python-unimr.memcachedlock-0.1rc2_r144828-2.lbn19.noarch

unimr.memcachedlock implements a distributed "soft" locking mechanism using memcached. It provides factory functions and decorators for a primitive locking, a reentrant locking and a special locking for zeo-clients.

The native locking methods of python's threading module supports thread safe locking and therefore, provides only full locking support for single zope installations. However, zeo-clients have no locking mechanism beetween each other for concurrent write operations on identical objects (e.g. Catalog) and are unnecessarily stressed to resolve ConflictErrors on heavy load. The reason for this problem is the optimistic concurrency control of the ZODB which primarly prepares the changes of an object (in many cases expensive calculations) and thereafter checks the validity of the object for the write process. The higher the number of writes on the same object the higher the risk that a concurrent zeo-client has already invalidated the object while another zeo-client has still this object in use. The client with the invalidated object is constrained to roll back its changes and to recalculate the changes based on the refreshed object. At worst, this state goes in circles and results in a ConflictError. The optimistic concurrency control therefore perfectly fits only concurrent write operations on distinct objects.

Memcache locking overcomes this problem because it extends the regular concurrency mechanism by a shared locking beetween all involved zeo-clients by serializing the concurrent write operations before a ConflictError is provoked. This mechanism is also known as pessimistic concurrency control.