Hi All

Due to my usecase I have the need to access the native Lucene APIs. Is it safe to access the directory without using a hibernate session? For example in order to get an index reader at present I need to get a FullTextSession which wraps an Hibernate Session and then make my way to the Directory. Is it ok for me to create a IndexReader with just using the directory path and then closing the IndexReader after I've used it?

Thanks
Amin

Hi Amin,
yes provided you use the IndexReader in read only mode it would be safe, but you actually get the same object you would from the FullTextSession, but this one as some benefits:

• in case you enable sharding, you would still receive the correct one or a multireader wrapping the correct ones
• Search manages the IndexReader to provide very efficient reopening strategies, it won't be as fast just opening one yourself
• opening a Hibernate Session doesn't open a database connection until you need it, so it's a cheap operation
• first search of Lucene index on a new reader will be slower as the internal fieldcaches must be filled. using the Search provided reader this might be pooled and so it might have the caches filled in already
• you don't need to configure externally where these directories are
• you would have an hard time getting the reference to any non-fs directory, so it would be hard to try for example the new Infinispan based index

you have other reasons to open one yourself?
also you actually only need a reference to the SearchFactory, you could retrieve this once and keep the reference without opening a Session everytime you need (although that's not slow at all).

Hi Sanne

Thanks for your detailed reply. I really like the idea of storing the reference of the SearchFactory on initialisation and re use that as I normally would. It makes sense to use HSearch to get to the underlying index. I presume there will be no side effect (not getting up to date data) if I store the reference of the SearchFactory.

The main reason for asking this is that we started to notice we were running out of db connections, although this might be unrelated to search, I have more control of the search code so i'd like to take some defensive measures.

Amin

that should be unrelated, actually using Search more should be an effective strategy to reduce the need for database connections.
Check for leaks in your application or connection pool, if you deploy it on JBoss you get warnings logged about unclosed connections.

Hi Sanne

We found the culprit, it turns out we have a custom datasource factory which is built using Spring, however there is a fundamental bug in the code that is the root cause.

Thanks
Amin