling thread until the process whose :meth:`join` method is
  243. called terminates or until the optional timeout occurs.
  244. If *timeout* is ``None`` then there is no timeout.
  245. A process can be joined many times.
  246. A process cannot join itself because this would cause a deadlock. It is
  247. an error to attempt to join a process before it has been started.
  248. .. attribute:: name
  249. The process's name.
  250. The name is a string used for identification purposes only. It has no
  251. semantics. Multiple processes may be given the same name. The initial
  252. name is set by the constructor.
  253. .. method:: is_alive
  254. Return whether the process is alive.
  255. Roughly, a process object is alive from the moment the :meth:`start`
  256. method returns until the child process terminates.
  257. .. attribute:: daemon
  258. The process's daemon flag, a Boolean value. This must be set before
  259. :meth:`start` is called.
  260. The initial value is inherited from the creating process.
  261. When a process exits, it attempts to terminate all of its daemonic child
  262. processes.
  263. Note that a daemonic process is not allowed to create child processes.
  264. Otherwise a daemonic process would leave its children orphaned if it gets
  265. terminated when its parent process exits.
  266. In addition to the :class:`Threading.Thread` API, :class:`Process` objects
  267. also support the following attributes and methods:
  268. .. attribute:: pid
  269. Return the process ID. Before the process is spawned, this will be
  270. ``None``.
  271. .. attribute:: exitcode
  272. The child's exit code. This will be ``None`` if the process has not yet
  273. terminated. A negative value *-N* indicates that the child was terminated
  274. by signal *N*.
  275. .. attribute:: authkey
  276. The process's authentication key (a byte string).
  277. When :mod:`multiprocessing` is initialized the main process is assigned a
  278. random string using :func:`os.random`.
  279. When a :class:`Process` object is created, it will inherit the
  280. authentication key of its parent process, although this may be changed by
  281. setting :attr:`authkey` to another byte string.
  282. See :ref:`multiprocessing-auth-keys`.
  283. .. method:: terminate()
  284. Terminate the process. On Unix this is done using the ``SIGTERM`` signal;
  285. on Windows :cfunc:`TerminateProcess` is used. Note that exit handlers and
  286. finally clauses, etc., will not be executed.
  287. Note that descendant processes of the process will *not* be terminated --
  288. they will simply become orphaned.
  289. .. warning::
  290. If this method is used when the associated process is using a pipe or
  291. queue then the pipe or queue is liable to become corrupted and may
  292. become unusable by other process. Similarly, if the process has
  293. acquired a lock or semaphore etc. then terminating it is liable to
  294. cause other processes to deadlock.
  295. Note that the :meth:`start`, :meth:`join`, :meth:`is_alive` and
  296. :attr:`exit_code` methods should only be called by the process that created
  297. the process object.
  298. Example usage of some of the methods of :class:`Process`:
  299. .. doctest::
  300. >>> import multiprocessing, time, signal
  301. >>> p = multiprocessing.Process(target=time.sleep, args=(1000,))
  302. >>> print p, p.is_alive()
  303. <Process(Process-1, initial)> False
  304. >>> p.start()
  305. >>> print p, p.is_alive()
  306. <Process(Process-1, started)> True
  307. >>> p.terminate()
  308. >>> time.sleep(0.1)
  309. >>> print p, p.is_alive()
  310. <Process(Process-1, stopped[SIGTERM])> False
  311. >>> p.exitcode == -signal.SIGTERM
  312. True
  313. .. exception:: BufferTooShort
  314. Exception raised by :meth:`Connection.recv_bytes_into()` when the supplied
  315. buffer object is too small for the message read.
  316. If ``e`` is an instance of :exc:`BufferTooShort` then ``e.args[0]`` will give
  317. the message as a byte string.
  318. Pipes and Queues
  319. ~~~~~~~~~~~~~~~~
  320. When using multiple processes, one generally uses message passing for
  321. communication between processes and avoids having to use any synchronization
  322. primitives like locks.
  323. For passing messages one can use :func:`Pipe` (for a connection between two
  324. processes) or a queue (which allows multiple producers and consumers).
  325. The :class:`Queue` and :class:`JoinableQueue` types are multi-producer,
  326. multi-consumer FIFO queues modelled on the :class:`Queue.Queue` class in the
  327. standard library. They differ in that :class:`Queue` lacks the
  328. :meth:`~Queue.Queue.task_done` and :meth:`~Queue.Queue.join` methods introduced
  329. into Python 2.5's :class:`Queue.Queue` class.
  330. If you use :class:`JoinableQueue` then you **must** call
  331. :meth:`JoinableQueue.task_done` for each task removed from the queue or else the
  332. semaphore used to count the number of unfinished tasks may eventually overflow
  333. raising an exception.
  334. Note that one can also create a shared queue by using a manager object -- see
  335. :ref:`multiprocessing-managers`.
  336. .. note::
  337. :mod:`multiprocessing` uses the usual :exc:`Queue.Empty` and
  338. :exc:`Queue.Full` exceptions to signal a timeout. They are not available in
  339. the :mod:`multiprocessing` namespace so you need to import them from
  340. :mod:`Queue`.
  341. .. warning::
  342. If a process is killed using :meth:`Process.terminate` or :func:`os.kill`
  343. while it is trying to use a :class:`Queue`, then the data in the queue is
  344. likely to become corrupted. This may cause any other processes to get an
  345. exception when it tries to use the queue later on.
  346. .. warning::
  347. As mentioned above, if a child process has put items on a queue (and it has
  348. not used :meth:`JoinableQueue.cancel_join_thread`), then that process will
  349. not terminate until all buffered items have been flushed to the pipe.
  350. This means that if you try joining that process you may get a deadlock unless
  351. you are sure that all items which have been put on the queue have been
  352. consumed. Similarly, if the child process is non-daemonic then the parent
  353. process may hang on exit when it tries to join all its non-daemonic children.
  354. Note that a queue created using a manager does not have this issue. See
  355. :ref:`multiprocessing-programming`.
  356. For an example of the usage of queues for interprocess communication see
  357. :ref:`multiprocessing-examples`.
  358. .. function:: Pipe([duplex])
  359. Returns a pair ``(conn1, conn2)`` of :class:`Connection` objects representing
  360. the ends of a pipe.
  361. If *duplex* is ``True`` (the default) then the pipe is bidirectional. If
  362. *duplex* is ``False`` then the pipe is unidirectional: ``conn1`` can only be
  363. used for receiving messages and ``conn2`` can only be used for sending
  364. messages.
  365. .. class:: Queue([maxsize])
  366. Returns a process shared queue implemented using a pipe and a few
  367. locks/semaphores. When a process first puts an item on the queue a feeder
  368. thread is started which transfers objects from a buffer into the pipe.
  369. The usual :exc:`Queue.Empty` and :exc:`Queue.Full` exceptions from the
  370. standard library's :mod:`Queue` module are raised to signal timeouts.
  371. :class:`Queue` implements all the methods of :class:`Queue.Queue` except for
  372. :meth:`~Queue.Queue.task_done` and :meth:`~Queue.Queue.join`.
  373. .. method:: qsize()
  374. Return the approximate size of the queue. Because of
  375. multithreading/multiprocessing semantics, this number is not reliable.
  376. Note that this may raise :exc:`NotImplementedError` on Unix platforms like
  377. Mac OS X where ``sem_getvalue()`` is not implemented.
  378. .. method:: empty()
  379. Return ``True`` if the queue is empty, ``False`` otherwise. Because of
  380. multithreading/multiprocessing semantics, this is not reliable.
  381. .. method:: full()
  382. Return ``True`` if the queue is full, ``False`` otherwise. Because of
  383. multithreading/multiprocessing semantics, this is not reliable.
  384. .. method:: put(item[, block[, timeout]])
  385. Put item into the queue. If the optional argument *block* is ``True``
  386. (the default) and *timeout* is ``None`` (the default), block if necessary until
  387. a free slot is available. If *timeout* is a positive number, it blocks at
  388. most *timeout* seconds and raises the :exc:`Queue.Full` exception if no
  389. free slot was available within that time. Otherwise (*block* is
  390. ``False``), put an item on the queue if a free slot is immediately
  391. available, else raise the :exc:`Queue.Full` exception (*timeout* is
  392. ignored in that case).
