Question

Suppose I have a local cache on each sever using a LRU Cache or something similar.
If I invalidate an object at one node then other nodes should be noticed.
I have seen this implemented with jgroups.

I would like to know if a ‘cache invalidation queue’ using jms (lets say ActiveMQ) would fit the application. Not sure about response times and so.

Answer 1

It depends on your constraints, namely how long you can bear stale reads (i.e. the cache returning outdated data).

If this doesn’t hurt at all, JMS is a nice and clean way to notify all members since it’s easy to make sure every member gets the “flush” message eventually.

If you need to make sure obsolete data is flushed immediately, you can use JMS to coordinate the flush by adding transactions and request/response cycles to the game. But that means you can run into deadlocks or starving processes if you’re not careful – that’s the price you have to pay for synchronization.

If this bothers you, the question is what other technology you could use to achieve the same result. You could use RMI or sockets but then, you’d have to reinvent the wheel (which means you will spend a lot of time to get where JMS already is today). But that would only replace one transport technology with another – it wouldn’t help with synchronization issues.

Or you could use a cache that supports clustering. You can try Terracotta’s Enterprise Ehcache but under the hood, it will use a similar technique to keep caches in sync. The big advantage is that Ehcache has been used (= debugged) for years now, so you will only run into well-known problems.