I found this patch helpful but inadequate. It addressed the issue of unique column alias being generated with a given jvm but not across the cluster. With 100 jvms each generating distinct looking sql for the jpql our buffer caches were being overrun. I think the root of the issue the random nature the jpa files are loaded across the cluster. I needed the table index to be constant so I created a property file that defined all tables and their index. When the Hibernate config loads it references this property file. If there is no entry, falls back to the incremental counter.