multiprocessing.shared_memory — Shared memory for direct access across processes¶
Source code: Lib/multiprocessing/shared_memory.py
New in version 3.8.
This module provides a class,
SharedMemory, for the allocation
and management of shared memory to be accessed by one or more processes
on a multicore or symmetric multiprocessor (SMP) machine. To assist with
the life-cycle management of shared memory especially across distinct
SharedMemoryManager, is also provided in the
In this module, shared memory refers to “System V style” shared memory blocks (though is not necessarily implemented explicitly as such) and does not refer to “distributed shared memory”. This style of shared memory permits distinct processes to potentially read and write to a common (or shared) region of volatile memory. Processes are conventionally limited to only have access to their own process memory space but shared memory permits the sharing of data between processes, avoiding the need to instead send messages between processes containing that data. Sharing data directly via memory can provide significant performance benefits compared to sharing data via disk or socket or other communications requiring the serialization/deserialization and copying of data.
- class multiprocessing.shared_memory.SharedMemory(name=None, create=False, size=0)¶
Creates a new shared memory block or attaches to an existing shared memory block. Each shared memory block is assigned a unique name. In this way, one process can create a shared memory block with a particular name and a different process can attach to that same shared memory block using that same name.
As a resource for sharing data across processes, shared memory blocks may outlive the original process that created them. When one process no longer needs access to a shared memory block that might still be needed by other processes, the
close()method should be called. When a shared memory block is no longer needed by any process, the
unlink()method should be called to ensure proper cleanup.
name is the unique name for the requested shared memory, specified as a string. When creating a new shared memory block, if
None(the default) is supplied for the name, a novel name will be generated.
create controls whether a new shared memory block is created (
True) or an existing shared memory block is attached (
size specifies the requested number of bytes when creating a new shared memory block. Because some platforms choose to allocate chunks of memory based upon that platform’s memory page size, the exact size of the shared memory block may be larger or equal to the size requested. When attaching to an existing shared memory block, the
sizeparameter is ignored.
Closes access to the shared memory from this instance. In order to ensure proper cleanup of resources, all instances should call
close()once the instance is no longer needed. Note that calling
close()does not cause the shared memory block itself to be destroyed.
Requests that the underlying shared memory block be destroyed. In order to ensure proper cleanup of resources,
unlink()should be called once (and only once) across all processes which have need for the shared memory block. After requesting its destruction, a shared memory block may or may not be immediately destroyed and this behavior may differ across platforms. Attempts to access data inside the shared memory block after
unlink()has been called may result in memory access errors. Note: the last process relinquishing its hold on a shared memory block may call
close()in either order.
A memoryview of contents of the shared memory block.
Read-only access to the unique name of the shared memory block.
Read-only access to size in bytes of the shared memory block.
