Integrating BTM with Hibernate

Hibernate can be integrated straight with any JTA transaction manager. These instructions have been verified against BTM 1.3 and Hibernate 3.3.0.SP1.

The biggest added value (omitting the fact that you can use Hibernate and two databases) is Hibernate's Current Session context management with JTA. You do not have to take care about opening nor closing Session as Hibernate will automatically bind them to the JTA transaction's lifecycle. You just have to make sure JTA transactions are properly started and ended.

JPA, Hibernate and BTM The example discussed here uses the Hibernate API but the JPA API could be used as well. You just need to use the EntityManager but the same configuration applies.

Contents

JTA datasources

Hibernate cannot directly create a BTM PoolingDataSource. You will have to create them yourself (either via the API or the Resource Loader).

Setting up the BTM JNDI server

You have to bind the datasources and the transaction manager to some JNDI server. You can use any one you wish, but BTM 1.3 ships with one you might find more convenient to use.

It is very easy to use it in a standalone J2SE application. Just create a jndi.properties file at the root of your classpath. It should only contain this line:

You can now just create a InitialContext with the no-args constructor to have access to it.

Multiple JNDI providers If you are running your application in a context in which there already is a JNDI provider installed (for instance in a web application) you can tell Hibernate to perform lookups in a specific JNDI environment instead by setting the hibernate.jndi.class property to bitronix.tm.jndi.BitronixInitialContextFactory in Hibernate's configuration instead.

API way: Creating the datasources

As you can expect, you will need to create one PoolingDataSource per database. Say that you want to use two Embedded Derby databases, and configure them via the BTM API. Here is what your code would look like:

Datasource's unique name and JNDI location correspondence The BTM JNDI provider will automatically bind the datasources under their unique name. In this case, you can look up jdbc/testDS1 or jdbc/testDS2 as soon as the transaction manager started without having anything else to configure.

Resource Loader way: Creating the datasources

You can use BTM's Resource Loader instead of the BTM API. It is usually a good idea when you want to create a fully standalone application as you can get rid of the datasources creation and shutdown code.

Create a datasources.properties file in the current directory containing these properties:

Resource Loader and JNDI binding As with the API, the datasources will be available in JNDI under their unique name.

In your application code, you will have to configure BTM to use the resource loader:

This has the exact same behavior as creating the PoolingDataSource objects yourself. It is just more convenient.

Hibernate Session factories

You need to configure exactly one SessionFactory per datasource.

Datasource JNDI location

You have to tell Hibernate where to get the BTM datasource via JNDI. Add a connection.datasource property and set its value to the JNDI location of your datasource:

Current session context

You have to set current_session_context_class to jta.

Transaction factory class

You have to set transaction.factory_class to org.hibernate.transaction.JTATransactionFactory.

JPA / EJB3 API When you use Hibernate via the JPA / EJB3 API (also known as Hibernate EntityManager) you should not set the transaction.factory_class property as the Ejb3Configuration class will configure a special one which is JTA compatible if you configure your PersistenceUnit's transaction type to JTA.

Transaction manager lookup class

You have to set transaction.manager_lookup_class to an implementation of TransactionManagerLookup. Hibernate ships with one that can lookup BTM since version 3.3.

add this property to your config to use it:

BTMTransactionManagerLookup and JNDI The BTMTransactionManagerLookup implementation which ships with Hibernate has a limitation regarding JNDI lookup of the transaction manager. If you're going to use it within an application server that already contains a JNDI server you might encounter problems. An improved version which solves that issue has been submitted to the Hibernate project, see: http://opensource.atlassian.com/projects/hibernate/browse/HHH-3739 Also remember to change the jndiUserTransactionName value to something else, for instance: bitronixTransactionManager.

SessionFactory XML configuration files

Here is what the hibernate_testDS1.cfg.xml file will look like for the first datasource. Some other mandatory properties also have to be added, like the dialect, cache.provider_class and of course the required object mappings.

And here is the hibernate_testDS2.cfg.xml for the second datasource:

Hibernate connection release mode There currently is a bug in BTM's connection pool that impacts all versions up to and including 1.3.2. You must set connection.release_mode to after_statement to workaround the bug (see Hibernate's documentation). If you don't, you might end up seeing BTM throwing exceptions and leaking connections. This is bug BTM-33 which has been fixed in BTM 1.3.3.

End result

Now that Hibernate and BTM are properly configured, you can simply use the JTA and Hibernate APIs in your application.

Application code

Here is what your code will look like when you want to update the content of both databases atomically:

Say that persistUser() creates a new user, in no way will a user be created in one database and not in the other.

Download

You can download a sample runnable application putting these explanations in practice. It contains all the code that has been skipped for clarity in this page. Both the API and Resource Loader ways are implemented so you can try both and see which one you prefer.

You can download this demo here: HibernateBTM13.zip.

There is an ant build.xml file included as well as a the necessary batch and shell scripts required to run the application from Windows or Unix.

Before you run the application, you have to create the Derby database. Just run the included derby-create.sh or derby-create.bat script to do so, it will create two directories called users1 and users2. Then you can start the demo by either running run_api.sh or run_api.bat for the API version, run_rl.sh or run_rl.bat for the Resource Loader version.

Here is the list of JAR files with version required to run this demo. They're all included in the downloadable ZIP file.