multiprocessing is a package that supports spawning processes using an API similar to the threading module. The multiprocessing package offers both local and remote concurrency, effectively side-stepping the Global Interpreter Lock by using subprocesses instead of threads. Due to this, the multiprocessing module allows the programmer to fully leverage multiple processors on a given machine. It runs on both Unix and Windows.
Note
Some of this package’s functionality requires a functioning shared semaphore implementation on the host operating system. Without one, the multiprocessing.synchronize module will be disabled, and attempts to import it will result in an ImportError. See issue 3770 for additional information.
Note
Functionality within this package requires that the __main__ module be importable by the children. This is covered in Programming guidelines however it is worth pointing out here. This means that some examples, such as the multiprocessing.pool.Pool examples will not work in the interactive interpreter. For example:
>>> from multiprocessing import Pool
>>> p = Pool(5)
>>> def f(x):
... return x*x
...
>>> p.map(f, [1,2,3])
Process PoolWorker-1:
Process PoolWorker-2:
Process PoolWorker-3:
Traceback (most recent call last):
AttributeError: 'module' object has no attribute 'f'
AttributeError: 'module' object has no attribute 'f'
AttributeError: 'module' object has no attribute 'f'
(If you try this it will actually output three full tracebacks interleaved in a semi-random fashion, and then you may have to stop the master process somehow.)
In multiprocessing, processes are spawned by creating a Process object and then calling its start() method. Process follows the API of threading.Thread. A trivial example of a multiprocess program is
from multiprocessing import Process
def f(name):
print('hello', name)
if __name__ == '__main__':
p = Process(target=f, args=('bob',))
p.start()
p.join()
To show the individual process IDs involved, here is an expanded example:
from multiprocessing import Process
import os
def info(title):
print(title)
print('module name:', __name__)
if hasattr(os, 'getppid'): # only available on Unix
print('parent process:', os.getppid())
print('process id:', os.getpid())
def f(name):
info('function f')
print('hello', name)
if __name__ == '__main__':
info('main line')
p = Process(target=f, args=('bob',))
p.start()
p.join()
For an explanation of why the if __name__ == '__main__' part is necessary, see Programming guidelines.
Depending on the platform, multiprocessing supports three ways to start a process. These start methods are
- spawn
The parent process starts a fresh python interpreter process. The child process will only inherit those resources necessary to run the process objects run() method. In particular, unnecessary file descriptors and handles from the parent process will not be inherited. Starting a process using this method is rather slow compared to using fork or forkserver.
Available on Unix and Windows. The default on Windows.
- fork
The parent process uses os.fork() to fork the Python interpreter. The child process, when it begins, is effectively identical to the parent process. All resources of the parent are inherited by the child process. Note that safely forking a multithreaded process is problematic.
Available on Unix only. The default on Unix.
- forkserver
When the program starts and selects the forkserver start method, a server process is started. From then on, whenever a new process is needed, the parent process connects to the server and requests that it fork a new process. The fork server process is single threaded so it is safe for it to use os.fork(). No unnecessary resources are inherited.
Available on Unix platforms which support passing file descriptors over Unix pipes.
Changed in version 3.4: spawn added on all unix platforms, and forkserver added for some unix platforms. Child processes no longer inherit all of the parents inheritable handles on Windows.
On Unix using the spawn or forkserver start methods will also start a semaphore tracker process which tracks the unlinked named semaphores created by processes of the program. When all processes have exited the semaphore tracker unlinks any remaining semaphores. Usually there should be none, but if a process was killed by a signal there may some “leaked” semaphores. (Unlinking the named semaphores is a serious matter since the system allows only a limited number, and they will not be automatically unlinked until the next reboot.)
To select a start method you use the set_start_method() in the if __name__ == '__main__' clause of the main module. For example:
import multiprocessing as mp
def foo(q):
q.put('hello')
if __name__ == '__main__':
mp.set_start_method('spawn')
q = mp.Queue()
p = mp.Process(target=foo, args=(q,))
p.start()
print(q.get())
p.join()
set_start_method() should not be used more than once in the program.
Alternatively, you can use get_context() to obtain a context object. Context objects have the same API as the multiprocessing module, and allow one to use multiple start methods in the same program.
import multiprocessing as mp
def foo(q):
q.put('hello')
if __name__ == '__main__':
ctx = mp.get_context('spawn')
q = ctx.Queue()
p = ctx.Process(target=foo, args=(q,))
p.start()
print(q.get())
p.join()
Note that objects related to one context may not be compatible with processes for a different context. In particular, locks created using the fork context cannot be passed to a processes started using the spawn or forkserver start methods.
A library which wants to use a particular start method should probably use get_context() to avoid interfering with the choice of the library user.
multiprocessing supports two types of communication channel between processes:
Queues
The Queue class is a near clone of queue.Queue. For example:
from multiprocessing import Process, Queue def f(q): q.put([42, None, 'hello']) if __name__ == '__main__': q = Queue() p = Process(target=f, args=(q,)) p.start() print(q.get()) # prints "[42, None, 'hello']" p.join()Queues are thread and process safe.
Pipes
The Pipe() function returns a pair of connection objects connected by a pipe which by default is duplex (two-way). For example:
from multiprocessing import Process, Pipe def f(conn): conn.send([42, None, 'hello']) conn.close() if __name__ == '__main__': parent_conn, child_conn = Pipe() p = Process(target=f, args=(child_conn,)) p.start() print(parent_conn.recv()) # prints "[42, None, 'hello']" p.join()The two connection objects returned by Pipe() represent the two ends of the pipe. Each connection object has send() and recv() methods (among others). Note that data in a pipe may become corrupted if two processes (or threads) try to read from or write to the same end of the pipe at the same time. Of course there is no risk of corruption from processes using different ends of the pipe at the same time.
multiprocessing contains equivalents of all the synchronization primitives from threading. For instance one can use a lock to ensure that only one process prints to standard output at a time:
from multiprocessing import Process, Lock
def f(l, i):
l.acquire()
try:
print('hello world', i)
finally:
l.release()
if __name__ == '__main__':
lock = Lock()
for num in range(10):
Process(target=f, args=(lock, num)).start()
Without using the lock output from the different processes is liable to get all mixed up.
As mentioned above, when doing concurrent programming it is usually best to avoid using shared state as far as possible. This is particularly true when using multiple processes.
However, if you really do need to use some shared data then multiprocessing provides a couple of ways of doing so.
Shared memory
Data can be stored in a shared memory map using Value or Array. For example, the following code
from multiprocessing import Process, Value, Array def f(n, a): n.value = 3.1415927 for i in range(len(a)): a[i] = -a[i] if __name__ == '__main__': num = Value('d', 0.0) arr = Array('i', range(10)) p = Process(target=f, args=(num, arr)) p.start() p.join() print(num.value) print(arr[:])will print
3.1415927 [0, -1, -2, -3, -4, -5, -6, -7, -8, -9]The 'd' and 'i' arguments used when creating num and arr are typecodes of the kind used by the array module: 'd' indicates a double precision float and 'i' indicates a signed integer. These shared objects will be process and thread-safe.
For more flexibility in using shared memory one can use the multiprocessing.sharedctypes module which supports the creation of arbitrary ctypes objects allocated from shared memory.
Server process
A manager object returned by Manager() controls a server process which holds Python objects and allows other processes to manipulate them using proxies.
