Antoine Pitrou | 64a467d | 2010-12-12 20:34:49 +0000 | [diff] [blame] | 1 | :mod:`multiprocessing` --- Process-based parallelism |
| 2 | ==================================================== |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 3 | |
| 4 | .. module:: multiprocessing |
Antoine Pitrou | 64a467d | 2010-12-12 20:34:49 +0000 | [diff] [blame] | 5 | :synopsis: Process-based parallelism. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 6 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 7 | |
| 8 | Introduction |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 9 | ------------ |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 10 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 11 | :mod:`multiprocessing` is a package that supports spawning processes using an |
| 12 | API similar to the :mod:`threading` module. The :mod:`multiprocessing` package |
| 13 | offers both local and remote concurrency, effectively side-stepping the |
| 14 | :term:`Global Interpreter Lock` by using subprocesses instead of threads. Due |
| 15 | to this, the :mod:`multiprocessing` module allows the programmer to fully |
| 16 | leverage multiple processors on a given machine. It runs on both Unix and |
| 17 | Windows. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 18 | |
Antoine Pitrou | 73dd030 | 2015-01-11 15:05:29 +0100 | [diff] [blame] | 19 | The :mod:`multiprocessing` module also introduces APIs which do not have |
| 20 | analogs in the :mod:`threading` module. A prime example of this is the |
| 21 | :class:`~multiprocessing.pool.Pool` object which offers a convenient means of |
| 22 | parallelizing the execution of a function across multiple input values, |
| 23 | distributing the input data across processes (data parallelism). The following |
| 24 | example demonstrates the common practice of defining such functions in a module |
| 25 | so that child processes can successfully import that module. This basic example |
| 26 | of data parallelism using :class:`~multiprocessing.pool.Pool`, :: |
Benjamin Peterson | e5384b0 | 2008-10-04 22:00:42 +0000 | [diff] [blame] | 27 | |
Antoine Pitrou | 73dd030 | 2015-01-11 15:05:29 +0100 | [diff] [blame] | 28 | from multiprocessing import Pool |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 29 | |
Antoine Pitrou | 73dd030 | 2015-01-11 15:05:29 +0100 | [diff] [blame] | 30 | def f(x): |
| 31 | return x*x |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 32 | |
Antoine Pitrou | 73dd030 | 2015-01-11 15:05:29 +0100 | [diff] [blame] | 33 | if __name__ == '__main__': |
| 34 | with Pool(5) as p: |
| 35 | print(p.map(f, [1, 2, 3])) |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 36 | |
Antoine Pitrou | 73dd030 | 2015-01-11 15:05:29 +0100 | [diff] [blame] | 37 | will print to standard output :: |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 38 | |
Antoine Pitrou | 73dd030 | 2015-01-11 15:05:29 +0100 | [diff] [blame] | 39 | [1, 4, 9] |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 40 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 41 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 42 | The :class:`Process` class |
| 43 | ~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 44 | |
| 45 | In :mod:`multiprocessing`, processes are spawned by creating a :class:`Process` |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 46 | object and then calling its :meth:`~Process.start` method. :class:`Process` |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 47 | follows the API of :class:`threading.Thread`. A trivial example of a |
| 48 | multiprocess program is :: |
| 49 | |
Georg Brandl | b3959bd | 2010-04-08 06:33:16 +0000 | [diff] [blame] | 50 | from multiprocessing import Process |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 51 | |
| 52 | def f(name): |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 53 | print('hello', name) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 54 | |
Georg Brandl | b3959bd | 2010-04-08 06:33:16 +0000 | [diff] [blame] | 55 | if __name__ == '__main__': |
| 56 | p = Process(target=f, args=('bob',)) |
| 57 | p.start() |
| 58 | p.join() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 59 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 60 | To show the individual process IDs involved, here is an expanded example:: |
| 61 | |
| 62 | from multiprocessing import Process |
| 63 | import os |
| 64 | |
| 65 | def info(title): |
Ezio Melotti | 985e24d | 2009-09-13 07:54:02 +0000 | [diff] [blame] | 66 | print(title) |
| 67 | print('module name:', __name__) |
Berker Peksag | 44e4b11 | 2015-09-21 06:12:50 +0300 | [diff] [blame] | 68 | print('parent process:', os.getppid()) |
Ezio Melotti | 985e24d | 2009-09-13 07:54:02 +0000 | [diff] [blame] | 69 | print('process id:', os.getpid()) |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 70 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 71 | def f(name): |
| 72 | info('function f') |
Ezio Melotti | 985e24d | 2009-09-13 07:54:02 +0000 | [diff] [blame] | 73 | print('hello', name) |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 74 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 75 | if __name__ == '__main__': |
| 76 | info('main line') |
| 77 | p = Process(target=f, args=('bob',)) |
| 78 | p.start() |
| 79 | p.join() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 80 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 81 | For an explanation of why the ``if __name__ == '__main__'`` part is |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 82 | necessary, see :ref:`multiprocessing-programming`. |
| 83 | |
| 84 | |
| 85 | |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 86 | Contexts and start methods |
| 87 | ~~~~~~~~~~~~~~~~~~~~~~~~~~ |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 88 | |
R David Murray | ac18622 | 2013-12-20 17:23:57 -0500 | [diff] [blame] | 89 | .. _multiprocessing-start-methods: |
| 90 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 91 | Depending on the platform, :mod:`multiprocessing` supports three ways |
| 92 | to start a process. These *start methods* are |
| 93 | |
| 94 | *spawn* |
| 95 | The parent process starts a fresh python interpreter process. The |
| 96 | child process will only inherit those resources necessary to run |
| 97 | the process objects :meth:`~Process.run` method. In particular, |
| 98 | unnecessary file descriptors and handles from the parent process |
| 99 | will not be inherited. Starting a process using this method is |
| 100 | rather slow compared to using *fork* or *forkserver*. |
| 101 | |
| 102 | Available on Unix and Windows. The default on Windows. |
| 103 | |
| 104 | *fork* |
| 105 | The parent process uses :func:`os.fork` to fork the Python |
| 106 | interpreter. The child process, when it begins, is effectively |
| 107 | identical to the parent process. All resources of the parent are |
| 108 | inherited by the child process. Note that safely forking a |
| 109 | multithreaded process is problematic. |
| 110 | |
| 111 | Available on Unix only. The default on Unix. |
| 112 | |
| 113 | *forkserver* |
| 114 | When the program starts and selects the *forkserver* start method, |
| 115 | a server process is started. From then on, whenever a new process |
Georg Brandl | 213ef6e | 2013-10-09 15:51:57 +0200 | [diff] [blame] | 116 | is needed, the parent process connects to the server and requests |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 117 | that it fork a new process. The fork server process is single |
| 118 | threaded so it is safe for it to use :func:`os.fork`. No |
| 119 | unnecessary resources are inherited. |
| 120 | |
| 121 | Available on Unix platforms which support passing file descriptors |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 122 | over Unix pipes. |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 123 | |
Larry Hastings | 3732ed2 | 2014-03-15 21:13:56 -0700 | [diff] [blame] | 124 | .. versionchanged:: 3.4 |
| 125 | *spawn* added on all unix platforms, and *forkserver* added for |
Georg Brandl | df48b97 | 2014-03-24 09:06:18 +0100 | [diff] [blame] | 126 | some unix platforms. |
Larry Hastings | 3732ed2 | 2014-03-15 21:13:56 -0700 | [diff] [blame] | 127 | Child processes no longer inherit all of the parents inheritable |
Georg Brandl | df48b97 | 2014-03-24 09:06:18 +0100 | [diff] [blame] | 128 | handles on Windows. |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 129 | |
| 130 | On Unix using the *spawn* or *forkserver* start methods will also |
| 131 | start a *semaphore tracker* process which tracks the unlinked named |
| 132 | semaphores created by processes of the program. When all processes |
| 133 | have exited the semaphore tracker unlinks any remaining semaphores. |
| 134 | Usually there should be none, but if a process was killed by a signal |
| 135 | there may some "leaked" semaphores. (Unlinking the named semaphores |
| 136 | is a serious matter since the system allows only a limited number, and |
| 137 | they will not be automatically unlinked until the next reboot.) |
| 138 | |
R David Murray | ac18622 | 2013-12-20 17:23:57 -0500 | [diff] [blame] | 139 | To select a start method you use the :func:`set_start_method` in |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 140 | the ``if __name__ == '__main__'`` clause of the main module. For |
| 141 | example:: |
| 142 | |
| 143 | import multiprocessing as mp |
| 144 | |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 145 | def foo(q): |
| 146 | q.put('hello') |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 147 | |
| 148 | if __name__ == '__main__': |
| 149 | mp.set_start_method('spawn') |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 150 | q = mp.Queue() |
| 151 | p = mp.Process(target=foo, args=(q,)) |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 152 | p.start() |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 153 | print(q.get()) |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 154 | p.join() |
| 155 | |
| 156 | :func:`set_start_method` should not be used more than once in the |
| 157 | program. |
| 158 | |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 159 | Alternatively, you can use :func:`get_context` to obtain a context |
| 160 | object. Context objects have the same API as the multiprocessing |
| 161 | module, and allow one to use multiple start methods in the same |
| 162 | program. :: |
| 163 | |
| 164 | import multiprocessing as mp |
| 165 | |
| 166 | def foo(q): |
| 167 | q.put('hello') |
| 168 | |
| 169 | if __name__ == '__main__': |
| 170 | ctx = mp.get_context('spawn') |
| 171 | q = ctx.Queue() |
| 172 | p = ctx.Process(target=foo, args=(q,)) |
| 173 | p.start() |
| 174 | print(q.get()) |
| 175 | p.join() |
| 176 | |
| 177 | Note that objects related to one context may not be compatible with |
| 178 | processes for a different context. In particular, locks created using |
| 179 | the *fork* context cannot be passed to a processes started using the |
| 180 | *spawn* or *forkserver* start methods. |
| 181 | |
| 182 | A library which wants to use a particular start method should probably |
| 183 | use :func:`get_context` to avoid interfering with the choice of the |
| 184 | library user. |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 185 | |
| 186 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 187 | Exchanging objects between processes |
| 188 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 189 | |
| 190 | :mod:`multiprocessing` supports two types of communication channel between |
| 191 | processes: |
| 192 | |
| 193 | **Queues** |
| 194 | |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 195 | The :class:`Queue` class is a near clone of :class:`queue.Queue`. For |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 196 | example:: |
| 197 | |
| 198 | from multiprocessing import Process, Queue |
| 199 | |
| 200 | def f(q): |
| 201 | q.put([42, None, 'hello']) |
| 202 | |
Georg Brandl | 1f01deb | 2009-01-03 22:47:39 +0000 | [diff] [blame] | 203 | if __name__ == '__main__': |
| 204 | q = Queue() |
| 205 | p = Process(target=f, args=(q,)) |
| 206 | p.start() |
| 207 | print(q.get()) # prints "[42, None, 'hello']" |
| 208 | p.join() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 209 | |
Antoine Pitrou | fc6accc | 2012-05-18 13:57:04 +0200 | [diff] [blame] | 210 | Queues are thread and process safe. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 211 | |
| 212 | **Pipes** |
| 213 | |
| 214 | The :func:`Pipe` function returns a pair of connection objects connected by a |
| 215 | pipe which by default is duplex (two-way). For example:: |
| 216 | |
| 217 | from multiprocessing import Process, Pipe |
| 218 | |
| 219 | def f(conn): |
| 220 | conn.send([42, None, 'hello']) |
| 221 | conn.close() |
| 222 | |
| 223 | if __name__ == '__main__': |
| 224 | parent_conn, child_conn = Pipe() |
| 225 | p = Process(target=f, args=(child_conn,)) |
| 226 | p.start() |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 227 | print(parent_conn.recv()) # prints "[42, None, 'hello']" |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 228 | p.join() |
| 229 | |
| 230 | The two connection objects returned by :func:`Pipe` represent the two ends of |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 231 | the pipe. Each connection object has :meth:`~Connection.send` and |
| 232 | :meth:`~Connection.recv` methods (among others). Note that data in a pipe |
| 233 | may become corrupted if two processes (or threads) try to read from or write |
| 234 | to the *same* end of the pipe at the same time. Of course there is no risk |
| 235 | of corruption from processes using different ends of the pipe at the same |
| 236 | time. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 237 | |
| 238 | |
| 239 | Synchronization between processes |
| 240 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 241 | |
| 242 | :mod:`multiprocessing` contains equivalents of all the synchronization |
| 243 | primitives from :mod:`threading`. For instance one can use a lock to ensure |
| 244 | that only one process prints to standard output at a time:: |
| 245 | |
| 246 | from multiprocessing import Process, Lock |
| 247 | |
| 248 | def f(l, i): |
| 249 | l.acquire() |
Andrew Svetlov | ee750d8 | 2014-07-02 07:21:03 +0300 | [diff] [blame] | 250 | try: |
| 251 | print('hello world', i) |
| 252 | finally: |
| 253 | l.release() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 254 | |
| 255 | if __name__ == '__main__': |
| 256 | lock = Lock() |
| 257 | |
| 258 | for num in range(10): |
| 259 | Process(target=f, args=(lock, num)).start() |
| 260 | |
| 261 | Without using the lock output from the different processes is liable to get all |
| 262 | mixed up. |
| 263 | |
| 264 | |
| 265 | Sharing state between processes |
| 266 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 267 | |
| 268 | As mentioned above, when doing concurrent programming it is usually best to |
| 269 | avoid using shared state as far as possible. This is particularly true when |
| 270 | using multiple processes. |
| 271 | |
| 272 | However, if you really do need to use some shared data then |
| 273 | :mod:`multiprocessing` provides a couple of ways of doing so. |
| 274 | |
| 275 | **Shared memory** |
| 276 | |
| 277 | Data can be stored in a shared memory map using :class:`Value` or |
| 278 | :class:`Array`. For example, the following code :: |
| 279 | |
| 280 | from multiprocessing import Process, Value, Array |
| 281 | |
| 282 | def f(n, a): |
| 283 | n.value = 3.1415927 |
| 284 | for i in range(len(a)): |
| 285 | a[i] = -a[i] |
| 286 | |
| 287 | if __name__ == '__main__': |
| 288 | num = Value('d', 0.0) |
| 289 | arr = Array('i', range(10)) |
| 290 | |
| 291 | p = Process(target=f, args=(num, arr)) |
| 292 | p.start() |
| 293 | p.join() |
| 294 | |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 295 | print(num.value) |
| 296 | print(arr[:]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 297 | |
| 298 | will print :: |
| 299 | |
| 300 | 3.1415927 |
| 301 | [0, -1, -2, -3, -4, -5, -6, -7, -8, -9] |
| 302 | |
| 303 | The ``'d'`` and ``'i'`` arguments used when creating ``num`` and ``arr`` are |
| 304 | typecodes of the kind used by the :mod:`array` module: ``'d'`` indicates a |
Georg Brandl | 2ee470f | 2008-07-16 12:55:28 +0000 | [diff] [blame] | 305 | double precision float and ``'i'`` indicates a signed integer. These shared |
Georg Brandl | f285bcc | 2010-10-19 21:07:16 +0000 | [diff] [blame] | 306 | objects will be process and thread-safe. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 307 | |
| 308 | For more flexibility in using shared memory one can use the |
| 309 | :mod:`multiprocessing.sharedctypes` module which supports the creation of |
| 310 | arbitrary ctypes objects allocated from shared memory. |
| 311 | |
| 312 | **Server process** |
| 313 | |
| 314 | A manager object returned by :func:`Manager` controls a server process which |
Georg Brandl | 2ee470f | 2008-07-16 12:55:28 +0000 | [diff] [blame] | 315 | holds Python objects and allows other processes to manipulate them using |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 316 | proxies. |
| 317 | |
Richard Oudkerk | 3730a17 | 2012-06-15 18:26:07 +0100 | [diff] [blame] | 318 | A manager returned by :func:`Manager` will support types |
| 319 | :class:`list`, :class:`dict`, :class:`Namespace`, :class:`Lock`, |
| 320 | :class:`RLock`, :class:`Semaphore`, :class:`BoundedSemaphore`, |
| 321 | :class:`Condition`, :class:`Event`, :class:`Barrier`, |
| 322 | :class:`Queue`, :class:`Value` and :class:`Array`. For example, :: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 323 | |
| 324 | from multiprocessing import Process, Manager |
| 325 | |
| 326 | def f(d, l): |
| 327 | d[1] = '1' |
| 328 | d['2'] = 2 |
| 329 | d[0.25] = None |
| 330 | l.reverse() |
| 331 | |
| 332 | if __name__ == '__main__': |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 333 | with Manager() as manager: |
| 334 | d = manager.dict() |
| 335 | l = manager.list(range(10)) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 336 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 337 | p = Process(target=f, args=(d, l)) |
| 338 | p.start() |
| 339 | p.join() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 340 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 341 | print(d) |
| 342 | print(l) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 343 | |
| 344 | will print :: |
| 345 | |
| 346 | {0.25: None, 1: '1', '2': 2} |
| 347 | [9, 8, 7, 6, 5, 4, 3, 2, 1, 0] |
| 348 | |
| 349 | Server process managers are more flexible than using shared memory objects |
| 350 | because they can be made to support arbitrary object types. Also, a single |
| 351 | manager can be shared by processes on different computers over a network. |
| 352 | They are, however, slower than using shared memory. |
| 353 | |
| 354 | |
| 355 | Using a pool of workers |
| 356 | ~~~~~~~~~~~~~~~~~~~~~~~ |
| 357 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 358 | The :class:`~multiprocessing.pool.Pool` class represents a pool of worker |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 359 | processes. It has methods which allows tasks to be offloaded to the worker |
| 360 | processes in a few different ways. |
| 361 | |
| 362 | For example:: |
| 363 | |
| 364 | from multiprocessing import Pool |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 365 | from time import sleep |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 366 | |
| 367 | def f(x): |
| 368 | return x*x |
| 369 | |
| 370 | if __name__ == '__main__': |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 371 | # start 4 worker processes |
| 372 | with Pool(processes=4) as pool: |
| 373 | |
| 374 | # print "[0, 1, 4,..., 81]" |
| 375 | print(pool.map(f, range(10))) |
| 376 | |
| 377 | # print same numbers in arbitrary order |
| 378 | for i in pool.imap_unordered(f, range(10)): |
| 379 | print(i) |
| 380 | |
| 381 | # evaluate "f(10)" asynchronously |
| 382 | res = pool.apply_async(f, [10]) |
| 383 | print(res.get(timeout=1)) # prints "100" |
| 384 | |
| 385 | # make worker sleep for 10 secs |
Terry Jan Reedy | 9f5388f | 2014-07-23 20:30:29 -0400 | [diff] [blame] | 386 | res = pool.apply_async(sleep, [10]) |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 387 | print(res.get(timeout=1)) # raises multiprocessing.TimeoutError |
| 388 | |
