1
:mod:`multiprocessing` --- Process-based "threading" interface
2
==============================================================
4
.. module:: multiprocessing
5
:synopsis: Process-based "threading" interface.
11
----------------------
13
:mod:`multiprocessing` is a package that supports spawning processes using an
14
API similar to the :mod:`threading` module. The :mod:`multiprocessing` package
15
offers both local and remote concurrency, effectively side-stepping the
16
:term:`Global Interpreter Lock` by using subprocesses instead of threads. Due
17
to this, the :mod:`multiprocessing` module allows the programmer to fully
18
leverage multiple processors on a given machine. It runs on both Unix and
22
The :class:`Process` class
23
~~~~~~~~~~~~~~~~~~~~~~~~~~
25
In :mod:`multiprocessing`, processes are spawned by creating a :class:`Process`
26
object and then calling its :meth:`~Process.start` method. :class:`Process`
27
follows the API of :class:`threading.Thread`. A trivial example of a
28
multiprocess program is ::
30
from multiprocessing import Process
35
if __name__ == '__main__':
36
p = Process(target=f, args=('bob',))
40
Here the function ``f`` is run in a child process.
42
For an explanation of why (on Windows) the ``if __name__ == '__main__'`` part is
43
necessary, see :ref:`multiprocessing-programming`.
47
Exchanging objects between processes
48
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
50
:mod:`multiprocessing` supports two types of communication channel between
55
The :class:`Queue` class is a near clone of :class:`Queue.Queue`. For
58
from multiprocessing import Process, Queue
61
q.put([42, None, 'hello'])
63
if __name__ == '__main__':
65
p = Process(target=f, args=(q,))
67
print q.get() # prints "[42, None, 'hello']"
70
Queues are thread and process safe.
74
The :func:`Pipe` function returns a pair of connection objects connected by a
75
pipe which by default is duplex (two-way). For example::
77
from multiprocessing import Process, Pipe
80
conn.send([42, None, 'hello'])
83
if __name__ == '__main__':
84
parent_conn, child_conn = Pipe()
85
p = Process(target=f, args=(child_conn,))
87
print parent_conn.recv() # prints "[42, None, 'hello']"
90
The two connection objects returned by :func:`Pipe` represent the two ends of
91
the pipe. Each connection object has :meth:`~Connection.send` and
92
:meth:`~Connection.recv` methods (among others). Note that data in a pipe
93
may become corrupted if two processes (or threads) try to read from or write
94
to the *same* end of the pipe at the same time. Of course there is no risk
95
of corruption from processes using different ends of the pipe at the same
99
Synchronization between processes
100
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
102
:mod:`multiprocessing` contains equivalents of all the synchronization
103
primitives from :mod:`threading`. For instance one can use a lock to ensure
104
that only one process prints to standard output at a time::
106
from multiprocessing import Process, Lock
110
print 'hello world', i
113
if __name__ == '__main__':
116
for num in range(10):
117
Process(target=f, args=(lock, num)).start()
119
Without using the lock output from the different processes is liable to get all
123
Sharing state between processes
124
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
126
As mentioned above, when doing concurrent programming it is usually best to
127
avoid using shared state as far as possible. This is particularly true when
128
using multiple processes.
130
However, if you really do need to use some shared data then
131
:mod:`multiprocessing` provides a couple of ways of doing so.
135
Data can be stored in a shared memory map using :class:`Value` or
136
:class:`Array`. For example, the following code ::
138
from multiprocessing import Process, Value, Array
142
for i in range(len(a)):
145
if __name__ == '__main__':
146
num = Value('d', 0.0)
147
arr = Array('i', range(10))
149
p = Process(target=f, args=(num, arr))
159
[0, -1, -2, -3, -4, -5, -6, -7, -8, -9]
161
The ``'d'`` and ``'i'`` arguments used when creating ``num`` and ``arr`` are
162
typecodes of the kind used by the :mod:`array` module: ``'d'`` indicates a
163
double precision float and ``'i'`` indicates a signed integer. These shared
164
objects will be process and thread safe.
166
For more flexibility in using shared memory one can use the
167
:mod:`multiprocessing.sharedctypes` module which supports the creation of
168
arbitrary ctypes objects allocated from shared memory.
172
A manager object returned by :func:`Manager` controls a server process which
173
holds Python objects and allows other processes to manipulate them using
176
A manager returned by :func:`Manager` will support types :class:`list`,
177
:class:`dict`, :class:`Namespace`, :class:`Lock`, :class:`RLock`,
178
:class:`Semaphore`, :class:`BoundedSemaphore`, :class:`Condition`,
179
:class:`Event`, :class:`Queue`, :class:`Value` and :class:`Array`. For
182
from multiprocessing import Process, Manager
190
if __name__ == '__main__':
194
l = manager.list(range(10))
196
p = Process(target=f, args=(d, l))
205
{0.25: None, 1: '1', '2': 2}
206
[9, 8, 7, 6, 5, 4, 3, 2, 1, 0]
208
Server process managers are more flexible than using shared memory objects
209
because they can be made to support arbitrary object types. Also, a single
210
manager can be shared by processes on different computers over a network.
211
They are, however, slower than using shared memory.
214
Using a pool of workers
215
~~~~~~~~~~~~~~~~~~~~~~~
217
The :class:`~multiprocessing.pool.Pool` class represents a pool of worker
218
processes. It has methods which allows tasks to be offloaded to the worker
219
processes in a few different ways.
223
from multiprocessing import Pool
228
if __name__ == '__main__':
229
pool = Pool(processes=4) # start 4 worker processes
230
result = pool.applyAsync(f, [10]) # evaluate "f(10)" asynchronously
231
print result.get(timeout=1) # prints "100" unless your computer is *very* slow
232
print pool.map(f, range(10)) # prints "[0, 1, 4,..., 81]"
238
The :mod:`multiprocessing` package mostly replicates the API of the
239
:mod:`threading` module.
242
:class:`Process` and exceptions
243
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
245
.. class:: Process([group[, target[, name[, args[, kwargs]]]]])
247
Process objects represent activity that is run in a separate process. The
248
:class:`Process` class has equivalents of all the methods of
249
:class:`threading.Thread`.
251
The constructor should always be called with keyword arguments. *group*
252
should always be ``None``; it exists solely for compatibility with
253
:class:`threading.Thread`. *target* is the callable object to be invoked by
254
the :meth:`run()` method. It defaults to ``None``, meaning nothing is
255
called. *name* is the process name. By default, a unique name is constructed
256
of the form 'Process-N\ :sub:`1`:N\ :sub:`2`:...:N\ :sub:`k`' where N\
257
:sub:`1`,N\ :sub:`2`,...,N\ :sub:`k` is a sequence of integers whose length
258
is determined by the *generation* of the process. *args* is the argument
259
tuple for the target invocation. *kwargs* is a dictionary of keyword
260
arguments for the target invocation. By default, no arguments are passed to
263
If a subclass overrides the constructor, it must make sure it invokes the
264
base class constructor (:meth:`Process.__init__`) before doing anything else
269
Method representing the process's activity.
271
You may override this method in a subclass. The standard :meth:`run`
272
method invokes the callable object passed to the object's constructor as
273
the target argument, if any, with sequential and keyword arguments taken
274
from the *args* and *kwargs* arguments, respectively.
278
Start the process's activity.
280
This must be called at most once per process object. It arranges for the
281
object's :meth:`run` method to be invoked in a separate process.
283
.. method:: join([timeout])
285
Block the calling thread until the process whose :meth:`join` method is
286
called terminates or until the optional timeout occurs.
288
If *timeout* is ``None`` then there is no timeout.
290
A process can be joined many times.
292
A process cannot join itself because this would cause a deadlock. It is
293
an error to attempt to join a process before it has been started.