A manager returned by Manager() will support types list, dict, Namespace, Lock, RLock, Semaphore, BoundedSemaphore, Condition, Event, Barrier, Queue, Value and Array. For example,
from multiprocessing import Process, Manager def f(d, l): d[1] = '1' d['2'] = 2 d[0.25] = None l.reverse() if __name__ == '__main__': with Manager() as manager: d = manager.dict() l = manager.list(range(10)) p = Process(target=f, args=(d, l)) p.start() p.join() print(d) print(l)will print
{0.25: None, 1: '1', '2': 2} [9, 8, 7, 6, 5, 4, 3, 2, 1, 0]Server process managers are more flexible than using shared memory objects because they can be made to support arbitrary object types. Also, a single manager can be shared by processes on different computers over a network. They are, however, slower than using shared memory.
The Pool class represents a pool of worker processes. It has methods which allows tasks to be offloaded to the worker processes in a few different ways.
For example:
from multiprocessing import Pool
from time import sleep
def f(x):
return x*x
if __name__ == '__main__':
# start 4 worker processes
with Pool(processes=4) as pool:
# print "[0, 1, 4,..., 81]"
print(pool.map(f, range(10)))
# print same numbers in arbitrary order
for i in pool.imap_unordered(f, range(10)):
print(i)
# evaluate "f(10)" asynchronously
res = pool.apply_async(f, [10])
print(res.get(timeout=1)) # prints "100"
# make worker sleep for 10 secs
res = pool.apply_async(sleep, [10])
print(res.get(timeout=1)) # raises multiprocessing.TimeoutError
# exiting the 'with'-block has stopped the pool
Note that the methods of a pool should only ever be used by the process which created it.
The multiprocessing package mostly replicates the API of the threading module.
Process objects represent activity that is run in a separate process. The Process class has equivalents of all the methods of threading.Thread.
The constructor should always be called with keyword arguments. group should always be None; it exists solely for compatibility with threading.Thread. target is the callable object to be invoked by the run() method. It defaults to None, meaning nothing is called. name is the process name (see name for more details). args is the argument tuple for the target invocation. kwargs is a dictionary of keyword arguments for the target invocation. If provided, the keyword-only daemon argument sets the process daemon flag to True or False. If None (the default), this flag will be inherited from the creating process.
By default, no arguments are passed to target.
If a subclass overrides the constructor, it must make sure it invokes the base class constructor (Process.__init__()) before doing anything else to the process.
Changed in version 3.3: Added the daemon argument.
Method representing the process’s activity.
You may override this method in a subclass. The standard run() method invokes the callable object passed to the object’s constructor as the target argument, if any, with sequential and keyword arguments taken from the args and kwargs arguments, respectively.
Start the process’s activity.
This must be called at most once per process object. It arranges for the object’s run() method to be invoked in a separate process.
If the optional argument timeout is None (the default), the method blocks until the process whose join() method is called terminates. If timeout is a positive number, it blocks at most timeout seconds.
A process can be joined many times.
A process cannot join itself because this would cause a deadlock. It is an error to attempt to join a process before it has been started.
The process’s name. The name is a string used for identification purposes only. It has no semantics. Multiple processes may be given the same name.
The initial name is set by the constructor. If no explicit name is provided to the constructor, a name of the form ‘Process-N1:N2:...:Nk‘ is constructed, where each Nk is the N-th child of its parent.
Return whether the process is alive.
Roughly, a process object is alive from the moment the start() method returns until the child process terminates.
The process’s daemon flag, a Boolean value. This must be set before start() is called.
The initial value is inherited from the creating process.
When a process exits, it attempts to terminate all of its daemonic child processes.
Note that a daemonic process is not allowed to create child processes. Otherwise a daemonic process would leave its children orphaned if it gets terminated when its parent process exits. Additionally, these are not Unix daemons or services, they are normal processes that will be terminated (and not joined) if non-daemonic processes have exited.
In addition to the threading.Thread API, Process objects also support the following attributes and methods:
Return the process ID. Before the process is spawned, this will be None.
The child’s exit code. This will be None if the process has not yet terminated. A negative value -N indicates that the child was terminated by signal N.
The process’s authentication key (a byte string).
When multiprocessing is initialized the main process is assigned a random string using os.urandom().
When a Process object is created, it will inherit the authentication key of its parent process, although this may be changed by setting authkey to another byte string.
See Authentication keys.
A numeric handle of a system object which will become “ready” when the process ends.
You can use this value if you want to wait on several events at once using multiprocessing.connection.wait(). Otherwise calling join() is simpler.
On Windows, this is an OS handle usable with the WaitForSingleObject and WaitForMultipleObjects family of API calls. On Unix, this is a file descriptor usable with primitives from the select module.
New in version 3.3.
Terminate the process. On Unix this is done using the SIGTERM signal; on Windows TerminateProcess() is used. Note that exit handlers and finally clauses, etc., will not be executed.
Note that descendant processes of the process will not be terminated – they will simply become orphaned.
Warning
If this method is used when the associated process is using a pipe or queue then the pipe or queue is liable to become corrupted and may become unusable by other process. Similarly, if the process has acquired a lock or semaphore etc. then terminating it is liable to cause other processes to deadlock.
Note that the start(), join(), is_alive(), terminate() and exitcode methods should only be called by the process that created the process object.
Example usage of some of the methods of Process:
>>> import multiprocessing, time, signal
>>> p = multiprocessing.Process(target=time.sleep, args=(1000,))
>>> print(p, p.is_alive())
<Process(Process-1, initial)> False
>>> p.start()
>>> print(p, p.is_alive())
<Process(Process-1, started)> True
>>> p.terminate()
>>> time.sleep(0.1)
>>> print(p, p.is_alive())
<Process(Process-1, stopped[SIGTERM])> False
>>> p.exitcode == -signal.SIGTERM
True
The base class of all multiprocessing exceptions.
Exception raised by Connection.recv_bytes_into() when the supplied buffer object is too small for the message read.
If e is an instance of BufferTooShort then e.args[0] will give the message as a byte string.
Raised when there is an authentication error.
Raised by methods with a timeout when the timeout expires.
When using multiple processes, one generally uses message passing for communication between processes and avoids having to use any synchronization primitives like locks.
For passing messages one can use Pipe() (for a connection between two processes) or a queue (which allows multiple producers and consumers).
The Queue, SimpleQueue and JoinableQueue types are multi-producer, multi-consumer FIFO queues modelled on the queue.Queue class in the standard library. They differ in that Queue lacks the task_done() and join() methods introduced into Python 2.5’s queue.Queue class.
If you use JoinableQueue then you must call JoinableQueue.task_done() for each task removed from the queue or else the semaphore used to count the number of unfinished tasks may eventually overflow, raising an exception.
Note that one can also create a shared queue by using a manager object – see Managers.
Note
multiprocessing uses the usual queue.Empty and queue.Full exceptions to signal a timeout. They are not available in the multiprocessing namespace so you need to import them from queue.
Note
When an object is put on a queue, the object is pickled and a background thread later flushes the pickled data to an underlying pipe. This has some consequences which are a little surprising, but should not cause any practical difficulties – if they really bother you then you can instead use a queue created with a manager.
Warning
If a process is killed using Process.terminate() or os.kill() while it is trying to use a Queue, then the data in the queue is likely to become corrupted. This may cause any other process to get an exception when it tries to use the queue later on.
Warning
As mentioned above, if a child process has put items on a queue (and it has not used JoinableQueue.cancel_join_thread), then that process will not terminate until all buffered items have been flushed to the pipe.
This means that if you try joining that process you may get a deadlock unless you are sure that all items which have been put on the queue have been consumed. Similarly, if the child process is non-daemonic then the parent process may hang on exit when it tries to join all its non-daemonic children.
Note that a queue created using a manager does not have this issue. See Programming guidelines.
