The openejb dynamic datasource api aims to allow to use multiple data sources as one from an application point of view.
It can be useful for technical reasons (load balancing for example) or more generally functionnal reasons (filtering, aggregation, enriching...).
In this example the implementation simply use a datasource from its name and needs to be set before using any JPA operation in the transaction (to keep the logic simple in the example).
Our router has two configuration parameters:
The interface Router (org.apache.openejb.resource.jdbc.Router) have only one method to get the datasource and is simply implemented in org.superbiz.dynamicdatasourcerouting.DeterminedRouter class.
It uses a ThreadLocal to manage the currently used datasource. Keep in mind JPA used more than once the getDatasource() method for one operation. To change the datasource in one transaction is dangerous and should be avoid.
To be able to use your router as a resource you need to provide a service configuration. It is done in a file you can find in META-INF/org.router/ and called service-jar.xml (for your implementation you can of course change the package name).
It contains the following code:
<ServiceJar> <ServiceProvider id="DeterminedRouter" <!-- the name you want to use --> service="Resource" type="org.apache.openejb.resource.jdbc.Router" class-name="org.superbiz.dynamicdatasourcerouting.DeterminedRouter"> <!-- implementation class --> # the parameters DataSourceNames DefaultDataSourceName </ServiceProvider> </ServiceJar>
Using the conf/openejb.xml file the following configuration could have been used:
<!-- Router and datasource --> <Resource id="My Router" type="org.apache.openejb.router.test.DynamicDataSourceTest$DeterminedRouter" provider="org.routertest:DeterminedRouter"> DatasourceNames = database1 database2 database3 DefaultDataSourceName = database1 </Resource> <Resource id="Routed Datasource" type="org.apache.openejb.resource.jdbc.Router" provider="org.router:RoutedDataSource"> Router = My Router </Resource> <!-- real datasources --> <Resource id="database1" type="DataSource"> JdbcDriver = org.hsqldb.jdbcDriver JdbcUrl = jdbc:hsqldb:mem:db1 UserName = sa Password JtaManaged = true </Resource> <Resource id="database2" type="DataSource"> JdbcDriver = org.hsqldb.jdbcDriver JdbcUrl = jdbc:hsqldb:mem:db2 UserName = sa Password JtaManaged = true </Resource> <Resource id="database3" type="DataSource"> JdbcDriver = org.hsqldb.jdbcDriver JdbcUrl = jdbc:hsqldb:mem:db3 UserName = sa Password JtaManaged = true </Resource>
In test mode and using property style configuration the foolowing configuration is used:
String[] databases = new String[] { "database1", "database2", "database3" }; Properties properties = new Properties(); properties.setProperty(Context.INITIAL_CONTEXT_FACTORY, LocalInitialContextFactory.class.getName()); // resources // datasources for (int i = 1; i <= databases.length; i++) { String dbName = databases[i - 1]; properties.setProperty(dbName, "new://Resource?type=DataSource"); dbName += "."; properties.setProperty(dbName + "JdbcDriver", "org.hsqldb.jdbcDriver"); properties.setProperty(dbName + "JdbcDriver", "org.hsqldb.jdbcDriver"); properties.setProperty(dbName + "JdbcUrl", "jdbc:hsqldb:mem:db" + i); properties.setProperty(dbName + "UserName", "sa"); properties.setProperty(dbName + "Password", ""); properties.setProperty(dbName + "JtaManaged", "true"); } // router properties.setProperty("My Router", "new://Resource?provider=org.router:DeterminedRouter&type=" + DeterminedRouter.class.getName()); properties.setProperty("My Router.DatasourceNames", "database1 database2 database3"); properties.setProperty("My Router.DefaultDataSourceName", "database1"); // routed datasource properties.setProperty("Routed Datasource", "new://Resource?provider=RoutedDataSource&type=" + Router.class.getName()); properties.setProperty("Routed Datasource.Router", "My Router");
It is absolutely the same for the application.
Using more than one datasource behind one EntityManager means the databases are already created. If it is not the case, the JPA provider has to create the datasource at boot time.
Hibernate do it so if you declare your databases it will work. However with OpenJPA (the default JPA provider for OpenEJB), the creation is lazy and it happens only once so when you'll switch of database it will no more work.
Of course OpenEJB provides @Singleton and @Startup features of Java EE 6 and we can do a bean just making a simple find, even on none existing entities, just to force the database creation:
@Startup @Singleton public class BoostrapUtility { // inject all real databases @PersistenceContext(unitName = "db1") private EntityManager em1; @PersistenceContext(unitName = "db2") private EntityManager em2; @PersistenceContext(unitName = "db3") private EntityManager em3; // force database creation @PostConstruct @TransactionAttribute(TransactionAttributeType.SUPPORTS) public void initDatabase() { em1.find(Person.class, 0); em2.find(Person.class, 0); em3.find(Person.class, 0); } }
Now you configured the way you want to route your JPA operation, you registered the resources and you initialized your databases you can use it and see how it is simple:
@Stateless public class RoutedPersister { // injection of the "proxied" datasource @PersistenceContext(unitName = "router") private EntityManager em; // injection of the router you need it to configured the database @Resource(name = "My Router", type = DeterminedRouter.class) private DeterminedRouter router; public void persist(int id, String name, String ds) { router.setDataSource(ds); // configuring the database for the current transaction em.persist(new Person(id, name)); // will use ds database automatically } }