Well, that's what invalidation does. :-) No replication at all, just invalidates keys that have changed on remote instances.
Thanks for the reply, Manik.
I guess I was assuming too much, then. When the entry is invalidated, I thought the cache would see this the next time the entry is referenced, and get an updated entry from another cache in the cluster (presumably one that has a non-invalidated entry).
If I'm not using a persistent store, how do I leverage the invalidation to find a valid entry elsewhere in the cluster? In the doc, I see a brief reference to shared cache loaders, but nothing definitive on how you would set this up when not using persistence, or how you would figure out which cache you should get a non-invalidated entry from. Any pointers here would be greatly appreciated.
If I'm not using a persistent store, how do I leverage the invalidation to find a valid entry elsewhere in the cluster?
Take a look at TcpDelegatingCacheLoader, looks like this would solve your problem.
Or even the ClusteredCacheLoader will help - the CCL lazily loads entries from its neighbours if it is needed and cannot be found locally.