For an example of the usage of queues for interprocess communication see Examples.
Returns a pair (conn1, conn2) of Connection objects representing the ends of a pipe.
If duplex is True (the default) then the pipe is bidirectional. If duplex is False then the pipe is unidirectional: conn1 can only be used for receiving messages and conn2 can only be used for sending messages.
Returns a process shared queue implemented using a pipe and a few locks/semaphores. When a process first puts an item on the queue a feeder thread is started which transfers objects from a buffer into the pipe.
The usual queue.Empty and queue.Full exceptions from the standard library’s queue module are raised to signal timeouts.
Queue implements all the methods of queue.Queue except for task_done() and join().
Return the approximate size of the queue. Because of multithreading/multiprocessing semantics, this number is not reliable.
Note that this may raise NotImplementedError on Unix platforms like Mac OS X where sem_getvalue() is not implemented.
Return True if the queue is empty, False otherwise. Because of multithreading/multiprocessing semantics, this is not reliable.
Return True if the queue is full, False otherwise. Because of multithreading/multiprocessing semantics, this is not reliable.
Put obj into the queue. If the optional argument block is True (the default) and timeout is None (the default), block if necessary until a free slot is available. If timeout is a positive number, it blocks at most timeout seconds and raises the queue.Full exception if no free slot was available within that time. Otherwise (block is False), put an item on the queue if a free slot is immediately available, else raise the queue.Full exception (timeout is ignored in that case).
Equivalent to put(obj, False).
Remove and return an item from the queue. If optional args block is True (the default) and timeout is None (the default), block if necessary until an item is available. If timeout is a positive number, it blocks at most timeout seconds and raises the queue.Empty exception if no item was available within that time. Otherwise (block is False), return an item if one is immediately available, else raise the queue.Empty exception (timeout is ignored in that case).
Equivalent to get(False).
multiprocessing.Queue has a few additional methods not found in queue.Queue. These methods are usually unnecessary for most code:
Indicate that no more data will be put on this queue by the current process. The background thread will quit once it has flushed all buffered data to the pipe. This is called automatically when the queue is garbage collected.
Join the background thread. This can only be used after close() has been called. It blocks until the background thread exits, ensuring that all data in the buffer has been flushed to the pipe.
By default if a process is not the creator of the queue then on exit it will attempt to join the queue’s background thread. The process can call cancel_join_thread() to make join_thread() do nothing.
Prevent join_thread() from blocking. In particular, this prevents the background thread from being joined automatically when the process exits – see join_thread().
A better name for this method might be allow_exit_without_flush(). It is likely to cause enqueued data to lost, and you almost certainly will not need to use it. It is really only there if you need the current process to exit immediately without waiting to flush enqueued data to the underlying pipe, and you don’t care about lost data.
It is a simplified Queue type, very close to a locked Pipe.
Return True if the queue is empty, False otherwise.
Remove and return an item from the queue.
Put item into the queue.
JoinableQueue, a Queue subclass, is a queue which additionally has task_done() and join() methods.
Indicate that a formerly enqueued task is complete. Used by queue consumers. For each get() used to fetch a task, a subsequent call to task_done() tells the queue that the processing on the task is complete.
If a join() is currently blocking, it will resume when all items have been processed (meaning that a task_done() call was received for every item that had been put() into the queue).
Raises a ValueError if called more times than there were items placed in the queue.
Block until all items in the queue have been gotten and processed.
The count of unfinished tasks goes up whenever an item is added to the queue. The count goes down whenever a consumer calls task_done() to indicate that the item was retrieved and all work on it is complete. When the count of unfinished tasks drops to zero, join() unblocks.
Return list of all live children of the current process.
Calling this has the side effect of “joining” any processes which have already finished.
Return the number of CPUs in the system. May raise NotImplementedError.
See also
Return the Process object corresponding to the current process.
An analogue of threading.current_thread().
Add support for when a program which uses multiprocessing has been frozen to produce a Windows executable. (Has been tested with py2exe, PyInstaller and cx_Freeze.)
One needs to call this function straight after the if __name__ == '__main__' line of the main module. For example:
from multiprocessing import Process, freeze_support
def f():
print('hello world!')
if __name__ == '__main__':
freeze_support()
Process(target=f).start()
If the freeze_support() line is omitted then trying to run the frozen executable will raise RuntimeError.
If the module is being run normally by the Python interpreter then freeze_support() has no effect.
Returns a list of the supported start methods, the first of which is the default. The possible start methods are 'fork', 'spawn' and 'forkserver'. On Windows only 'spawn' is available. On Unix 'fork' and 'spawn' are always supported, with 'fork' being the default.
New in version 3.4.
Return a context object which has the same attributes as the multiprocessing module.
If method is None then the default context is returned. Otherwise method should be 'fork', 'spawn', 'forkserver'. ValueError is raised if the specified start method is not available.
New in version 3.4.
Return the name of start method used for starting processes.
If the start method has not been fixed and allow_none is false, then the start method is fixed to the default and the name is returned. If the start method has not been fixed and allow_none is true then None is returned.
The return value can be 'fork', 'spawn', 'forkserver' or None. 'fork' is the default on Unix, while 'spawn' is the default on Windows.
New in version 3.4.
Sets the path of the Python interpreter to use when starting a child process. (By default sys.executable is used). Embedders will probably need to do some thing like
set_executable(os.path.join(sys.exec_prefix, 'pythonw.exe'))
before they can create child processes.
Changed in version 3.4: Now supported on Unix when the 'spawn' start method is used.
Set the method which should be used to start child processes. method can be 'fork', 'spawn' or 'forkserver'.
Note that this should be called at most once, and it should be protected inside the if __name__ == '__main__' clause of the main module.
New in version 3.4.
Note
multiprocessing contains no analogues of threading.active_count(), threading.enumerate(), threading.settrace(), threading.setprofile(), threading.Timer, or threading.local.
Connection objects allow the sending and receiving of picklable objects or strings. They can be thought of as message oriented connected sockets.
Connection objects are usually created using Pipe() – see also Listeners and Clients.
Send an object to the other end of the connection which should be read using recv().
The object must be picklable. Very large pickles (approximately 32 MB+, though it depends on the OS) may raise a ValueError exception.
Return an object sent from the other end of the connection using send(). Blocks until there its something to receive. Raises EOFError if there is nothing left to receive and the other end was closed.
Return the file descriptor or handle used by the connection.
Close the connection.
This is called automatically when the connection is garbage collected.
Return whether there is any data available to be read.
If timeout is not specified then it will return immediately. If timeout is a number then this specifies the maximum time in seconds to block. If timeout is None then an infinite timeout is used.
Note that multiple connection objects may be polled at once by using multiprocessing.connection.wait().
Send byte data from a bytes-like object as a complete message.
If offset is given then data is read from that position in buffer. If size is given then that many bytes will be read from buffer. Very large buffers (approximately 32 MB+, though it depends on the OS) may raise a ValueError exception
Return a complete message of byte data sent from the other end of the connection as a string. Blocks until there is something to receive. Raises EOFError if there is nothing left to receive and the other end has closed.
If maxlength is specified and the message is longer than maxlength then OSError is raised and the connection will no longer be readable.
Read into buffer a complete message of byte data sent from the other end of the connection and return the number of bytes in the message. Blocks until there is something to receive. Raises EOFError if there is nothing left to receive and the other end was closed.
buffer must be a writable bytes-like object. If offset is given then the message will be written into the buffer from that position. Offset must be a non-negative integer less than the length of buffer (in bytes).
