Hibernate Envers with QueryDSL Update - spring

Question

Hibernate, Hibernate Envers and QueryDSL are configured and working correctly in a Spring boot 1.4.1.RELEASE.
The problem is when using UpdateClause<JPAUpdateClause> updateQueryBuilder = queryFactory.update(collectionTransaction); to build update query and execute that update query, Hibernate Envers does not pick up and audit those changes.
Following is the Spring Data JPA repository that implements QueryDSL
public class CollectionTransactionRepositoryImpl extends QueryDslRepositorySupport implements CollectionTransactionRepositoryCustom {
#Autowired
private JPAQueryFactory queryFactory;
public CollectionTransactionRepositoryImpl() {
super(CollectionTransaction.class);
}
#Override
public Collection<CollectionTransaction> updateCollectionTransaction(UpdateCollectionTransaction updateCollectionTransaction) {
QCollectionTransaction collectionTransaction = QCollectionTransaction.collectionTransaction;
UpdateClause<JPAUpdateClause> updateQueryBuilder = queryFactory.update(collectionTransaction);
.....//Code omitted for brevity
long updated = updateQueryBuilder.execute();
//.....
return ...
}
}
Is it possible for Hibernate Envers to pick up changes in this situation ?

Answer 1

This is a known concern outlined in JIRA HHH-10318.
Envers works based on Hibernate's event subsystem where Hibernate effectively notifies various callbacks that state for an entity has been modified in some way, and provides both the prior and new entity state. This state is precisely what Envers uses to determine what changed and insert audit change rows.
Lets take a trivial example:
UPDATE MyEntity e SET e.status = :status
Hibernate will perform the following tasks:
Flush the persistence context any any modifications.
Invalidate any cached instances of MyEntity.
Execute the bulk update operation.
No where in any of these steps does Hibernate load any existing state. It simply guarantees that current changes are flushed prior to the bulk update and that any subsequent operations will fetch from the datastore rather than a cache due to the bulk update.
Therefore from Envers perspective, it gets no callbacks and thus isn't aware that any operation took place because Hibernate ORM cannot provide any entity state for such an operation, it simply does not exist.
The big question here is how (if possible) to model and handle a change unit for such an operation.
It's difficult because Envers would effectively need some type of PreBulkOpEvent so that it can cache what it needs that is about to change and a PostBulkOpEvent to require and merge the two results to generate change log entries. The concern with such a concept really centers around how to do this effectively to avoid
Running out of memory due to large result-set manipulation.
Long execution time to load state from datastore for large result-set manipulation.
Anyway, you're welcomed to read over the JIRA and provide any feedback or ideas. But presently, its just something that falls outside the scope of what we can capture at the present moment.