The processing package mostly replicates the API of the threading module.
- class Process(group=None, target=None, name=None, args=(), kwargs={})
An analogue of threading.Thread.
See Process objects.
- exception ProcessExit
- Exception raised in a target process when the Process.stop() method is used. This is a subclass of SystemExit.
- exception BufferTooShort
Exception raise by the recvbytes_into() method of a connection object 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.
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 a pipe (for a connection between two processes) or a queue (which allows multiple producers and consumers).
Note that one can also create a shared queue by using a manager object -- see Managers.
For an example of the usage of queues for interprocess communication see test_workers.py.
- Pipe()
Returns a pair of connection objects representing the ends of a duplex pipe.
These connection objects can be inherited by child processes and have methods send() and recv() (among others) for sending and receiving picklable objects. (See Connection objects.) For example:
>>> from processing import Pipe >>> a, b = Pipe() >>> a.send([1, 'hello', None]) >>> b.recv() [1, 'hello', None] >>> b.sendbytes('thank you') >>> a.recvbytes() 'thank you'Note that it is not safe to have more than one process (or thread) reading or writing to the same end of a pipe at the same time.
On Windows this requires the _processing extension.
- Queue(maxsize=0)
Alias for either PosixQueue if available or else PipeQueue -- see below.
It differs from Python standard Queue.Queue type by only have finite capacity, so even when maxsize is specified as 0 the put() method might block. If you need to be sure that put() will not block then you should use BufferedQueue() instead.
- BufferedQueue()
Alias for either BufferedPosixQueue if available or else BufferedPipeQueue.
Differs from Queue() by guaranteeing that the put() method will not block. A buffered queue tends to be slower than a normal queue if you are putting items on the queue one by one (rather than using putmany()).
The first time a process puts an item on a buffered queue a thread is started whose job is to transfer items from a buffer onto the true interprocess queue. In addition to the usual queue methods BufferedQueue supports two extra:
- putmany(iterable)
- Adds all items in the iterable to the queue's buffer. So q.putmany(X) is a faster equivalent of for x in X: q.put(x).
- close()
- Flushes data from the buffer to the interprocess queue, then instructs the thread to stop and waits for it to do so. This will be called automatically when the queue is garbage collected or when the process exits.
- PipeQueue(maxsize=0)
Returns a near clone of Queue.Queue except that the qsize() method is not implemented. It is implemented using a pipe and some locks/semaphores.
On Unix if a client terminates while it is reading or writing from the queue, other clients reading from the queue may lose track of where messages boundaries are or may retrieve incomplete messages. At least on Windows a keyboard interrupt (SIGINT) or the use of a process's stop() method should not cause such a problem.
Placing an object on a PipeQueue can block because of lack of buffer space even if a zero timeout is used.
Requires support for native semaphore support from _processing.
- PosixQueue(maxsize=0, msgsize=0)
Returns a near clone of Queue.Queue implemented using a (Unix only) posix message queue.
If maxsize is non-zero it determines the maximum number of messages that can be in the queue and if msgsize is non-zero it determines the maximum size in bytes of a message. If either is zero then the system default (which is finite) is used. (For instance on my Linux system the default maximum number of messages in a queue is 10 and the maximum message size is 8192 bytes.) A PosixQueue object has attributes _maxmsg and _maxsize which give these limits for that queue.
Note that if maxsize or msgsize is larger than the system maximum then an OSError exception will be thrown. On Linux the system maximums can viewed and modified through the /proc filesystem --- see man 7 mq_overview.
Only available on Unix and only if support for posix queues was built in to _processing.
Generally synchronization primitives are not a necessary in a multiprocess program as they are in a mulithreaded program.
Note that one can also create synchronization primitves by using a manager object -- see Managers.
The following all require support for native sempahores from the _processing extension.
- BoundedSemaphore(value=1)
- Returns a bounded semaphore object: a clone of threading.BoundedSemaphore.
- Condition(lock=None)
Returns a condition variable: a clone of threading.Condition.
If lock is specified then it should be a Lock or RLock object from processing.
- Event()
- Returns an event object: a clone of threading.Event.
