Using a MOM though JMS to do what I call asynchonous indexing is actually our recommended approach to cluster full-text searches with Hibernate Search Chapter 10 covers this topic and demonstrates how to set that up in Java EE (JBoss AS 5 is used as an example).
Here is how it works, there are slave nodes and one master node. All slaves contain your application and can execute full text queries based on a loca version of the Lucene index (hence fast). When objects are changed, the changes are sent to the JMS queue instead of being applied to the Lcueen index. The master node does listen to the JMS queue (it's an MDB usually) and apply the changes to the master index. On a regular basis (configurable), slaves copy the new master index to get the updated version of the index. The copy is incremental so after the first time, the copy is quite minimal.
That's exactly what I wanted to know a way to avoid burdening the master index copy each and every time the updation is made, Rather work on the local copies, if any upadtes - send them to the MasterAwaited Queue and let the master do the work of sync when it has time to do so.
This very much addresses my concern (posted int he other thread) about having a ultra fast search engine rather than a very precisive one.
Surfs up space ponies, I'm making gravy without this lumpy, tiny ad:
Free, earth friendly heat - from the CodeRanch trailboss