Hi,

I am using JBoss as my application server where I am hosting by job service WAR. The indexing is a manual index where I have a java class that handles the update process of the index. I use a Spring scheduler that invokes the indexing class after every X minutes.

However I have a requirement to reindex everything once in every 24 hrs. The problem is this index is used by other applications. So when reindexing is happening the entire index will be purged and recreated. This will give the end user a very bad experience .

What I want is an ability to run the reindexing process in another system and later copy the index to this filesystem, However what I have observed that deleting the index in the current machine is not being allowed by Linux since the file system is being used by the application.

Is there a way I can delete an index and copy a new index to the same path.

Hi,

you really should use two directories. How else do you guarantee that Lucene will see a complete (non corrupt) index. Having to directories is also the approach which the JMS master/slave setup uses provided by Search. Have you considered the master/slave setup? Otherwise you can implement your own DirectoryProvider and use the existing directory providers as guide.

--Hardy

Thanks for the reply.

The thing is I cannot use JMS in my infrastructure. I will have the same VM. It would be best that I can use two directories, one will be acting as master and slave (but without JMS). All searches should happen on the slave directory while all updates should happen in the master.

IS there a way?

Thanks David

If it is in the same JVM why do you not just depend on the automatic indexing? And why a full rebuild every 24hrs? Couldn't you just re-index the entities which have changed (eg by checking a last modified field in the db)?

In your current approach you need to applications (one search and one index application). Having a master/slave setup truly within the same JVM and app is not possible. I guess that's not what you mean anyways.



One idea could be that you jusr use filesystem-slave and filesystem-master as directory providers for these two apps without setting the jms backend on the slave. This should just work fine, however, any index updates request triggered by the slave (your search app) would get ignored or would get applied into an index which gets overwritten in the next index sync.

--Hardy

Hi,
I agree with Hardy, using JMS seems like the clean approach. JMS is not limited only to be used on different servers communicating, it's also nice to use to loosely couple different applications running in the same JVM.

But in this simple case you don't need the JMS connection at all, just take the pattern we describe on the reference as the "JMS master/slave" for the indexes: essentially the user facing application deals with two indexes, one "being copied" and one used for searching, and then is able to switch between them dynamically. The master periodically copies the index.

As a backend, you can either:
- leave indexes enabled - might be nice for users.
- disable indexing
- use the blackhole backend
- use the new soft delete approach