299
The name is a string used for identification purposes only. It has no
300
semantics. Multiple processes may be given the same name. The initial
301
name is set by the constructor.
303
.. method:: is_alive()
305
Return whether the process is alive.
307
Roughly, a process object is alive from the moment the :meth:`start`
308
method returns until the child process terminates.
310
.. attribute:: daemon
312
The process's daemon flag, a Boolean value. This must be called before
313
:meth:`start` is called.
315
The initial value is inherited from the creating process.
317
When a process exits, it attempts to terminate all of its daemonic child
320
Note that a daemonic process is not allowed to create child processes.
321
Otherwise a daemonic process would leave its children orphaned if it gets
322
terminated when its parent process exits.
324
In addition to the :class:`Threading.Thread` API, :class:`Process` objects
325
also support the following attributes and methods:
329
Return the process ID. Before the process is spawned, this will be
332
.. attribute:: exitcode
334
The child's exit code. This will be ``None`` if the process has not yet
335
terminated. A negative value *-N* indicates that the child was terminated
338
.. attribute:: authkey
340
The process's authentication key (a byte string).
342
When :mod:`multiprocessing` is initialized the main process is assigned a
343
random string using :func:`os.random`.
345
When a :class:`Process` object is created, it will inherit the
346
authentication key of its parent process, although this may be changed by
347
setting :attr:`authkey` to another byte string.
349
See :ref:`multiprocessing-auth-keys`.
351
.. method:: terminate()
353
Terminate the process. On Unix this is done using the ``SIGTERM`` signal;
354
on Windows :cfunc:`TerminateProcess` is used. Note that exit handlers and
355
finally clauses, etc., will not be executed.
357
Note that descendant processes of the process will *not* be terminated --
358
they will simply become orphaned.
362
If this method is used when the associated process is using a pipe or
363
queue then the pipe or queue is liable to become corrupted and may
364
become unusable by other process. Similarly, if the process has
365
acquired a lock or semaphore etc. then terminating it is liable to
366
cause other processes to deadlock.
368
Note that the :meth:`start`, :meth:`join`, :meth:`is_alive` and
369
:attr:`exit_code` methods should only be called by the process that created
372
Example usage of some of the methods of :class:`Process`::
374
>>> import processing, time, signal
375
>>> p = processing.Process(target=time.sleep, args=(1000,))
376
>>> print p, p.is_alive()
377
<Process(Process-1, initial)> False
379
>>> print p, p.is_alive()
380
<Process(Process-1, started)> True
382
>>> print p, p.is_alive()
383
<Process(Process-1, stopped[SIGTERM])> False
384
>>> p.exitcode == -signal.SIGTERM
388
.. exception:: BufferTooShort
390
Exception raised by :meth:`Connection.recv_bytes_into()` when the supplied
391
buffer object is too small for the message read.
393
If ``e`` is an instance of :exc:`BufferTooShort` then ``e.args[0]`` will give
394
the message as a byte string.
400
When using multiple processes, one generally uses message passing for
401
communication between processes and avoids having to use any synchronization
402
primitives like locks.
404
For passing messages one can use :func:`Pipe` (for a connection between two
405
processes) or a queue (which allows multiple producers and consumers).
407
The :class:`Queue` and :class:`JoinableQueue` types are multi-producer,
408
multi-consumer FIFO queues modelled on the :class:`Queue.Queue` class in the
409
standard library. They differ in that :class:`Queue` lacks the
410
:meth:`~Queue.Queue.task_done` and :meth:`~Queue.Queue.join` methods introduced
411
into Python 2.5's :class:`Queue.Queue` class.
413
If you use :class:`JoinableQueue` then you **must** call
414
:meth:`JoinableQueue.task_done` for each task removed from the queue or else the
415
semaphore used to count the number of unfinished tasks may eventually overflow
416
raising an exception.
418
Note that one can also create a shared queue by using a manager object -- see
419
:ref:`multiprocessing-managers`.
423
:mod:`multiprocessing` uses the usual :exc:`Queue.Empty` and
424
:exc:`Queue.Full` exceptions to signal a timeout. They are not available in
425
the :mod:`multiprocessing` namespace so you need to import them from
431
If a process is killed using :meth:`Process.terminate` or :func:`os.kill`
432
while it is trying to use a :class:`Queue`, then the data in the queue is
433
likely to become corrupted. This may cause any other processes to get an
434
exception when it tries to use the queue later on.
438
As mentioned above, if a child process has put items on a queue (and it has
439
not used :meth:`JoinableQueue.cancel_join_thread`), then that process will
440
not terminate until all buffered items have been flushed to the pipe.
442
This means that if you try joining that process you may get a deadlock unless
443
you are sure that all items which have been put on the queue have been
444
consumed. Similarly, if the child process is non-daemonic then the parent
445
process may hang on exit when it tries to join all its non-daemonic children.
447
Note that a queue created using a manager does not have this issue. See
448
:ref:`multiprocessing-programming`.
450
For an example of the usage of queues for interprocess communication see
451
:ref:`multiprocessing-examples`.
454
.. function:: Pipe([duplex])
456
Returns a pair ``(conn1, conn2)`` of :class:`Connection` objects representing
459
If *duplex* is ``True`` (the default) then the pipe is bidirectional. If
460
*duplex* is ``False`` then the pipe is unidirectional: ``conn1`` can only be
461
used for receiving messages and ``conn2`` can only be used for sending
465
.. class:: Queue([maxsize])
467
Returns a process shared queue implemented using a pipe and a few
468
locks/semaphores. When a process first puts an item on the queue a feeder
469
thread is started which transfers objects from a buffer into the pipe.
471
The usual :exc:`Queue.Empty` and :exc:`Queue.Full` exceptions from the
472
standard library's :mod:`Queue` module are raised to signal timeouts.
474
:class:`Queue` implements all the methods of :class:`Queue.Queue` except for
475
:meth:`~Queue.Queue.task_done` and :meth:`~Queue.Queue.join`.
479
Return the approximate size of the queue. Because of
480
multithreading/multiprocessing semantics, this number is not reliable.
482
Note that this may raise :exc:`NotImplementedError` on Unix platforms like
483
Mac OS X where ``sem_getvalue()`` is not implemented.
487
Return ``True`` if the queue is empty, ``False`` otherwise. Because of
488
multithreading/multiprocessing semantics, this is not reliable.
492
Return ``True`` if the queue is full, ``False`` otherwise. Because of
493
multithreading/multiprocessing semantics, this is not reliable.
495
.. method:: put(item[, block[, timeout]])
497
Put item into the queue. If the optional argument *block* is ``True``
498
(the default) and *timeout* is ``None`` (the default), block if necessary until
499
a free slot is available. If *timeout* is a positive number, it blocks at
500
most *timeout* seconds and raises the :exc:`Queue.Full` exception if no
501
free slot was available within that time. Otherwise (*block* is
502
``False``), put an item on the queue if a free slot is immediately
503
available, else raise the :exc:`Queue.Full` exception (*timeout* is
504
ignored in that case).
506
.. method:: put_nowait(item)
508
Equivalent to ``put(item, False)``.
510
.. method:: get([block[, timeout]])
512
Remove and return an item from the queue. If optional args *block* is
513
``True`` (the default) and *timeout* is ``None`` (the default), block if
514
necessary until an item is available. If *timeout* is a positive number,
515
it blocks at most *timeout* seconds and raises the :exc:`Queue.Empty`
516
exception if no item was available within that time. Otherwise (block is
517
``False``), return an item if one is immediately available, else raise the
518
:exc:`Queue.Empty` exception (*timeout* is ignored in that case).
520
.. method:: get_nowait()
523
Equivalent to ``get(False)``.
