Python IPC Modules

My Python extensions posix_ipc and sysv_ipc are alternatives to the modules for IPC (inter-process communication) in the Python standard library.

They provide direct access to the low-level IPC primitives semaphores, shared memory and message queues. If you don't need low-level access, you're likely to find a higher level module like Python's multiprocessing easier to use.

I'm the author of posix_ipc and sysv_ipc and I maintain the other module mentioned here (shm, which is deprecated). They're all open source and do similar things. For the most part, they only work on Unix, but posix_ipc also works on Windows with Cygwin (see below).

Why Three Modules?

There's The Old Stuff...

Vladimir Marangozov wrote shm sometime in the 1990s. I adopted it in 2007. It provides access to System V semaphores and shared memory.

The interface has some un-Pythonic parts, so I added a Python wrapper to make it a little easier to work with. My wrapper is a bit of a kludge. I've made minimal changes to the C code, and I don't plan to make any more.

...And The New Stuff

In 2008 I wrote two new modules, sysv_ipc and posix_ipc.

The former is similar to shm in that it provides access to System V IPC objects. Since it's all new code (and all mine), it's easier for me to debug and extend.

The latter (posix_ipc) is totally new; I don't know of any previous module that made POSIX IPC objects available to Python programmers. Since System V IPC is available in more places than POSIX IPC, this addition isn't exactly revolutionary, but it is nice to have.

Which Should You Use?

Use posix_ipc if at all possible. I recommend it because POSIX IPC's design was informed by the strengths and weaknesses that years of use revealed in System V IPC. As a result, it's less quirky and less complicated, which means that your code and mine is less likely to contain bugs.

Since POSIX IPC is easier to work with, I implement new features there first and port them over to sysv_ipc as I have time.

For Windows users, posix_ipc compiles and runs its demos against the Cygwin 1.7 beta released in late September 2010.

Sysv_ipc is a good fallback if POSIX IPC is broken on a platform you need to support.

The semaphore and message queue interfaces of posix_ipc and sysv_ipc are similar so switching between one and the other should not be too difficult. The shared memory interface is pretty different, though.

Shm should be your choice of last resort. There's nothing wrong with it, but since it is not being developed anymore you'll eventually have to replace it with one of the other two modules. The only reason you might want to consider using shm versus one of the other modules is if you need to support Python ≤ 2.3 and you've proved that posix_ipc and sysv_ipc won't work.

Feature Matrix

All of the modules support basic use of inter-process semaphores and shared memory. The table below highlights some of the differences between the three modules.

  posix_ipc sysv_ipc shm
Oldest Python version supported Tested w/Python 2.4.4, may work with older versions Tested w/Python 2.4.4, may work with older versions Tested w/Python 2.3, may work with older versions
Works with Python 3.x Yes Yes No
Platforms supported Linux, OpenSolaris, OS X (partial)[1], FreeBSD ≥ 7.2 Linux, OpenSolaris, OS X (partial)[1], FreeBSD Linux, OpenSolaris, OS X[1], FreeBSD
Future development likely Yes Yes No
Message queues Yes Yes No
Semaphore timeouts Yes Yes No
Permit multiple semaphore acquisitions or releases in a single call No Yes No
Exposes a semaphore's "otime" Not applicable Yes No
Permits platform-specific creation flags Yes Yes No
Provides subclassable classes Yes Yes No
Follows PEP 8 naming conventions Yes Yes No (mea culpa)
Provides functions to map SysV keys to ids and test if a key is in use Not applicable No Yes

[1] - posix_ipc and sysv_ipc are only partially supported under OS X because they offer timed semaphores which OS X does not support. Timed semaphores are available on other platforms with these modules. shm, by contrast, does not offer timed semaphores on any platform and is thus no less capable on OS X than on other platforms.