  393. .. method:: put_nowait(item)
  394. Equivalent to ``put(item, False)``.
  395. .. method:: get([block[, timeout]])
  396. Remove and return an item from the queue. If optional args *block* is
  397. ``True`` (the default) and *timeout* is ``None`` (the default), block if
  398. necessary until an item is available. If *timeout* is a positive number,
  399. it blocks at most *timeout* seconds and raises the :exc:`Queue.Empty`
  400. exception if no item was available within that time. Otherwise (block is
  401. ``False``), return an item if one is immediately available, else raise the
  402. :exc:`Queue.Empty` exception (*timeout* is ignored in that case).
  403. .. method:: get_nowait()
  404. get_no_wait()
  405. Equivalent to ``get(False)``.
  406. :class:`multiprocessing.Queue` has a few additional methods not found in
  407. :class:`Queue.Queue`. These methods are usually unnecessary for most
  408. code:
  409. .. method:: close()
  410. Indicate that no more data will be put on this queue by the current
  411. process. The background thread will quit once it has flushed all buffered
  412. data to the pipe. This is called automatically when the queue is garbage
  413. collected.
  414. .. method:: join_thread()
  415. Join the background thread. This can only be used after :meth:`close` has
  416. been called. It blocks until the background thread exits, ensuring that
  417. all data in the buffer has been flushed to the pipe.
  418. By default if a process is not the creator of the queue then on exit it
  419. will attempt to join the queue's background thread. The process can call
  420. :meth:`cancel_join_thread` to make :meth:`join_thread` do nothing.
  421. .. method:: cancel_join_thread()
  422. Prevent :meth:`join_thread` from blocking. In particular, this prevents
  423. the background thread from being joined automatically when the process
  424. exits -- see :meth:`join_thread`.
  425. .. class:: JoinableQueue([maxsize])
  426. :class:`JoinableQueue`, a :class:`Queue` subclass, is a queue which
  427. additionally has :meth:`task_done` and :meth:`join` methods.
  428. .. method:: task_done()
  429. Indicate that a formerly enqueued task is complete. Used by queue consumer
  430. threads. For each :meth:`~Queue.get` used to fetch a task, a subsequent
  431. call to :meth:`task_done` tells the queue that the processing on the task
  432. is complete.
  433. If a :meth:`~Queue.join` is currently blocking, it will resume when all
  434. items have been processed (meaning that a :meth:`task_done` call was
  435. received for every item that had been :meth:`~Queue.put` into the queue).
  436. Raises a :exc:`ValueError` if called more times than there were items
  437. placed in the queue.
  438. .. method:: join()
  439. Block until all items in the queue have been gotten and processed.
  440. The count of unfinished tasks goes up whenever an item is added to the
  441. queue. The count goes down whenever a consumer thread calls
  442. :meth:`task_done` to indicate that the item was retrieved and all work on
  443. it is complete. When the count of unfinished tasks drops to zero,
  444. :meth:`~Queue.join` unblocks.
  445. Miscellaneous
  446. ~~~~~~~~~~~~~
  447. .. function:: active_children()
  448. Return list of all live children of the current process.
  449. Calling this has the side affect of "joining" any processes which have
  450. already finished.
  451. .. function:: cpu_count()
  452. Return the number of CPUs in the system. May raise
  453. :exc:`NotImplementedError`.
  454. .. function:: current_process()
  455. Return the :class:`Process` object corresponding to the current process.
  456. An analogue of :func:`threading.current_thread`.
  457. .. function:: freeze_support()
  458. Add support for when a program which uses :mod:`multiprocessing` has been
  459. frozen to produce a Windows executable. (Has been tested with **py2exe**,
  460. **PyInstaller** and **cx_Freeze**.)
  461. One needs to call this function straight after the ``if __name__ ==
  462. '__main__'`` line of the main module. For example::
  463. from multiprocessing import Process, freeze_support
  464. def f():
  465. print 'hello world!'
  466. if __name__ == '__main__':
  467. freeze_support()
  468. Process(target=f).start()
  469. If the ``freeze_support()`` line is omitted then trying to run the frozen
  470. executable will raise :exc:`RuntimeError`.
  471. If the module is being run normally by the Python interpreter then
  472. :func:`freeze_support` has no effect.
  473. .. function:: set_executable()
  474. Sets the path of the python interpreter to use when starting a child process.
  475. (By default :data:`sys.executable` is used). Embedders will probably need to
  476. do some thing like ::
  477. setExecutable(os.path.join(sys.exec_prefix, 'pythonw.exe'))
  478. before they can create child processes. (Windows only)
  479. .. note::
  480. :mod:`multiprocessing` contains no analogues of
  481. :func:`threading.active_count`, :func:`threading.enumerate`,
  482. :func:`threading.settrace`, :func:`threading.setprofile`,
  483. :class:`threading.Timer`, or :class:`threading.local`.
  484. Connection Objects
  485. ~~~~~~~~~~~~~~~~~~
  486. Connection objects allow the sending and receiving of picklable objects or
  487. strings. They can be thought of as message oriented connected sockets.
  488. Connection objects usually created using :func:`Pipe` -- see also
  489. :ref:`multiprocessing-listeners-clients`.
  490. .. class:: Connection
  491. .. method:: send(obj)
  492. Send an object to the other end of the connection which should be read
  493. using :meth:`recv`.
  494. The object must be picklable.
  495. .. method:: recv()
  496. Return an object sent from the other end of the connection using
  497. :meth:`send`. Raises :exc:`EOFError` if there is nothing left to receive
  498. and the other end was closed.
  499. .. method:: fileno()
  500. Returns the file descriptor or handle used by the connection.
  501. .. method:: close()
  502. Close the connection.
  503. This is called automatically when the connection is garbage collected.
  504. .. method:: poll([timeout])
  505. Return whether there is any data available to be read.
  506. If *timeout* is not specified then it will return immediately. If
  507. *timeout* is a number then this specifies the maximum time in seconds to
  508. block. If *timeout* is ``None`` then an infinite timeout is used.
  509. .. method:: send_bytes(buffer[, offset[, size]])
  510. Send byte data from an object supporting the buffer interface as a
  511. complete message.
  512. If *offset* is given then data is read from that position in *buffer*. If
  513. *size* is given then that many bytes will be read from buffer.
  514. .. method:: recv_bytes([maxlength])
  515. Return a complete message of byte data sent from the other end of the
  516. connection as a string. Raises :exc:`EOFError` if there is nothing left
  517. to receive and the other end has closed.
  518. If *maxlength* is specified and the message is longer than *maxlength*
  519. then :exc:`IOError` is raised and the connection will no longer be
  520. readable.
  521. .. method:: recv_bytes_into(buffer[, offset])
  522. Read into *buffer* a complete message of byte data sent from the other end
  523. of the connection and return the number of bytes in the message. Raises
  524. :exc:`EOFError` if there is nothing left to receive and the other end was
  525. closed.
  526. *buffer* must be an object satisfying the writable buffer interface. If
  527. *offset* is given then the message will be written into the buffer from
  528. that position. Offset must be a non-negative integer less than the
  529. length of *buffer* (in bytes).
  530. If the buffer is too short then a :exc:`BufferTooShort` exception is
  531. raised and the complete message is available as ``e.args[0]`` where ``e``
  532. is the exception instance.
  533. For example:
  534. .. doctest::
  535. >>> from multiprocessing import Pipe
  536. >>> a, b = Pipe()
  537. >>> a.send([1, 'hello', None])
  538. >>> b.recv()
  539. [1, 'hello', None]
  540. >>> b.send_bytes('thank you')
  541. >>> a.recv_bytes()
  542. 'thank you'
  543. >>> import array
  544. >>> arr1 = array.array('i', range(5))
  545. >>> arr2 = array.array('i', [0] * 10)
  546. >>> a.send_bytes(arr1)
  547. >>> count = b.recv_bytes_into(arr2)
  548. >>> assert count == len(arr1) * arr1.itemsize
  549. >>> arr2
  550. array('i', [0, 1, 2, 3, 4, 0, 0, 0, 0, 0])
  551. .. warning::
  552. The :meth:`Connection.recv` method automatically unpickles the data it
  553. receives, which can be a security risk unless you can trust the process
  554. which sent the message.
  555. Therefore, unless the connection object was produced using :func:`Pipe` you
  556. should only use the :meth:`~Connection.recv` and :meth:`~Connection.send`
  557. methods after performing some sort of authentication. See
  558. :ref:`multiprocessing-auth-keys`.
  559. .. warning::
  560. If a process is killed while it is trying to read or write to a pipe then
  561. the data in the pipe is likely to become corrupted, because it may become
  562. impossible to be sure where the message boundaries lie.