If the buffer is too short then a BufferTooShort exception is raised and the complete message is available as e.args[0] where e is the exception instance.
Changed in version 3.3: Connection objects themselves can now be transferred between processes using Connection.send() and Connection.recv().
New in version 3.3: Connection objects now support the context management protocol – see Context Manager Types. __enter__() returns the connection object, and __exit__() calls close().
For example:
>>> from multiprocessing import Pipe
>>> a, b = Pipe()
>>> a.send([1, 'hello', None])
>>> b.recv()
[1, 'hello', None]
>>> b.send_bytes(b'thank you')
>>> a.recv_bytes()
b'thank you'
>>> import array
>>> arr1 = array.array('i', range(5))
>>> arr2 = array.array('i', [0] * 10)
>>> a.send_bytes(arr1)
>>> count = b.recv_bytes_into(arr2)
>>> assert count == len(arr1) * arr1.itemsize
>>> arr2
array('i', [0, 1, 2, 3, 4, 0, 0, 0, 0, 0])
Warning
The Connection.recv() method automatically unpickles the data it receives, which can be a security risk unless you can trust the process which sent the message.
Therefore, unless the connection object was produced using Pipe() you should only use the recv() and send() methods after performing some sort of authentication. See Authentication keys.
Warning
If a process is killed while it is trying to read or write to a pipe then the data in the pipe is likely to become corrupted, because it may become impossible to be sure where the message boundaries lie.
Generally synchronization primitives are not as necessary in a multiprocess program as they are in a multithreaded program. See the documentation for threading module.
Note that one can also create synchronization primitives by using a manager object – see Managers.
A barrier object: a clone of threading.Barrier.
New in version 3.3.
A bounded semaphore object: a clone of threading.BoundedSemaphore.
(On Mac OS X, this is indistinguishable from Semaphore because sem_getvalue() is not implemented on that platform).
A condition variable: an alias for threading.Condition.
If lock is specified then it should be a Lock or RLock object from multiprocessing.
Changed in version 3.3: The wait_for() method was added.
A clone of threading.Event.
A non-recursive lock object: a clone of threading.Lock.
A recursive lock object: a clone of threading.RLock.
A semaphore object: a clone of threading.Semaphore.
Note
The acquire() and wait() methods of each of these types treat negative timeouts as zero timeouts. This differs from threading where, since version 3.2, the equivalent acquire() methods treat negative timeouts as infinite timeouts.
On Mac OS X, sem_timedwait is unsupported, so calling acquire() with a timeout will emulate that function’s behavior using a sleeping loop.
Note
If the SIGINT signal generated by Ctrl-C arrives while the main thread is blocked by a call to BoundedSemaphore.acquire(), Lock.acquire(), RLock.acquire(), Semaphore.acquire(), Condition.acquire() or Condition.wait() then the call will be immediately interrupted and KeyboardInterrupt will be raised.
This differs from the behaviour of threading where SIGINT will be ignored while the equivalent blocking calls are in progress.
Managers provide a way to create data which can be shared between different processes, including sharing over a network between processes running on different machines. A manager object controls a server process which manages shared objects. Other processes can access the shared objects by using proxies.
Returns a started SyncManager object which can be used for sharing objects between processes. The returned manager object corresponds to a spawned child process and has methods which will create shared objects and return corresponding proxies.
Manager processes will be shutdown as soon as they are garbage collected or their parent process exits. The manager classes are defined in the multiprocessing.managers module:
Create a BaseManager object.
Once created one should call start() or get_server().serve_forever() to ensure that the manager object refers to a started manager process.
address is the address on which the manager process listens for new connections. If address is None then an arbitrary one is chosen.
authkey is the authentication key which will be used to check the validity of incoming connections to the server process. If authkey is None then current_process().authkey is used. Otherwise authkey is used and it must be a byte string.
Start a subprocess to start the manager. If initializer is not None then the subprocess will call initializer(*initargs) when it starts.
Returns a Server object which represents the actual server under the control of the Manager. The Server object supports the serve_forever() method:
>>> from multiprocessing.managers import BaseManager
>>> manager = BaseManager(address=('', 50000), authkey=b'abc')
>>> server = manager.get_server()
>>> server.serve_forever()
Server additionally has an address attribute.
Connect a local manager object to a remote manager process:
>>> from multiprocessing.managers import BaseManager
>>> m = BaseManager(address=('127.0.0.1', 5000), authkey=b'abc')
>>> m.connect()
Stop the process used by the manager. This is only available if start() has been used to start the server process.
This can be called multiple times.
A classmethod which can be used for registering a type or callable with the manager class.
typeid is a “type identifier” which is used to identify a particular type of shared object. This must be a string.
callable is a callable used for creating objects for this type identifier. If a manager instance will be connected to the server using the connect() method, or if the create_method argument is False then this can be left as None.
proxytype is a subclass of BaseProxy which is used to create proxies for shared objects with this typeid. If None then a proxy class is created automatically.
exposed is used to specify a sequence of method names which proxies for this typeid should be allowed to access using BaseProxy._callmethod(). (If exposed is None then proxytype._exposed_ is used instead if it exists.) In the case where no exposed list is specified, all “public methods” of the shared object will be accessible. (Here a “public method” means any attribute which has a __call__() method and whose name does not begin with '_'.)
method_to_typeid is a mapping used to specify the return type of those exposed methods which should return a proxy. It maps method names to typeid strings. (If method_to_typeid is None then proxytype._method_to_typeid_ is used instead if it exists.) If a method’s name is not a key of this mapping or if the mapping is None then the object returned by the method will be copied by value.
create_method determines whether a method should be created with name typeid which can be used to tell the server process to create a new shared object and return a proxy for it. By default it is True.
BaseManager instances also have one read-only property:
The address used by the manager.
Changed in version 3.3: Manager objects support the context management protocol – see Context Manager Types. __enter__() starts the server process (if it has not already started) and then returns the manager object. __exit__() calls shutdown().
In previous versions __enter__() did not start the manager’s server process if it was not already started.
A subclass of BaseManager which can be used for the synchronization of processes. Objects of this type are returned by multiprocessing.Manager().
It also supports creation of shared lists and dictionaries.
Create a shared threading.Barrier object and return a proxy for it.
New in version 3.3.
Create a shared threading.BoundedSemaphore object and return a proxy for it.
Create a shared threading.Condition object and return a proxy for it.
If lock is supplied then it should be a proxy for a threading.Lock or threading.RLock object.
Changed in version 3.3: The wait_for() method was added.
Create a shared threading.Event object and return a proxy for it.
Create a shared threading.Lock object and return a proxy for it.
Create a shared queue.Queue object and return a proxy for it.
Create a shared threading.RLock object and return a proxy for it.
Create a shared threading.Semaphore object and return a proxy for it.
Create an array and return a proxy for it.
Create an object with a writable value attribute and return a proxy for it.
Create a shared dict object and return a proxy for it.
Create a shared list object and return a proxy for it.
Note
Modifications to mutable values or items in dict and list proxies will not be propagated through the manager, because the proxy has no way of knowing when its values or items are modified. To modify such an item, you can re-assign the modified object to the container proxy:
# create a list proxy and append a mutable object (a dictionary)
lproxy = manager.list()
lproxy.append({})
# now mutate the dictionary
d = lproxy[0]
d['a'] = 1
d['b'] = 2
# at this point, the changes to d are not yet synced, but by
# reassigning the dictionary, the proxy is notified of the change
lproxy[0] = d
A namespace object has no public methods, but does have writable attributes. Its representation shows the values of its attributes.