525
:class:`multiprocessing.Queue` has a few additional methods not found in
526
:class:`Queue.Queue`. These methods are usually unnecessary for most
531
Indicate that no more data will be put on this queue by the current
532
process. The background thread will quit once it has flushed all buffered
533
data to the pipe. This is called automatically when the queue is garbage
536
.. method:: join_thread()
538
Join the background thread. This can only be used after :meth:`close` has
539
been called. It blocks until the background thread exits, ensuring that
540
all data in the buffer has been flushed to the pipe.
542
By default if a process is not the creator of the queue then on exit it
543
will attempt to join the queue's background thread. The process can call
544
:meth:`cancel_join_thread` to make :meth:`join_thread` do nothing.
546
.. method:: cancel_join_thread()
548
Prevent :meth:`join_thread` from blocking. In particular, this prevents
549
the background thread from being joined automatically when the process
550
exits -- see :meth:`join_thread`.
553
.. class:: JoinableQueue([maxsize])
555
:class:`JoinableQueue`, a :class:`Queue` subclass, is a queue which
556
additionally has :meth:`task_done` and :meth:`join` methods.
558
.. method:: task_done()
560
Indicate that a formerly enqueued task is complete. Used by queue consumer
561
threads. For each :meth:`~Queue.get` used to fetch a task, a subsequent
562
call to :meth:`task_done` tells the queue that the processing on the task
565
If a :meth:`~Queue.join` is currently blocking, it will resume when all
566
items have been processed (meaning that a :meth:`task_done` call was
567
received for every item that had been :meth:`~Queue.put` into the queue).
569
Raises a :exc:`ValueError` if called more times than there were items
575
Block until all items in the queue have been gotten and processed.
577
The count of unfinished tasks goes up whenever an item is added to the
578
queue. The count goes down whenever a consumer thread calls
579
:meth:`task_done` to indicate that the item was retrieved and all work on
580
it is complete. When the count of unfinished tasks drops to zero,
581
:meth:`~Queue.join` unblocks.
587
.. function:: active_children()
589
Return list of all live children of the current process.
591
Calling this has the side affect of "joining" any processes which have
594
.. function:: cpu_count()
596
Return the number of CPUs in the system. May raise
597
:exc:`NotImplementedError`.
599
.. function:: current_process()
601
Return the :class:`Process` object corresponding to the current process.
603
An analogue of :func:`threading.current_thread`.
605
.. function:: freeze_support()
607
Add support for when a program which uses :mod:`multiprocessing` has been
608
frozen to produce a Windows executable. (Has been tested with **py2exe**,
609
**PyInstaller** and **cx_Freeze**.)
611
One needs to call this function straight after the ``if __name__ ==
612
'__main__'`` line of the main module. For example::
614
from multiprocessing import Process, freeze_support
619
if __name__ == '__main__':
621
Process(target=f).start()
623
If the ``freeze_support()`` line is missed out then trying to run the frozen
624
executable will raise :exc:`RuntimeError`.
626
If the module is being run normally by the Python interpreter then
627
:func:`freeze_support` has no effect.
629
.. function:: set_executable()
631
Sets the path of the python interpreter to use when starting a child process.
632
(By default :data:`sys.executable` is used). Embedders will probably need to
633
do some thing like ::
635
setExecutable(os.path.join(sys.exec_prefix, 'pythonw.exe'))
637
before they can create child processes. (Windows only)
642
:mod:`multiprocessing` contains no analogues of
643
:func:`threading.active_count`, :func:`threading.enumerate`,
644
:func:`threading.settrace`, :func:`threading.setprofile`,
645
:class:`threading.Timer`, or :class:`threading.local`.
651
Connection objects allow the sending and receiving of picklable objects or
652
strings. They can be thought of as message oriented connected sockets.
654
Connection objects usually created using :func:`Pipe` -- see also
655
:ref:`multiprocessing-listeners-clients`.
657
.. class:: Connection
659
.. method:: send(obj)
661
Send an object to the other end of the connection which should be read
664
The object must be picklable.
668
Return an object sent from the other end of the connection using
669
:meth:`send`. Raises :exc:`EOFError` if there is nothing left to receive
670
and the other end was closed.
674
Returns the file descriptor or handle used by the connection.
678
Close the connection.
680
This is called automatically when the connection is garbage collected.
682
.. method:: poll([timeout])
684
Return whether there is any data available to be read.
686
If *timeout* is not specified then it will return immediately. If
687
*timeout* is a number then this specifies the maximum time in seconds to
688
block. If *timeout* is ``None`` then an infinite timeout is used.
690
.. method:: send_bytes(buffer[, offset[, size]])
692
Send byte data from an object supporting the buffer interface as a
695
If *offset* is given then data is read from that position in *buffer*. If
696
*size* is given then that many bytes will be read from buffer.
698
.. method:: recv_bytes([maxlength])
700
Return a complete message of byte data sent from the other end of the
701
connection as a string. Raises :exc:`EOFError` if there is nothing left
702
to receive and the other end has closed.
704
If *maxlength* is specified and the message is longer than *maxlength*
705
then :exc:`IOError` is raised and the connection will no longer be
708
.. method:: recv_bytes_into(buffer[, offset])
710
Read into *buffer* a complete message of byte data sent from the other end
711
of the connection and return the number of bytes in the message. Raises
712
:exc:`EOFError` if there is nothing left to receive and the other end was
715
*buffer* must be an object satisfying the writable buffer interface. If
716
*offset* is given then the message will be written into the buffer from
717
*that position. Offset must be a non-negative integer less than the
718
*length of *buffer* (in bytes).
720
If the buffer is too short then a :exc:`BufferTooShort` exception is
721
raised and the complete message is available as ``e.args[0]`` where ``e``
722
is the exception instance.
727
>>> from multiprocessing import Pipe
729
>>> a.send([1, 'hello', None])
732
>>> b.send_bytes('thank you')
736
>>> arr1 = array.array('i', range(5))
737
>>> arr2 = array.array('i', [0] * 10)
738
>>> a.send_bytes(arr1)
739
>>> count = b.recv_bytes_into(arr2)
740
>>> assert count == len(arr1) * arr1.itemsize
742
array('i', [0, 1, 2, 3, 4, 0, 0, 0, 0, 0])
747
The :meth:`Connection.recv` method automatically unpickles the data it
748
receives, which can be a security risk unless you can trust the process
749
which sent the message.
751
Therefore, unless the connection object was produced using :func:`Pipe` you
752
should only use the :meth:`~Connection.recv` and :meth:`~Connection.send`
753
methods after performing some sort of authentication. See
754
:ref:`multiprocessing-auth-keys`.
758
If a process is killed while it is trying to read or write to a pipe then
759
the data in the pipe is likely to become corrupted, because it may become
760
impossible to be sure where the message boundaries lie.
763
Synchronization primitives
764
~~~~~~~~~~~~~~~~~~~~~~~~~~
766
Generally synchronization primitives are not as necessary in a multiprocess
767
program as they are in a multithreaded program. See the documentation for
768
:mod:`threading` module.
770
Note that one can also create synchronization primitives by using a manager
771
object -- see :ref:`multiprocessing-managers`.
773
.. class:: BoundedSemaphore([value])
775
A bounded semaphore object: a clone of :class:`threading.BoundedSemaphore`.
777
(On Mac OS X this is indistinguishable from :class:`Semaphore` because
778
``sem_getvalue()`` is not implemented on that platform).
780
.. class:: Condition([lock])
782
A condition variable: a clone of :class:`threading.Condition`.
784
If *lock* is specified then it should be a :class:`Lock` or :class:`RLock`
785
object from :mod:`multiprocessing`.
789
A clone of :class:`threading.Event`.
793
A non-recursive lock object: a clone of :class:`threading.Lock`.
797
A recursive lock object: a clone of :class:`threading.RLock`.
799
.. class:: Semaphore([value])
801
A bounded semaphore object: a clone of :class:`threading.Semaphore`.
