Concurrent usage of dbm causes cache to go stale

Migrated issue, originally created by David Gardner ()

Ran into this issue in production where users were reporting items in the cache being an hour old for a cache region configured with a one minute expiration time. Problem first observed with version 0.5.7, and reproduced with version 0.6.1.

I was able to reproduce the issue in a simple test of a function that returns datetime.now() (dogpile-test.py), if I run about 10 concurrent instances of the script after about a minute or so they will start reporting stale data.

However I noticed as a work-around if I use the MutexLock class from the lock_factory documentation, I don't run into the problem:
http://dogpilecache.readthedocs.io/en/latest/api.html?highlight=dogpile.cache.dbm#dogpile.cache.backends.file.DBMBackend.params.lock_factory

Attachments: dogpile-test.py

Start with the attached dogpile-test.py reproduction and the DBMBackend lock_factory documentation, then compare concurrent DBM usage with the MutexLock workaround. Reproduce the stale values under concurrent access and identify the expected expiration behavior; done means the DBM cache no longer serves expired data in that scenario.