| 389 | # exiting the 'with'-block has stopped the pool |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 390 | |
Richard Oudkerk | b3c4b98 | 2013-07-02 12:32:00 +0100 | [diff] [blame] | 391 | Note that the methods of a pool should only ever be used by the |
| 392 | process which created it. |
| 393 | |
Antoine Pitrou | 73dd030 | 2015-01-11 15:05:29 +0100 | [diff] [blame] | 394 | .. note:: |
| 395 | |
| 396 | Functionality within this package requires that the ``__main__`` module be |
| 397 | importable by the children. This is covered in :ref:`multiprocessing-programming` |
| 398 | however it is worth pointing out here. This means that some examples, such |
| 399 | as the :class:`multiprocessing.pool.Pool` examples will not work in the |
| 400 | interactive interpreter. For example:: |
| 401 | |
| 402 | >>> from multiprocessing import Pool |
| 403 | >>> p = Pool(5) |
| 404 | >>> def f(x): |
| 405 | ... return x*x |
| 406 | ... |
| 407 | >>> p.map(f, [1,2,3]) |
| 408 | Process PoolWorker-1: |
| 409 | Process PoolWorker-2: |
| 410 | Process PoolWorker-3: |
| 411 | Traceback (most recent call last): |
| 412 | Traceback (most recent call last): |
| 413 | Traceback (most recent call last): |
| 414 | AttributeError: 'module' object has no attribute 'f' |
| 415 | AttributeError: 'module' object has no attribute 'f' |
| 416 | AttributeError: 'module' object has no attribute 'f' |
| 417 | |
| 418 | (If you try this it will actually output three full tracebacks |
| 419 | interleaved in a semi-random fashion, and then you may have to |
| 420 | stop the master process somehow.) |
| 421 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 422 | |
| 423 | Reference |
| 424 | --------- |
| 425 | |
| 426 | The :mod:`multiprocessing` package mostly replicates the API of the |
| 427 | :mod:`threading` module. |
| 428 | |
| 429 | |
| 430 | :class:`Process` and exceptions |
| 431 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 432 | |
Ezio Melotti | 8429b67 | 2012-09-14 06:35:09 +0300 | [diff] [blame] | 433 | .. class:: Process(group=None, target=None, name=None, args=(), kwargs={}, \ |
| 434 | *, daemon=None) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 435 | |
| 436 | Process objects represent activity that is run in a separate process. The |
| 437 | :class:`Process` class has equivalents of all the methods of |
| 438 | :class:`threading.Thread`. |
| 439 | |
| 440 | The constructor should always be called with keyword arguments. *group* |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 441 | should always be ``None``; it exists solely for compatibility with |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 442 | :class:`threading.Thread`. *target* is the callable object to be invoked by |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 443 | the :meth:`run()` method. It defaults to ``None``, meaning nothing is |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 444 | called. *name* is the process name (see :attr:`name` for more details). |
| 445 | *args* is the argument tuple for the target invocation. *kwargs* is a |
| 446 | dictionary of keyword arguments for the target invocation. If provided, |
| 447 | the keyword-only *daemon* argument sets the process :attr:`daemon` flag |
| 448 | to ``True`` or ``False``. If ``None`` (the default), this flag will be |
| 449 | inherited from the creating process. |
Antoine Pitrou | 0bd4deb | 2011-02-25 22:07:43 +0000 | [diff] [blame] | 450 | |
| 451 | By default, no arguments are passed to *target*. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 452 | |
| 453 | If a subclass overrides the constructor, it must make sure it invokes the |
| 454 | base class constructor (:meth:`Process.__init__`) before doing anything else |
| 455 | to the process. |
| 456 | |
Antoine Pitrou | 0bd4deb | 2011-02-25 22:07:43 +0000 | [diff] [blame] | 457 | .. versionchanged:: 3.3 |
| 458 | Added the *daemon* argument. |
| 459 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 460 | .. method:: run() |
| 461 | |
| 462 | Method representing the process's activity. |
| 463 | |
| 464 | You may override this method in a subclass. The standard :meth:`run` |
| 465 | method invokes the callable object passed to the object's constructor as |
| 466 | the target argument, if any, with sequential and keyword arguments taken |
| 467 | from the *args* and *kwargs* arguments, respectively. |
| 468 | |
| 469 | .. method:: start() |
| 470 | |
| 471 | Start the process's activity. |
| 472 | |
| 473 | This must be called at most once per process object. It arranges for the |
| 474 | object's :meth:`run` method to be invoked in a separate process. |
| 475 | |
| 476 | .. method:: join([timeout]) |
| 477 | |
Charles-François Natali | acd9f7c | 2011-07-25 18:35:49 +0200 | [diff] [blame] | 478 | If the optional argument *timeout* is ``None`` (the default), the method |
| 479 | blocks until the process whose :meth:`join` method is called terminates. |
| 480 | If *timeout* is a positive number, it blocks at most *timeout* seconds. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 481 | |
| 482 | A process can be joined many times. |
| 483 | |
| 484 | A process cannot join itself because this would cause a deadlock. It is |
| 485 | an error to attempt to join a process before it has been started. |
| 486 | |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 487 | .. attribute:: name |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 488 | |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 489 | The process's name. The name is a string used for identification purposes |
| 490 | only. It has no semantics. Multiple processes may be given the same |
| 491 | name. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 492 | |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 493 | The initial name is set by the constructor. If no explicit name is |
| 494 | provided to the constructor, a name of the form |
| 495 | 'Process-N\ :sub:`1`:N\ :sub:`2`:...:N\ :sub:`k`' is constructed, where |
| 496 | each N\ :sub:`k` is the N-th child of its parent. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 497 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 498 | .. method:: is_alive |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 499 | |
| 500 | Return whether the process is alive. |
| 501 | |
| 502 | Roughly, a process object is alive from the moment the :meth:`start` |
| 503 | method returns until the child process terminates. |
| 504 | |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 505 | .. attribute:: daemon |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 506 | |
Benjamin Peterson | da10d3b | 2009-01-01 00:23:30 +0000 | [diff] [blame] | 507 | The process's daemon flag, a Boolean value. This must be set before |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 508 | :meth:`start` is called. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 509 | |
| 510 | The initial value is inherited from the creating process. |
| 511 | |
| 512 | When a process exits, it attempts to terminate all of its daemonic child |
| 513 | processes. |
| 514 | |
| 515 | Note that a daemonic process is not allowed to create child processes. |
| 516 | Otherwise a daemonic process would leave its children orphaned if it gets |
Alexandre Vassalotti | 260484d | 2009-07-17 11:43:26 +0000 | [diff] [blame] | 517 | terminated when its parent process exits. Additionally, these are **not** |
| 518 | Unix daemons or services, they are normal processes that will be |
Georg Brandl | 6faee4e | 2010-09-21 14:48:28 +0000 | [diff] [blame] | 519 | terminated (and not joined) if non-daemonic processes have exited. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 520 | |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 521 | In addition to the :class:`threading.Thread` API, :class:`Process` objects |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 522 | also support the following attributes and methods: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 523 | |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 524 | .. attribute:: pid |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 525 | |
| 526 | Return the process ID. Before the process is spawned, this will be |
| 527 | ``None``. |
| 528 | |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 529 | .. attribute:: exitcode |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 530 | |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 531 | The child's exit code. This will be ``None`` if the process has not yet |
| 532 | terminated. A negative value *-N* indicates that the child was terminated |
| 533 | by signal *N*. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 534 | |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 535 | .. attribute:: authkey |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 536 | |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 537 | The process's authentication key (a byte string). |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 538 | |
| 539 | When :mod:`multiprocessing` is initialized the main process is assigned a |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 540 | random string using :func:`os.urandom`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 541 | |
| 542 | When a :class:`Process` object is created, it will inherit the |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 543 | authentication key of its parent process, although this may be changed by |
| 544 | setting :attr:`authkey` to another byte string. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 545 | |
| 546 | See :ref:`multiprocessing-auth-keys`. |
| 547 | |
Antoine Pitrou | 176f07d | 2011-06-06 19:35:31 +0200 | [diff] [blame] | 548 | .. attribute:: sentinel |
| 549 | |
| 550 | A numeric handle of a system object which will become "ready" when |
| 551 | the process ends. |
| 552 | |
Antoine Pitrou | bdb1cf1 | 2012-03-05 19:28:37 +0100 | [diff] [blame] | 553 | You can use this value if you want to wait on several events at |
| 554 | once using :func:`multiprocessing.connection.wait`. Otherwise |
| 555 | calling :meth:`join()` is simpler. |
| 556 | |
Antoine Pitrou | 176f07d | 2011-06-06 19:35:31 +0200 | [diff] [blame] | 557 | On Windows, this is an OS handle usable with the ``WaitForSingleObject`` |
| 558 | and ``WaitForMultipleObjects`` family of API calls. On Unix, this is |
| 559 | a file descriptor usable with primitives from the :mod:`select` module. |
| 560 | |
Antoine Pitrou | 176f07d | 2011-06-06 19:35:31 +0200 | [diff] [blame] | 561 | .. versionadded:: 3.3 |
| 562 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 563 | .. method:: terminate() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 564 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 565 | Terminate the process. On Unix this is done using the ``SIGTERM`` signal; |
Georg Brandl | 60203b4 | 2010-10-06 10:11:56 +0000 | [diff] [blame] | 566 | on Windows :c:func:`TerminateProcess` is used. Note that exit handlers and |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 567 | finally clauses, etc., will not be executed. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 568 | |
| 569 | Note that descendant processes of the process will *not* be terminated -- |
| 570 | they will simply become orphaned. |
| 571 | |
| 572 | .. warning:: |
| 573 | |
| 574 | If this method is used when the associated process is using a pipe or |
| 575 | queue then the pipe or queue is liable to become corrupted and may |
| 576 | become unusable by other process. Similarly, if the process has |
| 577 | acquired a lock or semaphore etc. then terminating it is liable to |
| 578 | cause other processes to deadlock. |
| 579 | |
Ask Solem | ff7ffdd | 2010-11-09 21:52:33 +0000 | [diff] [blame] | 580 | Note that the :meth:`start`, :meth:`join`, :meth:`is_alive`, |
Richard Oudkerk | 64c25b4 | 2013-06-24 15:42:00 +0100 | [diff] [blame] | 581 | :meth:`terminate` and :attr:`exitcode` methods should only be called by |
Ask Solem | ff7ffdd | 2010-11-09 21:52:33 +0000 | [diff] [blame] | 582 | the process that created the process object. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 583 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 584 | Example usage of some of the methods of :class:`Process`: |
| 585 | |
| 586 | .. doctest:: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 587 | |
Benjamin Peterson | 206e307 | 2008-10-19 14:07:49 +0000 | [diff] [blame] | 588 | >>> import multiprocessing, time, signal |
| 589 | >>> p = multiprocessing.Process(target=time.sleep, args=(1000,)) |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 590 | >>> print(p, p.is_alive()) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 591 | <Process(Process-1, initial)> False |
| 592 | >>> p.start() |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 593 | >>> print(p, p.is_alive()) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 594 | <Process(Process-1, started)> True |
| 595 | >>> p.terminate() |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 596 | >>> time.sleep(0.1) |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 597 | >>> print(p, p.is_alive()) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 598 | <Process(Process-1, stopped[SIGTERM])> False |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 599 | >>> p.exitcode == -signal.SIGTERM |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 600 | True |
| 601 | |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 602 | .. exception:: ProcessError |
| 603 | |
| 604 | The base class of all :mod:`multiprocessing` exceptions. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 605 | |
| 606 | .. exception:: BufferTooShort |
| 607 | |
| 608 | Exception raised by :meth:`Connection.recv_bytes_into()` when the supplied |
| 609 | buffer object is too small for the message read. |
| 610 | |
| 611 | If ``e`` is an instance of :exc:`BufferTooShort` then ``e.args[0]`` will give |
| 612 | the message as a byte string. |
| 613 | |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 614 | .. exception:: AuthenticationError |
| 615 | |
| 616 | Raised when there is an authentication error. |
| 617 | |
| 618 | .. exception:: TimeoutError |
| 619 | |
| 620 | Raised by methods with a timeout when the timeout expires. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 621 | |
| 622 | Pipes and Queues |
| 623 | ~~~~~~~~~~~~~~~~ |
| 624 | |
| 625 | When using multiple processes, one generally uses message passing for |
| 626 | communication between processes and avoids having to use any synchronization |
| 627 | primitives like locks. |
| 628 | |
| 629 | For passing messages one can use :func:`Pipe` (for a connection between two |
| 630 | processes) or a queue (which allows multiple producers and consumers). |
| 631 | |
Sandro Tosi | cd77815 | 2012-02-15 23:27:00 +0100 | [diff] [blame] | 632 | The :class:`Queue`, :class:`SimpleQueue` and :class:`JoinableQueue` types are multi-producer, |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 633 | multi-consumer FIFO queues modelled on the :class:`queue.Queue` class in the |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 634 | standard library. They differ in that :class:`Queue` lacks the |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 635 | :meth:`~queue.Queue.task_done` and :meth:`~queue.Queue.join` methods introduced |
| 636 | into Python 2.5's :class:`queue.Queue` class. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 637 | |
| 638 | If you use :class:`JoinableQueue` then you **must** call |
| 639 | :meth:`JoinableQueue.task_done` for each task removed from the queue or else the |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 640 | semaphore used to count the number of unfinished tasks may eventually overflow, |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 641 | raising an exception. |
| 642 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 643 | Note that one can also create a shared queue by using a manager object -- see |
| 644 | :ref:`multiprocessing-managers`. |
| 645 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 646 | .. note:: |
| 647 | |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 648 | :mod:`multiprocessing` uses the usual :exc:`queue.Empty` and |
| 649 | :exc:`queue.Full` exceptions to signal a timeout. They are not available in |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 650 | the :mod:`multiprocessing` namespace so you need to import them from |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 651 | :mod:`queue`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 652 | |
Richard Oudkerk | 95fe1a7 | 2013-06-24 14:48:07 +0100 | [diff] [blame] | 653 | .. note:: |
| 654 | |
| 655 | When an object is put on a queue, the object is pickled and a |
| 656 | background thread later flushes the pickled data to an underlying |
| 657 | pipe. This has some consequences which are a little surprising, |
Richard Oudkerk | 7b69da7 | 2013-06-24 18:12:57 +0100 | [diff] [blame] | 658 | but should not cause any practical difficulties -- if they really |
| 659 | bother you then you can instead use a queue created with a |
| 660 | :ref:`manager <multiprocessing-managers>`. |
Richard Oudkerk | 95fe1a7 | 2013-06-24 14:48:07 +0100 | [diff] [blame] | 661 | |
| 662 | (1) After putting an object on an empty queue there may be an |
Richard Oudkerk | 2b310dd | 2013-06-24 20:38:46 +0100 | [diff] [blame] | 663 | infinitesimal delay before the queue's :meth:`~Queue.empty` |
Richard Oudkerk | 95fe1a7 | 2013-06-24 14:48:07 +0100 | [diff] [blame] | 664 | method returns :const:`False` and :meth:`~Queue.get_nowait` can |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 665 | return without raising :exc:`queue.Empty`. |
Richard Oudkerk | 95fe1a7 | 2013-06-24 14:48:07 +0100 | [diff] [blame] | 666 | |
| 667 | (2) If multiple processes are enqueuing objects, it is possible for |
| 668 | the objects to be received at the other end out-of-order. |
| 669 | However, objects enqueued by the same process will always be in |
| 670 | the expected order with respect to each other. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 671 | |
| 672 | .. warning:: |
| 673 | |
| 674 | If a process is killed using :meth:`Process.terminate` or :func:`os.kill` |
| 675 | while it is trying to use a :class:`Queue`, then the data in the queue is |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 676 | likely to become corrupted. This may cause any other process to get an |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 677 | exception when it tries to use the queue later on. |
| 678 | |
| 679 | .. warning:: |
| 680 | |
| 681 | As mentioned above, if a child process has put items on a queue (and it has |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 682 | not used :meth:`JoinableQueue.cancel_join_thread |
| 683 | <multiprocessing.Queue.cancel_join_thread>`), then that process will |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 684 | not terminate until all buffered items have been flushed to the pipe. |
| 685 | |
| 686 | This means that if you try joining that process you may get a deadlock unless |
| 687 | you are sure that all items which have been put on the queue have been |
| 688 | consumed. Similarly, if the child process is non-daemonic then the parent |
Georg Brandl | 2ee470f | 2008-07-16 12:55:28 +0000 | [diff] [blame] | 689 | process may hang on exit when it tries to join all its non-daemonic children. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 690 | |
| 691 | Note that a queue created using a manager does not have this issue. See |
| 692 | :ref:`multiprocessing-programming`. |
| 693 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 694 | For an example of the usage of queues for interprocess communication see |
| 695 | :ref:`multiprocessing-examples`. |
| 696 | |
| 697 | |
| 698 | .. function:: Pipe([duplex]) |
| 699 | |
| 700 | Returns a pair ``(conn1, conn2)`` of :class:`Connection` objects representing |
| 701 | the ends of a pipe. |
| 702 | |
| 703 | If *duplex* is ``True`` (the default) then the pipe is bidirectional. If |
| 704 | *duplex* is ``False`` then the pipe is unidirectional: ``conn1`` can only be |
| 705 | used for receiving messages and ``conn2`` can only be used for sending |
| 706 | messages. |
| 707 | |
| 708 | |
| 709 | .. class:: Queue([maxsize]) |
| 710 | |
