Hibernate version: Hibernate Search 3.0

MySql 5.0

I am using hibernate search in JBOSS 4.2.1 Cluster configuration. We have an MDB deploy on the server to process the slave index updates. We need to know is there a failover support if the master server where the MDB is deployed goes down??

Since the slave indexes would not get processed until the master is up, will all next slave index searches be on stale indexes?

Well, I thought about whether we need some kind of fail over but I'm not convinced.

If your master go down, all the needed index operations will still be queued through JMS and will not be processed until after you bring the master back on track. I personally don't find that too bothering.
What kind of solution were you envisioning / expecting?

I also think it is not a problem. If at all it is a question whether you can setup your MBean in clustered mode. Only if the first Master node stops processing indexing requests the second node takes over. I believe this is possible with most application servers. However, you probably need to share the same filesystem (NFS, SAN, ...?). What I mean both master node would have to access the same master index. Given that you need some sort of shared file system anyway to copy the master index to the slaves this might not be an issue.

However, is it really worth the effort?

Yes Emmanuel you are correct that once the master node comes up it would update the master indexes and do the job. I was thinking if there is some kind of failover support where in some other node in the cluster start processing the slave indexes might not be stale for much longer time. I do understand that you need to expect some kind of stale resules when you are using indexing on clustering but was thinking if there is already a way we can minimize this.

Thanks a lot for your reply.

[quote="hardy.ferentschik"]I also think it is not a problem. If at all it is a question whether you can setup your MBean in clustered mode. Only if the first Master node stops processing indexing requests the second node takes over. I believe this is possible with most application servers. However, you probably need to share the same filesystem (NFS, SAN, ...?). What I mean both master node would have to access the same master index. Given that you need some sort of shared file system anyway to copy the master index to the slaves this might not be an issue.

However, is it really worth the effort?

Yes we would be having a NFS fs and all the slave node would read the master index copy. However I did not got what you meant by What I mean both master node would have to access the same master index

My question was mostly towards how can we give a failover so that the master index copy gets update by other node that becomes the master on fail over.

Here is Hardy's idea
Let's assume you have n slaves, they all copy the index from the master copy and all send their updates to the JMS queue

Now if instead of 1 master, you have 2 (one active, one passive), master 1 can process the queue and if it fails, you could activate (manually or automatically) master 2 to process the queue. In order to do that, you need to share the master directory between master 1 and 2 hence probably use a NFS which will slow down the indexing overall.

That's exactly what I ment :)

If you want to increase the reliability on the master side you have to make the master redundant, but that is really a JMS question.