1. What is the recommended way to copy the lucene index from master to slave in the architecture specified in section 2.2.2?

2. What are the limitations put on such a copying mechanism?

3. What are the options?

Some options:
* use FSMasterDirectoryProvider on the master and have sourceBase point to a NFS.

* use rsync like Solr does. Possibly reusing some of its scripts.

* use katta or some other hapood based/like solution

1. the one described here http://www.hibernate.org/hib_docs/search/reference/en/html/search-configuration.html#jms-backend

2. there is a (configurable) delay between the database change and it's visibility in searches. Usually this is acceptable.

3. I personally only have tested the first option but any is possible, you might have to write a custom DirectoryProvider to accomodate them.

Katta seems for very large sets of data. Are you sure you fit into this infrastructure. Also Katta solves a different problem. It's not replication but sharding of the index (which we also do at a different level though).