Hello, I'm using Hibernate Search 3.0 GA.

Due to the complexity of our architecture we have to rebuild the lucene indexes every night.

The indexes are built by a standalone java app, on a very fast machine.

Now we want to update all the nodes in our cluster (for the web app using the indexes) with a local copy of the latest index. How can we do that?

In other word we have 3 machines, one is the master rebuilding everynight the master index, and 2 machines are web app servers clustered with a copy of the master index. We do not need JMS since nobody update the master index (except the standalone app scheduled to run every night)

I tried to setup my 2 nodes like this:
<!-- Hibernate Search -->
<property name="hibernate.search.default.directory_provider">org.hibernate.search.store.FSSlaveDirectoryProvider</property>
<property name="hibernate.search.default.indexBase">C:\Projects\index</property>
<property name="hibernate.search.default.sourceBase">\\masterServer\lucene\masterindex</property>
<property name="hibernate.search.default.refresh">1800</property>


But I get the following exception:

org.hibernate.HibernateException: Unable to initialize: ContractAltId
at org.hibernate.search.store.DirectoryProviderFactory.createDirectoryProvider(DirectoryProviderFactory.java:147)
at org.hibernate.search.store.DirectoryProviderFactory.createDirectoryProviders(DirectoryProviderFactory.java:75)
at org.hibernate.search.impl.SearchFactoryImpl.initDocumentBuilders(SearchFactoryImpl.java:260)
at org.hibernate.search.impl.SearchFactoryImpl.<init>(SearchFactoryImpl.java:94)
at org.hibernate.search.impl.SearchFactoryImpl.getSearchFactory(SearchFactoryImpl.java:172)
Truncated. see log file for complete stacktrace
java.lang.IllegalStateException: No current marker in source directory
at org.hibernate.search.store.FSSlaveDirectoryProvider.initialize(FSSlaveDirectoryProvider.java:56)
at org.hibernate.search.store.DirectoryProviderFactory.createDirectoryProvider(DirectoryProviderFactory.java:144)
at org.hibernate.search.store.DirectoryProviderFactory.createDirectoryProviders(DirectoryProviderFactory.java:75)
at org.hibernate.search.impl.SearchFactoryImpl.initDocumentBuilders(SearchFactoryImpl.java:260)
at org.hibernate.search.impl.SearchFactoryImpl.<init>(SearchFactoryImpl.java:94)

Any suggestion on how to deploy properly this configuration?

Thanks!
-Christophe

I need to document that some day.
To avoid copying a locked or inconsistent directory, the master keeps 2 copies of the index. One is available for the slaves, the second is available for the master to update it. When a master update is done, the 2 directories switch.

So you basically have
masterDir/1 (directory)
masterDir(2 (directory)

To differentiate an active directory (for slave copy) from the other one, a marker file is read.

masterDir/current1 if the directory 1 is the available one
masterDir/current2 if the diectory 2 is the available one

So you will need to follow this pattern for your master.

Note that you probably also add hibernate.search.indexing_strategy manual to avoid the event system to work, since your slaves are read only

Thanks Emmanuel, that really helped me to understand how the master/slave config works. That's really clever.

I have the similar problem on the slave side. My configuration file on the slave side is as follow:


I assume the Create, Update, & Delete operations on the slave side with JPA/Entity Manager will register index creation to the master via JMS.

But running the slave part with J2SE triggers the exception.


Could you please tell me what is wrong in my config? The code is okay when I use FSDirectoryProvider.

Hi,

first of all you will have to make sure that the master node will start first - at the very least for the very first time. The master node is responsible for creating the master index from which the slaves nodes will take their local copy. In your case this master index does not seem to exist.

Also, you are using c:/temp/lucene/mastercopy are source directory for the master. This will only work if you are running master and slave on the same machine. Normally the sourceBase is on a shared file system since master and slave reside on different machines.

--Hardy