| 711 | Returns a process shared queue implemented using a pipe and a few |
| 712 | locks/semaphores. When a process first puts an item on the queue a feeder |
| 713 | thread is started which transfers objects from a buffer into the pipe. |
| 714 | |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 715 | The usual :exc:`queue.Empty` and :exc:`queue.Full` exceptions from the |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 716 | standard library's :mod:`queue` module are raised to signal timeouts. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 717 | |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 718 | :class:`Queue` implements all the methods of :class:`queue.Queue` except for |
| 719 | :meth:`~queue.Queue.task_done` and :meth:`~queue.Queue.join`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 720 | |
| 721 | .. method:: qsize() |
| 722 | |
| 723 | Return the approximate size of the queue. Because of |
| 724 | multithreading/multiprocessing semantics, this number is not reliable. |
| 725 | |
| 726 | Note that this may raise :exc:`NotImplementedError` on Unix platforms like |
Georg Brandl | c575c90 | 2008-09-13 17:46:05 +0000 | [diff] [blame] | 727 | Mac OS X where ``sem_getvalue()`` is not implemented. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 728 | |
| 729 | .. method:: empty() |
| 730 | |
| 731 | Return ``True`` if the queue is empty, ``False`` otherwise. Because of |
| 732 | multithreading/multiprocessing semantics, this is not reliable. |
| 733 | |
| 734 | .. method:: full() |
| 735 | |
| 736 | Return ``True`` if the queue is full, ``False`` otherwise. Because of |
| 737 | multithreading/multiprocessing semantics, this is not reliable. |
| 738 | |
Senthil Kumaran | e969a21 | 2011-09-06 00:21:30 +0800 | [diff] [blame] | 739 | .. method:: put(obj[, block[, timeout]]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 740 | |
Senthil Kumaran | e969a21 | 2011-09-06 00:21:30 +0800 | [diff] [blame] | 741 | Put obj into the queue. If the optional argument *block* is ``True`` |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 742 | (the default) and *timeout* is ``None`` (the default), block if necessary until |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 743 | a free slot is available. If *timeout* is a positive number, it blocks at |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 744 | most *timeout* seconds and raises the :exc:`queue.Full` exception if no |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 745 | free slot was available within that time. Otherwise (*block* is |
| 746 | ``False``), put an item on the queue if a free slot is immediately |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 747 | available, else raise the :exc:`queue.Full` exception (*timeout* is |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 748 | ignored in that case). |
| 749 | |
Senthil Kumaran | e969a21 | 2011-09-06 00:21:30 +0800 | [diff] [blame] | 750 | .. method:: put_nowait(obj) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 751 | |
Senthil Kumaran | e969a21 | 2011-09-06 00:21:30 +0800 | [diff] [blame] | 752 | Equivalent to ``put(obj, False)``. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 753 | |
| 754 | .. method:: get([block[, timeout]]) |
| 755 | |
| 756 | Remove and return an item from the queue. If optional args *block* is |
| 757 | ``True`` (the default) and *timeout* is ``None`` (the default), block if |
| 758 | necessary until an item is available. If *timeout* is a positive number, |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 759 | it blocks at most *timeout* seconds and raises the :exc:`queue.Empty` |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 760 | exception if no item was available within that time. Otherwise (block is |
| 761 | ``False``), return an item if one is immediately available, else raise the |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 762 | :exc:`queue.Empty` exception (*timeout* is ignored in that case). |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 763 | |
| 764 | .. method:: get_nowait() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 765 | |
| 766 | Equivalent to ``get(False)``. |
| 767 | |
| 768 | :class:`multiprocessing.Queue` has a few additional methods not found in |
Georg Brandl | 2ee470f | 2008-07-16 12:55:28 +0000 | [diff] [blame] | 769 | :class:`queue.Queue`. These methods are usually unnecessary for most |
| 770 | code: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 771 | |
| 772 | .. method:: close() |
| 773 | |
| 774 | Indicate that no more data will be put on this queue by the current |
| 775 | process. The background thread will quit once it has flushed all buffered |
| 776 | data to the pipe. This is called automatically when the queue is garbage |
| 777 | collected. |
| 778 | |
| 779 | .. method:: join_thread() |
| 780 | |
| 781 | Join the background thread. This can only be used after :meth:`close` has |
| 782 | been called. It blocks until the background thread exits, ensuring that |
| 783 | all data in the buffer has been flushed to the pipe. |
| 784 | |
| 785 | By default if a process is not the creator of the queue then on exit it |
| 786 | will attempt to join the queue's background thread. The process can call |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 787 | :meth:`cancel_join_thread` to make :meth:`join_thread` do nothing. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 788 | |
| 789 | .. method:: cancel_join_thread() |
| 790 | |
| 791 | Prevent :meth:`join_thread` from blocking. In particular, this prevents |
| 792 | the background thread from being joined automatically when the process |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 793 | exits -- see :meth:`join_thread`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 794 | |
Richard Oudkerk | d7d3f37 | 2013-07-02 12:59:55 +0100 | [diff] [blame] | 795 | A better name for this method might be |
| 796 | ``allow_exit_without_flush()``. It is likely to cause enqueued |
| 797 | data to lost, and you almost certainly will not need to use it. |
| 798 | It is really only there if you need the current process to exit |
| 799 | immediately without waiting to flush enqueued data to the |
| 800 | underlying pipe, and you don't care about lost data. |
| 801 | |
Berker Peksag | 7ecfc82 | 2015-04-08 17:56:30 +0300 | [diff] [blame] | 802 | .. note:: |
| 803 | |
| 804 | This class's functionality requires a functioning shared semaphore |
| 805 | implementation on the host operating system. Without one, the |
| 806 | functionality in this class will be disabled, and attempts to |
| 807 | instantiate a :class:`Queue` will result in an :exc:`ImportError`. See |
| 808 | :issue:`3770` for additional information. The same holds true for any |
| 809 | of the specialized queue types listed below. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 810 | |
Sandro Tosi | cd77815 | 2012-02-15 23:27:00 +0100 | [diff] [blame] | 811 | .. class:: SimpleQueue() |
Sandro Tosi | 5cb522c | 2012-02-15 23:14:21 +0100 | [diff] [blame] | 812 | |
| 813 | It is a simplified :class:`Queue` type, very close to a locked :class:`Pipe`. |
| 814 | |
| 815 | .. method:: empty() |
| 816 | |
| 817 | Return ``True`` if the queue is empty, ``False`` otherwise. |
| 818 | |
| 819 | .. method:: get() |
| 820 | |
| 821 | Remove and return an item from the queue. |
| 822 | |
| 823 | .. method:: put(item) |
| 824 | |
| 825 | Put *item* into the queue. |
| 826 | |
| 827 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 828 | .. class:: JoinableQueue([maxsize]) |
| 829 | |
| 830 | :class:`JoinableQueue`, a :class:`Queue` subclass, is a queue which |
| 831 | additionally has :meth:`task_done` and :meth:`join` methods. |
| 832 | |
| 833 | .. method:: task_done() |
| 834 | |
Eli Bendersky | 78da3bc | 2012-07-13 10:10:05 +0300 | [diff] [blame] | 835 | Indicate that a formerly enqueued task is complete. Used by queue |
| 836 | consumers. For each :meth:`~Queue.get` used to fetch a task, a subsequent |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 837 | call to :meth:`task_done` tells the queue that the processing on the task |
| 838 | is complete. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 839 | |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 840 | If a :meth:`~queue.Queue.join` is currently blocking, it will resume when all |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 841 | items have been processed (meaning that a :meth:`task_done` call was |
| 842 | received for every item that had been :meth:`~Queue.put` into the queue). |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 843 | |
| 844 | Raises a :exc:`ValueError` if called more times than there were items |
| 845 | placed in the queue. |
| 846 | |
| 847 | |
| 848 | .. method:: join() |
| 849 | |
| 850 | Block until all items in the queue have been gotten and processed. |
| 851 | |
| 852 | The count of unfinished tasks goes up whenever an item is added to the |
Eli Bendersky | 78da3bc | 2012-07-13 10:10:05 +0300 | [diff] [blame] | 853 | queue. The count goes down whenever a consumer calls |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 854 | :meth:`task_done` to indicate that the item was retrieved and all work on |
| 855 | it is complete. When the count of unfinished tasks drops to zero, |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 856 | :meth:`~queue.Queue.join` unblocks. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 857 | |
| 858 | |
| 859 | Miscellaneous |
| 860 | ~~~~~~~~~~~~~ |
| 861 | |
| 862 | .. function:: active_children() |
| 863 | |
| 864 | Return list of all live children of the current process. |
| 865 | |
Zachary Ware | 7280561 | 2014-10-03 10:55:12 -0500 | [diff] [blame] | 866 | Calling this has the side effect of "joining" any processes which have |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 867 | already finished. |
| 868 | |
| 869 | .. function:: cpu_count() |
| 870 | |
| 871 | Return the number of CPUs in the system. May raise |
| 872 | :exc:`NotImplementedError`. |
| 873 | |
Charles-Francois Natali | 44feda3 | 2013-05-20 14:40:46 +0200 | [diff] [blame] | 874 | .. seealso:: |
| 875 | :func:`os.cpu_count` |
| 876 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 877 | .. function:: current_process() |
| 878 | |
| 879 | Return the :class:`Process` object corresponding to the current process. |
| 880 | |
| 881 | An analogue of :func:`threading.current_thread`. |
| 882 | |
| 883 | .. function:: freeze_support() |
| 884 | |
| 885 | Add support for when a program which uses :mod:`multiprocessing` has been |
| 886 | frozen to produce a Windows executable. (Has been tested with **py2exe**, |
| 887 | **PyInstaller** and **cx_Freeze**.) |
| 888 | |
| 889 | One needs to call this function straight after the ``if __name__ == |
| 890 | '__main__'`` line of the main module. For example:: |
| 891 | |
| 892 | from multiprocessing import Process, freeze_support |
| 893 | |
| 894 | def f(): |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 895 | print('hello world!') |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 896 | |
| 897 | if __name__ == '__main__': |
| 898 | freeze_support() |
| 899 | Process(target=f).start() |
| 900 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 901 | If the ``freeze_support()`` line is omitted then trying to run the frozen |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 902 | executable will raise :exc:`RuntimeError`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 903 | |
| 904 | If the module is being run normally by the Python interpreter then |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 905 | :func:`freeze_support` has no effect. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 906 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 907 | .. function:: get_all_start_methods() |
| 908 | |
| 909 | Returns a list of the supported start methods, the first of which |
| 910 | is the default. The possible start methods are ``'fork'``, |
| 911 | ``'spawn'`` and ``'forkserver'``. On Windows only ``'spawn'`` is |
| 912 | available. On Unix ``'fork'`` and ``'spawn'`` are always |
| 913 | supported, with ``'fork'`` being the default. |
| 914 | |
| 915 | .. versionadded:: 3.4 |
| 916 | |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 917 | .. function:: get_context(method=None) |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 918 | |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 919 | Return a context object which has the same attributes as the |
| 920 | :mod:`multiprocessing` module. |
| 921 | |
| 922 | If *method* is *None* then the default context is returned. |
| 923 | Otherwise *method* should be ``'fork'``, ``'spawn'``, |
| 924 | ``'forkserver'``. :exc:`ValueError` is raised if the specified |
| 925 | start method is not available. |
| 926 | |
| 927 | .. versionadded:: 3.4 |
| 928 | |
| 929 | .. function:: get_start_method(allow_none=False) |
| 930 | |
| 931 | Return the name of start method used for starting processes. |
| 932 | |
| 933 | If the start method has not been fixed and *allow_none* is false, |
| 934 | then the start method is fixed to the default and the name is |
| 935 | returned. If the start method has not been fixed and *allow_none* |
| 936 | is true then *None* is returned. |
| 937 | |
| 938 | The return value can be ``'fork'``, ``'spawn'``, ``'forkserver'`` |
| 939 | or *None*. ``'fork'`` is the default on Unix, while ``'spawn'`` is |
| 940 | the default on Windows. |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 941 | |
| 942 | .. versionadded:: 3.4 |
| 943 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 944 | .. function:: set_executable() |
| 945 | |
Ezio Melotti | 0639d5a | 2009-12-19 23:26:38 +0000 | [diff] [blame] | 946 | Sets the path of the Python interpreter to use when starting a child process. |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 947 | (By default :data:`sys.executable` is used). Embedders will probably need to |
| 948 | do some thing like :: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 949 | |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 950 | set_executable(os.path.join(sys.exec_prefix, 'pythonw.exe')) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 951 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 952 | before they can create child processes. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 953 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 954 | .. versionchanged:: 3.4 |
| 955 | Now supported on Unix when the ``'spawn'`` start method is used. |
| 956 | |
| 957 | .. function:: set_start_method(method) |
| 958 | |
| 959 | Set the method which should be used to start child processes. |
| 960 | *method* can be ``'fork'``, ``'spawn'`` or ``'forkserver'``. |
| 961 | |
| 962 | Note that this should be called at most once, and it should be |
| 963 | protected inside the ``if __name__ == '__main__'`` clause of the |
| 964 | main module. |
| 965 | |
| 966 | .. versionadded:: 3.4 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 967 | |
| 968 | .. note:: |
| 969 | |
| 970 | :mod:`multiprocessing` contains no analogues of |
| 971 | :func:`threading.active_count`, :func:`threading.enumerate`, |
| 972 | :func:`threading.settrace`, :func:`threading.setprofile`, |
| 973 | :class:`threading.Timer`, or :class:`threading.local`. |
| 974 | |
| 975 | |
| 976 | Connection Objects |
| 977 | ~~~~~~~~~~~~~~~~~~ |
| 978 | |
| 979 | Connection objects allow the sending and receiving of picklable objects or |
| 980 | strings. They can be thought of as message oriented connected sockets. |
| 981 | |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 982 | Connection objects are usually created using :func:`Pipe` -- see also |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 983 | :ref:`multiprocessing-listeners-clients`. |
| 984 | |
| 985 | .. class:: Connection |
| 986 | |
| 987 | .. method:: send(obj) |
| 988 | |
| 989 | Send an object to the other end of the connection which should be read |
| 990 | using :meth:`recv`. |
| 991 | |
Benjamin Peterson | 965ce87 | 2009-04-05 21:24:58 +0000 | [diff] [blame] | 992 | The object must be picklable. Very large pickles (approximately 32 MB+, |
| 993 | though it depends on the OS) may raise a ValueError exception. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 994 | |
| 995 | .. method:: recv() |
| 996 | |
| 997 | Return an object sent from the other end of the connection using |
Sandro Tosi | b52e7a9 | 2012-01-07 17:56:58 +0100 | [diff] [blame] | 998 | :meth:`send`. Blocks until there its something to receive. Raises |
| 999 | :exc:`EOFError` if there is nothing left to receive |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1000 | and the other end was closed. |
| 1001 | |
| 1002 | .. method:: fileno() |
| 1003 | |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 1004 | Return the file descriptor or handle used by the connection. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1005 | |
| 1006 | .. method:: close() |
| 1007 | |
| 1008 | Close the connection. |
| 1009 | |
| 1010 | This is called automatically when the connection is garbage collected. |
| 1011 | |
| 1012 | .. method:: poll([timeout]) |
| 1013 | |
| 1014 | Return whether there is any data available to be read. |
| 1015 | |
| 1016 | If *timeout* is not specified then it will return immediately. If |
| 1017 | *timeout* is a number then this specifies the maximum time in seconds to |
| 1018 | block. If *timeout* is ``None`` then an infinite timeout is used. |
| 1019 | |
Antoine Pitrou | bdb1cf1 | 2012-03-05 19:28:37 +0100 | [diff] [blame] | 1020 | Note that multiple connection objects may be polled at once by |
| 1021 | using :func:`multiprocessing.connection.wait`. |
| 1022 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1023 | .. method:: send_bytes(buffer[, offset[, size]]) |
| 1024 | |
Ezio Melotti | c228e96 | 2013-05-04 18:06:34 +0300 | [diff] [blame] | 1025 | Send byte data from a :term:`bytes-like object` as a complete message. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1026 | |
| 1027 | If *offset* is given then data is read from that position in *buffer*. If |
Benjamin Peterson | 965ce87 | 2009-04-05 21:24:58 +0000 | [diff] [blame] | 1028 | *size* is given then that many bytes will be read from buffer. Very large |
| 1029 | buffers (approximately 32 MB+, though it depends on the OS) may raise a |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 1030 | :exc:`ValueError` exception |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1031 | |
| 1032 | .. method:: recv_bytes([maxlength]) |
| 1033 | |
| 1034 | Return a complete message of byte data sent from the other end of the |
Sandro Tosi | b52e7a9 | 2012-01-07 17:56:58 +0100 | [diff] [blame] | 1035 | connection as a string. Blocks until there is something to receive. |
| 1036 | Raises :exc:`EOFError` if there is nothing left |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1037 | to receive and the other end has closed. |
| 1038 | |
| 1039 | If *maxlength* is specified and the message is longer than *maxlength* |
Antoine Pitrou | 62ab10a0 | 2011-10-12 20:10:51 +0200 | [diff] [blame] | 1040 | then :exc:`OSError` is raised and the connection will no longer be |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1041 | readable. |
| 1042 | |
Antoine Pitrou | 62ab10a0 | 2011-10-12 20:10:51 +0200 | [diff] [blame] | 1043 | .. versionchanged:: 3.3 |
Martin Panter | 7462b649 | 2015-11-02 03:37:02 +0000 | [diff] [blame] | 1044 | This function used to raise :exc:`IOError`, which is now an |
Antoine Pitrou | 62ab10a0 | 2011-10-12 20:10:51 +0200 | [diff] [blame] | 1045 | alias of :exc:`OSError`. |
| 1046 | |
| 1047 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1048 | .. method:: recv_bytes_into(buffer[, offset]) |
| 1049 | |
| 1050 | Read into *buffer* a complete message of byte data sent from the other end |
Sandro Tosi | b52e7a9 | 2012-01-07 17:56:58 +0100 | [diff] [blame] | 1051 | of the connection and return the number of bytes in the message. Blocks |
| 1052 | until there is something to receive. Raises |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1053 | :exc:`EOFError` if there is nothing left to receive and the other end was |
| 1054 | closed. |
| 1055 | |
