Introduction

The Persistence Context acts as a transactional-write behind cache for the incoming entity state transitions, and all changes are synchronized with the database during flushing.

Although both the Hibernate Session and the JPA EntityManager define a flush() method to manually trigger this process, it’s much more convenient to let Hibernate manage the Persistence Context flushing. Unfortunately, there’s is a major difference between how JPA and Hibernate define the automatic flushing mechanism.

When using the default AUTO flush mode, the Persistence Context should be flushed as follows:

• before the transaction is committed,
• before running a JPQL or HQL query
• before executing a native SQL query

As previously explained, Hibernate triggers the AUTO flush only for the first two events, and the native SQL queries must either override the ALWAYS flush mode using the Query#setFlushMode(FlushMode flushMode) method or add a table space synchronization (e.g. SQLQuery#addSynchronizedEntityClass(Class entityClass), SQLQuery#addSynchronizedEntityName(String entityName), SQLQuery#addSynchronizedQuerySpace(String querySpace)).

This is only required for the native Hibernate API when using a Session explicitly.

Since Hibernate 5.2, if you bootstrap Hibernate using JPA (e.g. persistence.xml), then even the Hibernate FlushType.AUTO will behave just like its JPA counterpart.
Only if you bootstrap Hibernate using the native mechanism, will the Hibernate Session use the legacy FlushType.AUTO behavior.

JPA AUTO flushing

JPA is more strict, and the AUTO flush mode must trigger a flush before any query. More the section 3.10.8 of the Java Persistence API specification says that the AUTO flush mode should ensure that all pending changes are visible by any executing query.

This can be demonstrated by executing the following method:

assertTrue(((Number) entityManager
    .createNativeQuery("select count(*) from Post")
    .getSingleResult()).intValue() == 0);
    
Post post = new Post("Hibernate");
post.setId(1L);
entityManager.persist(post);

assertTrue(((Number) entityManager
    .createNativeQuery("select count(*) from Post")
    .getSingleResult()).intValue() == 1);

When running this test, Hibernate generates the following SQL statements:

SELECT COUNT(*) FROM Post

INSERT INTO post (title, version, id) 
VALUES ('Hibernate', 0, 1)

SELECT COUNT(*) FROM Post

So, the flush was triggered and the INSERT statement was executed prior to running the SELECTstatement.

This doesn’t happens when using a Session:

assertTrue(((Number) entityManager
    .createQuery("select count(*) from Post")
    .getSingleResult()).intValue() == 0);

Post post = new Post("Hibernate");
post.setId(1L);
entityManager.persist(post);

Session session = entityManager.unwrap(Session.class);
assertTrue(((Number) session
    .createSQLQuery("select count(*) from Post")
    .uniqueResult()).intValue() == 0);

This time, Hibernate generates the following statements:

SELECT COUNT(*) FROM Post

SELECT COUNT(*) as col_0_0_ 
FROM post blogentity0_

INSERT INTO post (title, version, id) 
VALUES ('Hibernate', 0, 1)

If you enjoyed this article, I bet you are going to love my Book and Video Courses as well.

Conclusion

Flushing is a very important concept for any ORM too, and Hibernate is no different.
As a rule of thumb, it’s better to make sure that native SQL queries don’t return inconsistent results when using the Hibernate Session API.