  563. Synchronization primitives
  564. ~~~~~~~~~~~~~~~~~~~~~~~~~~
  565. Generally synchronization primitives are not as necessary in a multiprocess
  566. program as they are in a multithreaded program. See the documentation for
  567. :mod:`threading` module.
  568. Note that one can also create synchronization primitives by using a manager
  569. object -- see :ref:`multiprocessing-managers`.
  570. .. class:: BoundedSemaphore([value])
  571. A bounded semaphore object: a clone of :class:`threading.BoundedSemaphore`.
  572. (On Mac OS X this is indistinguishable from :class:`Semaphore` because
  573. ``sem_getvalue()`` is not implemented on that platform).
  574. .. class:: Condition([lock])
  575. A condition variable: a clone of :class:`threading.Condition`.
  576. If *lock* is specified then it should be a :class:`Lock` or :class:`RLock`
  577. object from :mod:`multiprocessing`.
  578. .. class:: Event()
  579. A clone of :class:`threading.Event`.
  580. .. class:: Lock()
  581. A non-recursive lock object: a clone of :class:`threading.Lock`.
  582. .. class:: RLock()
  583. A recursive lock object: a clone of :class:`threading.RLock`.
  584. .. class:: Semaphore([value])
  585. A bounded semaphore object: a clone of :class:`threading.Semaphore`.
  586. .. note::
  587. The :meth:`acquire` method of :class:`BoundedSemaphore`, :class:`Lock`,
  588. :class:`RLock` and :class:`Semaphore` has a timeout parameter not supported
  589. by the equivalents in :mod:`threading`. The signature is
  590. ``acquire(block=True, timeout=None)`` with keyword parameters being
  591. acceptable. If *block* is ``True`` and *timeout* is not ``None`` then it
  592. specifies a timeout in seconds. If *block* is ``False`` then *timeout* is
  593. ignored.
  594. .. note::
  595. On OS/X ``sem_timedwait`` is unsupported, so timeout arguments for the
  596. aforementioned :meth:`acquire` methods will be ignored on OS/X.
  597. .. note::
  598. If the SIGINT signal generated by Ctrl-C arrives while the main thread is
  599. blocked by a call to :meth:`BoundedSemaphore.acquire`, :meth:`Lock.acquire`,
  600. :meth:`RLock.acquire`, :meth:`Semaphore.acquire`, :meth:`Condition.acquire`
  601. or :meth:`Condition.wait` then the call will be immediately interrupted and
  602. :exc:`KeyboardInterrupt` will be raised.
  603. This differs from the behaviour of :mod:`threading` where SIGINT will be
  604. ignored while the equivalent blocking calls are in progress.
  605. Shared :mod:`ctypes` Objects
  606. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  607. It is possible to create shared objects using shared memory which can be
  608. inherited by child processes.
  609. .. function:: Value(typecode_or_type, *args[, lock])
  610. Return a :mod:`ctypes` object allocated from shared memory. By default the
  611. return value is actually a synchronized wrapper for the object.
  612. *typecode_or_type* determines the type of the returned object: it is either a
  613. ctypes type or a one character typecode of the kind used by the :mod:`array`
  614. module. *\*args* is passed on to the constructor for the type.
  615. If *lock* is ``True`` (the default) then a new lock object is created to
  616. synchronize access to the value. If *lock* is a :class:`Lock` or
  617. :class:`RLock` object then that will be used to synchronize access to the
  618. value. If *lock* is ``False`` then access to the returned object will not be
  619. automatically protected by a lock, so it will not necessarily be
  620. "process-safe".
  621. Note that *lock* is a keyword-only argument.
  622. .. function:: Array(typecode_or_type, size_or_initializer, *, lock=True)
  623. Return a ctypes array allocated from shared memory. By default the return
  624. value is actually a synchronized wrapper for the array.
  625. *typecode_or_type* determines the type of the elements of the returned array:
  626. it is either a ctypes type or a one character typecode of the kind used by
  627. the :mod:`array` module. If *size_or_initializer* is an integer, then it
  628. determines the length of the array, and the array will be initially zeroed.
  629. Otherwise, *size_or_initializer* is a sequence which is used to initialize
  630. the array and whose length determines the length of the array.
  631. If *lock* is ``True`` (the default) then a new lock object is created to
  632. synchronize access to the value. If *lock* is a :class:`Lock` or
  633. :class:`RLock` object then that will be used to synchronize access to the
  634. value. If *lock* is ``False`` then access to the returned object will not be
  635. automatically protected by a lock, so it will not necessarily be
  636. "process-safe".
  637. Note that *lock* is a keyword only argument.
  638. Note that an array of :data:`ctypes.c_char` has *value* and *raw*
  639. attributes which allow one to use it to store and retrieve strings.
  640. The :mod:`multiprocessing.sharedctypes` module
  641. >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
  642. .. module:: multiprocessing.sharedctypes
  643. :synopsis: Allocate ctypes objects from shared memory.
  644. The :mod:`multiprocessing.sharedctypes` module provides functions for allocating
  645. :mod:`ctypes` objects from shared memory which can be inherited by child
  646. processes.
  647. .. note::
  648. Although it is possible to store a pointer in shared memory remember that
  649. this will refer to a location in the address space of a specific process.
  650. However, the pointer is quite likely to be invalid in the context of a second
  651. process and trying to dereference the pointer from the second process may
  652. cause a crash.
  653. .. function:: RawArray(typecode_or_type, size_or_initializer)
  654. Return a ctypes array allocated from shared memory.
  655. *typecode_or_type* determines the type of the elements of the returned array:
  656. it is either a ctypes type or a one character typecode of the kind used by
  657. the :mod:`array` module. If *size_or_initializer* is an integer then it
  658. determines the length of the array, and the array will be initially zeroed.
  659. Otherwise *size_or_initializer* is a sequence which is used to initialize the
  660. array and whose length determines the length of the array.
  661. Note that setting and getting an element is potentially non-atomic -- use
  662. :func:`Array` instead to make sure that access is automatically synchronized
  663. using a lock.
  664. .. function:: RawValue(typecode_or_type, *args)
  665. Return a ctypes object allocated from shared memory.
  666. *typecode_or_type* determines the type of the returned object: it is either a
  667. ctypes type or a one character typecode of the kind used by the :mod:`array`
  668. module. *\*args* is passed on to the constructor for the type.
  669. Note that setting and getting the value is potentially non-atomic -- use
  670. :func:`Value` instead to make sure that access is automatically synchronized
  671. using a lock.
  672. Note that an array of :data:`ctypes.c_char` has ``value`` and ``raw``
  673. attributes which allow one to use it to store and retrieve strings -- see
  674. documentation for :mod:`ctypes`.
  675. .. function:: Array(typecode_or_type, size_or_initializer, *args[, lock])
  676. The same as :func:`RawArray` except that depending on the value of *lock* a
  677. process-safe synchronization wrapper may be returned instead of a raw ctypes
  678. array.
  679. If *lock* is ``True`` (the default) then a new lock object is created to
  680. synchronize access to the value. If *lock* is a :class:`Lock` or
  681. :class:`RLock` object then that will be used to synchronize access to the
  682. value. If *lock* is ``False`` then access to the returned object will not be
  683. automatically protected by a lock, so it will not necessarily be
  684. "process-safe".
  685. Note that *lock* is a keyword-only argument.
  686. .. function:: Value(typecode_or_type, *args[, lock])
  687. The same as :func:`RawValue` except that depending on the value of *lock* a
  688. process-safe synchronization wrapper may be returned instead of a raw ctypes
  689. object.
  690. If *lock* is ``True`` (the default) then a new lock object is created to
  691. synchronize access to the value. If *lock* is a :class:`Lock` or
  692. :class:`RLock` object then that will be used to synchronize access to the
  693. value. If *lock* is ``False`` then access to the returned object will not be
  694. automatically protected by a lock, so it will not necessarily be
  695. "process-safe".
  696. Note that *lock* is a keyword-only argument.
  697. .. function:: copy(obj)
  698. Return a ctypes object allocated from shared memory which is a copy of the
  699. ctypes object *obj*.
  700. .. function:: synchronized(obj[, lock])
  701. Return a process-safe wrapper object for a ctypes object which uses *lock* to
  702. synchronize access. If *lock* is ``None`` (the default) then a
  703. :class:`multiprocessing.RLock` object is created automatically.
  704. A synchronized wrapper will have two methods in addition to those of the
  705. object it wraps: :meth:`get_obj` returns the wrapped object and
  706. :meth:`get_lock` returns the lock object used for synchronization.