Ezio Melotti | c228e96 | 2013-05-04 18:06:34 +0300 | [diff] [blame] | 1056 | *buffer* must be a writable :term:`bytes-like object`. If |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1057 | *offset* is given then the message will be written into the buffer from |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1058 | that position. Offset must be a non-negative integer less than the |
| 1059 | length of *buffer* (in bytes). |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1060 | |
| 1061 | If the buffer is too short then a :exc:`BufferTooShort` exception is |
| 1062 | raised and the complete message is available as ``e.args[0]`` where ``e`` |
| 1063 | is the exception instance. |
| 1064 | |
Antoine Pitrou | 5438ed1 | 2012-04-24 22:56:57 +0200 | [diff] [blame] | 1065 | .. versionchanged:: 3.3 |
| 1066 | Connection objects themselves can now be transferred between processes |
| 1067 | using :meth:`Connection.send` and :meth:`Connection.recv`. |
| 1068 | |
Richard Oudkerk | d69cfe8 | 2012-06-18 17:47:52 +0100 | [diff] [blame] | 1069 | .. versionadded:: 3.3 |
Serhiy Storchaka | 1486799 | 2014-09-10 23:43:41 +0300 | [diff] [blame] | 1070 | Connection objects now support the context management protocol -- see |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1071 | :ref:`typecontextmanager`. :meth:`~contextmanager.__enter__` returns the |
| 1072 | connection object, and :meth:`~contextmanager.__exit__` calls :meth:`close`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1073 | |
| 1074 | For example: |
| 1075 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1076 | .. doctest:: |
| 1077 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1078 | >>> from multiprocessing import Pipe |
| 1079 | >>> a, b = Pipe() |
| 1080 | >>> a.send([1, 'hello', None]) |
| 1081 | >>> b.recv() |
| 1082 | [1, 'hello', None] |
Georg Brandl | 3017689 | 2010-10-29 05:22:17 +0000 | [diff] [blame] | 1083 | >>> b.send_bytes(b'thank you') |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1084 | >>> a.recv_bytes() |
Georg Brandl | 3017689 | 2010-10-29 05:22:17 +0000 | [diff] [blame] | 1085 | b'thank you' |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1086 | >>> import array |
| 1087 | >>> arr1 = array.array('i', range(5)) |
| 1088 | >>> arr2 = array.array('i', [0] * 10) |
| 1089 | >>> a.send_bytes(arr1) |
| 1090 | >>> count = b.recv_bytes_into(arr2) |
| 1091 | >>> assert count == len(arr1) * arr1.itemsize |
| 1092 | >>> arr2 |
| 1093 | array('i', [0, 1, 2, 3, 4, 0, 0, 0, 0, 0]) |
| 1094 | |
| 1095 | |
| 1096 | .. warning:: |
| 1097 | |
| 1098 | The :meth:`Connection.recv` method automatically unpickles the data it |
| 1099 | receives, which can be a security risk unless you can trust the process |
| 1100 | which sent the message. |
| 1101 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1102 | Therefore, unless the connection object was produced using :func:`Pipe` you |
| 1103 | should only use the :meth:`~Connection.recv` and :meth:`~Connection.send` |
| 1104 | methods after performing some sort of authentication. See |
| 1105 | :ref:`multiprocessing-auth-keys`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1106 | |
| 1107 | .. warning:: |
| 1108 | |
| 1109 | If a process is killed while it is trying to read or write to a pipe then |
| 1110 | the data in the pipe is likely to become corrupted, because it may become |
| 1111 | impossible to be sure where the message boundaries lie. |
| 1112 | |
| 1113 | |
| 1114 | Synchronization primitives |
| 1115 | ~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 1116 | |
| 1117 | Generally synchronization primitives are not as necessary in a multiprocess |
Georg Brandl | 2ee470f | 2008-07-16 12:55:28 +0000 | [diff] [blame] | 1118 | program as they are in a multithreaded program. See the documentation for |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1119 | :mod:`threading` module. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1120 | |
| 1121 | Note that one can also create synchronization primitives by using a manager |
| 1122 | object -- see :ref:`multiprocessing-managers`. |
| 1123 | |
Richard Oudkerk | 3730a17 | 2012-06-15 18:26:07 +0100 | [diff] [blame] | 1124 | .. class:: Barrier(parties[, action[, timeout]]) |
| 1125 | |
| 1126 | A barrier object: a clone of :class:`threading.Barrier`. |
| 1127 | |
| 1128 | .. versionadded:: 3.3 |
| 1129 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1130 | .. class:: BoundedSemaphore([value]) |
| 1131 | |
Berker Peksag | 407c497 | 2015-09-21 06:50:55 +0300 | [diff] [blame] | 1132 | A bounded semaphore object: a close analog of |
| 1133 | :class:`threading.BoundedSemaphore`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1134 | |
Berker Peksag | 407c497 | 2015-09-21 06:50:55 +0300 | [diff] [blame] | 1135 | A solitary difference from its close analog exists: its ``acquire`` method's |
| 1136 | first argument is named *block*, as is consistent with :meth:`Lock.acquire`. |
| 1137 | |
| 1138 | .. note:: |
| 1139 | On Mac OS X, this is indistinguishable from :class:`Semaphore` because |
| 1140 | ``sem_getvalue()`` is not implemented on that platform. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1141 | |
| 1142 | .. class:: Condition([lock]) |
| 1143 | |
R David Murray | ef4d286 | 2012-10-06 14:35:35 -0400 | [diff] [blame] | 1144 | A condition variable: an alias for :class:`threading.Condition`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1145 | |
| 1146 | If *lock* is specified then it should be a :class:`Lock` or :class:`RLock` |
| 1147 | object from :mod:`multiprocessing`. |
| 1148 | |
Charles-François Natali | c8ce715 | 2012-04-17 18:45:57 +0200 | [diff] [blame] | 1149 | .. versionchanged:: 3.3 |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1150 | The :meth:`~threading.Condition.wait_for` method was added. |
Charles-François Natali | c8ce715 | 2012-04-17 18:45:57 +0200 | [diff] [blame] | 1151 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1152 | .. class:: Event() |
| 1153 | |
| 1154 | A clone of :class:`threading.Event`. |
| 1155 | |
Berker Peksag | 407c497 | 2015-09-21 06:50:55 +0300 | [diff] [blame] | 1156 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1157 | .. class:: Lock() |
| 1158 | |
Berker Peksag | 407c497 | 2015-09-21 06:50:55 +0300 | [diff] [blame] | 1159 | A non-recursive lock object: a close analog of :class:`threading.Lock`. |
| 1160 | Once a process or thread has acquired a lock, subsequent attempts to |
| 1161 | acquire it from any process or thread will block until it is released; |
| 1162 | any process or thread may release it. The concepts and behaviors of |
| 1163 | :class:`threading.Lock` as it applies to threads are replicated here in |
| 1164 | :class:`multiprocessing.Lock` as it applies to either processes or threads, |
| 1165 | except as noted. |
| 1166 | |
| 1167 | Note that :class:`Lock` is actually a factory function which returns an |
| 1168 | instance of ``multiprocessing.synchronize.Lock`` initialized with a |
| 1169 | default context. |
| 1170 | |
| 1171 | :class:`Lock` supports the :term:`context manager` protocol and thus may be |
| 1172 | used in :keyword:`with` statements. |
| 1173 | |
| 1174 | .. method:: acquire(block=True, timeout=None) |
| 1175 | |
| 1176 | Acquire a lock, blocking or non-blocking. |
| 1177 | |
| 1178 | With the *block* argument set to ``True`` (the default), the method call |
| 1179 | will block until the lock is in an unlocked state, then set it to locked |
| 1180 | and return ``True``. Note that the name of this first argument differs |
| 1181 | from that in :meth:`threading.Lock.acquire`. |
| 1182 | |
| 1183 | With the *block* argument set to ``False``, the method call does not |
| 1184 | block. If the lock is currently in a locked state, return ``False``; |
| 1185 | otherwise set the lock to a locked state and return ``True``. |
| 1186 | |
| 1187 | When invoked with a positive, floating-point value for *timeout*, block |
| 1188 | for at most the number of seconds specified by *timeout* as long as |
| 1189 | the lock can not be acquired. Invocations with a negative value for |
| 1190 | *timeout* are equivalent to a *timeout* of zero. Invocations with a |
| 1191 | *timeout* value of ``None`` (the default) set the timeout period to |
| 1192 | infinite. Note that the treatment of negative or ``None`` values for |
| 1193 | *timeout* differs from the implemented behavior in |
| 1194 | :meth:`threading.Lock.acquire`. The *timeout* argument has no practical |
| 1195 | implications if the *block* argument is set to ``False`` and is thus |
| 1196 | ignored. Returns ``True`` if the lock has been acquired or ``False`` if |
| 1197 | the timeout period has elapsed. |
| 1198 | |
| 1199 | |
| 1200 | .. method:: release() |
| 1201 | |
| 1202 | Release a lock. This can be called from any process or thread, not only |
| 1203 | the process or thread which originally acquired the lock. |
| 1204 | |
| 1205 | Behavior is the same as in :meth:`threading.Lock.release` except that |
| 1206 | when invoked on an unlocked lock, a :exc:`ValueError` is raised. |
| 1207 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1208 | |
| 1209 | .. class:: RLock() |
| 1210 | |
Berker Peksag | 407c497 | 2015-09-21 06:50:55 +0300 | [diff] [blame] | 1211 | A recursive lock object: a close analog of :class:`threading.RLock`. A |
| 1212 | recursive lock must be released by the process or thread that acquired it. |
| 1213 | Once a process or thread has acquired a recursive lock, the same process |
| 1214 | or thread may acquire it again without blocking; that process or thread |
| 1215 | must release it once for each time it has been acquired. |
| 1216 | |
| 1217 | Note that :class:`RLock` is actually a factory function which returns an |
| 1218 | instance of ``multiprocessing.synchronize.RLock`` initialized with a |
| 1219 | default context. |
| 1220 | |
| 1221 | :class:`RLock` supports the :term:`context manager` protocol and thus may be |
| 1222 | used in :keyword:`with` statements. |
| 1223 | |
| 1224 | |
| 1225 | .. method:: acquire(block=True, timeout=None) |
| 1226 | |
| 1227 | Acquire a lock, blocking or non-blocking. |
| 1228 | |
| 1229 | When invoked with the *block* argument set to ``True``, block until the |
| 1230 | lock is in an unlocked state (not owned by any process or thread) unless |
| 1231 | the lock is already owned by the current process or thread. The current |
| 1232 | process or thread then takes ownership of the lock (if it does not |
| 1233 | already have ownership) and the recursion level inside the lock increments |
| 1234 | by one, resulting in a return value of ``True``. Note that there are |
| 1235 | several differences in this first argument's behavior compared to the |
| 1236 | implementation of :meth:`threading.RLock.acquire`, starting with the name |
| 1237 | of the argument itself. |
| 1238 | |
| 1239 | When invoked with the *block* argument set to ``False``, do not block. |
| 1240 | If the lock has already been acquired (and thus is owned) by another |
| 1241 | process or thread, the current process or thread does not take ownership |
| 1242 | and the recursion level within the lock is not changed, resulting in |
| 1243 | a return value of ``False``. If the lock is in an unlocked state, the |
| 1244 | current process or thread takes ownership and the recursion level is |
| 1245 | incremented, resulting in a return value of ``True``. |
| 1246 | |
| 1247 | Use and behaviors of the *timeout* argument are the same as in |
| 1248 | :meth:`Lock.acquire`. Note that some of these behaviors of *timeout* |
| 1249 | differ from the implemented behaviors in :meth:`threading.RLock.acquire`. |
| 1250 | |
| 1251 | |
| 1252 | .. method:: release() |
| 1253 | |
| 1254 | Release a lock, decrementing the recursion level. If after the |
| 1255 | decrement the recursion level is zero, reset the lock to unlocked (not |
| 1256 | owned by any process or thread) and if any other processes or threads |
| 1257 | are blocked waiting for the lock to become unlocked, allow exactly one |
| 1258 | of them to proceed. If after the decrement the recursion level is still |
| 1259 | nonzero, the lock remains locked and owned by the calling process or |
| 1260 | thread. |
| 1261 | |
| 1262 | Only call this method when the calling process or thread owns the lock. |
| 1263 | An :exc:`AssertionError` is raised if this method is called by a process |
| 1264 | or thread other than the owner or if the lock is in an unlocked (unowned) |
| 1265 | state. Note that the type of exception raised in this situation |
| 1266 | differs from the implemented behavior in :meth:`threading.RLock.release`. |
| 1267 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1268 | |
| 1269 | .. class:: Semaphore([value]) |
| 1270 | |
Berker Peksag | 407c497 | 2015-09-21 06:50:55 +0300 | [diff] [blame] | 1271 | A semaphore object: a close analog of :class:`threading.Semaphore`. |
| 1272 | |
| 1273 | A solitary difference from its close analog exists: its ``acquire`` method's |
| 1274 | first argument is named *block*, as is consistent with :meth:`Lock.acquire`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1275 | |
| 1276 | .. note:: |
| 1277 | |
Georg Brandl | 592296e | 2010-05-21 21:48:27 +0000 | [diff] [blame] | 1278 | On Mac OS X, ``sem_timedwait`` is unsupported, so calling ``acquire()`` with |
| 1279 | a timeout will emulate that function's behavior using a sleeping loop. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1280 | |
| 1281 | .. note:: |
| 1282 | |
Serhiy Storchaka | 0424eaf | 2015-09-12 17:45:25 +0300 | [diff] [blame] | 1283 | If the SIGINT signal generated by :kbd:`Ctrl-C` arrives while the main thread is |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1284 | blocked by a call to :meth:`BoundedSemaphore.acquire`, :meth:`Lock.acquire`, |
| 1285 | :meth:`RLock.acquire`, :meth:`Semaphore.acquire`, :meth:`Condition.acquire` |
| 1286 | or :meth:`Condition.wait` then the call will be immediately interrupted and |
| 1287 | :exc:`KeyboardInterrupt` will be raised. |
| 1288 | |
| 1289 | This differs from the behaviour of :mod:`threading` where SIGINT will be |
| 1290 | ignored while the equivalent blocking calls are in progress. |
| 1291 | |
Berker Peksag | 7ecfc82 | 2015-04-08 17:56:30 +0300 | [diff] [blame] | 1292 | .. note:: |
| 1293 | |
| 1294 | Some of this package's functionality requires a functioning shared semaphore |
| 1295 | implementation on the host operating system. Without one, the |
| 1296 | :mod:`multiprocessing.synchronize` module will be disabled, and attempts to |
| 1297 | import it will result in an :exc:`ImportError`. See |
| 1298 | :issue:`3770` for additional information. |
| 1299 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1300 | |
| 1301 | Shared :mod:`ctypes` Objects |
| 1302 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 1303 | |
| 1304 | It is possible to create shared objects using shared memory which can be |
| 1305 | inherited by child processes. |
| 1306 | |
Richard Oudkerk | 87ea780 | 2012-05-29 12:01:47 +0100 | [diff] [blame] | 1307 | .. function:: Value(typecode_or_type, *args, lock=True) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1308 | |
| 1309 | Return a :mod:`ctypes` object allocated from shared memory. By default the |
Eli Bendersky | 78da3bc | 2012-07-13 10:10:05 +0300 | [diff] [blame] | 1310 | return value is actually a synchronized wrapper for the object. The object |
| 1311 | itself can be accessed via the *value* attribute of a :class:`Value`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1312 | |
| 1313 | *typecode_or_type* determines the type of the returned object: it is either a |
| 1314 | ctypes type or a one character typecode of the kind used by the :mod:`array` |
| 1315 | module. *\*args* is passed on to the constructor for the type. |
| 1316 | |
Richard Oudkerk | edcf8da | 2013-11-17 17:00:38 +0000 | [diff] [blame] | 1317 | If *lock* is ``True`` (the default) then a new recursive lock |
| 1318 | object is created to synchronize access to the value. If *lock* is |
| 1319 | a :class:`Lock` or :class:`RLock` object then that will be used to |
| 1320 | synchronize access to the value. If *lock* is ``False`` then |
| 1321 | access to the returned object will not be automatically protected |
| 1322 | by a lock, so it will not necessarily be "process-safe". |
| 1323 | |
| 1324 | Operations like ``+=`` which involve a read and write are not |
| 1325 | atomic. So if, for instance, you want to atomically increment a |
| 1326 | shared value it is insufficient to just do :: |
| 1327 | |
| 1328 | counter.value += 1 |
| 1329 | |
| 1330 | Assuming the associated lock is recursive (which it is by default) |
| 1331 | you can instead do :: |
| 1332 | |
| 1333 | with counter.get_lock(): |
| 1334 | counter.value += 1 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1335 | |
| 1336 | Note that *lock* is a keyword-only argument. |
| 1337 | |
| 1338 | .. function:: Array(typecode_or_type, size_or_initializer, *, lock=True) |
| 1339 | |
| 1340 | Return a ctypes array allocated from shared memory. By default the return |
| 1341 | value is actually a synchronized wrapper for the array. |
| 1342 | |
| 1343 | *typecode_or_type* determines the type of the elements of the returned array: |
| 1344 | it is either a ctypes type or a one character typecode of the kind used by |
| 1345 | the :mod:`array` module. If *size_or_initializer* is an integer, then it |
| 1346 | determines the length of the array, and the array will be initially zeroed. |
| 1347 | Otherwise, *size_or_initializer* is a sequence which is used to initialize |
| 1348 | the array and whose length determines the length of the array. |
| 1349 | |
| 1350 | If *lock* is ``True`` (the default) then a new lock object is created to |
| 1351 | synchronize access to the value. If *lock* is a :class:`Lock` or |
| 1352 | :class:`RLock` object then that will be used to synchronize access to the |
| 1353 | value. If *lock* is ``False`` then access to the returned object will not be |
| 1354 | automatically protected by a lock, so it will not necessarily be |
| 1355 | "process-safe". |
| 1356 | |
| 1357 | Note that *lock* is a keyword only argument. |
| 1358 | |
Amaury Forgeot d'Arc | b0c2916 | 2008-11-22 22:18:04 +0000 | [diff] [blame] | 1359 | Note that an array of :data:`ctypes.c_char` has *value* and *raw* |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1360 | attributes which allow one to use it to store and retrieve strings. |
| 1361 | |
| 1362 | |
| 1363 | The :mod:`multiprocessing.sharedctypes` module |
| 1364 | >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> |
| 1365 | |
| 1366 | .. module:: multiprocessing.sharedctypes |
| 1367 | :synopsis: Allocate ctypes objects from shared memory. |
| 1368 | |
| 1369 | The :mod:`multiprocessing.sharedctypes` module provides functions for allocating |
| 1370 | :mod:`ctypes` objects from shared memory which can be inherited by child |
| 1371 | processes. |
| 1372 | |
| 1373 | .. note:: |
| 1374 | |
Georg Brandl | 2ee470f | 2008-07-16 12:55:28 +0000 | [diff] [blame] | 1375 | Although it is possible to store a pointer in shared memory remember that |
| 1376 | this will refer to a location in the address space of a specific process. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1377 | However, the pointer is quite likely to be invalid in the context of a second |
| 1378 | process and trying to dereference the pointer from the second process may |
| 1379 | cause a crash. |
| 1380 | |
| 1381 | .. function:: RawArray(typecode_or_type, size_or_initializer) |
| 1382 | |
| 1383 | Return a ctypes array allocated from shared memory. |
| 1384 | |
| 1385 | *typecode_or_type* determines the type of the elements of the returned array: |
| 1386 | it is either a ctypes type or a one character typecode of the kind used by |
| 1387 | the :mod:`array` module. If *size_or_initializer* is an integer then it |
| 1388 | determines the length of the array, and the array will be initially zeroed. |
| 1389 | Otherwise *size_or_initializer* is a sequence which is used to initialize the |
| 1390 | array and whose length determines the length of the array. |
| 1391 | |
| 1392 | Note that setting and getting an element is potentially non-atomic -- use |
| 1393 | :func:`Array` instead to make sure that access is automatically synchronized |
| 1394 | using a lock. |
| 1395 | |
| 1396 | .. function:: RawValue(typecode_or_type, *args) |
| 1397 | |
| 1398 | Return a ctypes object allocated from shared memory. |