However, when using a proxy for a namespace object, an attribute beginning with '_' will be an attribute of the proxy and not an attribute of the referent:
>>> manager = multiprocessing.Manager()
>>> Global = manager.Namespace()
>>> Global.x = 10
>>> Global.y = 'hello'
>>> Global._z = 12.3 # this is an attribute of the proxy
>>> print(Global)
Namespace(x=10, y='hello')
To create one’s own manager, one creates a subclass of BaseManager and uses the register() classmethod to register new types or callables with the manager class. For example:
from multiprocessing.managers import BaseManager
class MathsClass:
def add(self, x, y):
return x + y
def mul(self, x, y):
return x * y
class MyManager(BaseManager):
pass
MyManager.register('Maths', MathsClass)
if __name__ == '__main__':
with MyManager() as manager:
maths = manager.Maths()
print(maths.add(4, 3)) # prints 7
print(maths.mul(7, 8)) # prints 56
It is possible to run a manager server on one machine and have clients use it from other machines (assuming that the firewalls involved allow it).
Running the following commands creates a server for a single shared queue which remote clients can access:
>>> from multiprocessing.managers import BaseManager
>>> import queue
>>> queue = queue.Queue()
>>> class QueueManager(BaseManager): pass
>>> QueueManager.register('get_queue', callable=lambda:queue)
>>> m = QueueManager(address=('', 50000), authkey=b'abracadabra')
>>> s = m.get_server()
>>> s.serve_forever()
One client can access the server as follows:
>>> from multiprocessing.managers import BaseManager
>>> class QueueManager(BaseManager): pass
>>> QueueManager.register('get_queue')
>>> m = QueueManager(address=('foo.bar.org', 50000), authkey=b'abracadabra')
>>> m.connect()
>>> queue = m.get_queue()
>>> queue.put('hello')
Another client can also use it:
>>> from multiprocessing.managers import BaseManager
>>> class QueueManager(BaseManager): pass
>>> QueueManager.register('get_queue')
>>> m = QueueManager(address=('foo.bar.org', 50000), authkey=b'abracadabra')
>>> m.connect()
>>> queue = m.get_queue()
>>> queue.get()
'hello'
Local processes can also access that queue, using the code from above on the client to access it remotely:
>>> from multiprocessing import Process, Queue
>>> from multiprocessing.managers import BaseManager
>>> class Worker(Process):
... def __init__(self, q):
... self.q = q
... super(Worker, self).__init__()
... def run(self):
... self.q.put('local hello')
...
>>> queue = Queue()
>>> w = Worker(queue)
>>> w.start()
>>> class QueueManager(BaseManager): pass
...
>>> QueueManager.register('get_queue', callable=lambda: queue)
>>> m = QueueManager(address=('', 50000), authkey=b'abracadabra')
>>> s = m.get_server()
>>> s.serve_forever()
A proxy is an object which refers to a shared object which lives (presumably) in a different process. The shared object is said to be the referent of the proxy. Multiple proxy objects may have the same referent.
A proxy object has methods which invoke corresponding methods of its referent (although not every method of the referent will necessarily be available through the proxy). A proxy can usually be used in most of the same ways that its referent can:
>>> from multiprocessing import Manager
>>> manager = Manager()
>>> l = manager.list([i*i for i in range(10)])
>>> print(l)
[0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
>>> print(repr(l))
<ListProxy object, typeid 'list' at 0x...>
>>> l[4]
16
>>> l[2:5]
[4, 9, 16]
Notice that applying str() to a proxy will return the representation of the referent, whereas applying repr() will return the representation of the proxy.
An important feature of proxy objects is that they are picklable so they can be passed between processes. Note, however, that if a proxy is sent to the corresponding manager’s process then unpickling it will produce the referent itself. This means, for example, that one shared object can contain a second:
>>> a = manager.list()
>>> b = manager.list()
>>> a.append(b) # referent of a now contains referent of b
>>> print(a, b)
[[]] []
>>> b.append('hello')
>>> print(a, b)
[['hello']] ['hello']
Note
The proxy types in multiprocessing do nothing to support comparisons by value. So, for instance, we have:
>>> manager.list([1,2,3]) == [1,2,3]
False
One should just use a copy of the referent instead when making comparisons.
Proxy objects are instances of subclasses of BaseProxy.
Call and return the result of a method of the proxy’s referent.
If proxy is a proxy whose referent is obj then the expression
proxy._callmethod(methodname, args, kwds)
will evaluate the expression
getattr(obj, methodname)(*args, **kwds)
in the manager’s process.
The returned value will be a copy of the result of the call or a proxy to a new shared object – see documentation for the method_to_typeid argument of BaseManager.register().
If an exception is raised by the call, then is re-raised by _callmethod(). If some other exception is raised in the manager’s process then this is converted into a RemoteError exception and is raised by _callmethod().
Note in particular that an exception will be raised if methodname has not been exposed
An example of the usage of _callmethod():
>>> l = manager.list(range(10))
>>> l._callmethod('__len__')
10
>>> l._callmethod('__getslice__', (2, 7)) # equiv to `l[2:7]`
[2, 3, 4, 5, 6]
>>> l._callmethod('__getitem__', (20,)) # equiv to `l[20]`
Traceback (most recent call last):
...
IndexError: list index out of range
Return a copy of the referent.
If the referent is unpicklable then this will raise an exception.
Return a representation of the proxy object.
Return the representation of the referent.
A proxy object uses a weakref callback so that when it gets garbage collected it deregisters itself from the manager which owns its referent.
A shared object gets deleted from the manager process when there are no longer any proxies referring to it.
One can create a pool of processes which will carry out tasks submitted to it with the Pool class.
A process pool object which controls a pool of worker processes to which jobs can be submitted. It supports asynchronous results with timeouts and callbacks and has a parallel map implementation.
processes is the number of worker processes to use. If processes is None then the number returned by os.cpu_count() is used.
If initializer is not None then each worker process will call initializer(*initargs) when it starts.
maxtasksperchild is the number of tasks a worker process can complete before it will exit and be replaced with a fresh worker process, to enable unused resources to be freed. The default maxtasksperchild is None, which means worker processes will live as long as the pool.
context can be used to specify the context used for starting the worker processes. Usually a pool is created using the function multiprocessing.Pool() or the Pool() method of a context object. In both cases context is set appropriately.
Note that the methods of the pool object should only be called by the process which created the pool.
New in version 3.2: maxtasksperchild
New in version 3.4: context
Note
Worker processes within a Pool typically live for the complete duration of the Pool’s work queue. A frequent pattern found in other systems (such as Apache, mod_wsgi, etc) to free resources held by workers is to allow a worker within a pool to complete only a set amount of work before being exiting, being cleaned up and a new process spawned to replace the old one. The maxtasksperchild argument to the Pool exposes this ability to the end user.
Call func with arguments args and keyword arguments kwds. It blocks until the result is ready. Given this blocks, apply_async() is better suited for performing work in parallel. Additionally, func is only executed in one of the workers of the pool.
A variant of the apply() method which returns a result object.
If callback is specified then it should be a callable which accepts a single argument. When the result becomes ready callback is applied to it, that is unless the call failed, in which case the error_callback is applied instead
If error_callback is specified then it should be a callable which accepts a single argument. If the target function fails, then the error_callback is called with the exception instance.
Callbacks should complete immediately since otherwise the thread which handles the results will get blocked.
A parallel equivalent of the map() built-in function (it supports only one iterable argument though). It blocks until the result is ready.
This method chops the iterable into a number of chunks which it submits to the process pool as separate tasks. The (approximate) size of these chunks can be specified by setting chunksize to a positive integer.
