We have a CMP2.0 bean with some extra persistent state that the CMP engine cannot handle. To ensure that the extra state is saved to persistent storage we use call-ejb-store-on-clean=true to ensure that ejbStore is called as the spec requires. So far no problem.

The problem comes when we use this bean in a clustered environment (commit option B) and do the dual-deployment RW/RO pattern with cache invalidations.

In this case a cache invalidation is not sent out from the RW deployment to the RO deployment if the only persistent state changed is the non-CMP-managed persistent state.

To get around this problem we have written a new interceptor EntityBeanCacheBatchEagerInvalidatorInterceptor. Replacing the standard EntityBeanCacheBatchInvalidatorInterceptor with this interceptor ensures that cache invalidation messages are always sent unless the bean is read-only, or the entity method call is read-only.

I will attach this interceptor written for 3.2.6 (ie. a contribution).

I would like this - or something similar - to be part of future JBoss versions.