| 1399 | |
| 1400 | *typecode_or_type* determines the type of the returned object: it is either a |
| 1401 | ctypes type or a one character typecode of the kind used by the :mod:`array` |
Jesse Noller | b0516a6 | 2009-01-18 03:11:38 +0000 | [diff] [blame] | 1402 | module. *\*args* is passed on to the constructor for the type. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1403 | |
| 1404 | Note that setting and getting the value is potentially non-atomic -- use |
| 1405 | :func:`Value` instead to make sure that access is automatically synchronized |
| 1406 | using a lock. |
| 1407 | |
Amaury Forgeot d'Arc | b0c2916 | 2008-11-22 22:18:04 +0000 | [diff] [blame] | 1408 | Note that an array of :data:`ctypes.c_char` has ``value`` and ``raw`` |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1409 | attributes which allow one to use it to store and retrieve strings -- see |
| 1410 | documentation for :mod:`ctypes`. |
| 1411 | |
Richard Oudkerk | 87ea780 | 2012-05-29 12:01:47 +0100 | [diff] [blame] | 1412 | .. function:: Array(typecode_or_type, size_or_initializer, *, lock=True) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1413 | |
| 1414 | The same as :func:`RawArray` except that depending on the value of *lock* a |
| 1415 | process-safe synchronization wrapper may be returned instead of a raw ctypes |
| 1416 | array. |
| 1417 | |
| 1418 | If *lock* is ``True`` (the default) then a new lock object is created to |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1419 | synchronize access to the value. If *lock* is a |
| 1420 | :class:`~multiprocessing.Lock` or :class:`~multiprocessing.RLock` object |
| 1421 | then that will be used to synchronize access to the |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1422 | value. If *lock* is ``False`` then access to the returned object will not be |
| 1423 | automatically protected by a lock, so it will not necessarily be |
| 1424 | "process-safe". |
| 1425 | |
| 1426 | Note that *lock* is a keyword-only argument. |
| 1427 | |
Richard Oudkerk | 87ea780 | 2012-05-29 12:01:47 +0100 | [diff] [blame] | 1428 | .. function:: Value(typecode_or_type, *args, lock=True) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1429 | |
| 1430 | The same as :func:`RawValue` except that depending on the value of *lock* a |
| 1431 | process-safe synchronization wrapper may be returned instead of a raw ctypes |
| 1432 | object. |
| 1433 | |
| 1434 | If *lock* is ``True`` (the default) then a new lock object is created to |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1435 | synchronize access to the value. If *lock* is a :class:`~multiprocessing.Lock` or |
| 1436 | :class:`~multiprocessing.RLock` object then that will be used to synchronize access to the |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1437 | value. If *lock* is ``False`` then access to the returned object will not be |
| 1438 | automatically protected by a lock, so it will not necessarily be |
| 1439 | "process-safe". |
| 1440 | |
| 1441 | Note that *lock* is a keyword-only argument. |
| 1442 | |
| 1443 | .. function:: copy(obj) |
| 1444 | |
| 1445 | Return a ctypes object allocated from shared memory which is a copy of the |
| 1446 | ctypes object *obj*. |
| 1447 | |
| 1448 | .. function:: synchronized(obj[, lock]) |
| 1449 | |
| 1450 | Return a process-safe wrapper object for a ctypes object which uses *lock* to |
| 1451 | synchronize access. If *lock* is ``None`` (the default) then a |
| 1452 | :class:`multiprocessing.RLock` object is created automatically. |
| 1453 | |
| 1454 | A synchronized wrapper will have two methods in addition to those of the |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1455 | object it wraps: :meth:`get_obj` returns the wrapped object and |
| 1456 | :meth:`get_lock` returns the lock object used for synchronization. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1457 | |
| 1458 | Note that accessing the ctypes object through the wrapper can be a lot slower |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1459 | than accessing the raw ctypes object. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1460 | |
| 1461 | |
| 1462 | The table below compares the syntax for creating shared ctypes objects from |
| 1463 | shared memory with the normal ctypes syntax. (In the table ``MyStruct`` is some |
| 1464 | subclass of :class:`ctypes.Structure`.) |
| 1465 | |
| 1466 | ==================== ========================== =========================== |
| 1467 | ctypes sharedctypes using type sharedctypes using typecode |
| 1468 | ==================== ========================== =========================== |
| 1469 | c_double(2.4) RawValue(c_double, 2.4) RawValue('d', 2.4) |
| 1470 | MyStruct(4, 6) RawValue(MyStruct, 4, 6) |
| 1471 | (c_short * 7)() RawArray(c_short, 7) RawArray('h', 7) |
| 1472 | (c_int * 3)(9, 2, 8) RawArray(c_int, (9, 2, 8)) RawArray('i', (9, 2, 8)) |
| 1473 | ==================== ========================== =========================== |
| 1474 | |
| 1475 | |
| 1476 | Below is an example where a number of ctypes objects are modified by a child |
| 1477 | process:: |
| 1478 | |
| 1479 | from multiprocessing import Process, Lock |
| 1480 | from multiprocessing.sharedctypes import Value, Array |
| 1481 | from ctypes import Structure, c_double |
| 1482 | |
| 1483 | class Point(Structure): |
| 1484 | _fields_ = [('x', c_double), ('y', c_double)] |
| 1485 | |
| 1486 | def modify(n, x, s, A): |
| 1487 | n.value **= 2 |
| 1488 | x.value **= 2 |
| 1489 | s.value = s.value.upper() |
| 1490 | for a in A: |
| 1491 | a.x **= 2 |
| 1492 | a.y **= 2 |
| 1493 | |
| 1494 | if __name__ == '__main__': |
| 1495 | lock = Lock() |
| 1496 | |
| 1497 | n = Value('i', 7) |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1498 | x = Value(c_double, 1.0/3.0, lock=False) |
Richard Oudkerk | b517596 | 2012-09-10 13:00:33 +0100 | [diff] [blame] | 1499 | s = Array('c', b'hello world', lock=lock) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1500 | A = Array(Point, [(1.875,-6.25), (-5.75,2.0), (2.375,9.5)], lock=lock) |
| 1501 | |
| 1502 | p = Process(target=modify, args=(n, x, s, A)) |
| 1503 | p.start() |
| 1504 | p.join() |
| 1505 | |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 1506 | print(n.value) |
| 1507 | print(x.value) |
| 1508 | print(s.value) |
| 1509 | print([(a.x, a.y) for a in A]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1510 | |
| 1511 | |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 1512 | .. highlight:: none |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1513 | |
| 1514 | The results printed are :: |
| 1515 | |
| 1516 | 49 |
| 1517 | 0.1111111111111111 |
| 1518 | HELLO WORLD |
| 1519 | [(3.515625, 39.0625), (33.0625, 4.0), (5.640625, 90.25)] |
| 1520 | |
Ezio Melotti | f86b28e | 2012-04-13 20:50:48 -0600 | [diff] [blame] | 1521 | .. highlight:: python3 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1522 | |
| 1523 | |
| 1524 | .. _multiprocessing-managers: |
| 1525 | |
| 1526 | Managers |
| 1527 | ~~~~~~~~ |
| 1528 | |
| 1529 | Managers provide a way to create data which can be shared between different |
Eli Bendersky | 78da3bc | 2012-07-13 10:10:05 +0300 | [diff] [blame] | 1530 | processes, including sharing over a network between processes running on |
| 1531 | different machines. A manager object controls a server process which manages |
| 1532 | *shared objects*. Other processes can access the shared objects by using |
| 1533 | proxies. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1534 | |
| 1535 | .. function:: multiprocessing.Manager() |
| 1536 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1537 | Returns a started :class:`~multiprocessing.managers.SyncManager` object which |
| 1538 | can be used for sharing objects between processes. The returned manager |
| 1539 | object corresponds to a spawned child process and has methods which will |
| 1540 | create shared objects and return corresponding proxies. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1541 | |
| 1542 | .. module:: multiprocessing.managers |
| 1543 | :synopsis: Share data between process with shared objects. |
| 1544 | |
| 1545 | Manager processes will be shutdown as soon as they are garbage collected or |
| 1546 | their parent process exits. The manager classes are defined in the |
| 1547 | :mod:`multiprocessing.managers` module: |
| 1548 | |
| 1549 | .. class:: BaseManager([address[, authkey]]) |
| 1550 | |
| 1551 | Create a BaseManager object. |
| 1552 | |
Benjamin Peterson | 21896a3 | 2010-03-21 22:03:03 +0000 | [diff] [blame] | 1553 | Once created one should call :meth:`start` or ``get_server().serve_forever()`` to ensure |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1554 | that the manager object refers to a started manager process. |
| 1555 | |
| 1556 | *address* is the address on which the manager process listens for new |
| 1557 | connections. If *address* is ``None`` then an arbitrary one is chosen. |
| 1558 | |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 1559 | *authkey* is the authentication key which will be used to check the |
| 1560 | validity of incoming connections to the server process. If |
| 1561 | *authkey* is ``None`` then ``current_process().authkey`` is used. |
| 1562 | Otherwise *authkey* is used and it must be a byte string. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1563 | |
Benjamin Peterson | f47ed4a | 2009-04-11 20:45:40 +0000 | [diff] [blame] | 1564 | .. method:: start([initializer[, initargs]]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1565 | |
Benjamin Peterson | f47ed4a | 2009-04-11 20:45:40 +0000 | [diff] [blame] | 1566 | Start a subprocess to start the manager. If *initializer* is not ``None`` |
| 1567 | then the subprocess will call ``initializer(*initargs)`` when it starts. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1568 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1569 | .. method:: get_server() |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1570 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1571 | Returns a :class:`Server` object which represents the actual server under |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1572 | the control of the Manager. The :class:`Server` object supports the |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1573 | :meth:`serve_forever` method:: |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1574 | |
Georg Brandl | 1f01deb | 2009-01-03 22:47:39 +0000 | [diff] [blame] | 1575 | >>> from multiprocessing.managers import BaseManager |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 1576 | >>> manager = BaseManager(address=('', 50000), authkey=b'abc') |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1577 | >>> server = manager.get_server() |
| 1578 | >>> server.serve_forever() |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1579 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1580 | :class:`Server` additionally has an :attr:`address` attribute. |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1581 | |
| 1582 | .. method:: connect() |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1583 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1584 | Connect a local manager object to a remote manager process:: |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1585 | |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1586 | >>> from multiprocessing.managers import BaseManager |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 1587 | >>> m = BaseManager(address=('127.0.0.1', 5000), authkey=b'abc') |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1588 | >>> m.connect() |
| 1589 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1590 | .. method:: shutdown() |
| 1591 | |
| 1592 | Stop the process used by the manager. This is only available if |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1593 | :meth:`start` has been used to start the server process. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1594 | |
| 1595 | This can be called multiple times. |
| 1596 | |
| 1597 | .. method:: register(typeid[, callable[, proxytype[, exposed[, method_to_typeid[, create_method]]]]]) |
| 1598 | |
| 1599 | A classmethod which can be used for registering a type or callable with |
| 1600 | the manager class. |
| 1601 | |
| 1602 | *typeid* is a "type identifier" which is used to identify a particular |
| 1603 | type of shared object. This must be a string. |
| 1604 | |
| 1605 | *callable* is a callable used for creating objects for this type |
Richard Oudkerk | f0604fd | 2012-06-11 17:56:08 +0100 | [diff] [blame] | 1606 | identifier. If a manager instance will be connected to the |
| 1607 | server using the :meth:`connect` method, or if the |
| 1608 | *create_method* argument is ``False`` then this can be left as |
| 1609 | ``None``. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1610 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1611 | *proxytype* is a subclass of :class:`BaseProxy` which is used to create |
| 1612 | proxies for shared objects with this *typeid*. If ``None`` then a proxy |
| 1613 | class is created automatically. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1614 | |
| 1615 | *exposed* is used to specify a sequence of method names which proxies for |
| 1616 | this typeid should be allowed to access using |
Larry Hastings | 3732ed2 | 2014-03-15 21:13:56 -0700 | [diff] [blame] | 1617 | :meth:`BaseProxy._callmethod`. (If *exposed* is ``None`` then |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1618 | :attr:`proxytype._exposed_` is used instead if it exists.) In the case |
| 1619 | where no exposed list is specified, all "public methods" of the shared |
| 1620 | object will be accessible. (Here a "public method" means any attribute |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1621 | which has a :meth:`~object.__call__` method and whose name does not begin |
| 1622 | with ``'_'``.) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1623 | |
| 1624 | *method_to_typeid* is a mapping used to specify the return type of those |
| 1625 | exposed methods which should return a proxy. It maps method names to |
| 1626 | typeid strings. (If *method_to_typeid* is ``None`` then |
| 1627 | :attr:`proxytype._method_to_typeid_` is used instead if it exists.) If a |
| 1628 | method's name is not a key of this mapping or if the mapping is ``None`` |
| 1629 | then the object returned by the method will be copied by value. |
| 1630 | |
| 1631 | *create_method* determines whether a method should be created with name |
| 1632 | *typeid* which can be used to tell the server process to create a new |
| 1633 | shared object and return a proxy for it. By default it is ``True``. |
| 1634 | |
| 1635 | :class:`BaseManager` instances also have one read-only property: |
| 1636 | |
| 1637 | .. attribute:: address |
| 1638 | |
| 1639 | The address used by the manager. |
| 1640 | |
Richard Oudkerk | ac38571 | 2012-06-18 21:29:30 +0100 | [diff] [blame] | 1641 | .. versionchanged:: 3.3 |
Serhiy Storchaka | 1486799 | 2014-09-10 23:43:41 +0300 | [diff] [blame] | 1642 | Manager objects support the context management protocol -- see |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1643 | :ref:`typecontextmanager`. :meth:`~contextmanager.__enter__` starts the |
| 1644 | server process (if it has not already started) and then returns the |
| 1645 | manager object. :meth:`~contextmanager.__exit__` calls :meth:`shutdown`. |
Richard Oudkerk | ac38571 | 2012-06-18 21:29:30 +0100 | [diff] [blame] | 1646 | |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1647 | In previous versions :meth:`~contextmanager.__enter__` did not start the |
Richard Oudkerk | ac38571 | 2012-06-18 21:29:30 +0100 | [diff] [blame] | 1648 | manager's server process if it was not already started. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1649 | |
| 1650 | .. class:: SyncManager |
| 1651 | |
| 1652 | A subclass of :class:`BaseManager` which can be used for the synchronization |
| 1653 | of processes. Objects of this type are returned by |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1654 | :func:`multiprocessing.Manager`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1655 | |
| 1656 | It also supports creation of shared lists and dictionaries. |
| 1657 | |
Richard Oudkerk | 3730a17 | 2012-06-15 18:26:07 +0100 | [diff] [blame] | 1658 | .. method:: Barrier(parties[, action[, timeout]]) |
| 1659 | |
| 1660 | Create a shared :class:`threading.Barrier` object and return a |
| 1661 | proxy for it. |
| 1662 | |
| 1663 | .. versionadded:: 3.3 |
| 1664 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1665 | .. method:: BoundedSemaphore([value]) |
| 1666 | |
| 1667 | Create a shared :class:`threading.BoundedSemaphore` object and return a |
| 1668 | proxy for it. |
| 1669 | |
| 1670 | .. method:: Condition([lock]) |
| 1671 | |
| 1672 | Create a shared :class:`threading.Condition` object and return a proxy for |
| 1673 | it. |
| 1674 | |
| 1675 | If *lock* is supplied then it should be a proxy for a |
| 1676 | :class:`threading.Lock` or :class:`threading.RLock` object. |
| 1677 | |
Charles-François Natali | c8ce715 | 2012-04-17 18:45:57 +0200 | [diff] [blame] | 1678 | .. versionchanged:: 3.3 |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 1679 | The :meth:`~threading.Condition.wait_for` method was added. |
Charles-François Natali | c8ce715 | 2012-04-17 18:45:57 +0200 | [diff] [blame] | 1680 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1681 | .. method:: Event() |
| 1682 | |
| 1683 | Create a shared :class:`threading.Event` object and return a proxy for it. |
| 1684 | |
| 1685 | .. method:: Lock() |
| 1686 | |
| 1687 | Create a shared :class:`threading.Lock` object and return a proxy for it. |
| 1688 | |
| 1689 | .. method:: Namespace() |
| 1690 | |
| 1691 | Create a shared :class:`Namespace` object and return a proxy for it. |
| 1692 | |
| 1693 | .. method:: Queue([maxsize]) |
| 1694 | |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 1695 | Create a shared :class:`queue.Queue` object and return a proxy for it. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1696 | |
| 1697 | .. method:: RLock() |
| 1698 | |
| 1699 | Create a shared :class:`threading.RLock` object and return a proxy for it. |
| 1700 | |
| 1701 | .. method:: Semaphore([value]) |
| 1702 | |
| 1703 | Create a shared :class:`threading.Semaphore` object and return a proxy for |
| 1704 | it. |
| 1705 | |
| 1706 | .. method:: Array(typecode, sequence) |
| 1707 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1708 | Create an array and return a proxy for it. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1709 | |
| 1710 | .. method:: Value(typecode, value) |
| 1711 | |
| 1712 | Create an object with a writable ``value`` attribute and return a proxy |
| 1713 | for it. |
| 1714 | |
| 1715 | .. method:: dict() |
| 1716 | dict(mapping) |
| 1717 | dict(sequence) |
| 1718 | |
| 1719 | Create a shared ``dict`` object and return a proxy for it. |
| 1720 | |
| 1721 | .. method:: list() |
| 1722 | list(sequence) |
| 1723 | |
| 1724 | Create a shared ``list`` object and return a proxy for it. |
| 1725 | |
Georg Brandl | 3ed4114 | 2010-10-15 16:19:43 +0000 | [diff] [blame] | 1726 | .. note:: |
| 1727 | |
| 1728 | Modifications to mutable values or items in dict and list proxies will not |
| 1729 | be propagated through the manager, because the proxy has no way of knowing |
| 1730 | when its values or items are modified. To modify such an item, you can |
| 1731 | re-assign the modified object to the container proxy:: |
| 1732 | |
| 1733 | # create a list proxy and append a mutable object (a dictionary) |
| 1734 | lproxy = manager.list() |
| 1735 | lproxy.append({}) |
| 1736 | # now mutate the dictionary |
| 1737 | d = lproxy[0] |
| 1738 | d['a'] = 1 |
| 1739 | d['b'] = 2 |
| 1740 | # at this point, the changes to d are not yet synced, but by |
| 1741 | # reassigning the dictionary, the proxy is notified of the change |
| 1742 | lproxy[0] = d |
| 1743 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1744 | |
| 1745 | Namespace objects |
| 1746 | >>>>>>>>>>>>>>>>> |
| 1747 | |
| 1748 | A namespace object has no public methods, but does have writable attributes. |
| 1749 | Its representation shows the values of its attributes. |
| 1750 | |