  707. Note that accessing the ctypes object through the wrapper can be a lot slower
  708. than accessing the raw ctypes object.
  709. The table below compares the syntax for creating shared ctypes objects from
  710. shared memory with the normal ctypes syntax. (In the table ``MyStruct`` is some
  711. subclass of :class:`ctypes.Structure`.)
  712. ==================== ========================== ===========================
  713. ctypes sharedctypes using type sharedctypes using typecode
  714. ==================== ========================== ===========================
  715. c_double(2.4) RawValue(c_double, 2.4) RawValue('d', 2.4)
  716. MyStruct(4, 6) RawValue(MyStruct, 4, 6)
  717. (c_short * 7)() RawArray(c_short, 7) RawArray('h', 7)
  718. (c_int * 3)(9, 2, 8) RawArray(c_int, (9, 2, 8)) RawArray('i', (9, 2, 8))
  719. ==================== ========================== ===========================
  720. Below is an example where a number of ctypes objects are modified by a child
  721. process::
  722. from multiprocessing import Process, Lock
  723. from multiprocessing.sharedctypes import Value, Array
  724. from ctypes import Structure, c_double
  725. class Point(Structure):
  726. _fields_ = [('x', c_double), ('y', c_double)]
  727. def modify(n, x, s, A):
  728. n.value **= 2
  729. x.value **= 2
  730. s.value = s.value.upper()
  731. for a in A:
  732. a.x **= 2
  733. a.y **= 2
  734. if __name__ == '__main__':
  735. lock = Lock()
  736. n = Value('i', 7)
  737. x = Value(c_double, 1.0/3.0, lock=False)
  738. s = Array('c', 'hello world', lock=lock)
  739. A = Array(Point, [(1.875,-6.25), (-5.75,2.0), (2.375,9.5)], lock=lock)
  740. p = Process(target=modify, args=(n, x, s, A))
  741. p.start()
  742. p.join()
  743. print n.value
  744. print x.value
  745. print s.value
  746. print [(a.x, a.y) for a in A]
  747. .. highlightlang:: none
  748. The results printed are ::
  749. 49
  750. 0.1111111111111111
  751. HELLO WORLD
  752. [(3.515625, 39.0625), (33.0625, 4.0), (5.640625, 90.25)]
  753. .. highlightlang:: python
  754. .. _multiprocessing-managers:
  755. Managers
  756. ~~~~~~~~
  757. Managers provide a way to create data which can be shared between different
  758. processes. A manager object controls a server process which manages *shared
  759. objects*. Other processes can access the shared objects by using proxies.
  760. .. function:: multiprocessing.Manager()
  761. Returns a started :class:`~multiprocessing.managers.SyncManager` object which
  762. can be used for sharing objects between processes. The returned manager
  763. object corresponds to a spawned child process and has methods which will
  764. create shared objects and return corresponding proxies.
  765. .. module:: multiprocessing.managers
  766. :synopsis: Share data between process with shared objects.
  767. Manager processes will be shutdown as soon as they are garbage collected or
  768. their parent process exits. The manager classes are defined in the
  769. :mod:`multiprocessing.managers` module:
  770. .. class:: BaseManager([address[, authkey]])
  771. Create a BaseManager object.
  772. Once created one should call :meth:`start` or :meth:`serve_forever` to ensure
  773. that the manager object refers to a started manager process.
  774. *address* is the address on which the manager process listens for new
  775. connections. If *address* is ``None`` then an arbitrary one is chosen.
  776. *authkey* is the authentication key which will be used to check the validity
  777. of incoming connections to the server process. If *authkey* is ``None`` then
  778. ``current_process().authkey``. Otherwise *authkey* is used and it
  779. must be a string.
  780. .. method:: start()
  781. Start a subprocess to start the manager.
  782. .. method:: serve_forever()
  783. Run the server in the current process.
  784. .. method:: get_server()
  785. Returns a :class:`Server` object which represents the actual server under
  786. the control of the Manager. The :class:`Server` object supports the
  787. :meth:`serve_forever` method::
  788. >>> from multiprocessing.managers import BaseManager
  789. >>> manager = BaseManager(address=('', 50000), authkey='abc')
  790. >>> server = manager.get_server()
  791. >>> server.serve_forever()
  792. :class:`Server` additionally has an :attr:`address` attribute.
  793. .. method:: connect()
  794. Connect a local manager object to a remote manager process::
  795. >>> from multiprocessing.managers import BaseManager
  796. >>> m = BaseManager(address=('127.0.0.1', 5000), authkey='abc')
  797. >>> m.connect()
  798. .. method:: shutdown()
  799. Stop the process used by the manager. This is only available if
  800. :meth:`start` has been used to start the server process.
  801. This can be called multiple times.
  802. .. method:: register(typeid[, callable[, proxytype[, exposed[, method_to_typeid[, create_method]]]]])
  803. A classmethod which can be used for registering a type or callable with
  804. the manager class.
  805. *typeid* is a "type identifier" which is used to identify a particular
  806. type of shared object. This must be a string.
  807. *callable* is a callable used for creating objects for this type
  808. identifier. If a manager instance will be created using the
  809. :meth:`from_address` classmethod or if the *create_method* argument is
  810. ``False`` then this can be left as ``None``.
  811. *proxytype* is a subclass of :class:`BaseProxy` which is used to create
  812. proxies for shared objects with this *typeid*. If ``None`` then a proxy
  813. class is created automatically.
  814. *exposed* is used to specify a sequence of method names which proxies for
  815. this typeid should be allowed to access using
  816. :meth:`BaseProxy._callMethod`. (If *exposed* is ``None`` then
  817. :attr:`proxytype._exposed_` is used instead if it exists.) In the case
  818. where no exposed list is specified, all "public methods" of the shared
  819. object will be accessible. (Here a "public method" means any attribute
  820. which has a :meth:`__call__` method and whose name does not begin with
  821. ``'_'``.)
  822. *method_to_typeid* is a mapping used to specify the return type of those
  823. exposed methods which should return a proxy. It maps method names to
  824. typeid strings. (If *method_to_typeid* is ``None`` then
  825. :attr:`proxytype._method_to_typeid_` is used instead if it exists.) If a
  826. method's name is not a key of this mapping or if the mapping is ``None``
  827. then the object returned by the method will be copied by value.
  828. *create_method* determines whether a method should be created with name
  829. *typeid* which can be used to tell the server process to create a new
  830. shared object and return a proxy for it. By default it is ``True``.
  831. :class:`BaseManager` instances also have one read-only property:
  832. .. attribute:: address
  833. The address used by the manager.
  834. .. class:: SyncManager
  835. A subclass of :class:`BaseManager` which can be used for the synchronization
  836. of processes. Objects of this type are returned by
  837. :func:`multiprocessing.Manager`.
  838. It also supports creation of shared lists and dictionaries.
  839. .. method:: BoundedSemaphore([value])
  840. Create a shared :class:`threading.BoundedSemaphore` object and return a
  841. proxy for it.
  842. .. method:: Condition([lock])
  843. Create a shared :class:`threading.Condition` object and return a proxy for
  844. it.
  845. If *lock* is supplied then it should be a proxy for a
  846. :class:`threading.Lock` or :class:`threading.RLock` object.
  847. .. method:: Event()
  848. Create a shared :class:`threading.Event` object and return a proxy for it.
  849. .. method:: Lock()
  850. Create a shared :class:`threading.Lock` object and return a proxy for it.
  851. .. method:: Namespace()
  852. Create a shared :class:`Namespace` object and return a proxy for it.
  853. .. method:: Queue([maxsize])
  854. Create a shared :class:`Queue.Queue` object and return a proxy for it.
  855. .. method:: RLock()
  856. Create a shared :class:`threading.RLock` object and return a proxy for it.
  857. .. method:: Semaphore([value])
  858. Create a shared :class:`threading.Semaphore` object and return a proxy for
  859. it.
  860. .. method:: Array(typecode, sequence)
  861. Create an array and return a proxy for it.
  862. .. method:: Value(typecode, value)
  863. Create an object with a writable ``value`` attribute and return a proxy
  864. for it.
