Sanne Grinovero This is and has always been the case at least in my mind. Let me clarify.

I always expected the master to be a MDB and thus be involved in transactions. So if a JMS message can not be consumed, it goes back tot he queue. This is confirmed by the fact that AbstractJMSHibernateSearchController explicitly calls indexManager.performOperations.

Now, if a user set the backend as a hibernate.search.worker.execution async then indeed the actual indexing is not transactional. Do you want us to disallow such an option if we are running on the master? That would be a good idea I think especially with our IO optimization.