| 1751 | However, when using a proxy for a namespace object, an attribute beginning with |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1752 | ``'_'`` will be an attribute of the proxy and not an attribute of the referent: |
| 1753 | |
| 1754 | .. doctest:: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1755 | |
| 1756 | >>> manager = multiprocessing.Manager() |
| 1757 | >>> Global = manager.Namespace() |
| 1758 | >>> Global.x = 10 |
| 1759 | >>> Global.y = 'hello' |
| 1760 | >>> Global._z = 12.3 # this is an attribute of the proxy |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 1761 | >>> print(Global) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1762 | Namespace(x=10, y='hello') |
| 1763 | |
| 1764 | |
| 1765 | Customized managers |
| 1766 | >>>>>>>>>>>>>>>>>>> |
| 1767 | |
| 1768 | To create one's own manager, one creates a subclass of :class:`BaseManager` and |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 1769 | uses the :meth:`~BaseManager.register` classmethod to register new types or |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1770 | callables with the manager class. For example:: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1771 | |
| 1772 | from multiprocessing.managers import BaseManager |
| 1773 | |
Éric Araujo | 28053fb | 2010-11-22 03:09:19 +0000 | [diff] [blame] | 1774 | class MathsClass: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1775 | def add(self, x, y): |
| 1776 | return x + y |
| 1777 | def mul(self, x, y): |
| 1778 | return x * y |
| 1779 | |
| 1780 | class MyManager(BaseManager): |
| 1781 | pass |
| 1782 | |
| 1783 | MyManager.register('Maths', MathsClass) |
| 1784 | |
| 1785 | if __name__ == '__main__': |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 1786 | with MyManager() as manager: |
| 1787 | maths = manager.Maths() |
| 1788 | print(maths.add(4, 3)) # prints 7 |
| 1789 | print(maths.mul(7, 8)) # prints 56 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1790 | |
| 1791 | |
| 1792 | Using a remote manager |
| 1793 | >>>>>>>>>>>>>>>>>>>>>> |
| 1794 | |
| 1795 | It is possible to run a manager server on one machine and have clients use it |
| 1796 | from other machines (assuming that the firewalls involved allow it). |
| 1797 | |
| 1798 | Running the following commands creates a server for a single shared queue which |
| 1799 | remote clients can access:: |
| 1800 | |
| 1801 | >>> from multiprocessing.managers import BaseManager |
Benjamin Peterson | 257060a | 2008-06-28 01:42:41 +0000 | [diff] [blame] | 1802 | >>> import queue |
| 1803 | >>> queue = queue.Queue() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1804 | >>> class QueueManager(BaseManager): pass |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1805 | >>> QueueManager.register('get_queue', callable=lambda:queue) |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 1806 | >>> m = QueueManager(address=('', 50000), authkey=b'abracadabra') |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1807 | >>> s = m.get_server() |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1808 | >>> s.serve_forever() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1809 | |
| 1810 | One client can access the server as follows:: |
| 1811 | |
| 1812 | >>> from multiprocessing.managers import BaseManager |
| 1813 | >>> class QueueManager(BaseManager): pass |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1814 | >>> QueueManager.register('get_queue') |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 1815 | >>> m = QueueManager(address=('foo.bar.org', 50000), authkey=b'abracadabra') |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1816 | >>> m.connect() |
| 1817 | >>> queue = m.get_queue() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1818 | >>> queue.put('hello') |
| 1819 | |
| 1820 | Another client can also use it:: |
| 1821 | |
| 1822 | >>> from multiprocessing.managers import BaseManager |
| 1823 | >>> class QueueManager(BaseManager): pass |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1824 | >>> QueueManager.register('get_queue') |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 1825 | >>> m = QueueManager(address=('foo.bar.org', 50000), authkey=b'abracadabra') |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1826 | >>> m.connect() |
| 1827 | >>> queue = m.get_queue() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1828 | >>> queue.get() |
| 1829 | 'hello' |
| 1830 | |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1831 | Local processes can also access that queue, using the code from above on the |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1832 | client to access it remotely:: |
| 1833 | |
| 1834 | >>> from multiprocessing import Process, Queue |
| 1835 | >>> from multiprocessing.managers import BaseManager |
| 1836 | >>> class Worker(Process): |
| 1837 | ... def __init__(self, q): |
| 1838 | ... self.q = q |
| 1839 | ... super(Worker, self).__init__() |
| 1840 | ... def run(self): |
| 1841 | ... self.q.put('local hello') |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1842 | ... |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1843 | >>> queue = Queue() |
| 1844 | >>> w = Worker(queue) |
| 1845 | >>> w.start() |
| 1846 | >>> class QueueManager(BaseManager): pass |
Georg Brandl | 48310cd | 2009-01-03 21:18:54 +0000 | [diff] [blame] | 1847 | ... |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1848 | >>> QueueManager.register('get_queue', callable=lambda: queue) |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 1849 | >>> m = QueueManager(address=('', 50000), authkey=b'abracadabra') |
Jesse Noller | 4523968 | 2008-11-28 18:46:19 +0000 | [diff] [blame] | 1850 | >>> s = m.get_server() |
| 1851 | >>> s.serve_forever() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1852 | |
| 1853 | Proxy Objects |
| 1854 | ~~~~~~~~~~~~~ |
| 1855 | |
| 1856 | A proxy is an object which *refers* to a shared object which lives (presumably) |
| 1857 | in a different process. The shared object is said to be the *referent* of the |
| 1858 | proxy. Multiple proxy objects may have the same referent. |
| 1859 | |
| 1860 | A proxy object has methods which invoke corresponding methods of its referent |
| 1861 | (although not every method of the referent will necessarily be available through |
| 1862 | the proxy). A proxy can usually be used in most of the same ways that its |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1863 | referent can: |
| 1864 | |
| 1865 | .. doctest:: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1866 | |
| 1867 | >>> from multiprocessing import Manager |
| 1868 | >>> manager = Manager() |
| 1869 | >>> l = manager.list([i*i for i in range(10)]) |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 1870 | >>> print(l) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1871 | [0, 1, 4, 9, 16, 25, 36, 49, 64, 81] |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 1872 | >>> print(repr(l)) |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1873 | <ListProxy object, typeid 'list' at 0x...> |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1874 | >>> l[4] |
| 1875 | 16 |
| 1876 | >>> l[2:5] |
| 1877 | [4, 9, 16] |
| 1878 | |
| 1879 | Notice that applying :func:`str` to a proxy will return the representation of |
| 1880 | the referent, whereas applying :func:`repr` will return the representation of |
| 1881 | the proxy. |
| 1882 | |
| 1883 | An important feature of proxy objects is that they are picklable so they can be |
| 1884 | passed between processes. Note, however, that if a proxy is sent to the |
| 1885 | corresponding manager's process then unpickling it will produce the referent |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1886 | itself. This means, for example, that one shared object can contain a second: |
| 1887 | |
| 1888 | .. doctest:: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1889 | |
| 1890 | >>> a = manager.list() |
| 1891 | >>> b = manager.list() |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1892 | >>> a.append(b) # referent of a now contains referent of b |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 1893 | >>> print(a, b) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1894 | [[]] [] |
| 1895 | >>> b.append('hello') |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 1896 | >>> print(a, b) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1897 | [['hello']] ['hello'] |
| 1898 | |
| 1899 | .. note:: |
| 1900 | |
| 1901 | The proxy types in :mod:`multiprocessing` do nothing to support comparisons |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1902 | by value. So, for instance, we have: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1903 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1904 | .. doctest:: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1905 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1906 | >>> manager.list([1,2,3]) == [1,2,3] |
| 1907 | False |
| 1908 | |
| 1909 | One should just use a copy of the referent instead when making comparisons. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1910 | |
| 1911 | .. class:: BaseProxy |
| 1912 | |
| 1913 | Proxy objects are instances of subclasses of :class:`BaseProxy`. |
| 1914 | |
Benjamin Peterson | 6ebe78f | 2008-12-21 00:06:59 +0000 | [diff] [blame] | 1915 | .. method:: _callmethod(methodname[, args[, kwds]]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1916 | |
| 1917 | Call and return the result of a method of the proxy's referent. |
| 1918 | |
| 1919 | If ``proxy`` is a proxy whose referent is ``obj`` then the expression :: |
| 1920 | |
Benjamin Peterson | 6ebe78f | 2008-12-21 00:06:59 +0000 | [diff] [blame] | 1921 | proxy._callmethod(methodname, args, kwds) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1922 | |
| 1923 | will evaluate the expression :: |
| 1924 | |
| 1925 | getattr(obj, methodname)(*args, **kwds) |
| 1926 | |
| 1927 | in the manager's process. |
| 1928 | |
| 1929 | The returned value will be a copy of the result of the call or a proxy to |
| 1930 | a new shared object -- see documentation for the *method_to_typeid* |
| 1931 | argument of :meth:`BaseManager.register`. |
| 1932 | |
Ezio Melotti | e130a52 | 2011-10-19 10:58:56 +0300 | [diff] [blame] | 1933 | If an exception is raised by the call, then is re-raised by |
Benjamin Peterson | 6ebe78f | 2008-12-21 00:06:59 +0000 | [diff] [blame] | 1934 | :meth:`_callmethod`. If some other exception is raised in the manager's |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1935 | process then this is converted into a :exc:`RemoteError` exception and is |
Benjamin Peterson | 6ebe78f | 2008-12-21 00:06:59 +0000 | [diff] [blame] | 1936 | raised by :meth:`_callmethod`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1937 | |
| 1938 | Note in particular that an exception will be raised if *methodname* has |
Martin Panter | d21e0b5 | 2015-10-10 10:36:22 +0000 | [diff] [blame] | 1939 | not been *exposed*. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1940 | |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 1941 | An example of the usage of :meth:`_callmethod`: |
| 1942 | |
| 1943 | .. doctest:: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1944 | |
| 1945 | >>> l = manager.list(range(10)) |
Benjamin Peterson | 6ebe78f | 2008-12-21 00:06:59 +0000 | [diff] [blame] | 1946 | >>> l._callmethod('__len__') |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1947 | 10 |
Serhiy Storchaka | a60c2fe | 2015-03-12 21:56:08 +0200 | [diff] [blame] | 1948 | >>> l._callmethod('__getitem__', (slice(2, 7),)) # equiv to `l[2:7]` |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1949 | [2, 3, 4, 5, 6] |
Benjamin Peterson | 6ebe78f | 2008-12-21 00:06:59 +0000 | [diff] [blame] | 1950 | >>> l._callmethod('__getitem__', (20,)) # equiv to `l[20]` |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1951 | Traceback (most recent call last): |
| 1952 | ... |
| 1953 | IndexError: list index out of range |
| 1954 | |
Benjamin Peterson | 6ebe78f | 2008-12-21 00:06:59 +0000 | [diff] [blame] | 1955 | .. method:: _getvalue() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1956 | |
| 1957 | Return a copy of the referent. |
| 1958 | |
| 1959 | If the referent is unpicklable then this will raise an exception. |
| 1960 | |
| 1961 | .. method:: __repr__ |
| 1962 | |
| 1963 | Return a representation of the proxy object. |
| 1964 | |
| 1965 | .. method:: __str__ |
| 1966 | |
| 1967 | Return the representation of the referent. |
| 1968 | |
| 1969 | |
| 1970 | Cleanup |
| 1971 | >>>>>>> |
| 1972 | |
| 1973 | A proxy object uses a weakref callback so that when it gets garbage collected it |
| 1974 | deregisters itself from the manager which owns its referent. |
| 1975 | |
| 1976 | A shared object gets deleted from the manager process when there are no longer |
| 1977 | any proxies referring to it. |
| 1978 | |
| 1979 | |
| 1980 | Process Pools |
| 1981 | ~~~~~~~~~~~~~ |
| 1982 | |
| 1983 | .. module:: multiprocessing.pool |
| 1984 | :synopsis: Create pools of processes. |
| 1985 | |
| 1986 | One can create a pool of processes which will carry out tasks submitted to it |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 1987 | with the :class:`Pool` class. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1988 | |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 1989 | .. class:: Pool([processes[, initializer[, initargs[, maxtasksperchild [, context]]]]]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1990 | |
| 1991 | A process pool object which controls a pool of worker processes to which jobs |
| 1992 | can be submitted. It supports asynchronous results with timeouts and |
| 1993 | callbacks and has a parallel map implementation. |
| 1994 | |
| 1995 | *processes* is the number of worker processes to use. If *processes* is |
Larry Hastings | 3732ed2 | 2014-03-15 21:13:56 -0700 | [diff] [blame] | 1996 | ``None`` then the number returned by :func:`os.cpu_count` is used. |
| 1997 | |
| 1998 | If *initializer* is not ``None`` then each worker process will call |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 1999 | ``initializer(*initargs)`` when it starts. |
| 2000 | |
Larry Hastings | 3732ed2 | 2014-03-15 21:13:56 -0700 | [diff] [blame] | 2001 | *maxtasksperchild* is the number of tasks a worker process can complete |
| 2002 | before it will exit and be replaced with a fresh worker process, to enable |
| 2003 | unused resources to be freed. The default *maxtasksperchild* is None, which |
| 2004 | means worker processes will live as long as the pool. |
| 2005 | |
| 2006 | *context* can be used to specify the context used for starting |
| 2007 | the worker processes. Usually a pool is created using the |
| 2008 | function :func:`multiprocessing.Pool` or the :meth:`Pool` method |
| 2009 | of a context object. In both cases *context* is set |
| 2010 | appropriately. |
| 2011 | |
Richard Oudkerk | b3c4b98 | 2013-07-02 12:32:00 +0100 | [diff] [blame] | 2012 | Note that the methods of the pool object should only be called by |
| 2013 | the process which created the pool. |
| 2014 | |
Georg Brandl | 17ef0d5 | 2010-10-17 06:21:59 +0000 | [diff] [blame] | 2015 | .. versionadded:: 3.2 |
Larry Hastings | 3732ed2 | 2014-03-15 21:13:56 -0700 | [diff] [blame] | 2016 | *maxtasksperchild* |
Jesse Noller | 1f0b658 | 2010-01-27 03:36:01 +0000 | [diff] [blame] | 2017 | |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 2018 | .. versionadded:: 3.4 |
Larry Hastings | 3732ed2 | 2014-03-15 21:13:56 -0700 | [diff] [blame] | 2019 | *context* |
Richard Oudkerk | b1694cf | 2013-10-16 16:41:56 +0100 | [diff] [blame] | 2020 | |
Jesse Noller | 1f0b658 | 2010-01-27 03:36:01 +0000 | [diff] [blame] | 2021 | .. note:: |
| 2022 | |
Georg Brandl | 17ef0d5 | 2010-10-17 06:21:59 +0000 | [diff] [blame] | 2023 | Worker processes within a :class:`Pool` typically live for the complete |
| 2024 | duration of the Pool's work queue. A frequent pattern found in other |
| 2025 | systems (such as Apache, mod_wsgi, etc) to free resources held by |
| 2026 | workers is to allow a worker within a pool to complete only a set |
| 2027 | amount of work before being exiting, being cleaned up and a new |
| 2028 | process spawned to replace the old one. The *maxtasksperchild* |
| 2029 | argument to the :class:`Pool` exposes this ability to the end user. |
Jesse Noller | 1f0b658 | 2010-01-27 03:36:01 +0000 | [diff] [blame] | 2030 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2031 | .. method:: apply(func[, args[, kwds]]) |
| 2032 | |
Benjamin Peterson | 37d2fe0 | 2008-10-24 22:28:58 +0000 | [diff] [blame] | 2033 | Call *func* with arguments *args* and keyword arguments *kwds*. It blocks |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 2034 | until the result is ready. Given this blocks, :meth:`apply_async` is |
| 2035 | better suited for performing work in parallel. Additionally, *func* |
| 2036 | is only executed in one of the workers of the pool. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2037 | |
Ask Solem | 1d3b893 | 2010-11-09 21:36:56 +0000 | [diff] [blame] | 2038 | .. method:: apply_async(func[, args[, kwds[, callback[, error_callback]]]]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2039 | |
| 2040 | A variant of the :meth:`apply` method which returns a result object. |
| 2041 | |
| 2042 | If *callback* is specified then it should be a callable which accepts a |
| 2043 | single argument. When the result becomes ready *callback* is applied to |
Ask Solem | 1d3b893 | 2010-11-09 21:36:56 +0000 | [diff] [blame] | 2044 | it, that is unless the call failed, in which case the *error_callback* |
Martin Panter | d21e0b5 | 2015-10-10 10:36:22 +0000 | [diff] [blame] | 2045 | is applied instead. |
Ask Solem | 1d3b893 | 2010-11-09 21:36:56 +0000 | [diff] [blame] | 2046 | |
| 2047 | If *error_callback* is specified then it should be a callable which |
| 2048 | accepts a single argument. If the target function fails, then |
| 2049 | the *error_callback* is called with the exception instance. |
| 2050 | |
| 2051 | Callbacks should complete immediately since otherwise the thread which |
| 2052 | handles the results will get blocked. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2053 | |
| 2054 | .. method:: map(func, iterable[, chunksize]) |
| 2055 | |
Georg Brandl | 22b3431 | 2009-07-26 14:54:51 +0000 | [diff] [blame] | 2056 | A parallel equivalent of the :func:`map` built-in function (it supports only |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 2057 | one *iterable* argument though). It blocks until the result is ready. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2058 | |
| 2059 | This method chops the iterable into a number of chunks which it submits to |
| 2060 | the process pool as separate tasks. The (approximate) size of these |
| 2061 | chunks can be specified by setting *chunksize* to a positive integer. |
| 2062 | |
Sandro Tosi | db79e95 | 2011-08-08 16:38:13 +0200 | [diff] [blame] | 2063 | .. method:: map_async(func, iterable[, chunksize[, callback[, error_callback]]]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2064 | |
Georg Brandl | 502d9a5 | 2009-07-26 15:02:41 +0000 | [diff] [blame] | 2065 | A variant of the :meth:`.map` method which returns a result object. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2066 | |
| 2067 | If *callback* is specified then it should be a callable which accepts a |
| 2068 | single argument. When the result becomes ready *callback* is applied to |
Ask Solem | 1d3b893 | 2010-11-09 21:36:56 +0000 | [diff] [blame] | 2069 | it, that is unless the call failed, in which case the *error_callback* |
Martin Panter | d21e0b5 | 2015-10-10 10:36:22 +0000 | [diff] [blame] | 2070 | is applied instead. |
Ask Solem | 1d3b893 | 2010-11-09 21:36:56 +0000 | [diff] [blame] | 2071 | |
| 2072 | If *error_callback* is specified then it should be a callable which |
| 2073 | accepts a single argument. If the target function fails, then |
| 2074 | the *error_callback* is called with the exception instance. |
| 2075 | |
| 2076 | Callbacks should complete immediately since otherwise the thread which |
| 2077 | handles the results will get blocked. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2078 | |
| 2079 | .. method:: imap(func, iterable[, chunksize]) |
| 2080 | |
Georg Brandl | 9290503 | 2008-11-22 08:51:39 +0000 | [diff] [blame] | 2081 | A lazier version of :meth:`map`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2082 | |