805
The :meth:`acquire` method of :class:`BoundedSemaphore`, :class:`Lock`,
806
:class:`RLock` and :class:`Semaphore` has a timeout parameter not supported
807
by the equivalents in :mod:`threading`. The signature is
808
``acquire(block=True, timeout=None)`` with keyword parameters being
809
acceptable. If *block* is ``True`` and *timeout* is not ``None`` then it
810
specifies a timeout in seconds. If *block* is ``False`` then *timeout* is
815
If the SIGINT signal generated by Ctrl-C arrives while the main thread is
816
blocked by a call to :meth:`BoundedSemaphore.acquire`, :meth:`Lock.acquire`,
817
:meth:`RLock.acquire`, :meth:`Semaphore.acquire`, :meth:`Condition.acquire`
818
or :meth:`Condition.wait` then the call will be immediately interrupted and
819
:exc:`KeyboardInterrupt` will be raised.
821
This differs from the behaviour of :mod:`threading` where SIGINT will be
822
ignored while the equivalent blocking calls are in progress.
825
Shared :mod:`ctypes` Objects
826
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
828
It is possible to create shared objects using shared memory which can be
829
inherited by child processes.
831
.. function:: Value(typecode_or_type[, *args, lock]])
833
Return a :mod:`ctypes` object allocated from shared memory. By default the
834
return value is actually a synchronized wrapper for the object.
836
*typecode_or_type* determines the type of the returned object: it is either a
837
ctypes type or a one character typecode of the kind used by the :mod:`array`
838
module. *\*args* is passed on to the constructor for the type.
840
If *lock* is ``True`` (the default) then a new lock object is created to
841
synchronize access to the value. If *lock* is a :class:`Lock` or
842
:class:`RLock` object then that will be used to synchronize access to the
843
value. If *lock* is ``False`` then access to the returned object will not be
844
automatically protected by a lock, so it will not necessarily be
847
Note that *lock* is a keyword-only argument.
849
.. function:: Array(typecode_or_type, size_or_initializer, *, lock=True)
851
Return a ctypes array allocated from shared memory. By default the return
852
value is actually a synchronized wrapper for the array.
854
*typecode_or_type* determines the type of the elements of the returned array:
855
it is either a ctypes type or a one character typecode of the kind used by
856
the :mod:`array` module. If *size_or_initializer* is an integer, then it
857
determines the length of the array, and the array will be initially zeroed.
858
Otherwise, *size_or_initializer* is a sequence which is used to initialize
859
the array and whose length determines the length of the array.
861
If *lock* is ``True`` (the default) then a new lock object is created to
862
synchronize access to the value. If *lock* is a :class:`Lock` or
863
:class:`RLock` object then that will be used to synchronize access to the
864
value. If *lock* is ``False`` then access to the returned object will not be
865
automatically protected by a lock, so it will not necessarily be
868
Note that *lock* is a keyword only argument.
870
Note that an array of :data:`ctypes.c_char` has *value* and *rawvalue*
871
attributes which allow one to use it to store and retrieve strings.
874
The :mod:`multiprocessing.sharedctypes` module
875
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
877
.. module:: multiprocessing.sharedctypes
878
:synopsis: Allocate ctypes objects from shared memory.
880
The :mod:`multiprocessing.sharedctypes` module provides functions for allocating
881
:mod:`ctypes` objects from shared memory which can be inherited by child
886
Although it is possible to store a pointer in shared memory remember that
887
this will refer to a location in the address space of a specific process.
888
However, the pointer is quite likely to be invalid in the context of a second
889
process and trying to dereference the pointer from the second process may
892
.. function:: RawArray(typecode_or_type, size_or_initializer)
894
Return a ctypes array allocated from shared memory.
896
*typecode_or_type* determines the type of the elements of the returned array:
897
it is either a ctypes type or a one character typecode of the kind used by
898
the :mod:`array` module. If *size_or_initializer* is an integer then it
899
determines the length of the array, and the array will be initially zeroed.
900
Otherwise *size_or_initializer* is a sequence which is used to initialize the
901
array and whose length determines the length of the array.
903
Note that setting and getting an element is potentially non-atomic -- use
904
:func:`Array` instead to make sure that access is automatically synchronized
907
.. function:: RawValue(typecode_or_type, *args)
909
Return a ctypes object allocated from shared memory.
911
*typecode_or_type* determines the type of the returned object: it is either a
912
ctypes type or a one character typecode of the kind used by the :mod:`array`
913
module. */*args* is passed on to the constructor for the type.
915
Note that setting and getting the value is potentially non-atomic -- use
916
:func:`Value` instead to make sure that access is automatically synchronized
919
Note that an array of :data:`ctypes.c_char` has ``value`` and ``rawvalue``
920
attributes which allow one to use it to store and retrieve strings -- see
921
documentation for :mod:`ctypes`.
923
.. function:: Array(typecode_or_type, size_or_initializer[, *args[, lock]])
925
The same as :func:`RawArray` except that depending on the value of *lock* a
926
process-safe synchronization wrapper may be returned instead of a raw ctypes
929
If *lock* is ``True`` (the default) then a new lock object is created to
930
synchronize access to the value. If *lock* is a :class:`Lock` or
931
:class:`RLock` object then that will be used to synchronize access to the
932
value. If *lock* is ``False`` then access to the returned object will not be
933
automatically protected by a lock, so it will not necessarily be
936
Note that *lock* is a keyword-only argument.
938
.. function:: Value(typecode_or_type, *args[, lock])
940
The same as :func:`RawValue` except that depending on the value of *lock* a
941
process-safe synchronization wrapper may be returned instead of a raw ctypes
944
If *lock* is ``True`` (the default) then a new lock object is created to
945
synchronize access to the value. If *lock* is a :class:`Lock` or
946
:class:`RLock` object then that will be used to synchronize access to the
947
value. If *lock* is ``False`` then access to the returned object will not be
948
automatically protected by a lock, so it will not necessarily be
951
Note that *lock* is a keyword-only argument.
953
.. function:: copy(obj)
955
Return a ctypes object allocated from shared memory which is a copy of the
958
.. function:: synchronized(obj[, lock])
960
Return a process-safe wrapper object for a ctypes object which uses *lock* to
961
synchronize access. If *lock* is ``None`` (the default) then a
962
:class:`multiprocessing.RLock` object is created automatically.
964
A synchronized wrapper will have two methods in addition to those of the
965
object it wraps: :meth:`get_obj` returns the wrapped object and
966
:meth:`get_lock` returns the lock object used for synchronization.
968
Note that accessing the ctypes object through the wrapper can be a lot slower
969
than accessing the raw ctypes object.
972
The table below compares the syntax for creating shared ctypes objects from
973
shared memory with the normal ctypes syntax. (In the table ``MyStruct`` is some
974
subclass of :class:`ctypes.Structure`.)
976
==================== ========================== ===========================
977
ctypes sharedctypes using type sharedctypes using typecode
978
==================== ========================== ===========================
979
c_double(2.4) RawValue(c_double, 2.4) RawValue('d', 2.4)
980
MyStruct(4, 6) RawValue(MyStruct, 4, 6)
981
(c_short * 7)() RawArray(c_short, 7) RawArray('h', 7)
982
(c_int * 3)(9, 2, 8) RawArray(c_int, (9, 2, 8)) RawArray('i', (9, 2, 8))
983
==================== ========================== ===========================
986
Below is an example where a number of ctypes objects are modified by a child
989
from multiprocessing import Process, Lock
990
from multiprocessing.sharedctypes import Value, Array
991
from ctypes import Structure, c_double
993
class Point(Structure):
994
_fields_ = [('x', c_double), ('y', c_double)]
996
def modify(n, x, s, A):
999
s.value = s.value.upper()
1004
if __name__ == '__main__':
1008
x = Value(ctypes.c_double, 1.0/3.0, lock=False)
1009
s = Array('c', 'hello world', lock=lock)
1010
A = Array(Point, [(1.875,-6.25), (-5.75,2.0), (2.375,9.5)], lock=lock)
1012
p = Process(target=modify, args=(n, x, s, A))
1019
print [(a.x, a.y) for a in A]
1022
.. highlightlang:: none
1024
The results printed are ::
1029
[(3.515625, 39.0625), (33.0625, 4.0), (5.640625, 90.25)]
1031
.. highlightlang:: python
1034
.. _multiprocessing-managers:
1039
Managers provide a way to create data which can be shared between different
1040
processes. A manager object controls a server process which manages *shared
1041
objects*. Other processes can access the shared objects by using proxies.