- Lock()
Returns a non-recursive lock object: a near clone of threading.Lock.
There are two differences from threading.Lock: trying to acquire a lock already owned by the current thread raises an exception instead of deadlocking; and trying to release a lock held by a different thread/process will raise and exception.
- RLock()
- Returns a recursive lock object: a clone of threading.RLock.
- Semaphore(value=1)
- Returns a bounded semaphore object: a clone of threading.Semaphore.
Managers provide a way to create data which can be shared between different processes.
- LocalManager()
Returns a manager object which uses shared memory instead of a server process. It has instance methods
SharedValue, SharedStruct, SharedArrayfor creating objects stored in shared memory map. Also has static methods
Lock, RLock, Semaphore, BoundedSemaphore, Condition, Event, Queuewhich are just aliases for other functions in the processing namespace. See LocalManager.
Requires support for native semaphores from _processing.
- Manager()
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.
The methods for creating shared objects are
list(), dict(), Namespace(), SharedValue(), SharedStruct(), SharedArray(), Lock(), RLock(), Semaphore(), BoundedSemaphore(), Condition(), Event(), Queue().For example:
>>> from processing import Manager >>> manager = Manager() >>> l = manager.list(range(10)) >>> l.reverse() >>> print l [9, 8, 7, 6, 5, 4, 3, 2, 1, 0] >>> print repr(l) <Proxy[list] object at 0x00E1B3B0>See SyncManager and Proxy objects.
One can create a pool of processes which will carry out tasks submitted to it.
- Pool(processes=None)
Returns 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.
If processes is None then the number returned by cpuCount() is used. See Pool objects.
Example:
from processing import Pool def f(x): return x*x if __name__ == '__main__': pool = Pool(processes=2) result1 = pool.apply_async(f, (10,)) result2 = pool.map_async(f, range(5)) print result1.get() # => "100" print result2.get(timeout=1) # => "[0, 1, 4, 9, 16]"Requires support for native semaphores from _processing.
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.
- enableLogging(level, HandlerType=None, handlerArgs=(), format=None)
Enables logging and sets the debug level to level -- see documentation for the logging package in the standard library.
If HandlerType is specified then a handler is created using HandlerType(*handlerArgs) and this will be used by the logger -- any previous handlers will be discarded. If format is specified then this will be used for the handler; otherwise format defaults to '[%(levelname)s/%(processName)s] %(message)s'. (The logger used by processing allows use of the non-standard '%(processName)s' format.)
If HandlerType is not specified and the logger has no handlers then a default one is created which prints to sys.stderr.
Note: on Windows a child process does not directly inherit its parent's logger; instead it will automatically call enableLogging() with the same arguments which were used when its parent process last called enableLogging() (if it ever did).
- getLogger()
- Returns the logger used by processing. If enableLogging() has not yet been called then None is returned.
Below is an example session with logging turned on:
>>> import processing, logging >>> processing.enableLogging(level=logging.INFO) >>> processing.getLogger().warn('doomed') [WARNING/MainProcess] doomed >>> m = processing.Manager() [INFO/SyncManager-1] process starting up [INFO/SyncManager-1] manager bound to '\\\\.\\pipe\\pyc-1352-0-r97d0b' >>> del m [INFO/MainProcess] sending shutdown message to manager [INFO/SyncManager-1] manager received shutdown message [INFO/SyncManager-1] running all "atexit" finalizers [INFO/SyncManager-1] process exiting with `os.exit(0)`
- activeChildren()
Return list of all live children of the current process.
Calling this has the side affect of "joining" any processes which have already finished.
- cpuCount()
- Returns the number of CPUs in the system. May raise NotImplementedError.
- currentProcess()
An analogue of threading.currentThread
Returns the object corresponding to the current process.
- freezeSupport()
Adds support for when a program which uses the processing package 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 processing import Process, freezeSupport def f(): print "hello world!" if __name__ == '__main__': freezeSupport() p = Process(target=f) p.start()If the freezeSupport() line is missed out then the frozen executable produced from this module would (on Windows) recursively create new processes.
If the module is being run normally by the python interpreter then freezeSupport() has no effect.
Note