A variant of the map() method which returns a result object.
If callback is specified then it should be a callable which accepts a single argument. When the result becomes ready callback is applied to it, that is unless the call failed, in which case the error_callback is applied instead
If error_callback is specified then it should be a callable which accepts a single argument. If the target function fails, then the error_callback is called with the exception instance.
Callbacks should complete immediately since otherwise the thread which handles the results will get blocked.
A lazier version of map().
The chunksize argument is the same as the one used by the map() method. For very long iterables using a large value for chunksize can make the job complete much faster than using the default value of 1.
Also if chunksize is 1 then the next() method of the iterator returned by the imap() method has an optional timeout parameter: next(timeout) will raise multiprocessing.TimeoutError if the result cannot be returned within timeout seconds.
The same as imap() except that the ordering of the results from the returned iterator should be considered arbitrary. (Only when there is only one worker process is the order guaranteed to be “correct”.)
Like map() except that the elements of the iterable are expected to be iterables that are unpacked as arguments.
Hence an iterable of [(1,2), (3, 4)] results in [func(1,2), func(3,4)].
New in version 3.3.
A combination of starmap() and map_async() that iterates over iterable of iterables and calls func with the iterables unpacked. Returns a result object.
New in version 3.3.
Prevents any more tasks from being submitted to the pool. Once all the tasks have been completed the worker processes will exit.
Stops the worker processes immediately without completing outstanding work. When the pool object is garbage collected terminate() will be called immediately.
Wait for the worker processes to exit. One must call close() or terminate() before using join().
New in version 3.3: Pool objects now support the context management protocol – see Context Manager Types. __enter__() returns the pool object, and __exit__() calls terminate().
The class of the result returned by Pool.apply_async() and Pool.map_async().
Return the result when it arrives. If timeout is not None and the result does not arrive within timeout seconds then multiprocessing.TimeoutError is raised. If the remote call raised an exception then that exception will be reraised by get().
Wait until the result is available or until timeout seconds pass.
Return whether the call has completed.
Return whether the call completed without raising an exception. Will raise AssertionError if the result is not ready.
The following example demonstrates the use of a pool:
from multiprocessing import Pool
def f(x):
return x*x
if __name__ == '__main__':
with Pool(processes=4) as pool: # start 4 worker processes
result = pool.apply_async(f, (10,)) # evaluate "f(10)" asynchronously
print(result.get(timeout=1)) # prints "100" unless your computer is *very* slow
print(pool.map(f, range(10))) # prints "[0, 1, 4,..., 81]"
it = pool.imap(f, range(10))
print(next(it)) # prints "0"
print(next(it)) # prints "1"
print(it.next(timeout=1)) # prints "4" unless your computer is *very* slow
import time
result = pool.apply_async(time.sleep, (10,))
print(result.get(timeout=1)) # raises TimeoutError
Usually message passing between processes is done using queues or by using Connection objects returned by Pipe().
However, the multiprocessing.connection module allows some extra flexibility. It basically gives a high level message oriented API for dealing with sockets or Windows named pipes. It also has support for digest authentication using the hmac module, and for polling multiple connections at the same time.
Send a randomly generated message to the other end of the connection and wait for a reply.
If the reply matches the digest of the message using authkey as the key then a welcome message is sent to the other end of the connection. Otherwise AuthenticationError is raised.
Receive a message, calculate the digest of the message using authkey as the key, and then send the digest back.
If a welcome message is not received, then AuthenticationError is raised.
Attempt to set up a connection to the listener which is using address address, returning a Connection.
The type of the connection is determined by family argument, but this can generally be omitted since it can usually be inferred from the format of address. (See Address Formats)
If authenticate is True or authkey is a byte string then digest authentication is used. The key used for authentication will be either authkey or current_process().authkey if authkey is None. If authentication fails then AuthenticationError is raised. See Authentication keys.
A wrapper for a bound socket or Windows named pipe which is ‘listening’ for connections.
address is the address to be used by the bound socket or named pipe of the listener object.
Note
If an address of ‘0.0.0.0’ is used, the address will not be a connectable end point on Windows. If you require a connectable end-point, you should use ‘127.0.0.1’.
family is the type of socket (or named pipe) to use. This can be one of the strings 'AF_INET' (for a TCP socket), 'AF_UNIX' (for a Unix domain socket) or 'AF_PIPE' (for a Windows named pipe). Of these only the first is guaranteed to be available. If family is None then the family is inferred from the format of address. If address is also None then a default is chosen. This default is the family which is assumed to be the fastest available. See Address Formats. Note that if family is 'AF_UNIX' and address is None then the socket will be created in a private temporary directory created using tempfile.mkstemp().
If the listener object uses a socket then backlog (1 by default) is passed to the listen() method of the socket once it has been bound.
If authenticate is True (False by default) or authkey is not None then digest authentication is used.
If authkey is a byte string then it will be used as the authentication key; otherwise it must be None.
If authkey is None and authenticate is True then current_process().authkey is used as the authentication key. If authkey is None and authenticate is False then no authentication is done. If authentication fails then AuthenticationError is raised. See Authentication keys.
Accept a connection on the bound socket or named pipe of the listener object and return a Connection object. If authentication is attempted and fails, then AuthenticationError is raised.
Close the bound socket or named pipe of the listener object. This is called automatically when the listener is garbage collected. However it is advisable to call it explicitly.
Listener objects have the following read-only properties:
The address which is being used by the Listener object.
The address from which the last accepted connection came. If this is unavailable then it is None.
New in version 3.3: Listener objects now support the context management protocol – see Context Manager Types. __enter__() returns the listener object, and __exit__() calls close().
Wait till an object in object_list is ready. Returns the list of those objects in object_list which are ready. If timeout is a float then the call blocks for at most that many seconds. If timeout is None then it will block for an unlimited period. A negative timeout is equivalent to a zero timeout.
For both Unix and Windows, an object can appear in object_list if it is
A connection or socket object is ready when there is data available to be read from it, or the other end has been closed.
Unix: wait(object_list, timeout) almost equivalent select.select(object_list, [], [], timeout). The difference is that, if select.select() is interrupted by a signal, it can raise OSError with an error number of EINTR, whereas wait() will not.
Windows: An item in object_list must either be an integer handle which is waitable (according to the definition used by the documentation of the Win32 function WaitForMultipleObjects()) or it can be an object with a fileno() method which returns a socket handle or pipe handle. (Note that pipe handles and socket handles are not waitable handles.)
New in version 3.3.
Examples
The following server code creates a listener which uses 'secret password' as an authentication key. It then waits for a connection and sends some data to the client:
from multiprocessing.connection import Listener
from array import array
address = ('localhost', 6000) # family is deduced to be 'AF_INET'
with Listener(address, authkey=b'secret password') as listener:
with listener.accept() as conn:
print('connection accepted from', listener.last_accepted)
conn.send([2.25, None, 'junk', float])
conn.send_bytes(b'hello')
conn.send_bytes(array('i', [42, 1729]))
The following code connects to the server and receives some data from the server:
from multiprocessing.connection import Client
from array import array
address = ('localhost', 6000)
with Client(address, authkey=b'secret password') as conn:
print(conn.recv()) # => [2.25, None, 'junk', float]
print(conn.recv_bytes()) # => 'hello'
arr = array('i', [0, 0, 0, 0, 0])
print(conn.recv_bytes_into(arr)) # => 8
print(arr) # => array('i', [42, 1729, 0, 0, 0])
The following code uses wait() to wait for messages from multiple processes at once:
import time, random
from multiprocessing import Process, Pipe, current_process
from multiprocessing.connection import wait
def foo(w):
for i in range(10):
w.send((i, current_process().name))
w.close()
if __name__ == '__main__':
readers = []
for i in range(4):
r, w = Pipe(duplex=False)
readers.append(r)
p = Process(target=foo, args=(w,))
p.start()
# We close the writable end of the pipe now to be sure that
# p is the only process which owns a handle for it. This
# ensures that when p closes its handle for the writable end,
# wait() will promptly report the readable end as being ready.
w.close()
while readers:
for r in wait(readers):
try:
msg = r.recv()
except EOFError:
readers.remove(r)
else:
print(msg)
An 'AF_INET' address is a tuple of the form (hostname, port) where hostname is a string and port is an integer.