  865. .. method:: dict()
  866. dict(mapping)
  867. dict(sequence)
  868. Create a shared ``dict`` object and return a proxy for it.
  869. .. method:: list()
  870. list(sequence)
  871. Create a shared ``list`` object and return a proxy for it.
  872. Namespace objects
  873. >>>>>>>>>>>>>>>>>
  874. A namespace object has no public methods, but does have writable attributes.
  875. Its representation shows the values of its attributes.
  876. However, when using a proxy for a namespace object, an attribute beginning with
  877. ``'_'`` will be an attribute of the proxy and not an attribute of the referent:
  878. .. doctest::
  879. >>> manager = multiprocessing.Manager()
  880. >>> Global = manager.Namespace()
  881. >>> Global.x = 10
  882. >>> Global.y = 'hello'
  883. >>> Global._z = 12.3 # this is an attribute of the proxy
  884. >>> print Global
  885. Namespace(x=10, y='hello')
  886. Customized managers
  887. >>>>>>>>>>>>>>>>>>>
  888. To create one's own manager, one creates a subclass of :class:`BaseManager` and
  889. use the :meth:`~BaseManager.register` classmethod to register new types or
  890. callables with the manager class. For example::
  891. from multiprocessing.managers import BaseManager
  892. class MathsClass(object):
  893. def add(self, x, y):
  894. return x + y
  895. def mul(self, x, y):
  896. return x * y
  897. class MyManager(BaseManager):
  898. pass
  899. MyManager.register('Maths', MathsClass)
  900. if __name__ == '__main__':
  901. manager = MyManager()
  902. manager.start()
  903. maths = manager.Maths()
  904. print maths.add(4, 3) # prints 7
  905. print maths.mul(7, 8) # prints 56
  906. Using a remote manager
  907. >>>>>>>>>>>>>>>>>>>>>>
  908. It is possible to run a manager server on one machine and have clients use it
  909. from other machines (assuming that the firewalls involved allow it).
  910. Running the following commands creates a server for a single shared queue which
  911. remote clients can access::
  912. >>> from multiprocessing.managers import BaseManager
  913. >>> import Queue
  914. >>> queue = Queue.Queue()
  915. >>> class QueueManager(BaseManager): pass
  916. >>> QueueManager.register('get_queue', callable=lambda:queue)
  917. >>> m = QueueManager(address=('', 50000), authkey='abracadabra')
  918. >>> s = m.get_server()
  919. >>> s.serve_forever()
  920. One client can access the server as follows::
  921. >>> from multiprocessing.managers import BaseManager
  922. >>> class QueueManager(BaseManager): pass
  923. >>> QueueManager.register('get_queue')
  924. >>> m = QueueManager(address=('foo.bar.org', 50000), authkey='abracadabra')
  925. >>> m.connect()
  926. >>> queue = m.get_queue()
  927. >>> queue.put('hello')
  928. Another client can also use it::
  929. >>> from multiprocessing.managers import BaseManager
  930. >>> class QueueManager(BaseManager): pass
  931. >>> QueueManager.register('get_queue')
  932. >>> m = QueueManager(address=('foo.bar.org', 50000), authkey='abracadabra')
  933. >>> m.connect()
  934. >>> queue = m.get_queue()
  935. >>> queue.get()
  936. 'hello'
  937. Local processes can also access that queue, using the code from above on the
  938. client to access it remotely::
  939. >>> from multiprocessing import Process, Queue
  940. >>> from multiprocessing.managers import BaseManager
  941. >>> class Worker(Process):
  942. ... def __init__(self, q):
  943. ... self.q = q
  944. ... super(Worker, self).__init__()
  945. ... def run(self):
  946. ... self.q.put('local hello')
  947. ...
  948. >>> queue = Queue()
  949. >>> w = Worker(queue)
  950. >>> w.start()
  951. >>> class QueueManager(BaseManager): pass
  952. ...
  953. >>> QueueManager.register('get_queue', callable=lambda: queue)
  954. >>> m = QueueManager(address=('', 50000), authkey='abracadabra')
  955. >>> s = m.get_server()
  956. >>> s.serve_forever()
  957. Proxy Objects
  958. ~~~~~~~~~~~~~
  959. A proxy is an object which *refers* to a shared object which lives (presumably)
  960. in a different process. The shared object is said to be the *referent* of the
  961. proxy. Multiple proxy objects may have the same referent.
  962. A proxy object has methods which invoke corresponding methods of its referent
  963. (although not every method of the referent will necessarily be available through
  964. the proxy). A proxy can usually be used in most of the same ways that its
  965. referent can:
  966. .. doctest::
  967. >>> from multiprocessing import Manager
  968. >>> manager = Manager()
  969. >>> l = manager.list([i*i for i in range(10)])
  970. >>> print l
  971. [0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
  972. >>> print repr(l)
  973. <ListProxy object, typeid 'list' at 0x...>
  974. >>> l[4]
  975. 16
  976. >>> l[2:5]
  977. [4, 9, 16]
  978. Notice that applying :func:`str` to a proxy will return the representation of
  979. the referent, whereas applying :func:`repr` will return the representation of
  980. the proxy.
  981. An important feature of proxy objects is that they are picklable so they can be
  982. passed between processes. Note, however, that if a proxy is sent to the
  983. corresponding manager's process then unpickling it will produce the referent
  984. itself. This means, for example, that one shared object can contain a second:
  985. .. doctest::
  986. >>> a = manager.list()
  987. >>> b = manager.list()
  988. >>> a.append(b) # referent of a now contains referent of b
  989. >>> print a, b
  990. [[]] []
  991. >>> b.append('hello')
  992. >>> print a, b
  993. [['hello']] ['hello']
  994. .. note::
  995. The proxy types in :mod:`multiprocessing` do nothing to support comparisons
  996. by value. So, for instance, we have:
  997. .. doctest::
  998. >>> manager.list([1,2,3]) == [1,2,3]
  999. False
  1000. One should just use a copy of the referent instead when making comparisons.
  1001. .. class:: BaseProxy
  1002. Proxy objects are instances of subclasses of :class:`BaseProxy`.
  1003. .. method:: _callmethod(methodname[, args[, kwds]])
  1004. Call and return the result of a method of the proxy's referent.
  1005. If ``proxy`` is a proxy whose referent is ``obj`` then the expression ::
  1006. proxy._callmethod(methodname, args, kwds)
  1007. will evaluate the expression ::
  1008. getattr(obj, methodname)(*args, **kwds)
  1009. in the manager's process.
  1010. The returned value will be a copy of the result of the call or a proxy to
  1011. a new shared object -- see documentation for the *method_to_typeid*
  1012. argument of :meth:`BaseManager.register`.
  1013. If an exception is raised by the call, then then is re-raised by
  1014. :meth:`_callmethod`. If some other exception is raised in the manager's
  1015. process then this is converted into a :exc:`RemoteError` exception and is
  1016. raised by :meth:`_callmethod`.
  1017. Note in particular that an exception will be raised if *methodname* has
  1018. not been *exposed*
  1019. An example of the usage of :meth:`_callmethod`:
  1020. .. doctest::
  1021. >>> l = manager.list(range(10))
  1022. >>> l._callmethod('__len__')
  1023. 10
  1024. >>> l._callmethod('__getslice__', (2, 7)) # equiv to `l[2:7]`
  1025. [2, 3, 4, 5, 6]
  1026. >>> l._callmethod('__getitem__', (20,)) # equiv to `l[20]`
  1027. Traceback (most recent call last):
  1028. ...
  1029. IndexError: list index out of range
  1030. .. method:: _getvalue()
  1031. Return a copy of the referent.
  1032. If the referent is unpicklable then this will raise an exception.
  1033. .. method:: __repr__
  1034. Return a representation of the proxy object.
  1035. .. method:: __str__
  1036. Return the representation of the referent.
  1037. Cleanup
  1038. >>>>>>>
  1039. A proxy object uses a weakref callback so that when it gets garbage collected it
  1040. deregisters itself from the manager which owns its referent.
  1041. A shared object gets deleted from the manager process when there are no longer
  1042. any proxies referring to it.
  1043. Process Pools
  1044. ~~~~~~~~~~~~~
  1045. .. module:: multiprocessing.pool
  1046. :synopsis: Create pools of processes.
  1047. One can create a pool of processes which will carry out tasks submitted to it
  1048. with the :class:`Pool` class.
  1049. .. class:: multiprocessing.Pool([processes[, initializer[, initargs]]])
  1050. A process pool object which controls a pool of worker processes to which jobs
  1051. can be submitted. It supports asynchronous results with timeouts and
  1052. callbacks and has a parallel map implementation.