1043
.. function:: multiprocessing.Manager()
1045
Returns a started :class:`~multiprocessing.managers.SyncManager` object which
1046
can be used for sharing objects between processes. The returned manager
1047
object corresponds to a spawned child process and has methods which will
1048
create shared objects and return corresponding proxies.
1050
.. module:: multiprocessing.managers
1051
:synopsis: Share data between process with shared objects.
1053
Manager processes will be shutdown as soon as they are garbage collected or
1054
their parent process exits. The manager classes are defined in the
1055
:mod:`multiprocessing.managers` module:
1057
.. class:: BaseManager([address[, authkey]])
1059
Create a BaseManager object.
1061
Once created one should call :meth:`start` or :meth:`serve_forever` to ensure
1062
that the manager object refers to a started manager process.
1064
*address* is the address on which the manager process listens for new
1065
connections. If *address* is ``None`` then an arbitrary one is chosen.
1067
*authkey* is the authentication key which will be used to check the validity
1068
of incoming connections to the server process. If *authkey* is ``None`` then
1069
``current_process().authkey``. Otherwise *authkey* is used and it
1074
Start a subprocess to start the manager.
1076
.. method:: serve_forever()
1078
Run the server in the current process.
1080
.. method:: from_address(address, authkey)
1082
A class method which creates a manager object referring to a pre-existing
1083
server process which is using the given address and authentication key.
1085
.. method:: shutdown()
1087
Stop the process used by the manager. This is only available if
1088
:meth:`start` has been used to start the server process.
1090
This can be called multiple times.
1092
.. method:: register(typeid[, callable[, proxytype[, exposed[, method_to_typeid[, create_method]]]]])
1094
A classmethod which can be used for registering a type or callable with
1097
*typeid* is a "type identifier" which is used to identify a particular
1098
type of shared object. This must be a string.
1100
*callable* is a callable used for creating objects for this type
1101
identifier. If a manager instance will be created using the
1102
:meth:`from_address` classmethod or if the *create_method* argument is
1103
``False`` then this can be left as ``None``.
1105
*proxytype* is a subclass of :class:`BaseProxy` which is used to create
1106
proxies for shared objects with this *typeid*. If ``None`` then a proxy
1107
class is created automatically.
1109
*exposed* is used to specify a sequence of method names which proxies for
1110
this typeid should be allowed to access using
1111
:meth:`BaseProxy._callMethod`. (If *exposed* is ``None`` then
1112
:attr:`proxytype._exposed_` is used instead if it exists.) In the case
1113
where no exposed list is specified, all "public methods" of the shared
1114
object will be accessible. (Here a "public method" means any attribute
1115
which has a :meth:`__call__` method and whose name does not begin with
1118
*method_to_typeid* is a mapping used to specify the return type of those
1119
exposed methods which should return a proxy. It maps method names to
1120
typeid strings. (If *method_to_typeid* is ``None`` then
1121
:attr:`proxytype._method_to_typeid_` is used instead if it exists.) If a
1122
method's name is not a key of this mapping or if the mapping is ``None``
1123
then the object returned by the method will be copied by value.
1125
*create_method* determines whether a method should be created with name
1126
*typeid* which can be used to tell the server process to create a new
1127
shared object and return a proxy for it. By default it is ``True``.
1129
:class:`BaseManager` instances also have one read-only property:
1131
.. attribute:: address
1133
The address used by the manager.
1136
.. class:: SyncManager
1138
A subclass of :class:`BaseManager` which can be used for the synchronization
1139
of processes. Objects of this type are returned by
1140
:func:`multiprocessing.Manager`.
1142
It also supports creation of shared lists and dictionaries.
1144
.. method:: BoundedSemaphore([value])
1146
Create a shared :class:`threading.BoundedSemaphore` object and return a
1149
.. method:: Condition([lock])
1151
Create a shared :class:`threading.Condition` object and return a proxy for
1154
If *lock* is supplied then it should be a proxy for a
1155
:class:`threading.Lock` or :class:`threading.RLock` object.
1159
Create a shared :class:`threading.Event` object and return a proxy for it.
1163
Create a shared :class:`threading.Lock` object and return a proxy for it.
1165
.. method:: Namespace()
1167
Create a shared :class:`Namespace` object and return a proxy for it.
1169
.. method:: Queue([maxsize])
1171
Create a shared :class:`Queue.Queue` object and return a proxy for it.
1175
Create a shared :class:`threading.RLock` object and return a proxy for it.
1177
.. method:: Semaphore([value])
1179
Create a shared :class:`threading.Semaphore` object and return a proxy for
1182
.. method:: Array(typecode, sequence)
1184
Create an array and return a proxy for it.
1186
.. method:: Value(typecode, value)
1188
Create an object with a writable ``value`` attribute and return a proxy
1195
Create a shared ``dict`` object and return a proxy for it.
1200
Create a shared ``list`` object and return a proxy for it.
1206
A namespace object has no public methods, but does have writable attributes.
1207
Its representation shows the values of its attributes.
1209
However, when using a proxy for a namespace object, an attribute beginning with
1210
``'_'`` will be an attribute of the proxy and not an attribute of the referent::
1212
>>> manager = multiprocessing.Manager()
1213
>>> Global = manager.Namespace()
1215
>>> Global.y = 'hello'
1216
>>> Global._z = 12.3 # this is an attribute of the proxy
1218
Namespace(x=10, y='hello')
1224
To create one's own manager, one creates a subclass of :class:`BaseManager` and
1225
use the :meth:`~BaseManager.resgister` classmethod to register new types or
1226
callables with the manager class. For example::
1228
from multiprocessing.managers import BaseManager
1230
class MathsClass(object):
1231
def add(self, x, y):
1233
def mul(self, x, y):
1236
class MyManager(BaseManager):
1239
MyManager.register('Maths', MathsClass)
1241
if __name__ == '__main__':
1242
manager = MyManager()
1244
maths = manager.Maths()
1245
print maths.add(4, 3) # prints 7
1246
print maths.mul(7, 8) # prints 56
1249
Using a remote manager
1250
>>>>>>>>>>>>>>>>>>>>>>
1252
It is possible to run a manager server on one machine and have clients use it
1253
from other machines (assuming that the firewalls involved allow it).
1255
Running the following commands creates a server for a single shared queue which
1256
remote clients can access::
1258
>>> from multiprocessing.managers import BaseManager
1260
>>> queue = Queue.Queue()
1261
>>> class QueueManager(BaseManager): pass
1263
>>> QueueManager.register('getQueue', callable=lambda:queue)
1264
>>> m = QueueManager(address=('', 50000), authkey='abracadabra')
1265
>>> m.serveForever()
1267
One client can access the server as follows::
1269
>>> from multiprocessing.managers import BaseManager
1270
>>> class QueueManager(BaseManager): pass
1272
>>> QueueManager.register('getQueue')
1273
>>> m = QueueManager.from_address(address=('foo.bar.org', 50000),
1274
>>> authkey='abracadabra')
1275
>>> queue = m.getQueue()
1276
>>> queue.put('hello')
1278
Another client can also use it::
1280
>>> from multiprocessing.managers import BaseManager
1281
>>> class QueueManager(BaseManager): pass
1283
>>> QueueManager.register('getQueue')
1284
>>> m = QueueManager.from_address(address=('foo.bar.org', 50000), authkey='abracadabra')
1285
>>> queue = m.getQueue()
1293
A proxy is an object which *refers* to a shared object which lives (presumably)
1294
in a different process. The shared object is said to be the *referent* of the
1295
proxy. Multiple proxy objects may have the same referent.
