Indeed, it turns out to be very useful filter caches. This would have a very positive effect on the speed of executing commands. All in all it would not be difficult to implement. Even it would be possible to make L1 groats based on tens by default. And time-stable L2 caches using infinispan. You can also think about how to replicate query results between jboss server nodes. In the present case, using the cache would not be difficult. Just calculate the key based on the filter parameters and save the partial result of the query.

You can initially replicate the previous functionality. Then think if it makes sense to create two-level caches. The first level can be useful when performing similar queries in transactions or in nested transactions based on maps of small memory maps.

The second level would be to record the result of the query in a temporarily more durable cache. Of course, as is the case in hibernate, user-defined. It will probably be infinispan eventually. You can choose something lighter for testing.