Please excuse my english, since it's not my native language.

Hibernate version:

Hibernate core 3.2.4.sp1
Hibernate search : 3.0.0.GA
Hibernate annotations : 3.3.0.GA

I'm currently using Hibernate in a end-user application. I'm currently running on a problem whose cause I can't understand.

Each time I launch my application, all the indexes created by lucene get cleared, and as such, if I don't manually index my data again, I can't use it. The fact is : I think manually indexing the database again could be too long to be bearable by the user, and in fact, I don't understand if the deletion of the previously created indexes is a feature of hibernate search (or lucene), or if I'm just doing something wrong. Shouldn't the old indexes be reused ?

The connection to the database is programmaticaly configured, does it play any role here ? I don't know where to look first, and didn't find anything on indexes deletion or reusing in the manual.

then again, I could have missed it, I'm lost here. I don't even know what information could be relevant to get help. :/

Hibernate Search does not erase the Lucene directories, so it come from somewhere else.
Are you sure you use a FSDirectoryProvider (and not a RAMDirectoryProvider)?
Check the indexBase property (check the doc for the full name), to see if the index is persisted while your app is running

Thanks for the answer.

I do use a FSDirectoryProvider, and I configured its base directory. When I go in one of the subdirectoriez containing the indexes with Explorer, I do see many files being created, updated, and increasing in size as long as the application does its work. When I quit the application, the files are still there. The problem occur when I launch the application again : the files disappear, and instead I find two little files which seems to be empty indexes.

It seems to me that hibernate search or lucene doesn't like its old indexes and replace them with new, cleaner ones, yet I don't see why.

edit : here is more info :

the new files created at launch are named "segments.gen" and "segments_12r".

When I try to open the index with Luke, I get an error : "invalid or corrupted index".

Still, if I reindex manually the database, with session.index(), I do have a correct index that I can open and read with Luke.

my config for hibernate search is :

hum I really don't see.

Maybe try to change

hibernate.search.default.indexBase = .focus_search_indexes

to

hibernate.search.default.indexBase = ./focus_search_indexes (or \ on windows)

If that makes it, there might be a bug in the directory discovery or the vm, let me know. If it still does not work, give the debugger a try, the directory thing happen in the FSDirectoryProvider

Thanks, I'll try all that. I'm using a .directory name in order to have it hidden under unix file management system.

I'm also willing to try without underscores in the name (why not ?)

I'll keep you informed.

So I tried renaming the file without dots not underscores, with no success.

Then I launched my debugger, and I realized that in the FSDirectoryProvider.initialize() method, it always went through this



create was always true because my indexBase was configured like this :

.focus_search_indexes

Don't see it ? me neither, until I checked in the debugger : there were 3 trailing spaces at the end of the config line, just after the directory name.

Still, I wonder if it means that something should be done on the hibernate side of the code to prevent such errors again, since it worked partially and never returned a configuration error. Do you usually trim this kind of Strings ?

That said, thanks a lot for the guidelines on how to find my error. It's really nice to get answers so quickly and effectively.

I would be inclined to do a trim, but actually 'mydir ' is a perfectly valid directory name AFAIK.
At least we should provide better log infos, please open a JIRA issue to enhance the logs

Thank you very much.

http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-130

Still I would like to discuss that last part.

If trailing space are not a problem for a valid directory name, why is it that HibernateSearch didn't create the directory with trailing space and used it, but used the one without those trailing spaces ? It used the one in which older indexes already existed, and it recreated them, clearing all existing content.

I can understand if it's on the JVM or the OS side that the problem resides, but still, Hibernate Search behaviour here seems quite illogical to me.

But yes, I know, I'm the one at fault here. :)