  1053. *processes* is the number of worker processes to use. If *processes* is
  1054. ``None`` then the number returned by :func:`cpu_count` is used. If
  1055. *initializer* is not ``None`` then each worker process will call
  1056. ``initializer(*initargs)`` when it starts.
  1057. .. method:: apply(func[, args[, kwds]])
  1058. Equivalent of the :func:`apply` builtin function. It blocks till the
  1059. result is ready.
  1060. .. method:: apply_async(func[, args[, kwds[, callback]]])
  1061. A variant of the :meth:`apply` method which returns a result object.
  1062. If *callback* is specified then it should be a callable which accepts a
  1063. single argument. When the result becomes ready *callback* is applied to
  1064. it (unless the call failed). *callback* should complete immediately since
  1065. otherwise the thread which handles the results will get blocked.
  1066. .. method:: map(func, iterable[, chunksize])
  1067. A parallel equivalent of the :func:`map` builtin function (it supports only
  1068. one *iterable* argument though). It blocks till the result is ready.
  1069. This method chops the iterable into a number of chunks which it submits to
  1070. the process pool as separate tasks. The (approximate) size of these
  1071. chunks can be specified by setting *chunksize* to a positive integer.
  1072. .. method:: map_async(func, iterable[, chunksize[, callback]])
  1073. A variant of the :meth:`map` method which returns a result object.
  1074. If *callback* is specified then it should be a callable which accepts a
  1075. single argument. When the result becomes ready *callback* is applied to
  1076. it (unless the call failed). *callback* should complete immediately since
  1077. otherwise the thread which handles the results will get blocked.
  1078. .. method:: imap(func, iterable[, chunksize])
  1079. An equivalent of :func:`itertools.imap`.
  1080. The *chunksize* argument is the same as the one used by the :meth:`.map`
  1081. method. For very long iterables using a large value for *chunksize* can
  1082. make make the job complete **much** faster than using the default value of
  1083. ``1``.
  1084. Also if *chunksize* is ``1`` then the :meth:`next` method of the iterator
  1085. returned by the :meth:`imap` method has an optional *timeout* parameter:
  1086. ``next(timeout)`` will raise :exc:`multiprocessing.TimeoutError` if the
  1087. result cannot be returned within *timeout* seconds.
  1088. .. method:: imap_unordered(func, iterable[, chunksize])
  1089. The same as :meth:`imap` except that the ordering of the results from the
  1090. returned iterator should be considered arbitrary. (Only when there is
  1091. only one worker process is the order guaranteed to be "correct".)
  1092. .. method:: close()
  1093. Prevents any more tasks from being submitted to the pool. Once all the
  1094. tasks have been completed the worker processes will exit.
  1095. .. method:: terminate()
  1096. Stops the worker processes immediately without completing outstanding
  1097. work. When the pool object is garbage collected :meth:`terminate` will be
  1098. called immediately.
  1099. .. method:: join()
  1100. Wait for the worker processes to exit. One must call :meth:`close` or
  1101. :meth:`terminate` before using :meth:`join`.
  1102. .. class:: AsyncResult
  1103. The class of the result returned by :meth:`Pool.apply_async` and
  1104. :meth:`Pool.map_async`.
  1105. .. method:: get([timeout])
  1106. Return the result when it arrives. If *timeout* is not ``None`` and the
  1107. result does not arrive within *timeout* seconds then
  1108. :exc:`multiprocessing.TimeoutError` is raised. If the remote call raised
  1109. an exception then that exception will be reraised by :meth:`get`.
  1110. .. method:: wait([timeout])
  1111. Wait until the result is available or until *timeout* seconds pass.
  1112. .. method:: ready()
  1113. Return whether the call has completed.
  1114. .. method:: successful()
  1115. Return whether the call completed without raising an exception. Will
  1116. raise :exc:`AssertionError` if the result is not ready.
  1117. The following example demonstrates the use of a pool::
  1118. from multiprocessing import Pool
  1119. def f(x):
  1120. return x*x
  1121. if __name__ == '__main__':
  1122. pool = Pool(processes=4) # start 4 worker processes
  1123. result = pool.apply_async(f, (10,)) # evaluate "f(10)" asynchronously
  1124. print result.get(timeout=1) # prints "100" unless your computer is *very* slow
  1125. print pool.map(f, range(10)) # prints "[0, 1, 4,..., 81]"
  1126. it = pool.imap(f, range(10))
  1127. print it.next() # prints "0"
  1128. print it.next() # prints "1"
  1129. print it.next(timeout=1) # prints "4" unless your computer is *very* slow
  1130. import time
  1131. result = pool.apply_async(time.sleep, (10,))
  1132. print result.get(timeout=1) # raises TimeoutError
  1133. .. _multiprocessing-listeners-clients:
  1134. Listeners and Clients
  1135. ~~~~~~~~~~~~~~~~~~~~~
  1136. .. module:: multiprocessing.connection
  1137. :synopsis: API for dealing with sockets.
  1138. Usually message passing between processes is done using queues or by using
  1139. :class:`Connection` objects returned by :func:`Pipe`.
  1140. However, the :mod:`multiprocessing.connection` module allows some extra
  1141. flexibility. It basically gives a high level message oriented API for dealing
  1142. with sockets or Windows named pipes, and also has support for *digest
  1143. authentication* using the :mod:`hmac` module.
  1144. .. function:: deliver_challenge(connection, authkey)
  1145. Send a randomly generated message to the other end of the connection and wait
  1146. for a reply.
  1147. If the reply matches the digest of the message using *authkey* as the key
  1148. then a welcome message is sent to the other end of the connection. Otherwise
  1149. :exc:`AuthenticationError` is raised.
  1150. .. function:: answerChallenge(connection, authkey)
  1151. Receive a message, calculate the digest of the message using *authkey* as the
  1152. key, and then send the digest back.
  1153. If a welcome message is not received, then :exc:`AuthenticationError` is
  1154. raised.
  1155. .. function:: Client(address[, family[, authenticate[, authkey]]])
  1156. Attempt to set up a connection to the listener which is using address
  1157. *address*, returning a :class:`~multiprocessing.Connection`.
  1158. The type of the connection is determined by *family* argument, but this can
  1159. generally be omitted since it can usually be inferred from the format of
  1160. *address*. (See :ref:`multiprocessing-address-formats`)
  1161. If *authentication* is ``True`` or *authkey* is a string then digest
  1162. authentication is used. The key used for authentication will be either
  1163. *authkey* or ``current_process().authkey)`` if *authkey* is ``None``.
  1164. If authentication fails then :exc:`AuthenticationError` is raised. See
  1165. :ref:`multiprocessing-auth-keys`.
  1166. .. class:: Listener([address[, family[, backlog[, authenticate[, authkey]]]]])
  1167. A wrapper for a bound socket or Windows named pipe which is 'listening' for
  1168. connections.
  1169. *address* is the address to be used by the bound socket or named pipe of the
  1170. listener object.
  1171. .. note::
  1172. If an address of '0.0.0.0' is used, the address will not be a connectable
  1173. end point on Windows. If you require a connectable end-point,
  1174. you should use '127.0.0.1'.
  1175. *family* is the type of socket (or named pipe) to use. This can be one of
  1176. the strings ``'AF_INET'`` (for a TCP socket), ``'AF_UNIX'`` (for a Unix
  1177. domain socket) or ``'AF_PIPE'`` (for a Windows named pipe). Of these only
  1178. the first is guaranteed to be available. If *family* is ``None`` then the
  1179. family is inferred from the format of *address*. If *address* is also
  1180. ``None`` then a default is chosen. This default is the family which is
  1181. assumed to be the fastest available. See
  1182. :ref:`multiprocessing-address-formats`. Note that if *family* is
  1183. ``'AF_UNIX'`` and address is ``None`` then the socket will be created in a
  1184. private temporary directory created using :func:`tempfile.mkstemp`.
  1185. If the listener object uses a socket then *backlog* (1 by default) is passed
  1186. to the :meth:`listen` method of the socket once it has been bound.
  1187. If *authenticate* is ``True`` (``False`` by default) or *authkey* is not
  1188. ``None`` then digest authentication is used.
  1189. If *authkey* is a string then it will be used as the authentication key;
  1190. otherwise it must be *None*.
  1191. If *authkey* is ``None`` and *authenticate* is ``True`` then
  1192. ``current_process().authkey`` is used as the authentication key. If
  1193. *authkey* is ``None`` and *authentication* is ``False`` then no
  1194. authentication is done. If authentication fails then
  1195. :exc:`AuthenticationError` is raised. See :ref:`multiprocessing-auth-keys`.