An 'AF_UNIX' address is a string representing a filename on the filesystem.
r'\\.\pipe\PipeName'. To use Client() to connect to a named pipe on a remote computer called ServerName one should use an address of the form r'\\ServerName\pipe\PipeName' instead.
Note that any string beginning with two backslashes is assumed by default to be an 'AF_PIPE' address rather than an 'AF_UNIX' address.
When one uses Connection.recv, the data received is automatically unpickled. Unfortunately unpickling data from an untrusted source is a security risk. Therefore Listener and Client() use the hmac module to provide digest authentication.
An authentication key is a byte string which can be thought of as a password: once a connection is established both ends will demand proof that the other knows the authentication key. (Demonstrating that both ends are using the same key does not involve sending the key over the connection.)
If authentication is requested but no authentication key is specified then the return value of current_process().authkey is used (see Process). This value will automatically inherited by any Process object that the current process creates. This means that (by default) all processes of a multi-process program will share a single authentication key which can be used when setting up connections between themselves.
Suitable authentication keys can also be generated by using os.urandom().
Some support for logging is available. Note, however, that the logging package does not use process shared locks so it is possible (depending on the handler type) for messages from different processes to get mixed up.
Returns the logger used by multiprocessing. If necessary, a new one will be created.
When first created the logger has level logging.NOTSET and no default handler. Messages sent to this logger will not by default propagate to the root logger.
Note that on Windows child processes will only inherit the level of the parent process’s logger – any other customization of the logger will not be inherited.
This function performs a call to get_logger() but in addition to returning the logger created by get_logger, it adds a handler which sends output to sys.stderr using format '[%(levelname)s/%(processName)s] %(message)s'.
Below is an example session with logging turned on:
>>> import multiprocessing, logging
>>> logger = multiprocessing.log_to_stderr()
>>> logger.setLevel(logging.INFO)
>>> logger.warning('doomed')
[WARNING/MainProcess] doomed
>>> m = multiprocessing.Manager()
[INFO/SyncManager-...] child process calling self.run()
[INFO/SyncManager-...] created temp directory /.../pymp-...
[INFO/SyncManager-...] manager serving at '/.../listener-...'
>>> del m
[INFO/MainProcess] sending shutdown message to manager
[INFO/SyncManager-...] manager exiting with exitcode 0
For a full table of logging levels, see the logging module.
multiprocessing.dummy replicates the API of multiprocessing but is no more than a wrapper around the threading module.
There are certain guidelines and idioms which should be adhered to when using multiprocessing.
The following applies to all start methods.
Avoid shared state
As far as possible one should try to avoid shifting large amounts of data between processes.
It is probably best to stick to using queues or pipes for communication between processes rather than using the lower level synchronization primitives.
Picklability
Ensure that the arguments to the methods of proxies are picklable.
Thread safety of proxies
Do not use a proxy object from more than one thread unless you protect it with a lock.
(There is never a problem with different processes using the same proxy.)
Joining zombie processes
On Unix when a process finishes but has not been joined it becomes a zombie. There should never be very many because each time a new process starts (or active_children() is called) all completed processes which have not yet been joined will be joined. Also calling a finished process’s Process.is_alive will join the process. Even so it is probably good practice to explicitly join all the processes that you start.
Better to inherit than pickle/unpickle
When using the spawn or forkserver start methods many types from multiprocessing need to be picklable so that child processes can use them. However, one should generally avoid sending shared objects to other processes using pipes or queues. Instead you should arrange the program so that a process which needs access to a shared resource created elsewhere can inherit it from an ancestor process.
Avoid terminating processes
Using the Process.terminate method to stop a process is liable to cause any shared resources (such as locks, semaphores, pipes and queues) currently being used by the process to become broken or unavailable to other processes.
Therefore it is probably best to only consider using Process.terminate on processes which never use any shared resources.
Joining processes that use queues
Bear in mind that a process that has put items in a queue will wait before terminating until all the buffered items are fed by the “feeder” thread to the underlying pipe. (The child process can call the Queue.cancel_join_thread method of the queue to avoid this behaviour.)
This means that whenever you use a queue you need to make sure that all items which have been put on the queue will eventually be removed before the process is joined. Otherwise you cannot be sure that processes which have put items on the queue will terminate. Remember also that non-daemonic processes will be joined automatically.
An example which will deadlock is the following:
from multiprocessing import Process, Queue def f(q): q.put('X' * 1000000) if __name__ == '__main__': queue = Queue() p = Process(target=f, args=(queue,)) p.start() p.join() # this deadlocks obj = queue.get()A fix here would be to swap the last two lines (or simply remove the p.join() line).
Explicitly pass resources to child processes
On Unix using the fork start method, a child process can make use of a shared resource created in a parent process using a global resource. However, it is better to pass the object as an argument to the constructor for the child process.
Apart from making the code (potentially) compatible with Windows and the other start methods this also ensures that as long as the child process is still alive the object will not be garbage collected in the parent process. This might be important if some resource is freed when the object is garbage collected in the parent process.
So for instance
from multiprocessing import Process, Lock def f(): ... do something using "lock" ... if __name__ == '__main__': lock = Lock() for i in range(10): Process(target=f).start()should be rewritten as
from multiprocessing import Process, Lock def f(l): ... do something using "l" ... if __name__ == '__main__': lock = Lock() for i in range(10): Process(target=f, args=(lock,)).start()
Beware of replacing sys.stdin with a “file like object”
multiprocessing originally unconditionally called:
os.close(sys.stdin.fileno())in the multiprocessing.Process._bootstrap() method — this resulted in issues with processes-in-processes. This has been changed to:
sys.stdin.close() sys.stdin = open(os.devnull)Which solves the fundamental issue of processes colliding with each other resulting in a bad file descriptor error, but introduces a potential danger to applications which replace sys.stdin() with a “file-like object” with output buffering. This danger is that if multiple processes call close() on this file-like object, it could result in the same data being flushed to the object multiple times, resulting in corruption.
If you write a file-like object and implement your own caching, you can make it fork-safe by storing the pid whenever you append to the cache, and discarding the cache when the pid changes. For example:
@property def cache(self): pid = os.getpid() if pid != self._pid: self._pid = pid self._cache = [] return self._cacheFor more information, see issue 5155, issue 5313 and issue 5331
There are a few extra restriction which don’t apply to the fork start method.
More picklability
Ensure that all arguments to Process.__init__() are picklable. This means, in particular, that bound or unbound methods cannot be used directly as the target (unless you use the fork start method) — just define a function and use that instead.
Also, if you subclass Process then make sure that instances will be picklable when the Process.start method is called.
Global variables
Bear in mind that if code run in a child process tries to access a global variable, then the value it sees (if any) may not be the same as the value in the parent process at the time that Process.start was called.
However, global variables which are just module level constants cause no problems.
