Question

From my understanding, each of these methods: get() and put() are atomic.

But, when accessing caches like ehcache & memcached, in a CheckThenAct fashion, race conditions will arise.

What should we synchronize on to ensure that if thread1 does a Get() and thread2 does a Get() followed by a Put(), that the first Get() will receive the new value added by thread2 ?

Answer 1

The way to handle this kind of situation is to not to cache writes from multiple places or to control access in some other way.

A simple solution to this is to not do any writes to the cache from your clients. Instead if you want to update the value you fire off a request to have the value updated. If that value is retrieved from somewhere else then this can work well. If it is calculated or otherwise comes from the client then it can be more problematic.

A common component to use with memcache—at least in the PHP world—is beanstalkd as a distributed work queue processor. In this scenario you can fire off an “update cache” request as a unit of work.

If you are doing an operation on the cache entry that uses the existing value as an input or in some other way would introduce a race condition if get-then-update weren’t an atomic operation, an alternative is to create a unit of work that includes the old value and the new value. That way, whatever process handles the updates can invalidate or otherwise handle those situations where the current value isn’t the expected value.

Incrementing and decrementing are handled by the memcached interface so if your accesses are of that type (or can be reduced to that kind of operation) then that solves the problem as well.