Skip to content


Subversion checkout URL

You can clone with
Download ZIP
A Python wrapper around the libmemcached interface from TangentOrg.
Tag: 1.0-alpha

Fetching latest commit…

Cannot retrieve the latest commit at this time

Failed to load latest commit information.


pylibmc is a quick and small Python client for memcached written in C.

It builds on the famous libmemcached C client from TangetOrg, notable for its speed and flexibility.

libmemcached must be installed separately, and be available to the compiler and linker.

Basic usage is that of python-memcached, like so:

>>> import pylibmc
>>> mc = pylibmc.Client([""])
>>> mc.set("foo", "Hello world!")
>>> mc.get("foo")
'Hello world!'

There's also support for some other features not present in other Python libraries, like the binary protocol:

>>> mc = pylibmc.Client([""], binary=True)


libmemcached has ways of telling it how to behave. You'll have to refer to its documentation on what the different behaviors do.

To change behaviors, quite simply:

>>> mc.behaviors["hash"] = "fnv1a_32"

For a list of the defined behavior key names, see what the keys of a client is. For example:

>>> mc.behaviors.keys()  # doctest: +NORMALIZE_WHITESPACE
['hash', 'connect timeout', 'cache lookups', 'buffer requests',
 'verify key', 'support cas', 'poll timeout', 'no block', 'tcp nodelay',
 'distribution', 'sort hosts']

The hash and distribution keys are mapped by the Python module to constant integer values used by libmemcached. See pylibmc.hashers and pylibmc.distributions.


In multithreaded environments, accessing the same memcached client object is both unsafe and counter-productive in terms of performance. libmemcached's take on this is to introduce pooling on C level, which is correspondingly mapped to pooling on Python level in pylibmc:

>>> mc = pylibmc.Client([""])
>>> pool = pylibmc.ThreadMappedPool(mc)
>>> # (in a thread...)
>>> with pool.reserve() as mc:
...     mc.set("hello", "world")

For more information on pooling, see my two long posts about it.

Comparison to other libraries

Why use pylibmc? Because it's fast.

See this (a bit old) speed comparison, or's comparison.


#sendapatch on

Change Log

New in version 0.9

  • Added a get_stats method, which behaves exactly like python-memcached's equivalent.
  • Gives the empty string for empty memcached values like python-memcached does.
  • Added exceptions for most libmemcached return codes.
  • Fixed an issue with Client.behaviors.update.

New in version 0.8

  • Pooling helpers are now available. See pooling.rst in the distribution.
  • The binary protocol is now properly exposed, simply pass binary=True to the constructor and there you go.
  • Call signatures now match libmemcached 0.32, but should work with older versions. Remember to run the tests!

New in version 0.7

  • Restructured some of the code, which should yield better performance (if not for that, it reads better.)
  • Fixed some memory leaks.
  • Integrated changes from, which should make pylibmc work under Snow Leopard.
  • Add support for the boolean datatype.
  • Improved test-runner -- now tests build/lib.*/ if available, and reports some version information.
  • Support for x86_64 should now work completely.
  • Builds with Python 2.4, tests run fine, but not officially supported.
  • Fixed critical bugs in behavior manipulation.

New in version 0.6

  • Added compatibility with libmemcached 0.26, WRT error return codes.
  • Added flush_all and disconnect_all methods.
  • Now using the latest pickling protocol.

New in version 0.5

  • Fixed lots of memory leaks, and added support for libmemcached 0.23.
  • Also made the code tighter in terms of compiler pedantics.

New in version 0.4

  • Renamed the C module to _pylibmc, and added lots of libmemcached constants to it, as well as implemented behaviors.
Something went wrong with that request. Please try again.