Safe importing of main module
Make sure that the main module can be safely imported by a new Python interpreter without causing unintended side effects (such a starting a new process).
For example, using the spawn or forkserver start method running the following module would fail with a RuntimeError:
from multiprocessing import Process def foo(): print('hello') p = Process(target=foo) p.start()Instead one should protect the “entry point” of the program by using if __name__ == '__main__': as follows:
from multiprocessing import Process, freeze_support, set_start_method def foo(): print('hello') if __name__ == '__main__': freeze_support() set_start_method('spawn') p = Process(target=foo) p.start()(The freeze_support() line can be omitted if the program will be run normally instead of frozen.)
This allows the newly spawned Python interpreter to safely import the module and then run the module’s foo() function.
Similar restrictions apply if a pool or manager is created in the main module.
Demonstration of how to create and use customized managers and proxies:
from multiprocessing import freeze_support
from multiprocessing.managers import BaseManager, BaseProxy
import operator
##
class Foo:
def f(self):
print('you called Foo.f()')
def g(self):
print('you called Foo.g()')
def _h(self):
print('you called Foo._h()')
# A simple generator function
def baz():
for i in range(10):
yield i*i
# Proxy type for generator objects
class GeneratorProxy(BaseProxy):
_exposed_ = ['__next__']
def __iter__(self):
return self
def __next__(self):
return self._callmethod('__next__')
# Function to return the operator module
def get_operator_module():
return operator
##
class MyManager(BaseManager):
pass
# register the Foo class; make `f()` and `g()` accessible via proxy
MyManager.register('Foo1', Foo)
# register the Foo class; make `g()` and `_h()` accessible via proxy
MyManager.register('Foo2', Foo, exposed=('g', '_h'))
# register the generator function baz; use `GeneratorProxy` to make proxies
MyManager.register('baz', baz, proxytype=GeneratorProxy)
# register get_operator_module(); make public functions accessible via proxy
MyManager.register('operator', get_operator_module)
##
def test():
manager = MyManager()
manager.start()
print('-' * 20)
f1 = manager.Foo1()
f1.f()
f1.g()
assert not hasattr(f1, '_h')
assert sorted(f1._exposed_) == sorted(['f', 'g'])
print('-' * 20)
f2 = manager.Foo2()
f2.g()
f2._h()
assert not hasattr(f2, 'f')
assert sorted(f2._exposed_) == sorted(['g', '_h'])
print('-' * 20)
it = manager.baz()
for i in it:
print('<%d>' % i, end=' ')
print()
print('-' * 20)
op = manager.operator()
print('op.add(23, 45) =', op.add(23, 45))
print('op.pow(2, 94) =', op.pow(2, 94))
print('op._exposed_ =', op._exposed_)
##
if __name__ == '__main__':
freeze_support()
test()
Using Pool:
import multiprocessing
import time
import random
import sys
#
# Functions used by test code
#
def calculate(func, args):
result = func(*args)
return '%s says that %s%s = %s' % (
multiprocessing.current_process().name,
func.__name__, args, result
)
def calculatestar(args):
return calculate(*args)
def mul(a, b):
time.sleep(0.5 * random.random())
return a * b
def plus(a, b):
time.sleep(0.5 * random.random())
return a + b
def f(x):
return 1.0 / (x - 5.0)
def pow3(x):
return x ** 3
def noop(x):
pass
#
# Test code
#
def test():
PROCESSES = 4
print('Creating pool with %d processes\n' % PROCESSES)
with multiprocessing.Pool(PROCESSES) as pool:
#
# Tests
#
TASKS = [(mul, (i, 7)) for i in range(10)] + \
[(plus, (i, 8)) for i in range(10)]
results = [pool.apply_async(calculate, t) for t in TASKS]
imap_it = pool.imap(calculatestar, TASKS)
imap_unordered_it = pool.imap_unordered(calculatestar, TASKS)
print('Ordered results using pool.apply_async():')
for r in results:
print('\t', r.get())
print()
print('Ordered results using pool.imap():')
for x in imap_it:
print('\t', x)
print()
print('Unordered results using pool.imap_unordered():')
for x in imap_unordered_it:
print('\t', x)
print()
print('Ordered results using pool.map() --- will block till complete:')
for x in pool.map(calculatestar, TASKS):
print('\t', x)
print()
#
# Test error handling
#
print('Testing error handling:')
try:
print(pool.apply(f, (5,)))
except ZeroDivisionError:
print('\tGot ZeroDivisionError as expected from pool.apply()')
else:
raise AssertionError('expected ZeroDivisionError')
try:
print(pool.map(f, list(range(10))))
except ZeroDivisionError:
print('\tGot ZeroDivisionError as expected from pool.map()')
else:
raise AssertionError('expected ZeroDivisionError')
try:
print(list(pool.imap(f, list(range(10)))))
except ZeroDivisionError:
print('\tGot ZeroDivisionError as expected from list(pool.imap())')
else:
raise AssertionError('expected ZeroDivisionError')
it = pool.imap(f, list(range(10)))
for i in range(10):
try:
x = next(it)
except ZeroDivisionError:
if i == 5:
pass
except StopIteration:
break
else:
if i == 5:
raise AssertionError('expected ZeroDivisionError')
assert i == 9
print('\tGot ZeroDivisionError as expected from IMapIterator.next()')
print()
#
# Testing timeouts
#
print('Testing ApplyResult.get() with timeout:', end=' ')
res = pool.apply_async(calculate, TASKS[0])
while 1:
sys.stdout.flush()
try:
sys.stdout.write('\n\t%s' % res.get(0.02))
break
except multiprocessing.TimeoutError:
sys.stdout.write('.')
print()
print()
print('Testing IMapIterator.next() with timeout:', end=' ')
it = pool.imap(calculatestar, TASKS)
while 1:
sys.stdout.flush()
try:
sys.stdout.write('\n\t%s' % it.next(0.02))
except StopIteration:
break
except multiprocessing.TimeoutError:
sys.stdout.write('.')
print()
print()
if __name__ == '__main__':
multiprocessing.freeze_support()
test()
An example showing how to use queues to feed tasks to a collection of worker processes and collect the results:
import time
import random
from multiprocessing import Process, Queue, current_process, freeze_support
#
# Function run by worker processes
#
def worker(input, output):
for func, args in iter(input.get, 'STOP'):
result = calculate(func, args)
output.put(result)
#
# Function used to calculate result
#
def calculate(func, args):
result = func(*args)
return '%s says that %s%s = %s' % \
(current_process().name, func.__name__, args, result)
#
# Functions referenced by tasks
#
def mul(a, b):
time.sleep(0.5*random.random())
return a * b
def plus(a, b):
time.sleep(0.5*random.random())
return a + b
#
#
#
def test():
NUMBER_OF_PROCESSES = 4
TASKS1 = [(mul, (i, 7)) for i in range(20)]
TASKS2 = [(plus, (i, 8)) for i in range(10)]
# Create queues
task_queue = Queue()
done_queue = Queue()
# Submit tasks
for task in TASKS1:
task_queue.put(task)
# Start worker processes
for i in range(NUMBER_OF_PROCESSES):
Process(target=worker, args=(task_queue, done_queue)).start()
# Get and print results
print('Unordered results:')
for i in range(len(TASKS1)):
print('\t', done_queue.get())
# Add more tasks using `put()`
for task in TASKS2:
task_queue.put(task)
# Get and print some more results
for i in range(len(TASKS2)):
print('\t', done_queue.get())
# Tell child processes to stop
for i in range(NUMBER_OF_PROCESSES):
task_queue.put('STOP')
if __name__ == '__main__':
freeze_support()
test()