We have a Service which is @Statefull. Most of the Data-Operations are atomic, but within a certain set of functions We want to run multiple native queries within one transaction.
We injected the EntityManager with a transaction scoped persistence context. When creating a "bunch" of normal Entities, using em.persist() everything is working fine.
But when using native queries (some tables are not represented by any @Entity) Hibernate does not run them within the same transaction but basically uses ONE transaction per query.
So, I already tried to use manual START TRANSACTION; and COMMIT; entries - but that seems to interfer with the transactions, hibernate is using to persist Entities, when mixing native queries and persistence calls.
@Statefull
class Service{
@PersistenceContext(unitName = "service")
private EntityManager em;
public void doSth(){
this.em.createNativeQuery("blabla").executeUpdate();
this.em.persist(SomeEntity);
this.em.createNativeQuery("blablubb").executeUpdate();
}
}
Everything inside this method should happen within one transaction. Is this possible with Hibernate?
When debugging it, it is clearly visible that every statement happens "independent" of any transaction. (I.e. Changes are flushed to the database right after every statement.)
i've tested the bellow given example with a minimum setup in order to elimnate any other factors on the problem (Strings are just for breakpoints to review the database after each query):
@Stateful
@TransactionManagement(value=TransactionManagementType.CONTAINER)
@TransactionAttribute(value=TransactionAttributeType.REQUIRED)
public class TestService {
@PersistenceContext(name = "test")
private EntityManager em;
public void transactionalCreation(){
em.createNativeQuery("INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','b','c')").executeUpdate();
String x = "test";
em.createNativeQuery("INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','c','b')").executeUpdate();
String y = "test2";
em.createNativeQuery("INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('c','b','a')").executeUpdate();
}
}
Hibernate is configured like this:
<persistence-unit name="test">
<provider>org.hibernate.jpa.HibernatePersistenceProvider</provider>
<jta-data-source>java:jboss/datasources/test</jta-data-source>
<properties>
<property name="hibernate.dialect" value="org.hibernate.dialect.MySQL5InnoDBDialect" />
<property name="hibernate.transaction.jta.platform"
value="org.hibernate.service.jta.platform.internal.JBossAppServerJtaPlatform" />
<property name="hibernate.archive.autodetection" value="true" />
<property name="hibernate.jdbc.batch_size" value="20" />
<property name="connection.autocommit" value="false"/>
</properties>
</persistence-unit>
And the outcome is the same as with autocommit mode: After every native query, the database (reviewing content from a second connection) is updated immediately.
The idea of using the transaction in a manuall way leads to the same result:
public void transactionalCreation(){
Session s = em.unwrap(Session.class);
Session s2 = s.getSessionFactory().openSession();
s2.setFlushMode(FlushMode.MANUAL);
s2.getTransaction().begin();
s2.createSQLQuery("INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','b','c')").executeUpdate();
String x = "test";
s2.createSQLQuery("INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','c','b')").executeUpdate();
String y = "test2";
s2.createSQLQuery("INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('c','b','a')").executeUpdate();
s2.getTransaction().commit();
s2.close();
}
In case you don't use container managed transactions then you need to add the transaction policy too:
@Stateful
@TransactionManagement(value=TransactionManagementType.CONTAINER)
@TransactionAttribute(value=REQUIRED)
I have only seen this phenomenon in two situations:
- the
DataSource is running in auto-commit mode, hence each statement is executed in a separate transaction
- the
EntityManager was not configured with @Transactional, but then only queries can be run since any DML operation would end-up throwing a transaction required exception.
Let's recap you have set the following Hibernate properties:
hibernate.current_session_context_class=JTA
transaction.factory_class=org.hibernate.transaction.JTATransactionFactory
jta.UserTransaction=java:comp/UserTransaction
Where the final property must be set with your Application Server UserTransaction JNDI naming key.
You could also use the:
hibernate.transaction.manager_lookup_class=org.hibernate.transaction.JBossTransactionManagerLookup
or some other strategy according to your current Java EE Application Server.