The following example demonstrates low-level use of
>>> from multiprocessing import shared_memory >>> shm_a = shared_memory.SharedMemory(create=True, size=10) >>> type(shm_a.buf) <class 'memoryview'> >>> buffer = shm_a.buf >>> len(buffer) 10 >>> buffer[:4] = bytearray([22, 33, 44, 55]) # Modify multiple at once >>> buffer = 100 # Modify single byte at a time >>> # Attach to an existing shared memory block >>> shm_b = shared_memory.SharedMemory(shm_a.name) >>> import array >>> array.array('b', shm_b.buf[:5]) # Copy the data into a new array.array array('b', [22, 33, 44, 55, 100]) >>> shm_b.buf[:5] = b'howdy' # Modify via shm_b using bytes >>> bytes(shm_a.buf[:5]) # Access via shm_a b'howdy' >>> shm_b.close() # Close each SharedMemory instance >>> shm_a.close() >>> shm_a.unlink() # Call unlink only once to release the shared memory
>>> # In the first Python interactive shell >>> import numpy as np >>> a = np.array([1, 1, 2, 3, 5, 8]) # Start with an existing NumPy array >>> from multiprocessing import shared_memory >>> shm = shared_memory.SharedMemory(create=True, size=a.nbytes) >>> # Now create a NumPy array backed by shared memory >>> b = np.ndarray(a.shape, dtype=a.dtype, buffer=shm.buf) >>> b[:] = a[:] # Copy the original data into shared memory >>> b array([1, 1, 2, 3, 5, 8]) >>> type(b) <class 'numpy.ndarray'> >>> type(a) <class 'numpy.ndarray'> >>> shm.name # We did not specify a name so one was chosen for us 'psm_21467_46075' >>> # In either the same shell or a new Python shell on the same machine >>> import numpy as np >>> from multiprocessing import shared_memory >>> # Attach to the existing shared memory block >>> existing_shm = shared_memory.SharedMemory(name='psm_21467_46075') >>> # Note that a.shape is (6,) and a.dtype is np.int64 in this example >>> c = np.ndarray((6,), dtype=np.int64, buffer=existing_shm.buf) >>> c array([1, 1, 2, 3, 5, 8]) >>> c[-1] = 888 >>> c array([ 1, 1, 2, 3, 5, 888]) >>> # Back in the first Python interactive shell, b reflects this change >>> b array([ 1, 1, 2, 3, 5, 888]) >>> # Clean up from within the second Python shell >>> del c # Unnecessary; merely emphasizing the array is no longer used >>> existing_shm.close() >>> # Clean up from within the first Python shell >>> del b # Unnecessary; merely emphasizing the array is no longer used >>> shm.close() >>> shm.unlink() # Free and release the shared memory block at the very end
- class multiprocessing.managers.SharedMemoryManager([address[, authkey]])¶
A subclass of
BaseManagerwhich can be used for the management of shared memory blocks across processes.
A call to
SharedMemoryManagerinstance causes a new process to be started. This new process’s sole purpose is to manage the life cycle of all shared memory blocks created through it. To trigger the release of all shared memory blocks managed by that process, call
shutdown()on the instance. This triggers a
SharedMemory.unlink()call on all of the
SharedMemoryobjects managed by that process and then stops the process itself. By creating
SharedMemoryinstances through a
SharedMemoryManager, we avoid the need to manually track and trigger the freeing of shared memory resources.
multiprocessing.managers.BaseManagerfor a description of the inherited address and authkey optional input arguments and how they may be used to connect to an existing
SharedMemoryManagerservice from other processes.
The following example demonstrates the basic mechanisms of a
>>> from multiprocessing.managers import SharedMemoryManager >>> smm = SharedMemoryManager() >>> smm.start() # Start the process that manages the shared memory blocks >>> sl = smm.ShareableList(range(4)) >>> sl ShareableList([0, 1, 2, 3], name='psm_6572_7512') >>> raw_shm = smm.SharedMemory(size=128) >>> another_sl = smm.ShareableList('alpha') >>> another_sl ShareableList(['a', 'l', 'p', 'h', 'a'], name='psm_6572_12221') >>> smm.shutdown() # Calls unlink() on sl, raw_shm, and another_sl
The following example depicts a potentially more convenient pattern for using
SharedMemoryManager objects via the
with statement to
ensure that all shared memory blocks are released after they are no longer
>>> with SharedMemoryManager() as smm: ... sl = smm.ShareableList(range(2000)) ... # Divide the work among two processes, storing partial results in sl ... p1 = Process(target=do_work, args=(sl, 0, 1000)) ... p2 = Process(target=do_work, args=(sl, 1000, 2000)) ... p1.start() ... p2.start() # A multiprocessing.Pool might be more efficient ... p1.join() ... p2.join() # Wait for all work to complete in both processes ... total_result = sum(sl) # Consolidate the partial results now in sl
- class multiprocessing.shared_memory.ShareableList(sequence=None, \*, name=None)¶
Provides a mutable list-like object where all values stored within are stored in a shared memory block. This constrains storable values to only the
str(less than 10M bytes each when encoded as utf-8),
bytes(less than 10M bytes each), and
Nonebuilt-in data types. It also notably differs from the built-in
listtype in that these lists can not change their overall length (i.e. no append, insert, etc.) and do not support the dynamic creation of new
ShareableListinstances via slicing.