1297
A proxy object has methods which invoke corresponding methods of its referent
1298
(although not every method of the referent will necessarily be available through
1299
the proxy). A proxy can usually be used in most of the same ways that its
1302
>>> from multiprocessing import Manager
1303
>>> manager = Manager()
1304
>>> l = manager.list([i*i for i in range(10)])
1306
[0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
1308
<ListProxy object, typeid 'list' at 0xb799974c>
1314
Notice that applying :func:`str` to a proxy will return the representation of
1315
the referent, whereas applying :func:`repr` will return the representation of
1318
An important feature of proxy objects is that they are picklable so they can be
1319
passed between processes. Note, however, that if a proxy is sent to the
1320
corresponding manager's process then unpickling it will produce the referent
1321
itself. This means, for example, that one shared object can contain a second::
1323
>>> a = manager.list()
1324
>>> b = manager.list()
1325
>>> a.append(b) # referent of a now contains referent of b
1328
>>> b.append('hello')
1330
[['hello']] ['hello']
1334
The proxy types in :mod:`multiprocessing` do nothing to support comparisons
1335
by value. So, for instance, ::
1337
manager.list([1,2,3]) == [1,2,3]
1339
will return ``False``. One should just use a copy of the referent instead
1340
when making comparisons.
1342
.. class:: BaseProxy
1344
Proxy objects are instances of subclasses of :class:`BaseProxy`.
1346
.. method:: _call_method(methodname[, args[, kwds]])
1348
Call and return the result of a method of the proxy's referent.
1350
If ``proxy`` is a proxy whose referent is ``obj`` then the expression ::
1352
proxy._call_method(methodname, args, kwds)
1354
will evaluate the expression ::
1356
getattr(obj, methodname)(*args, **kwds)
1358
in the manager's process.
1360
The returned value will be a copy of the result of the call or a proxy to
1361
a new shared object -- see documentation for the *method_to_typeid*
1362
argument of :meth:`BaseManager.register`.
1364
If an exception is raised by the call, then then is re-raised by
1365
:meth:`_call_method`. If some other exception is raised in the manager's
1366
process then this is converted into a :exc:`RemoteError` exception and is
1367
raised by :meth:`_call_method`.
1369
Note in particular that an exception will be raised if *methodname* has
1372
An example of the usage of :meth:`_call_method`::
1374
>>> l = manager.list(range(10))
1375
>>> l._call_method('__len__')
1377
>>> l._call_method('__getslice__', (2, 7)) # equiv to `l[2:7]`
1379
>>> l._call_method('__getitem__', (20,)) # equiv to `l[20]`
1380
Traceback (most recent call last):
1382
IndexError: list index out of range
1384
.. method:: _get_value()
1386
Return a copy of the referent.
1388
If the referent is unpicklable then this will raise an exception.
1390
.. method:: __repr__
1392
Return a representation of the proxy object.
1396
Return the representation of the referent.
1402
A proxy object uses a weakref callback so that when it gets garbage collected it
1403
deregisters itself from the manager which owns its referent.
1405
A shared object gets deleted from the manager process when there are no longer
1406
any proxies referring to it.
1412
.. module:: multiprocessing.pool
1413
:synopsis: Create pools of processes.
1415
One can create a pool of processes which will carry out tasks submitted to it
1416
with the :class:`Pool` class.
1418
.. class:: multiprocessing.Pool([processes[, initializer[, initargs]]])
1420
A process pool object which controls a pool of worker processes to which jobs
1421
can be submitted. It supports asynchronous results with timeouts and
1422
callbacks and has a parallel map implementation.
1424
*processes* is the number of worker processes to use. If *processes* is
1425
``None`` then the number returned by :func:`cpu_count` is used. If
1426
*initializer* is not ``None`` then each worker process will call
1427
``initializer(*initargs)`` when it starts.
1429
.. method:: apply(func[, args[, kwds]])
1431
Equivalent of the :func:`apply` builtin function. It blocks till the
1434
.. method:: apply_async(func[, args[, kwds[, callback]]])
1436
A variant of the :meth:`apply` method which returns a result object.
1438
If *callback* is specified then it should be a callable which accepts a
1439
single argument. When the result becomes ready *callback* is applied to
1440
it (unless the call failed). *callback* should complete immediately since
1441
otherwise the thread which handles the results will get blocked.
1443
.. method:: map(func, iterable[, chunksize])
1445
A parallel equivalent of the :func:`map` builtin function. It blocks till
1446
the result is ready.
1448
This method chops the iterable into a number of chunks which it submits to
1449
the process pool as separate tasks. The (approximate) size of these
1450
chunks can be specified by setting *chunksize* to a positive integer.
1452
.. method:: map_async(func, iterable[, chunksize[, callback]])
1454
A variant of the :meth:`map` method which returns a result object.
1456
If *callback* is specified then it should be a callable which accepts a
1457
single argument. When the result becomes ready *callback* is applied to
1458
it (unless the call failed). *callback* should complete immediately since
1459
otherwise the thread which handles the results will get blocked.
1461
.. method:: imap(func, iterable[, chunksize])
1463
An equivalent of :func:`itertools.imap`.
1465
The *chunksize* argument is the same as the one used by the :meth:`.map`
1466
method. For very long iterables using a large value for *chunksize* can
1467
make make the job complete **much** faster than using the default value of
1470
Also if *chunksize* is ``1`` then the :meth:`next` method of the iterator
1471
returned by the :meth:`imap` method has an optional *timeout* parameter:
1472
``next(timeout)`` will raise :exc:`multiprocessing.TimeoutError` if the
1473
result cannot be returned within *timeout* seconds.
1475
.. method:: imap_unordered(func, iterable[, chunksize])
1477
The same as :meth:`imap` except that the ordering of the results from the
1478
returned iterator should be considered arbitrary. (Only when there is
1479
only one worker process is the order guaranteed to be "correct".)
1483
Prevents any more tasks from being submitted to the pool. Once all the
1484
tasks have been completed the worker processes will exit.
1486
.. method:: terminate()
1488
Stops the worker processes immediately without completing outstanding
1489
work. When the pool object is garbage collected :meth:`terminate` will be
1494
Wait for the worker processes to exit. One must call :meth:`close` or
1495
:meth:`terminate` before using :meth:`join`.
1498
.. class:: AsyncResult
1500
The class of the result returned by :meth:`Pool.apply_async` and
1501
:meth:`Pool.map_async`.
1503
.. method:: get([timeout)
1505
Return the result when it arrives. If *timeout* is not ``None`` and the
1506
result does not arrive within *timeout* seconds then
1507
:exc:`multiprocessing.TimeoutError` is raised. If the remote call raised
1508
an exception then that exception will be reraised by :meth:`get`.
1510
.. method:: wait([timeout])
1512
Wait until the result is available or until *timeout* seconds pass.
1516
Return whether the call has completed.
1518
.. method:: successful()
1520
Return whether the call completed without raising an exception. Will
1521
raise :exc:`AssertionError` if the result is not ready.