| 2083 | The *chunksize* argument is the same as the one used by the :meth:`.map` |
| 2084 | method. For very long iterables using a large value for *chunksize* can |
Ezio Melotti | e130a52 | 2011-10-19 10:58:56 +0300 | [diff] [blame] | 2085 | make the job complete **much** faster than using the default value of |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2086 | ``1``. |
| 2087 | |
Georg Brandl | 502d9a5 | 2009-07-26 15:02:41 +0000 | [diff] [blame] | 2088 | Also if *chunksize* is ``1`` then the :meth:`!next` method of the iterator |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2089 | returned by the :meth:`imap` method has an optional *timeout* parameter: |
| 2090 | ``next(timeout)`` will raise :exc:`multiprocessing.TimeoutError` if the |
| 2091 | result cannot be returned within *timeout* seconds. |
| 2092 | |
| 2093 | .. method:: imap_unordered(func, iterable[, chunksize]) |
| 2094 | |
| 2095 | The same as :meth:`imap` except that the ordering of the results from the |
| 2096 | returned iterator should be considered arbitrary. (Only when there is |
| 2097 | only one worker process is the order guaranteed to be "correct".) |
| 2098 | |
Antoine Pitrou | de911b2 | 2011-12-21 11:03:24 +0100 | [diff] [blame] | 2099 | .. method:: starmap(func, iterable[, chunksize]) |
| 2100 | |
Georg Brandl | 6b4c847 | 2014-10-30 22:26:26 +0100 | [diff] [blame] | 2101 | Like :meth:`map` except that the elements of the *iterable* are expected |
Antoine Pitrou | de911b2 | 2011-12-21 11:03:24 +0100 | [diff] [blame] | 2102 | to be iterables that are unpacked as arguments. |
| 2103 | |
Georg Brandl | 6b4c847 | 2014-10-30 22:26:26 +0100 | [diff] [blame] | 2104 | Hence an *iterable* of ``[(1,2), (3, 4)]`` results in ``[func(1,2), |
| 2105 | func(3,4)]``. |
Antoine Pitrou | de911b2 | 2011-12-21 11:03:24 +0100 | [diff] [blame] | 2106 | |
| 2107 | .. versionadded:: 3.3 |
| 2108 | |
| 2109 | .. method:: starmap_async(func, iterable[, chunksize[, callback[, error_back]]]) |
| 2110 | |
| 2111 | A combination of :meth:`starmap` and :meth:`map_async` that iterates over |
Georg Brandl | 6b4c847 | 2014-10-30 22:26:26 +0100 | [diff] [blame] | 2112 | *iterable* of iterables and calls *func* with the iterables unpacked. |
Antoine Pitrou | de911b2 | 2011-12-21 11:03:24 +0100 | [diff] [blame] | 2113 | Returns a result object. |
| 2114 | |
| 2115 | .. versionadded:: 3.3 |
| 2116 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2117 | .. method:: close() |
| 2118 | |
| 2119 | Prevents any more tasks from being submitted to the pool. Once all the |
| 2120 | tasks have been completed the worker processes will exit. |
| 2121 | |
| 2122 | .. method:: terminate() |
| 2123 | |
| 2124 | Stops the worker processes immediately without completing outstanding |
| 2125 | work. When the pool object is garbage collected :meth:`terminate` will be |
| 2126 | called immediately. |
| 2127 | |
| 2128 | .. method:: join() |
| 2129 | |
| 2130 | Wait for the worker processes to exit. One must call :meth:`close` or |
| 2131 | :meth:`terminate` before using :meth:`join`. |
| 2132 | |
Richard Oudkerk | d69cfe8 | 2012-06-18 17:47:52 +0100 | [diff] [blame] | 2133 | .. versionadded:: 3.3 |
Serhiy Storchaka | 1486799 | 2014-09-10 23:43:41 +0300 | [diff] [blame] | 2134 | Pool objects now support the context management protocol -- see |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2135 | :ref:`typecontextmanager`. :meth:`~contextmanager.__enter__` returns the |
Georg Brandl | 325a1c2 | 2013-10-27 09:16:01 +0100 | [diff] [blame] | 2136 | pool object, and :meth:`~contextmanager.__exit__` calls :meth:`terminate`. |
Richard Oudkerk | d69cfe8 | 2012-06-18 17:47:52 +0100 | [diff] [blame] | 2137 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2138 | |
| 2139 | .. class:: AsyncResult |
| 2140 | |
| 2141 | The class of the result returned by :meth:`Pool.apply_async` and |
| 2142 | :meth:`Pool.map_async`. |
| 2143 | |
Georg Brandl | e3d70ae | 2008-11-22 08:54:21 +0000 | [diff] [blame] | 2144 | .. method:: get([timeout]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2145 | |
| 2146 | Return the result when it arrives. If *timeout* is not ``None`` and the |
| 2147 | result does not arrive within *timeout* seconds then |
| 2148 | :exc:`multiprocessing.TimeoutError` is raised. If the remote call raised |
| 2149 | an exception then that exception will be reraised by :meth:`get`. |
| 2150 | |
| 2151 | .. method:: wait([timeout]) |
| 2152 | |
| 2153 | Wait until the result is available or until *timeout* seconds pass. |
| 2154 | |
| 2155 | .. method:: ready() |
| 2156 | |
| 2157 | Return whether the call has completed. |
| 2158 | |
| 2159 | .. method:: successful() |
| 2160 | |
| 2161 | Return whether the call completed without raising an exception. Will |
| 2162 | raise :exc:`AssertionError` if the result is not ready. |
| 2163 | |
| 2164 | The following example demonstrates the use of a pool:: |
| 2165 | |
| 2166 | from multiprocessing import Pool |
| 2167 | |
| 2168 | def f(x): |
| 2169 | return x*x |
| 2170 | |
| 2171 | if __name__ == '__main__': |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2172 | with Pool(processes=4) as pool: # start 4 worker processes |
| 2173 | result = pool.apply_async(f, (10,)) # evaluate "f(10)" asynchronously |
| 2174 | print(result.get(timeout=1)) # prints "100" unless your computer is *very* slow |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2175 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2176 | print(pool.map(f, range(10))) # prints "[0, 1, 4,..., 81]" |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2177 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2178 | it = pool.imap(f, range(10)) |
| 2179 | print(next(it)) # prints "0" |
| 2180 | print(next(it)) # prints "1" |
| 2181 | print(it.next(timeout=1)) # prints "4" unless your computer is *very* slow |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2182 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2183 | import time |
| 2184 | result = pool.apply_async(time.sleep, (10,)) |
| 2185 | print(result.get(timeout=1)) # raises TimeoutError |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2186 | |
| 2187 | |
| 2188 | .. _multiprocessing-listeners-clients: |
| 2189 | |
| 2190 | Listeners and Clients |
| 2191 | ~~~~~~~~~~~~~~~~~~~~~ |
| 2192 | |
| 2193 | .. module:: multiprocessing.connection |
| 2194 | :synopsis: API for dealing with sockets. |
| 2195 | |
| 2196 | Usually message passing between processes is done using queues or by using |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2197 | :class:`~multiprocessing.Connection` objects returned by |
| 2198 | :func:`~multiprocessing.Pipe`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2199 | |
| 2200 | However, the :mod:`multiprocessing.connection` module allows some extra |
| 2201 | flexibility. It basically gives a high level message oriented API for dealing |
Antoine Pitrou | bdb1cf1 | 2012-03-05 19:28:37 +0100 | [diff] [blame] | 2202 | with sockets or Windows named pipes. It also has support for *digest |
| 2203 | authentication* using the :mod:`hmac` module, and for polling |
| 2204 | multiple connections at the same time. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2205 | |
| 2206 | |
| 2207 | .. function:: deliver_challenge(connection, authkey) |
| 2208 | |
| 2209 | Send a randomly generated message to the other end of the connection and wait |
| 2210 | for a reply. |
| 2211 | |
| 2212 | If the reply matches the digest of the message using *authkey* as the key |
| 2213 | then a welcome message is sent to the other end of the connection. Otherwise |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 2214 | :exc:`~multiprocessing.AuthenticationError` is raised. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2215 | |
Ezio Melotti | c09959a | 2013-04-10 17:59:20 +0300 | [diff] [blame] | 2216 | .. function:: answer_challenge(connection, authkey) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2217 | |
| 2218 | Receive a message, calculate the digest of the message using *authkey* as the |
| 2219 | key, and then send the digest back. |
| 2220 | |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 2221 | If a welcome message is not received, then |
| 2222 | :exc:`~multiprocessing.AuthenticationError` is raised. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2223 | |
| 2224 | .. function:: Client(address[, family[, authenticate[, authkey]]]) |
| 2225 | |
| 2226 | Attempt to set up a connection to the listener which is using address |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 2227 | *address*, returning a :class:`~multiprocessing.Connection`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2228 | |
| 2229 | The type of the connection is determined by *family* argument, but this can |
| 2230 | generally be omitted since it can usually be inferred from the format of |
| 2231 | *address*. (See :ref:`multiprocessing-address-formats`) |
| 2232 | |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 2233 | If *authenticate* is ``True`` or *authkey* is a byte string then digest |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2234 | authentication is used. The key used for authentication will be either |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 2235 | *authkey* or ``current_process().authkey`` if *authkey* is ``None``. |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 2236 | If authentication fails then |
| 2237 | :exc:`~multiprocessing.AuthenticationError` is raised. See |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2238 | :ref:`multiprocessing-auth-keys`. |
| 2239 | |
| 2240 | .. class:: Listener([address[, family[, backlog[, authenticate[, authkey]]]]]) |
| 2241 | |
| 2242 | A wrapper for a bound socket or Windows named pipe which is 'listening' for |
| 2243 | connections. |
| 2244 | |
| 2245 | *address* is the address to be used by the bound socket or named pipe of the |
| 2246 | listener object. |
| 2247 | |
Benjamin Peterson | d23f822 | 2009-04-05 19:13:16 +0000 | [diff] [blame] | 2248 | .. note:: |
| 2249 | |
| 2250 | If an address of '0.0.0.0' is used, the address will not be a connectable |
| 2251 | end point on Windows. If you require a connectable end-point, |
| 2252 | you should use '127.0.0.1'. |
| 2253 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2254 | *family* is the type of socket (or named pipe) to use. This can be one of |
| 2255 | the strings ``'AF_INET'`` (for a TCP socket), ``'AF_UNIX'`` (for a Unix |
| 2256 | domain socket) or ``'AF_PIPE'`` (for a Windows named pipe). Of these only |
| 2257 | the first is guaranteed to be available. If *family* is ``None`` then the |
| 2258 | family is inferred from the format of *address*. If *address* is also |
| 2259 | ``None`` then a default is chosen. This default is the family which is |
| 2260 | assumed to be the fastest available. See |
| 2261 | :ref:`multiprocessing-address-formats`. Note that if *family* is |
| 2262 | ``'AF_UNIX'`` and address is ``None`` then the socket will be created in a |
| 2263 | private temporary directory created using :func:`tempfile.mkstemp`. |
| 2264 | |
| 2265 | If the listener object uses a socket then *backlog* (1 by default) is passed |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2266 | to the :meth:`~socket.socket.listen` method of the socket once it has been |
| 2267 | bound. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2268 | |
| 2269 | If *authenticate* is ``True`` (``False`` by default) or *authkey* is not |
| 2270 | ``None`` then digest authentication is used. |
| 2271 | |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 2272 | If *authkey* is a byte string then it will be used as the |
| 2273 | authentication key; otherwise it must be *None*. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2274 | |
| 2275 | If *authkey* is ``None`` and *authenticate* is ``True`` then |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 2276 | ``current_process().authkey`` is used as the authentication key. If |
Alexandre Vassalotti | c57a84f | 2009-07-17 12:07:01 +0000 | [diff] [blame] | 2277 | *authkey* is ``None`` and *authenticate* is ``False`` then no |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2278 | authentication is done. If authentication fails then |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 2279 | :exc:`~multiprocessing.AuthenticationError` is raised. |
| 2280 | See :ref:`multiprocessing-auth-keys`. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2281 | |
| 2282 | .. method:: accept() |
| 2283 | |
| 2284 | Accept a connection on the bound socket or named pipe of the listener |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2285 | object and return a :class:`~multiprocessing.Connection` object. If |
| 2286 | authentication is attempted and fails, then |
Eli Bendersky | b674dcf | 2012-07-13 09:45:31 +0300 | [diff] [blame] | 2287 | :exc:`~multiprocessing.AuthenticationError` is raised. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2288 | |
| 2289 | .. method:: close() |
| 2290 | |
| 2291 | Close the bound socket or named pipe of the listener object. This is |
| 2292 | called automatically when the listener is garbage collected. However it |
| 2293 | is advisable to call it explicitly. |
| 2294 | |
| 2295 | Listener objects have the following read-only properties: |
| 2296 | |
| 2297 | .. attribute:: address |
| 2298 | |
| 2299 | The address which is being used by the Listener object. |
| 2300 | |
| 2301 | .. attribute:: last_accepted |
| 2302 | |
| 2303 | The address from which the last accepted connection came. If this is |
| 2304 | unavailable then it is ``None``. |
| 2305 | |
Richard Oudkerk | d69cfe8 | 2012-06-18 17:47:52 +0100 | [diff] [blame] | 2306 | .. versionadded:: 3.3 |
Serhiy Storchaka | 1486799 | 2014-09-10 23:43:41 +0300 | [diff] [blame] | 2307 | Listener objects now support the context management protocol -- see |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2308 | :ref:`typecontextmanager`. :meth:`~contextmanager.__enter__` returns the |
Georg Brandl | 325a1c2 | 2013-10-27 09:16:01 +0100 | [diff] [blame] | 2309 | listener object, and :meth:`~contextmanager.__exit__` calls :meth:`close`. |
Richard Oudkerk | d69cfe8 | 2012-06-18 17:47:52 +0100 | [diff] [blame] | 2310 | |
Antoine Pitrou | bdb1cf1 | 2012-03-05 19:28:37 +0100 | [diff] [blame] | 2311 | .. function:: wait(object_list, timeout=None) |
| 2312 | |
| 2313 | Wait till an object in *object_list* is ready. Returns the list of |
| 2314 | those objects in *object_list* which are ready. If *timeout* is a |
| 2315 | float then the call blocks for at most that many seconds. If |
| 2316 | *timeout* is ``None`` then it will block for an unlimited period. |
Richard Oudkerk | 59d5404 | 2012-05-10 16:11:12 +0100 | [diff] [blame] | 2317 | A negative timeout is equivalent to a zero timeout. |
Antoine Pitrou | bdb1cf1 | 2012-03-05 19:28:37 +0100 | [diff] [blame] | 2318 | |
| 2319 | For both Unix and Windows, an object can appear in *object_list* if |
| 2320 | it is |
| 2321 | |
| 2322 | * a readable :class:`~multiprocessing.Connection` object; |
| 2323 | * a connected and readable :class:`socket.socket` object; or |
| 2324 | * the :attr:`~multiprocessing.Process.sentinel` attribute of a |
| 2325 | :class:`~multiprocessing.Process` object. |
| 2326 | |
| 2327 | A connection or socket object is ready when there is data available |
| 2328 | to be read from it, or the other end has been closed. |
| 2329 | |
| 2330 | **Unix**: ``wait(object_list, timeout)`` almost equivalent |
| 2331 | ``select.select(object_list, [], [], timeout)``. The difference is |
| 2332 | that, if :func:`select.select` is interrupted by a signal, it can |
| 2333 | raise :exc:`OSError` with an error number of ``EINTR``, whereas |
| 2334 | :func:`wait` will not. |
| 2335 | |
| 2336 | **Windows**: An item in *object_list* must either be an integer |
| 2337 | handle which is waitable (according to the definition used by the |
| 2338 | documentation of the Win32 function ``WaitForMultipleObjects()``) |
| 2339 | or it can be an object with a :meth:`fileno` method which returns a |
| 2340 | socket handle or pipe handle. (Note that pipe handles and socket |
| 2341 | handles are **not** waitable handles.) |
| 2342 | |
| 2343 | .. versionadded:: 3.3 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2344 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2345 | |
| 2346 | **Examples** |
| 2347 | |
| 2348 | The following server code creates a listener which uses ``'secret password'`` as |
| 2349 | an authentication key. It then waits for a connection and sends some data to |
| 2350 | the client:: |
| 2351 | |
| 2352 | from multiprocessing.connection import Listener |
| 2353 | from array import array |
| 2354 | |
| 2355 | address = ('localhost', 6000) # family is deduced to be 'AF_INET' |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2356 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2357 | with Listener(address, authkey=b'secret password') as listener: |
| 2358 | with listener.accept() as conn: |
| 2359 | print('connection accepted from', listener.last_accepted) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2360 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2361 | conn.send([2.25, None, 'junk', float]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2362 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2363 | conn.send_bytes(b'hello') |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2364 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2365 | conn.send_bytes(array('i', [42, 1729])) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2366 | |
| 2367 | The following code connects to the server and receives some data from the |
| 2368 | server:: |
| 2369 | |
| 2370 | from multiprocessing.connection import Client |
| 2371 | from array import array |
| 2372 | |
| 2373 | address = ('localhost', 6000) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2374 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2375 | with Client(address, authkey=b'secret password') as conn: |
| 2376 | print(conn.recv()) # => [2.25, None, 'junk', float] |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2377 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2378 | print(conn.recv_bytes()) # => 'hello' |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2379 | |
Richard Oudkerk | 633c4d9 | 2012-06-18 21:29:36 +0100 | [diff] [blame] | 2380 | arr = array('i', [0, 0, 0, 0, 0]) |
| 2381 | print(conn.recv_bytes_into(arr)) # => 8 |
| 2382 | print(arr) # => array('i', [42, 1729, 0, 0, 0]) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2383 | |
Antoine Pitrou | bdb1cf1 | 2012-03-05 19:28:37 +0100 | [diff] [blame] | 2384 | The following code uses :func:`~multiprocessing.connection.wait` to |
| 2385 | wait for messages from multiple processes at once:: |
| 2386 | |
| 2387 | import time, random |
| 2388 | from multiprocessing import Process, Pipe, current_process |
| 2389 | from multiprocessing.connection import wait |
| 2390 | |
| 2391 | def foo(w): |
| 2392 | for i in range(10): |
| 2393 | w.send((i, current_process().name)) |
| 2394 | w.close() |
| 2395 | |
| 2396 | if __name__ == '__main__': |
| 2397 | readers = [] |
| 2398 | |
| 2399 | for i in range(4): |
| 2400 | r, w = Pipe(duplex=False) |
| 2401 | readers.append(r) |
| 2402 | p = Process(target=foo, args=(w,)) |
| 2403 | p.start() |
| 2404 | # We close the writable end of the pipe now to be sure that |
| 2405 | # p is the only process which owns a handle for it. This |
| 2406 | # ensures that when p closes its handle for the writable end, |
| 2407 | # wait() will promptly report the readable end as being ready. |
| 2408 | w.close() |
| 2409 | |
| 2410 | while readers: |
| 2411 | for r in wait(readers): |
| 2412 | try: |
| 2413 | msg = r.recv() |
| 2414 | except EOFError: |
| 2415 | readers.remove(r) |
| 2416 | else: |
| 2417 | print(msg) |
| 2418 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2419 | |
| 2420 | .. _multiprocessing-address-formats: |
| 2421 | |
| 2422 | Address Formats |
| 2423 | >>>>>>>>>>>>>>> |
| 2424 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 2425 | * An ``'AF_INET'`` address is a tuple of the form ``(hostname, port)`` where |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2426 | *hostname* is a string and *port* is an integer. |
