we are having a clustered enterprise application using JTA transaction and hibernate for database operations deployed on JBoss EAP.
To increase system performance we are planning to use Jboss data grid. This is how I plan to use jboss data grid:
- I am adding/replacing the object is cache whenever its inserted/updated in database using cache.put
- when object is deleted from database its deleted from cache using cache.remove
- while retrieving, first try to get the data from cache using key or query. If data is not present, load the data from database.
However, I have below questions on data grid:
- To query objects we are using hibernate criteria however data grid uses its own query builder. Can we avoid writing separate query for hibernate and datagrid?
- I want a list of objects to be returned matching a criteria. If one of the objects matching the criteria is evicted from cache, is it reloaded automatically from database?
- If the transactions is rolled back is it rolled back from data grid cache as well
- Are there any examples which I can refer for my implementation of data grid?
- which is better choice for my requirement infinispan as 2nd level cache or data grid in library or remote mode?
Galder's comment is right, the best practice is using Infinispan as the second-level cache provider. Trying to implement it on your own is very prone to timing issues (you'd have stale/non-updated entries in the cache).
Regarding queries: With 2LC query caching on the cache keeps a map of 'sql query' -> 'list of results'. However once you update any type that's used in a query, all such queries are invalidated (e.g. if the query lists people with age > 60, updating a newborn still invalidates that query). Therefore this should be on only when the queries prevail over updates.
Infinispan has its own query support but this is not exposed when using it as 2LC provider. It is assumed that the cache will hold only a (most frequently accessed) subset of the entities in the database and therefore the results of such queries would not be correct.
If you want to go for Infinispan but keep the DB persistence, an option might be using JPA cache store (and indexing). Note though that updates to DB that don't go through Infinispan would not be reflected in the cache, and the indexing may lag a bit (since it's asynchronous). You can split your dataset and use JPA for one part and Infinispan + JPA cache store for the other, too.
A third option is using Hibernate Search, which keeps the data in database but index is in Lucene (possibly stored in Infinispan caches, too) and you don't use the Criteria API but Hibernate Search API.