1523
The following example demonstrates the use of a pool::
1525
from multiprocessing import Pool
1530
if __name__ == '__main__':
1531
pool = Pool(processes=4) # start 4 worker processes
1533
result = pool.applyAsync(f, (10,)) # evaluate "f(10)" asynchronously
1534
print result.get(timeout=1) # prints "100" unless your computer is *very* slow
1536
print pool.map(f, range(10)) # prints "[0, 1, 4,..., 81]"
1538
it = pool.imap(f, range(10))
1539
print it.next() # prints "0"
1540
print it.next() # prints "1"
1541
print it.next(timeout=1) # prints "4" unless your computer is *very* slow
1544
result = pool.applyAsync(time.sleep, (10,))
1545
print result.get(timeout=1) # raises TimeoutError
1548
.. _multiprocessing-listeners-clients:
1550
Listeners and Clients
1551
~~~~~~~~~~~~~~~~~~~~~
1553
.. module:: multiprocessing.connection
1554
:synopsis: API for dealing with sockets.
1556
Usually message passing between processes is done using queues or by using
1557
:class:`Connection` objects returned by :func:`Pipe`.
1559
However, the :mod:`multiprocessing.connection` module allows some extra
1560
flexibility. It basically gives a high level message oriented API for dealing
1561
with sockets or Windows named pipes, and also has support for *digest
1562
authentication* using the :mod:`hmac` module.
1565
.. function:: deliver_challenge(connection, authkey)
1567
Send a randomly generated message to the other end of the connection and wait
1570
If the reply matches the digest of the message using *authkey* as the key
1571
then a welcome message is sent to the other end of the connection. Otherwise
1572
:exc:`AuthenticationError` is raised.
1574
.. function:: answerChallenge(connection, authkey)
1576
Receive a message, calculate the digest of the message using *authkey* as the
1577
key, and then send the digest back.
1579
If a welcome message is not received, then :exc:`AuthenticationError` is
1582
.. function:: Client(address[, family[, authenticate[, authkey]]])
1584
Attempt to set up a connection to the listener which is using address
1585
*address*, returning a :class:`~multiprocessing.Connection`.
1587
The type of the connection is determined by *family* argument, but this can
1588
generally be omitted since it can usually be inferred from the format of
1589
*address*. (See :ref:`multiprocessing-address-formats`)
1591
If *authentication* is ``True`` or *authkey* is a string then digest
1592
authentication is used. The key used for authentication will be either
1593
*authkey* or ``current_process().authkey)`` if *authkey* is ``None``.
1594
If authentication fails then :exc:`AuthenticationError` is raised. See
1595
:ref:`multiprocessing-auth-keys`.
1597
.. class:: Listener([address[, family[, backlog[, authenticate[, authkey]]]]])
1599
A wrapper for a bound socket or Windows named pipe which is 'listening' for
1602
*address* is the address to be used by the bound socket or named pipe of the
1605
*family* is the type of socket (or named pipe) to use. This can be one of
1606
the strings ``'AF_INET'`` (for a TCP socket), ``'AF_UNIX'`` (for a Unix
1607
domain socket) or ``'AF_PIPE'`` (for a Windows named pipe). Of these only
1608
the first is guaranteed to be available. If *family* is ``None`` then the
1609
family is inferred from the format of *address*. If *address* is also
1610
``None`` then a default is chosen. This default is the family which is
1611
assumed to be the fastest available. See
1612
:ref:`multiprocessing-address-formats`. Note that if *family* is
1613
``'AF_UNIX'`` and address is ``None`` then the socket will be created in a
1614
private temporary directory created using :func:`tempfile.mkstemp`.
1616
If the listener object uses a socket then *backlog* (1 by default) is passed
1617
to the :meth:`listen` method of the socket once it has been bound.
1619
If *authenticate* is ``True`` (``False`` by default) or *authkey* is not
1620
``None`` then digest authentication is used.
1622
If *authkey* is a string then it will be used as the authentication key;
1623
otherwise it must be *None*.
1625
If *authkey* is ``None`` and *authenticate* is ``True`` then
1626
``current_process().authkey`` is used as the authentication key. If
1627
*authkey* is ``None`` and *authentication* is ``False`` then no
1628
authentication is done. If authentication fails then
1629
:exc:`AuthenticationError` is raised. See :ref:`multiprocessing-auth-keys`.
1631
.. method:: accept()
1633
Accept a connection on the bound socket or named pipe of the listener
1634
object and return a :class:`Connection` object. If authentication is
1635
attempted and fails, then :exc:`AuthenticationError` is raised.
1639
Close the bound socket or named pipe of the listener object. This is
1640
called automatically when the listener is garbage collected. However it
1641
is advisable to call it explicitly.
1643
Listener objects have the following read-only properties:
1645
.. attribute:: address
1647
The address which is being used by the Listener object.
1649
.. attribute:: last_accepted
1651
The address from which the last accepted connection came. If this is
1652
unavailable then it is ``None``.
1655
The module defines two exceptions:
1657
.. exception:: AuthenticationError
1659
Exception raised when there is an authentication error.
1664
The following server code creates a listener which uses ``'secret password'`` as
1665
an authentication key. It then waits for a connection and sends some data to
1668
from multiprocessing.connection import Listener
1669
from array import array
1671
address = ('localhost', 6000) # family is deduced to be 'AF_INET'
1672
listener = Listener(address, authkey='secret password')
1674
conn = listener.accept()
1675
print 'connection accepted from', listener.last_accepted
1677
conn.send([2.25, None, 'junk', float])
1679
conn.send_bytes('hello')
1681
conn.send_bytes(array('i', [42, 1729]))
1686
The following code connects to the server and receives some data from the
1689
from multiprocessing.connection import Client
1690
from array import array
1692
address = ('localhost', 6000)
1693
conn = Client(address, authkey='secret password')
1695
print conn.recv() # => [2.25, None, 'junk', float]
1697
print conn.recv_bytes() # => 'hello'
1699
arr = array('i', [0, 0, 0, 0, 0])
1700
print conn.recv_bytes_into(arr) # => 8
1701
print arr # => array('i', [42, 1729, 0, 0, 0])
1706
.. _multiprocessing-address-formats:
1711
* An ``'AF_INET'`` address is a tuple of the form ``(hostname, port)`` where
1712
*hostname* is a string and *port* is an integer.
1714
* An ``'AF_UNIX'`` address is a string representing a filename on the
1717
* An ``'AF_PIPE'`` address is a string of the form
1718
``r'\\\\.\\pipe\\PipeName'``. To use :func:`Client` to connect to a named
1719
pipe on a remote computer called ServerName* one should use an address of the
1720
form ``r'\\\\ServerName\\pipe\\PipeName'`` instead.
1722
Note that any string beginning with two backslashes is assumed by default to be
1723
an ``'AF_PIPE'`` address rather than an ``'AF_UNIX'`` address.
1726
.. _multiprocessing-auth-keys:
1731
When one uses :meth:`Connection.recv`, the data received is automatically
1732
unpickled. Unfortunately unpickling data from an untrusted source is a security
1733
risk. Therefore :class:`Listener` and :func:`Client` use the :mod:`hmac` module
1734
to provide digest authentication.
1736
An authentication key is a string which can be thought of as a password: once a
1737
connection is established both ends will demand proof that the other knows the
1738
authentication key. (Demonstrating that both ends are using the same key does
1739
**not** involve sending the key over the connection.)
1741
If authentication is requested but do authentication key is specified then the
1742
return value of ``current_process().authkey`` is used (see
1743
:class:`~multiprocessing.Process`). This value will automatically inherited by
1744
any :class:`~multiprocessing.Process` object that the current process creates.
1745
This means that (by default) all processes of a multi-process program will share
1746
a single authentication key which can be used when setting up connections
1747
between the themselves.
1749
Suitable authentication keys can also be generated by using :func:`os.urandom`.
1755
Some support for logging is available. Note, however, that the :mod:`logging`
1756
package does not use process shared locks so it is possible (depending on the
1757
handler type) for messages from different processes to get mixed up.