| 2427 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 2428 | * An ``'AF_UNIX'`` address is a string representing a filename on the |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2429 | filesystem. |
| 2430 | |
| 2431 | * An ``'AF_PIPE'`` address is a string of the form |
Benjamin Peterson | da10d3b | 2009-01-01 00:23:30 +0000 | [diff] [blame] | 2432 | :samp:`r'\\\\.\\pipe\\{PipeName}'`. To use :func:`Client` to connect to a named |
Georg Brandl | 1f01deb | 2009-01-03 22:47:39 +0000 | [diff] [blame] | 2433 | pipe on a remote computer called *ServerName* one should use an address of the |
Benjamin Peterson | 28d88b4 | 2009-01-09 03:03:23 +0000 | [diff] [blame] | 2434 | form :samp:`r'\\\\{ServerName}\\pipe\\{PipeName}'` instead. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2435 | |
| 2436 | Note that any string beginning with two backslashes is assumed by default to be |
| 2437 | an ``'AF_PIPE'`` address rather than an ``'AF_UNIX'`` address. |
| 2438 | |
| 2439 | |
| 2440 | .. _multiprocessing-auth-keys: |
| 2441 | |
| 2442 | Authentication keys |
| 2443 | ~~~~~~~~~~~~~~~~~~~ |
| 2444 | |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2445 | When one uses :meth:`Connection.recv <multiprocessing.Connection.recv>`, the |
| 2446 | data received is automatically |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2447 | unpickled. Unfortunately unpickling data from an untrusted source is a security |
| 2448 | risk. Therefore :class:`Listener` and :func:`Client` use the :mod:`hmac` module |
| 2449 | to provide digest authentication. |
| 2450 | |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 2451 | An authentication key is a byte string which can be thought of as a |
| 2452 | password: once a connection is established both ends will demand proof |
| 2453 | that the other knows the authentication key. (Demonstrating that both |
| 2454 | ends are using the same key does **not** involve sending the key over |
| 2455 | the connection.) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2456 | |
Richard Oudkerk | 264e9ac | 2012-08-17 14:39:18 +0100 | [diff] [blame] | 2457 | If authentication is requested but no authentication key is specified then the |
Benjamin Peterson | a786b02 | 2008-08-25 21:05:21 +0000 | [diff] [blame] | 2458 | return value of ``current_process().authkey`` is used (see |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 2459 | :class:`~multiprocessing.Process`). This value will automatically inherited by |
| 2460 | any :class:`~multiprocessing.Process` object that the current process creates. |
| 2461 | This means that (by default) all processes of a multi-process program will share |
| 2462 | a single authentication key which can be used when setting up connections |
Benjamin Peterson | d23f822 | 2009-04-05 19:13:16 +0000 | [diff] [blame] | 2463 | between themselves. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2464 | |
| 2465 | Suitable authentication keys can also be generated by using :func:`os.urandom`. |
| 2466 | |
| 2467 | |
| 2468 | Logging |
| 2469 | ~~~~~~~ |
| 2470 | |
| 2471 | Some support for logging is available. Note, however, that the :mod:`logging` |
| 2472 | package does not use process shared locks so it is possible (depending on the |
| 2473 | handler type) for messages from different processes to get mixed up. |
| 2474 | |
| 2475 | .. currentmodule:: multiprocessing |
| 2476 | .. function:: get_logger() |
| 2477 | |
| 2478 | Returns the logger used by :mod:`multiprocessing`. If necessary, a new one |
| 2479 | will be created. |
| 2480 | |
Jesse Noller | 41faa54 | 2009-01-25 03:45:53 +0000 | [diff] [blame] | 2481 | When first created the logger has level :data:`logging.NOTSET` and no |
| 2482 | default handler. Messages sent to this logger will not by default propagate |
| 2483 | to the root logger. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2484 | |
| 2485 | Note that on Windows child processes will only inherit the level of the |
| 2486 | parent process's logger -- any other customization of the logger will not be |
| 2487 | inherited. |
| 2488 | |
Jesse Noller | 41faa54 | 2009-01-25 03:45:53 +0000 | [diff] [blame] | 2489 | .. currentmodule:: multiprocessing |
| 2490 | .. function:: log_to_stderr() |
| 2491 | |
| 2492 | This function performs a call to :func:`get_logger` but in addition to |
| 2493 | returning the logger created by get_logger, it adds a handler which sends |
| 2494 | output to :data:`sys.stderr` using format |
| 2495 | ``'[%(levelname)s/%(processName)s] %(message)s'``. |
| 2496 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2497 | Below is an example session with logging turned on:: |
| 2498 | |
Benjamin Peterson | 206e307 | 2008-10-19 14:07:49 +0000 | [diff] [blame] | 2499 | >>> import multiprocessing, logging |
Jesse Noller | 41faa54 | 2009-01-25 03:45:53 +0000 | [diff] [blame] | 2500 | >>> logger = multiprocessing.log_to_stderr() |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2501 | >>> logger.setLevel(logging.INFO) |
| 2502 | >>> logger.warning('doomed') |
| 2503 | [WARNING/MainProcess] doomed |
Benjamin Peterson | 206e307 | 2008-10-19 14:07:49 +0000 | [diff] [blame] | 2504 | >>> m = multiprocessing.Manager() |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 2505 | [INFO/SyncManager-...] child process calling self.run() |
| 2506 | [INFO/SyncManager-...] created temp directory /.../pymp-... |
| 2507 | [INFO/SyncManager-...] manager serving at '/.../listener-...' |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2508 | >>> del m |
| 2509 | [INFO/MainProcess] sending shutdown message to manager |
R. David Murray | 8e8099c | 2009-04-28 18:02:00 +0000 | [diff] [blame] | 2510 | [INFO/SyncManager-...] manager exiting with exitcode 0 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2511 | |
Jesse Noller | 41faa54 | 2009-01-25 03:45:53 +0000 | [diff] [blame] | 2512 | For a full table of logging levels, see the :mod:`logging` module. |
| 2513 | |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2514 | |
| 2515 | The :mod:`multiprocessing.dummy` module |
| 2516 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 2517 | |
| 2518 | .. module:: multiprocessing.dummy |
| 2519 | :synopsis: Dumb wrapper around threading. |
| 2520 | |
| 2521 | :mod:`multiprocessing.dummy` replicates the API of :mod:`multiprocessing` but is |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 2522 | no more than a wrapper around the :mod:`threading` module. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2523 | |
| 2524 | |
| 2525 | .. _multiprocessing-programming: |
| 2526 | |
| 2527 | Programming guidelines |
| 2528 | ---------------------- |
| 2529 | |
| 2530 | There are certain guidelines and idioms which should be adhered to when using |
| 2531 | :mod:`multiprocessing`. |
| 2532 | |
| 2533 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2534 | All start methods |
| 2535 | ~~~~~~~~~~~~~~~~~ |
| 2536 | |
| 2537 | The following applies to all start methods. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2538 | |
| 2539 | Avoid shared state |
| 2540 | |
| 2541 | As far as possible one should try to avoid shifting large amounts of data |
| 2542 | between processes. |
| 2543 | |
| 2544 | It is probably best to stick to using queues or pipes for communication |
| 2545 | between processes rather than using the lower level synchronization |
Eli Bendersky | 78da3bc | 2012-07-13 10:10:05 +0300 | [diff] [blame] | 2546 | primitives. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2547 | |
| 2548 | Picklability |
| 2549 | |
| 2550 | Ensure that the arguments to the methods of proxies are picklable. |
| 2551 | |
| 2552 | Thread safety of proxies |
| 2553 | |
| 2554 | Do not use a proxy object from more than one thread unless you protect it |
| 2555 | with a lock. |
| 2556 | |
| 2557 | (There is never a problem with different processes using the *same* proxy.) |
| 2558 | |
| 2559 | Joining zombie processes |
| 2560 | |
| 2561 | On Unix when a process finishes but has not been joined it becomes a zombie. |
| 2562 | There should never be very many because each time a new process starts (or |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2563 | :func:`~multiprocessing.active_children` is called) all completed processes |
| 2564 | which have not yet been joined will be joined. Also calling a finished |
| 2565 | process's :meth:`Process.is_alive <multiprocessing.Process.is_alive>` will |
| 2566 | join the process. Even so it is probably good |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2567 | practice to explicitly join all the processes that you start. |
| 2568 | |
| 2569 | Better to inherit than pickle/unpickle |
| 2570 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2571 | When using the *spawn* or *forkserver* start methods many types |
| 2572 | from :mod:`multiprocessing` need to be picklable so that child |
| 2573 | processes can use them. However, one should generally avoid |
| 2574 | sending shared objects to other processes using pipes or queues. |
| 2575 | Instead you should arrange the program so that a process which |
| 2576 | needs access to a shared resource created elsewhere can inherit it |
| 2577 | from an ancestor process. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2578 | |
| 2579 | Avoid terminating processes |
| 2580 | |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2581 | Using the :meth:`Process.terminate <multiprocessing.Process.terminate>` |
| 2582 | method to stop a process is liable to |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2583 | cause any shared resources (such as locks, semaphores, pipes and queues) |
| 2584 | currently being used by the process to become broken or unavailable to other |
| 2585 | processes. |
| 2586 | |
| 2587 | Therefore it is probably best to only consider using |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2588 | :meth:`Process.terminate <multiprocessing.Process.terminate>` on processes |
| 2589 | which never use any shared resources. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2590 | |
| 2591 | Joining processes that use queues |
| 2592 | |
| 2593 | Bear in mind that a process that has put items in a queue will wait before |
| 2594 | terminating until all the buffered items are fed by the "feeder" thread to |
| 2595 | the underlying pipe. (The child process can call the |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2596 | :meth:`Queue.cancel_join_thread <multiprocessing.Queue.cancel_join_thread>` |
| 2597 | method of the queue to avoid this behaviour.) |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2598 | |
| 2599 | This means that whenever you use a queue you need to make sure that all |
| 2600 | items which have been put on the queue will eventually be removed before the |
| 2601 | process is joined. Otherwise you cannot be sure that processes which have |
| 2602 | put items on the queue will terminate. Remember also that non-daemonic |
Zachary Ware | 7280561 | 2014-10-03 10:55:12 -0500 | [diff] [blame] | 2603 | processes will be joined automatically. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2604 | |
| 2605 | An example which will deadlock is the following:: |
| 2606 | |
| 2607 | from multiprocessing import Process, Queue |
| 2608 | |
| 2609 | def f(q): |
| 2610 | q.put('X' * 1000000) |
| 2611 | |
| 2612 | if __name__ == '__main__': |
| 2613 | queue = Queue() |
| 2614 | p = Process(target=f, args=(queue,)) |
| 2615 | p.start() |
| 2616 | p.join() # this deadlocks |
| 2617 | obj = queue.get() |
| 2618 | |
Zachary Ware | 7280561 | 2014-10-03 10:55:12 -0500 | [diff] [blame] | 2619 | A fix here would be to swap the last two lines (or simply remove the |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2620 | ``p.join()`` line). |
| 2621 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 2622 | Explicitly pass resources to child processes |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2623 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2624 | On Unix using the *fork* start method, a child process can make |
| 2625 | use of a shared resource created in a parent process using a |
| 2626 | global resource. However, it is better to pass the object as an |
| 2627 | argument to the constructor for the child process. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2628 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2629 | Apart from making the code (potentially) compatible with Windows |
| 2630 | and the other start methods this also ensures that as long as the |
| 2631 | child process is still alive the object will not be garbage |
| 2632 | collected in the parent process. This might be important if some |
| 2633 | resource is freed when the object is garbage collected in the |
| 2634 | parent process. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2635 | |
| 2636 | So for instance :: |
| 2637 | |
| 2638 | from multiprocessing import Process, Lock |
| 2639 | |
| 2640 | def f(): |
| 2641 | ... do something using "lock" ... |
| 2642 | |
| 2643 | if __name__ == '__main__': |
| 2644 | lock = Lock() |
| 2645 | for i in range(10): |
| 2646 | Process(target=f).start() |
| 2647 | |
| 2648 | should be rewritten as :: |
| 2649 | |
| 2650 | from multiprocessing import Process, Lock |
| 2651 | |
| 2652 | def f(l): |
| 2653 | ... do something using "l" ... |
| 2654 | |
| 2655 | if __name__ == '__main__': |
| 2656 | lock = Lock() |
| 2657 | for i in range(10): |
| 2658 | Process(target=f, args=(lock,)).start() |
| 2659 | |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 2660 | Beware of replacing :data:`sys.stdin` with a "file like object" |
Alexandre Vassalotti | c57a84f | 2009-07-17 12:07:01 +0000 | [diff] [blame] | 2661 | |
| 2662 | :mod:`multiprocessing` originally unconditionally called:: |
| 2663 | |
| 2664 | os.close(sys.stdin.fileno()) |
| 2665 | |
| 2666 | in the :meth:`multiprocessing.Process._bootstrap` method --- this resulted |
| 2667 | in issues with processes-in-processes. This has been changed to:: |
| 2668 | |
| 2669 | sys.stdin.close() |
| 2670 | sys.stdin = open(os.devnull) |
| 2671 | |
| 2672 | Which solves the fundamental issue of processes colliding with each other |
| 2673 | resulting in a bad file descriptor error, but introduces a potential danger |
| 2674 | to applications which replace :func:`sys.stdin` with a "file-like object" |
| 2675 | with output buffering. This danger is that if multiple processes call |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2676 | :meth:`~io.IOBase.close()` on this file-like object, it could result in the same |
Alexandre Vassalotti | c57a84f | 2009-07-17 12:07:01 +0000 | [diff] [blame] | 2677 | data being flushed to the object multiple times, resulting in corruption. |
| 2678 | |
| 2679 | If you write a file-like object and implement your own caching, you can |
| 2680 | make it fork-safe by storing the pid whenever you append to the cache, |
| 2681 | and discarding the cache when the pid changes. For example:: |
| 2682 | |
| 2683 | @property |
| 2684 | def cache(self): |
| 2685 | pid = os.getpid() |
| 2686 | if pid != self._pid: |
| 2687 | self._pid = pid |
| 2688 | self._cache = [] |
| 2689 | return self._cache |
| 2690 | |
| 2691 | For more information, see :issue:`5155`, :issue:`5313` and :issue:`5331` |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2692 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2693 | The *spawn* and *forkserver* start methods |
| 2694 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2695 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2696 | There are a few extra restriction which don't apply to the *fork* |
| 2697 | start method. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2698 | |
| 2699 | More picklability |
| 2700 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2701 | Ensure that all arguments to :meth:`Process.__init__` are |
| 2702 | picklable. This means, in particular, that bound or unbound |
| 2703 | methods cannot be used directly as the ``target`` (unless you use |
| 2704 | the *fork* start method) --- just define a function and use that |
| 2705 | instead. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2706 | |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2707 | Also, if you subclass :class:`~multiprocessing.Process` then make sure that |
| 2708 | instances will be picklable when the :meth:`Process.start |
| 2709 | <multiprocessing.Process.start>` method is called. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2710 | |
| 2711 | Global variables |
| 2712 | |
| 2713 | Bear in mind that if code run in a child process tries to access a global |
| 2714 | variable, then the value it sees (if any) may not be the same as the value |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2715 | in the parent process at the time that :meth:`Process.start |
| 2716 | <multiprocessing.Process.start>` was called. |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2717 | |
| 2718 | However, global variables which are just module level constants cause no |
| 2719 | problems. |
| 2720 | |
| 2721 | Safe importing of main module |
| 2722 | |
| 2723 | Make sure that the main module can be safely imported by a new Python |
| 2724 | interpreter without causing unintended side effects (such a starting a new |
| 2725 | process). |
| 2726 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2727 | For example, using the *spawn* or *forkserver* start method |
| 2728 | running the following module would fail with a |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2729 | :exc:`RuntimeError`:: |
| 2730 | |
| 2731 | from multiprocessing import Process |
| 2732 | |
| 2733 | def foo(): |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 2734 | print('hello') |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2735 | |
| 2736 | p = Process(target=foo) |
| 2737 | p.start() |
| 2738 | |
| 2739 | Instead one should protect the "entry point" of the program by using ``if |
| 2740 | __name__ == '__main__':`` as follows:: |
| 2741 | |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2742 | from multiprocessing import Process, freeze_support, set_start_method |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2743 | |
| 2744 | def foo(): |
Georg Brandl | 4970215 | 2008-09-29 06:43:45 +0000 | [diff] [blame] | 2745 | print('hello') |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2746 | |
| 2747 | if __name__ == '__main__': |
| 2748 | freeze_support() |
Richard Oudkerk | 84ed9a6 | 2013-08-14 15:35:41 +0100 | [diff] [blame] | 2749 | set_start_method('spawn') |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2750 | p = Process(target=foo) |
| 2751 | p.start() |
| 2752 | |
Benjamin Peterson | 5289b2b | 2008-06-28 00:40:54 +0000 | [diff] [blame] | 2753 | (The ``freeze_support()`` line can be omitted if the program will be run |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2754 | normally instead of frozen.) |
| 2755 | |
| 2756 | This allows the newly spawned Python interpreter to safely import the module |
| 2757 | and then run the module's ``foo()`` function. |
| 2758 | |
| 2759 | Similar restrictions apply if a pool or manager is created in the main |
| 2760 | module. |
| 2761 | |
| 2762 | |
| 2763 | .. _multiprocessing-examples: |
| 2764 | |
| 2765 | Examples |
| 2766 | -------- |
| 2767 | |
| 2768 | Demonstration of how to create and use customized managers and proxies: |
| 2769 | |
| 2770 | .. literalinclude:: ../includes/mp_newtype.py |
Ezio Melotti | f86b28e | 2012-04-13 20:50:48 -0600 | [diff] [blame] | 2771 | :language: python3 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2772 | |
| 2773 | |
Serhiy Storchaka | 9e0ae53 | 2013-08-24 00:23:38 +0300 | [diff] [blame] | 2774 | Using :class:`~multiprocessing.pool.Pool`: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2775 | |
| 2776 | .. literalinclude:: ../includes/mp_pool.py |
Ezio Melotti | f86b28e | 2012-04-13 20:50:48 -0600 | [diff] [blame] | 2777 | :language: python3 |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2778 | |
| 2779 | |
Georg Brandl | 0b37b33 | 2010-09-03 22:49:27 +0000 | [diff] [blame] | 2780 | An example showing how to use queues to feed tasks to a collection of worker |
Eli Bendersky | d08effe | 2011-12-31 07:20:26 +0200 | [diff] [blame] | 2781 | processes and collect the results: |
Benjamin Peterson | e711caf | 2008-06-11 16:44:04 +0000 | [diff] [blame] | 2782 | |
| 2783 | .. literalinclude:: ../includes/mp_workers.py |