Hello,
got 2 node clusters, jgroups + infinispan configuration, automatic master node selector, synchronized execution.

Inserting new record on slave node - the slave enters JgroupsBackendQueueTask.sendLuceneWorkList() method at which point I kill the master (simulating master node crash).

Such a scenario leads to index update message being silently dropped, the JgroupsBackendQueueTask.sendLuceneWorkList() completes with no error while the index is not updated.

I've traced the problem down to DispatchMessageSender.send() method. I have tried to provide custom implementation of message sender, but it is hard wired META-INF/services/org.hibernate.search.backend.jgroups.impl.MessageSenderService, not really configurable.

Would it be possible to throw an exception if no response is received in synchronous mode?

Vladimir