1759
.. currentmodule:: multiprocessing
1760
.. function:: get_logger()
1762
Returns the logger used by :mod:`multiprocessing`. If necessary, a new one
1765
When first created the logger has level :data:`logging.NOTSET` and has a
1766
handler which sends output to :data:`sys.stderr` using format
1767
``'[%(levelname)s/%(processName)s] %(message)s'``. (The logger allows use of
1768
the non-standard ``'%(processName)s'`` format.) Message sent to this logger
1769
will not by default propagate to the root logger.
1771
Note that on Windows child processes will only inherit the level of the
1772
parent process's logger -- any other customization of the logger will not be
1775
Below is an example session with logging turned on::
1777
>>> import processing, logging
1778
>>> logger = processing.getLogger()
1779
>>> logger.setLevel(logging.INFO)
1780
>>> logger.warning('doomed')
1781
[WARNING/MainProcess] doomed
1782
>>> m = processing.Manager()
1783
[INFO/SyncManager-1] child process calling self.run()
1784
[INFO/SyncManager-1] manager bound to '\\\\.\\pipe\\pyc-2776-0-lj0tfa'
1786
[INFO/MainProcess] sending shutdown message to manager
1787
[INFO/SyncManager-1] manager exiting with exitcode 0
1790
The :mod:`multiprocessing.dummy` module
1791
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1793
.. module:: multiprocessing.dummy
1794
:synopsis: Dumb wrapper around threading.
1796
:mod:`multiprocessing.dummy` replicates the API of :mod:`multiprocessing` but is
1797
no more than a wrapper around the :mod:`threading` module.
1800
.. _multiprocessing-programming:
1802
Programming guidelines
1803
----------------------
1805
There are certain guidelines and idioms which should be adhered to when using
1806
:mod:`multiprocessing`.
1814
As far as possible one should try to avoid shifting large amounts of data
1817
It is probably best to stick to using queues or pipes for communication
1818
between processes rather than using the lower level synchronization
1819
primitives from the :mod:`threading` module.
1823
Ensure that the arguments to the methods of proxies are picklable.
1825
Thread safety of proxies
1827
Do not use a proxy object from more than one thread unless you protect it
1830
(There is never a problem with different processes using the *same* proxy.)
1832
Joining zombie processes
1834
On Unix when a process finishes but has not been joined it becomes a zombie.
1835
There should never be very many because each time a new process starts (or
1836
:func:`active_children` is called) all completed processes which have not
1837
yet been joined will be joined. Also calling a finished process's
1838
:meth:`Process.is_alive` will join the process. Even so it is probably good
1839
practice to explicitly join all the processes that you start.
1841
Better to inherit than pickle/unpickle
1843
On Windows many types from :mod:`multiprocessing` need to be picklable so
1844
that child processes can use them. However, one should generally avoid
1845
sending shared objects to other processes using pipes or queues. Instead
1846
you should arrange the program so that a process which need access to a
1847
shared resource created elsewhere can inherit it from an ancestor process.
1849
Avoid terminating processes
1851
Using the :meth:`Process.terminate` method to stop a process is liable to
1852
cause any shared resources (such as locks, semaphores, pipes and queues)
1853
currently being used by the process to become broken or unavailable to other
1856
Therefore it is probably best to only consider using
1857
:meth:`Process.terminate` on processes which never use any shared resources.
1859
Joining processes that use queues
1861
Bear in mind that a process that has put items in a queue will wait before
1862
terminating until all the buffered items are fed by the "feeder" thread to
1863
the underlying pipe. (The child process can call the
1864
:meth:`Queue.cancel_join_thread` method of the queue to avoid this behaviour.)
1866
This means that whenever you use a queue you need to make sure that all
1867
items which have been put on the queue will eventually be removed before the
1868
process is joined. Otherwise you cannot be sure that processes which have
1869
put items on the queue will terminate. Remember also that non-daemonic
1870
processes will be automatically be joined.
1872
An example which will deadlock is the following::
1874
from multiprocessing import Process, Queue
1877
q.put('X' * 1000000)
1879
if __name__ == '__main__':
1881
p = Process(target=f, args=(queue,))
1883
p.join() # this deadlocks
1886
A fix here would be to swap the last two lines round (or simply remove the
1889
Explicitly pass resources to child processes
1891
On Unix a child process can make use of a shared resource created in a
1892
parent process using a global resource. However, it is better to pass the
1893
object as an argument to the constructor for the child process.
1895
Apart from making the code (potentially) compatible with Windows this also
1896
ensures that as long as the child process is still alive the object will not
1897
be garbage collected in the parent process. This might be important if some
1898
resource is freed when the object is garbage collected in the parent
1903
from multiprocessing import Process, Lock
1906
... do something using "lock" ...
1908
if __name__ == '__main__':
1911
Process(target=f).start()
1913
should be rewritten as ::
1915
from multiprocessing import Process, Lock
1918
... do something using "l" ...
1920
if __name__ == '__main__':
1923
Process(target=f, args=(lock,)).start()
1929
Since Windows lacks :func:`os.fork` it has a few extra restrictions:
1933
Ensure that all arguments to :meth:`Process.__init__` are picklable. This
1934
means, in particular, that bound or unbound methods cannot be used directly
1935
as the ``target`` argument on Windows --- just define a function and use
1938
Also, if you subclass :class:`Process` then make sure that instances will be
1939
picklable when the :meth:`Process.start` method is called.
1943
Bear in mind that if code run in a child process tries to access a global
1944
variable, then the value it sees (if any) may not be the same as the value
1945
in the parent process at the time that :meth:`Process.start` was called.
1947
However, global variables which are just module level constants cause no
1950
Safe importing of main module
1952
Make sure that the main module can be safely imported by a new Python
1953
interpreter without causing unintended side effects (such a starting a new
1956
For example, under Windows running the following module would fail with a
1957
:exc:`RuntimeError`::
1959
from multiprocessing import Process
1964
p = Process(target=foo)
1967
Instead one should protect the "entry point" of the program by using ``if
1968
__name__ == '__main__':`` as follows::
1970
from multiprocessing import Process, freeze_support
1975
if __name__ == '__main__':
1977
p = Process(target=foo)
1980
(The ``freeze_support()`` line can be omitted if the program will be run
1981
normally instead of frozen.)
1983
This allows the newly spawned Python interpreter to safely import the module
1984
and then run the module's ``foo()`` function.
1986
Similar restrictions apply if a pool or manager is created in the main
1990
.. _multiprocessing-examples:
1995
Demonstration of how to create and use customized managers and proxies:
1997
.. literalinclude:: ../includes/mp_newtype.py
2000
Using :class:`Pool`:
2002
.. literalinclude:: ../includes/mp_pool.py
2005
Synchronization types like locks, conditions and queues:
2007
.. literalinclude:: ../includes/mp_synchronize.py
2010
An showing how to use queues to feed tasks to a collection of worker process and
2011
collect the results:
2013
.. literalinclude:: ../includes/mp_workers.py
2016
An example of how a pool of worker processes can each run a
2017
:class:`SimpleHTTPServer.HttpServer` instance while sharing a single listening
2020
.. literalinclude:: ../includes/mp_webserver.py
2023
Some simple benchmarks comparing :mod:`multiprocessing` with :mod:`threading`:
2025
.. literalinclude:: ../includes/mp_benchmarks.py
2027
An example/demo of how to use the :class:`managers.SyncManager`, :class:`Process`
2028
and others to build a system which can distribute processes and work via a
2029
distributed queue to a "cluster" of machines on a network, accessible via SSH.
2030
You will need to have private key authentication for all hosts configured for
2033
.. literalinclude:: ../includes/mp_distributing.py
added
removed
Doc/library/multiprocessing.rst