After reading about the topic for another bunch of hours while playing around with every configuration property and/or annotation I could find a working solution for my usecase. It might not be the best or only solution, but since the question has received some bookmarks and upvotes, i'd like to share what i have so far:
At first, there was no way to get it working as expected when running the persistence-unit in managed mode. (<persistence-unit name="test" transaction-type="JTA"> - JTA is default if no value given.)
I decided to add another persistence-unit to the persistence xml, which is configured to run in unmanaged mode: <persistence-unit name="test2" transaction-type="RESOURCE_LOCAL">.
(Note: The waring about Multiple Persistence Units is just cause eclipse can't handle. It has no functional impact at all)
The unmanaged persitence-context requires local configuration of the database, since it is no longer container-provided:
<persistence-unit name="test2" transaction-type="RESOURCE_LOCAL">
<provider>org.hibernate.jpa.HibernatePersistenceProvider</provider>
<class>test.AEntity</class>
<properties>
<property name="hibernate.connection.url" value="jdbc:mysql://localhost/test"/>
<property name="hibernate.dialect" value="org.hibernate.dialect.MySQL5InnoDBDialect" />
<property name="hibernate.connection.driver_class" value="com.mysql.jdbc.Driver"/>
<property name="hibernate.connection.password" value="1234"/>
<property name="hibernate.connection.username" value="root"/>
<property name="hibernate.hbm2ddl.auto" value="update" />
<property name="hibernate.show_sql" value="true" />
<property name="hibernate.archive.autodetection" value="true" />
<property name="hibernate.jdbc.batch_size" value="20" />
<property name="hibernate.connection.autocommit" value="false" />
</properties>
</persistence-unit>
A change required to the project would now be, that you add an unitName, whenever you use the @PersistenceContext annotation to retrieve a managed instance of the EntityManager.
But be aware, that you can only use @PersistenceContext for the managed persistence-unit. For the unmanaged one, you could implement a simple Producer and Inject the EntityManager using CDI whenever required:
@ApplicationScoped
public class Resources {
private static EntityManagerFactory emf;
static {
emf = Persistence.createEntityManagerFactory("test2");
}
@Produces
public static EntityManager createEm(){
return emf.createEntityManager();
}
}
Now, in the example given in the original Post, you need to Inject the EntityManager and manually take care about transactions.
@Stateful
public class TestService {
@Inject
private EntityManager em;
public void transactionalCreation() throws Exception {
em.getTransaction().begin();
try {
em.createNativeQuery(
"INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','b','a')")
.executeUpdate();
em.createNativeQuery(
"INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','b','b')")
.executeUpdate();
em.createNativeQuery(
"INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','b','c')")
.executeUpdate();
em.createNativeQuery(
"INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','b','d')")
.executeUpdate();
AEntity a = new AEntity();
a.setName("TestEntity1");
em.persist(a);
// force unique key violation, rollback should appear.
// em.createNativeQuery(
// "INSERT INTO `ttest` (`name`,`state`,`constraintCol`)VALUES('a','b','d')")
// .executeUpdate();
em.getTransaction().commit();
} catch (Exception e) {
em.getTransaction().rollback();
}
}
}
My tests so far showed that mixing of native queries and persistence calls lead to the desired result: Either everything is commited or the transaction is rolledback as a whole.
For now, the solution seems to work. I will continue to validate it's functionality in the main project and check if there are any other sideeffects.
Another thing I need to verify is if it would be save to:
- Inject both Versions of the EM into one Bean and mix usage. (First checks seem to work, even when using both ems at the same time on the same table(s))
- Having both Versions of the EM operating on the same datasource. (Same data source would most likely be no problem, same tables I assume could lead to unexpected problems.)
ps.: This is Draft 1. I will continue to improve the answer and point out problems and/or drawbacks I'm going to find.
You have to add <hibernate.connection.release_mode key="hibernate.connection.release_mode" value="after_transaction" /> to your properties. After a restart should the Transaction handling work.