  1196. .. method:: accept()
  1197. Accept a connection on the bound socket or named pipe of the listener
  1198. object and return a :class:`Connection` object. If authentication is
  1199. attempted and fails, then :exc:`AuthenticationError` is raised.
  1200. .. method:: close()
  1201. Close the bound socket or named pipe of the listener object. This is
  1202. called automatically when the listener is garbage collected. However it
  1203. is advisable to call it explicitly.
  1204. Listener objects have the following read-only properties:
  1205. .. attribute:: address
  1206. The address which is being used by the Listener object.
  1207. .. attribute:: last_accepted
  1208. The address from which the last accepted connection came. If this is
  1209. unavailable then it is ``None``.
  1210. The module defines two exceptions:
  1211. .. exception:: AuthenticationError
  1212. Exception raised when there is an authentication error.
  1213. **Examples**
  1214. The following server code creates a listener which uses ``'secret password'`` as
  1215. an authentication key. It then waits for a connection and sends some data to
  1216. the client::
  1217. from multiprocessing.connection import Listener
  1218. from array import array
  1219. address = ('localhost', 6000) # family is deduced to be 'AF_INET'
  1220. listener = Listener(address, authkey='secret password')
  1221. conn = listener.accept()
  1222. print 'connection accepted from', listener.last_accepted
  1223. conn.send([2.25, None, 'junk', float])
  1224. conn.send_bytes('hello')
  1225. conn.send_bytes(array('i', [42, 1729]))
  1226. conn.close()
  1227. listener.close()
  1228. The following code connects to the server and receives some data from the
  1229. server::
  1230. from multiprocessing.connection import Client
  1231. from array import array
  1232. address = ('localhost', 6000)
  1233. conn = Client(address, authkey='secret password')
  1234. print conn.recv() # => [2.25, None, 'junk', float]
  1235. print conn.recv_bytes() # => 'hello'
  1236. arr = array('i', [0, 0, 0, 0, 0])
  1237. print conn.recv_bytes_into(arr) # => 8
  1238. print arr # => array('i', [42, 1729, 0, 0, 0])
  1239. conn.close()
  1240. .. _multiprocessing-address-formats:
  1241. Address Formats
  1242. >>>>>>>>>>>>>>>
  1243. * An ``'AF_INET'`` address is a tuple of the form ``(hostname, port)`` where
  1244. *hostname* is a string and *port* is an integer.
  1245. * An ``'AF_UNIX'`` address is a string representing a filename on the
  1246. filesystem.
  1247. * An ``'AF_PIPE'`` address is a string of the form
  1248. :samp:`r'\\\\.\\pipe\\{PipeName}'`. To use :func:`Client` to connect to a named
  1249. pipe on a remote computer called *ServerName* one should use an address of the
  1250. form :samp:`r'\\\\{ServerName}\\pipe\\{PipeName}'` instead.
  1251. Note that any string beginning with two backslashes is assumed by default to be
  1252. an ``'AF_PIPE'`` address rather than an ``'AF_UNIX'`` address.
  1253. .. _multiprocessing-auth-keys:
  1254. Authentication keys
  1255. ~~~~~~~~~~~~~~~~~~~
  1256. When one uses :meth:`Connection.recv`, the data received is automatically
  1257. unpickled. Unfortunately unpickling data from an untrusted source is a security
  1258. risk. Therefore :class:`Listener` and :func:`Client` use the :mod:`hmac` module
  1259. to provide digest authentication.
  1260. An authentication key is a string which can be thought of as a password: once a
  1261. connection is established both ends will demand proof that the other knows the
  1262. authentication key. (Demonstrating that both ends are using the same key does
  1263. **not** involve sending the key over the connection.)
  1264. If authentication is requested but do authentication key is specified then the
  1265. return value of ``current_process().authkey`` is used (see
  1266. :class:`~multiprocessing.Process`). This value will automatically inherited by
  1267. any :class:`~multiprocessing.Process` object that the current process creates.
  1268. This means that (by default) all processes of a multi-process program will share
  1269. a single authentication key which can be used when setting up connections
  1270. between themselves.
  1271. Suitable authentication keys can also be generated by using :func:`os.urandom`.
  1272. Logging
  1273. ~~~~~~~
  1274. Some support for logging is available. Note, however, that the :mod:`logging`
  1275. package does not use process shared locks so it is possible (depending on the
  1276. handler type) for messages from different processes to get mixed up.
  1277. .. currentmodule:: multiprocessing
  1278. .. function:: get_logger()
  1279. Returns the logger used by :mod:`multiprocessing`. If necessary, a new one
  1280. will be created.
  1281. When first created the logger has level :data:`logging.NOTSET` and no
  1282. default handler. Messages sent to this logger will not by default propagate
  1283. to the root logger.
  1284. Note that on Windows child processes will only inherit the level of the
  1285. parent process's logger -- any other customization of the logger will not be
  1286. inherited.
  1287. .. currentmodule:: multiprocessing
  1288. .. function:: log_to_stderr()
  1289. This function performs a call to :func:`get_logger` but in addition to
  1290. returning the logger created by get_logger, it adds a handler which sends
  1291. output to :data:`sys.stderr` using format
  1292. ``'[%(levelname)s/%(processName)s] %(message)s'``.
  1293. Below is an example session with logging turned on::
  1294. >>> import multiprocessing, logging
  1295. >>> logger = multiprocessing.log_to_stderr()
  1296. >>> logger.setLevel(logging.INFO)
  1297. >>> logger.warning('doomed')
  1298. [WARNING/MainProcess] doomed
  1299. >>> m = multiprocessing.Manager()
  1300. [INFO/SyncManager-...] child process calling self.run()
  1301. [INFO/SyncManager-...] created temp directory /.../pymp-...
  1302. [INFO/SyncManager-...] manager serving at '/.../listener-...'
  1303. >>> del m
  1304. [INFO/MainProcess] sending shutdown message to manager
  1305. [INFO/SyncManager-...] manager exiting with exitcode 0
  1306. In addition to having these two logging functions, the multiprocessing also
  1307. exposes two additional logging level attributes. These are :const:`SUBWARNING`
  1308. and :const:`SUBDEBUG`. The table below illustrates where theses fit in the
  1309. normal level hierarchy.
  1310. +----------------+----------------+
  1311. | Level | Numeric value |
  1312. +================+================+
  1313. | ``SUBWARNING`` | 25 |
  1314. +----------------+----------------+
  1315. | ``SUBDEBUG`` | 5 |
  1316. +----------------+----------------+
  1317. For a full table of logging levels, see the :mod:`logging` module.
  1318. These additional logging levels are used primarily for certain debug messages
  1319. within the multiprocessing module. Below is the same example as above, except
  1320. with :const:`SUBDEBUG` enabled::
  1321. >>> import multiprocessing, logging
  1322. >>> logger = multiprocessing.log_to_stderr()
  1323. >>> logger.setLevel(multiprocessing.SUBDEBUG)
  1324. >>> logger.warning('doomed')
  1325. [WARNING/MainProcess] doomed
  1326. >>> m = multiprocessing.Manager()
  1327. [INFO/SyncManager-...] child process calling self.run()
  1328. [INFO/SyncManager-...] created temp directory /.../pymp-...
  1329. [INFO/SyncManager-...] manager serving at '/.../pymp-djGBXN/listener-...'
  1330. >>> del m
  1331. [SUBDEBUG/MainProcess] finalizer calling ...
  1332. [INFO/MainProcess] sending shutdown message to manager
  1333. [DEBUG/SyncManager-...] manager received shutdown message
  1334. [SUBDEBUG/SyncManager-...] calling <Finalize object, callback=unlink, ...
  1335. [SUBDEBUG/SyncManager-...] finalizer calling <built-in function unlink> ...
  1336. [SUBDEBUG/SyncManager-...] calling <Finalize object, dead>
  1337. [SUBDEBUG/SyncManager-...] finalizer calling <function rmtree at 0x5aa730> ...
  1338. [INFO/SyncManager-...] manager exiting with exitcode 0
  1339. The :mod:`multiprocessing.dummy` module
  1340. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  1341. .. module:: multiprocessing.dummy
  1342. :synopsis: Dumb wrapper around threading.
  1343. :mod:`multiprocessing.dummy` replicates the API of :mod:`multiprocessing` but is
  1344. no more than a wrapper around the :mod:`threading` module.
  1345. .. _multiprocessing-programming:
  1346. Programming guidelines
  1347. ----------------------
  1348. There are certain guidelines and idioms which should be adhered to when using
  1349. :mod:`multiprocessing`.
