Hello,

I've a CMS system based in Lucene engine. We have several repository kind (photograph, video, audio, text, etc). Each one has differents fields to be indexed.
The administrator can create new repositories and the CMS creates new indexes on demand for the new repositories. And the users can insert documents and searchs for them.
I try to figure how can I create index on demand with Hibernate Search. I dive into documentation, books and tutorials and I can't to find anything about this feature.

Can anyone help me?

Hibernate Search will automatically start and create needed indexes when it starts.

Are each of these "CMS repositories" handled by a different Hibernate SessionFactory? If so, you don't need to do anything.

If you want to use a single SessionFactory and use sharding for the indexes, then you need dynamic sharding and that's not implemented yet, it will likely be available in Hibernate Search v.4.0 (next release).

link to the JIRA issue: https://hibernate.onjira.com/browse/HSEARCH-472

I think I have not explained well ...

When the administrator creates a new repository, the CMS creates a new index "on the fly" not when the system starts. Regardless of another indexes of the same class exists, for each new repository a new index is created. Also, when the user searches for in a repository, the CMS searchs in the index associated, not in all indexes.

I've not seen would be possible do that with Hibernate Search because, as you said, the indexes are created in start time and the same class is associated with only one index.

For example, there are three photograph repositories, A, B and C. The document class (PhotographBean.java) is the same for them. The system creates three differents indexes, each one with its documents.

This architecture gives us several adventages, one of this is when the index is corrupted and its must be reindexed. The system only reconstruct this one, not a big index with about 2 million of documents.

The user can works with the other repositories, meanwhile the corrupted is reindexed.

Is there any manner to replicate this architecture with Hibernate Search?

Follow with my example...

Could be possible create classes "on-the-fly" with BCEL o similar package?

I mean... create subclasses PhotographBeanA.java, PhotographBeanB.java, PhotographBeanC.java inherited from PhotographBean.java with
@Indexed(name="A"), Indexed(name="B"), @Indexed(name="C") for each class and restart (in run-time) SessionFactory o EntityFactory in order to HSearch creates the new indexes.

What do you think about this solution?

All of this, with the added problem our system is clustered ;)

you don't need BCEL, it can be much simpler.
Why can't you start a new instance of Hibernate Search when "the administrator creates a new repository", "on the fly" as you wrote. that's one option, likely the simplest one.

Another option is you use Sharding, use a single PhotographBean class and filter on shards. The only complexity here is that sharding is not dynamic, but you can stop the SearchFactory and start a new one.

Also consider you can use the programmatic configuration API, indexes don't need to be defined via annotations.

I've seen Programmatic API and it seems to adjust more to my needs.

Thanks a lot.