Yes, I use the nfs mount to share the index for other search instance and all the instances have same lock directory configured, but the only the difference is that nfs mount is read-only mount, so I have to disable the lock mechanism for search instances, only lock is enabled for index modification instance. We have 6 jboss cluster for our application. so 5 instances of jboss search on the same index and the 6th instance used for index update.

OK unfortunately this won't work.

Well, it will "work" but you'll hit occasional FileNotFoundExceptions on your searchers, whenever a searcher tries to restart itself while the updater is writing a new segments file.

Even though the searcher's are read only, they still need to briefly hold the commit lock to ensure the updater doesn't write a new segments file while the searcher is reading it (and opening each segment).

We are working towards a fix for lock files over NFS mounts, first by decoupling locking from directory implementation (http://issues.apache.org/jira/browse/LUCENE-635) and second by creating better LockFactory implementations for different cases (eg at least a locking implementation based on native OS locks). But this is still in process...

I think the best workaround for now is to take an approach like Solr:

  http://incubator.apache.org/solr/features.html
  http://incubator.apache.org/solr/tutorial.html

whereby the single writer will occasionally (at a known safe time) make a snapshot of its index, and then the multiple searchers can switch to that index once it's safe.

Mike

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to