  1350. All platforms
  1351. ~~~~~~~~~~~~~
  1352. Avoid shared state
  1353. As far as possible one should try to avoid shifting large amounts of data
  1354. between processes.
  1355. It is probably best to stick to using queues or pipes for communication
  1356. between processes rather than using the lower level synchronization
  1357. primitives from the :mod:`threading` module.
  1358. Picklability
  1359. Ensure that the arguments to the methods of proxies are picklable.
  1360. Thread safety of proxies
  1361. Do not use a proxy object from more than one thread unless you protect it
  1362. with a lock.
  1363. (There is never a problem with different processes using the *same* proxy.)
  1364. Joining zombie processes
  1365. On Unix when a process finishes but has not been joined it becomes a zombie.
  1366. There should never be very many because each time a new process starts (or
  1367. :func:`active_children` is called) all completed processes which have not
  1368. yet been joined will be joined. Also calling a finished process's
  1369. :meth:`Process.is_alive` will join the process. Even so it is probably good
  1370. practice to explicitly join all the processes that you start.
  1371. Better to inherit than pickle/unpickle
  1372. On Windows many types from :mod:`multiprocessing` need to be picklable so
  1373. that child processes can use them. However, one should generally avoid
  1374. sending shared objects to other processes using pipes or queues. Instead
  1375. you should arrange the program so that a process which need access to a
  1376. shared resource created elsewhere can inherit it from an ancestor process.
  1377. Avoid terminating processes
  1378. Using the :meth:`Process.terminate` method to stop a process is liable to
  1379. cause any shared resources (such as locks, semaphores, pipes and queues)
  1380. currently being used by the process to become broken or unavailable to other
  1381. processes.
  1382. Therefore it is probably best to only consider using
  1383. :meth:`Process.terminate` on processes which never use any shared resources.
  1384. Joining processes that use queues
  1385. Bear in mind that a process that has put items in a queue will wait before
  1386. terminating until all the buffered items are fed by the "feeder" thread to
  1387. the underlying pipe. (The child process can call the
  1388. :meth:`Queue.cancel_join_thread` method of the queue to avoid this behaviour.)
  1389. This means that whenever you use a queue you need to make sure that all
  1390. items which have been put on the queue will eventually be removed before the
  1391. process is joined. Otherwise you cannot be sure that processes which have
  1392. put items on the queue will terminate. Remember also that non-daemonic
  1393. processes will be automatically be joined.
  1394. An example which will deadlock is the following::
  1395. from multiprocessing import Process, Queue
  1396. def f(q):
  1397. q.put('X' * 1000000)
  1398. if __name__ == '__main__':
  1399. queue = Queue()
  1400. p = Process(target=f, args=(queue,))
  1401. p.start()
  1402. p.join() # this deadlocks
  1403. obj = queue.get()
  1404. A fix here would be to swap the last two lines round (or simply remove the
  1405. ``p.join()`` line).
  1406. Explicitly pass resources to child processes
  1407. On Unix a child process can make use of a shared resource created in a
  1408. parent process using a global resource. However, it is better to pass the
  1409. object as an argument to the constructor for the child process.
  1410. Apart from making the code (potentially) compatible with Windows this also
  1411. ensures that as long as the child process is still alive the object will not
  1412. be garbage collected in the parent process. This might be important if some
  1413. resource is freed when the object is garbage collected in the parent
  1414. process.
  1415. So for instance ::
  1416. from multiprocessing import Process, Lock
  1417. def f():
  1418. ... do something using "lock" ...
  1419. if __name__ == '__main__':
  1420. lock = Lock()
  1421. for i in range(10):
  1422. Process(target=f).start()
  1423. should be rewritten as ::
  1424. from multiprocessing import Process, Lock
  1425. def f(l):
  1426. ... do something using "l" ...
  1427. if __name__ == '__main__':
  1428. lock = Lock()
  1429. for i in range(10):
  1430. Process(target=f, args=(lock,)).start()
  1431. Beware replacing sys.stdin with a "file like object"
  1432. :mod:`multiprocessing` originally unconditionally called::
  1433. os.close(sys.stdin.fileno())
  1434. in the :meth:`multiprocessing.Process._bootstrap` method --- this resulted
  1435. in issues with processes-in-processes. This has been changed to::
  1436. sys.stdin.close()
  1437. sys.stdin = open(os.devnull)
  1438. Which solves the fundamental issue of processes colliding with each other
  1439. resulting in a bad file descriptor error, but introduces a potential danger
  1440. to applications which replace :func:`sys.stdin` with a "file-like object"
  1441. with output buffering. This danger is that if multiple processes call
  1442. :func:`close()` on this file-like object, it could result in the same
  1443. data being flushed to the object multiple times, resulting in corruption.
  1444. If you write a file-like object and implement your own caching, you can
  1445. make it fork-safe by storing the pid whenever you append to the cache,
  1446. and discarding the cache when the pid changes. For example::
  1447. @property
  1448. def cache(self):
  1449. pid = os.getpid()
  1450. if pid != self._pid:
  1451. self._pid = pid
  1452. self._cache = []
  1453. return self._cache
  1454. For more information, see :issue:`5155`, :issue:`5313` and :issue:`5331`
  1455. Windows
  1456. ~~~~~~~
  1457. Since Windows lacks :func:`os.fork` it has a few extra restrictions:
  1458. More picklability
  1459. Ensure that all arguments to :meth:`Process.__init__` are picklable. This
  1460. means, in particular, that bound or unbound methods cannot be used directly
  1461. as the ``target`` argument on Windows --- just define a function and use
  1462. that instead.
  1463. Also, if you subclass :class:`Process` then make sure that instances will be
  1464. picklable when the :meth:`Process.start` method is called.
  1465. Global variables
  1466. Bear in mind that if code run in a child process tries to access a global
  1467. variable, then the value it sees (if any) may not be the same as the value
  1468. in the parent process at the time that :meth:`Process.start` was called.
  1469. However, global variables which are just module level constants cause no
  1470. problems.
  1471. Safe importing of main module
  1472. Make sure that the main module can be safely imported by a new Python
  1473. interpreter without causing unintended side effects (such a starting a new
  1474. process).
  1475. For example, under Windows running the following module would fail with a
  1476. :exc:`RuntimeError`::
  1477. from multiprocessing import Process
  1478. def foo():
  1479. print 'hello'
  1480. p = Process(target=foo)
  1481. p.start()
  1482. Instead one should protect the "entry point" of the program by using ``if
  1483. __name__ == '__main__':`` as follows::
  1484. from multiprocessing import Process, freeze_support
  1485. def foo():
  1486. print 'hello'
  1487. if __name__ == '__main__':
  1488. freeze_support()
  1489. p = Process(target=foo)
  1490. p.start()
  1491. (The ``freeze_support()`` line can be omitted if the program will be run
  1492. normally instead of frozen.)
  1493. This allows the newly spawned Python interpreter to safely import the module
  1494. and then run the module's ``foo()`` function.
  1495. Similar restrictions apply if a pool or manager is created in the main
  1496. module.
  1497. .. _multiprocessing-examples:
  1498. Examples
  1499. --------
  1500. Demonstration of how to create and use customized managers and proxies:
  1501. .. literalinclude:: ../includes/mp_newtype.py
  1502. Using :class:`Pool`:
  1503. .. literalinclude:: ../includes/mp_pool.py
  1504. Synchronization types like locks, conditions and queues:
  1505. .. literalinclude:: ../includes/mp_synchronize.py
  1506. An showing how to use queues to feed tasks to a collection of worker process and
  1507. collect the results:
  1508. .. literalinclude:: ../includes/mp_workers.py
  1509. An example of how a pool of worker processes can each run a
  1510. :class:`SimpleHTTPServer.HttpServer` instance while sharing a single listening
  1511. socket.
  1512. .. literalinclude:: ../includes/mp_webserver.py
  1513. Some simple benchmarks comparing :mod:`multiprocessing` with :mod:`threading`:
  1514. .. literalinclude:: ../includes/mp_benchmarks.py
  1515. An example/demo of how to use the :class:`managers.SyncManager`, :class:`Process`
  1516. and others to build a system which can distribute processes and work via a
  1517. distributed queue to a "cluster" of machines on a network, accessible via SSH.
  1518. You will need to have private key authentication for all hosts configured for
  1519. this to work.
  1520. .. literalinclude:: ../includes/mp_distributing.py