Hello,

I would like to be able to configure the clustering support provided with Hibernate Search so that it doesn't require an explicit pointer to the master queue.

Any examples on how this is done?

I also want to be able to failover so that a slave can take the place of a master that fails.

In example configurations I have seen as well as in the book they explicitly define the <property name="hibernate.search.worker.jndi.url" value="jnp://localhost:1099"/> property.

Basically I want to be able to get rid of that option so that failover does not require a reconfiguration of the slaves to point to a new master.

Thanks

Hi,


How would you like this to work then? Using JMS, the slaves need to be able to find the queue to which do add the indexing jobs.


This idea has several problems. Firstly, the slaves don't really know if the master goes down. They actually don't know the master. All they know is a JMS queue they post their work to. The only thing they can detect is whether the JMS queue fails. This does not mean that the master node itself is not working. Depending on your chosen JMS provider there are sveral ways to cope with failing JMS queues.
Secondly, even if a slave could detect whether a master went down or not, you would have a implement some sort of protocol to decide which slave takes over the role of the master.
The latest version of the book suggest to have a backup master node on standby which can either be manually or automatically (via some monitoring scripts) be activated. If you choose the automatic way you have to make sure of course that the original node does not come online after the backup has taken over.


As said, that is taken care of already. Master and slaves are already decoupled by the JMS queue.

--Hardy