sequence is used in populating a new
ShareableListfull of values. Set to
Noneto instead attach to an already existing
ShareableListby its unique shared memory name.
name is the unique name for the requested shared memory, as described in the definition for
SharedMemory. When attaching to an existing
ShareableList, specify its shared memory block’s unique name while leaving
A known issue exists for
strvalues. If they end with
\x00nul bytes or characters, those may be silently stripped when fetching them by index from the
.rstrip(b'\x00')behavior is considered a bug and may go away in the future. See gh-106939.
For applications where rstripping of trailing nulls is a problem, work around it by always unconditionally appending an extra non-0 byte to the end of such values when storing and unconditionally removing it when fetching:
>>> from multiprocessing import shared_memory >>> nul_bug_demo = shared_memory.ShareableList(['?\x00', b'\x03\x02\x01\x00\x00\x00']) >>> nul_bug_demo '?' >>> nul_bug_demo b'\x03\x02\x01' >>> nul_bug_demo.shm.unlink() >>> padded = shared_memory.ShareableList(['?\x00\x07', b'\x03\x02\x01\x00\x00\x00\x07']) >>> padded[:-1] '?\x00' >>> padded[:-1] b'\x03\x02\x01\x00\x00\x00' >>> padded.shm.unlink()
Returns the number of occurrences of
Read-only attribute containing the
structpacking format used by all currently stored values.
The following example demonstrates basic use of a
>>> from multiprocessing import shared_memory >>> a = shared_memory.ShareableList(['howdy', b'HoWdY', -273.154, 100, None, True, 42]) >>> [ type(entry) for entry in a ] [<class 'str'>, <class 'bytes'>, <class 'float'>, <class 'int'>, <class 'NoneType'>, <class 'bool'>, <class 'int'>] >>> a -273.154 >>> a = -78.5 >>> a -78.5 >>> a = 'dry ice' # Changing data types is supported as well >>> a 'dry ice' >>> a = 'larger than previously allocated storage space' Traceback (most recent call last): ... ValueError: exceeds available storage for existing str >>> a 'dry ice' >>> len(a) 7 >>> a.index(42) 6 >>> a.count(b'howdy') 0 >>> a.count(b'HoWdY') 1 >>> a.shm.close() >>> a.shm.unlink() >>> del a # Use of a ShareableList after call to unlink() is unsupported
The following example depicts how one, two, or many processes may access the
ShareableList by supplying the name of the shared memory block
>>> b = shared_memory.ShareableList(range(5)) # In a first process >>> c = shared_memory.ShareableList(name=b.shm.name) # In a second process >>> c ShareableList([0, 1, 2, 3, 4], name='...') >>> c[-1] = -999 >>> b[-1] -999 >>> b.shm.close() >>> c.shm.close() >>> c.shm.unlink()
The following examples demonstrates that
can be pickled and unpickled if needed.
Note, that it will still be the same shared object.
This happens, because the deserialized object has
the same unique name and is just attached to an existing
object with the same name (if the object is still alive):
>>> import pickle >>> from multiprocessing import shared_memory >>> sl = shared_memory.ShareableList(range(10)) >>> list(sl) [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
>>> deserialized_sl = pickle.loads(pickle.dumps(sl)) >>> list(deserialized_sl) [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
>>> sl = -1 >>> deserialized_sl = -2 >>> list(sl) [-1, -2, 2, 3, 4, 5, 6, 7, 8, 9] >>> list(deserialized_sl) [-1, -2, 2, 3, 4, 5, 6, 7, 8, 9]
>>> sl.shm.close() >>> sl.shm.unlink()