Conditional insert with Spring JPA / Hibernate

Question

I'm working on a project that runs in a clustered environment, where there are many nodes and a single database. The project uses Spring-data-JPA (1.9.0) and Hibernate (5.0.1). I'm having trouble resolving how to prevent duplicate row issues.

For sake of example, here's a simple table

@Entity
@Table(name = "scheduled_updates")
public class ScheduledUpdateData {
    public enum UpdateType {
        TYPE_A,
        TYPE_B
    }

    @Id
    @GeneratedValue(strategy = GenerationType.AUTO)
    @Column(name = "id")
    private UUID id;

    @Column(name = "type", nullable = false)
    @Enumerated(EnumType.STRING)
    private UpdateType type;

    @Column(name = "source", nullable = false)
    private UUID source;
}

The important part is that there is a UNIQUE(type, source) constraint.

And of course, matching example repository:

@Repository
public class ScheduledUpdateRepository implements JpaRepository<ScheduledUpdateData, UUID> {
    ScheduledUpdateData findOneByTypeAndSource(final UpdateType type, final UUID source);

    //...
}

The idea for this example is that parts of the system can insert rows to be schedule for something that runs periodically, any number of times between said runs. When whatever that something is actually runs, it doesn't have to worry about operating on the same thing twice.

How can I write a service method that would conditionally insert into this table? A few things I've tried that don't work are:

1. Find > Act - The service method would use the repository to see if a entry already exists, and then either update the found entry or save a new one as needed. This does not work.
2. Try insert > Update if fail - The service method would try to insert, catch the exception due to the unique constraint, and then do an update instead. This does not work since the transaction will already be in a rolled-back state and no further operations can be done in it.

3. Native query with "INSERT INTO ... WHERE NOT EXISTS ..."* - The repository has a new native query:

   @Repository
   public class ScheduledUpdateRepository implements JpaRepository<ScheduledUpdateData, UUID> {
       // ...
   
       @Modifying
       @Query(nativeQuery = true, value = "INSERT INTO scheduled_updates (type, source)" +
                                          " SELECT :type, :src" +
                                          " WHERE NOT EXISTS (SELECT * FROM scheduled_updates WHERE type = :type AND source = :src)")
       void insertUniquely(@Param("type") final String type, @Param("src") final UUID source);
   }
   

   This unfortunately also does not work, as Hibernate appears to perform the SELECT used by the WHERE clause on its own first - which means in the end multiple inserts are tried, causing a unique constraint violation.

I definitely don't know a lot of the finer points of JTA, JPA, or Hibernate. Any suggestions on how insert into tables with unique constraints (beyond just the primary key) across multiple JVMs?

Edit 2016-02-02

With Postgres (2.3) as a database, tried using Isolation level SERIALIZABLE - sadly by itself this still caused constraint violation exceptions.

Answer 1

You are trying to ensure that only 1 node can perform this operation at a time. The best (or at least most DB-agnostic) way to do this is with a 'lock' table. This table will have a single row, and will act as a semaphore to ensure serial access.

Make sure that this method is wrapped in a transaction

// this line will block if any other thread already has a lock
// until that thread's transaction commits
Lock lock = entityManager.find(Lock.class, Lock.ID, LockModeType.PESSIMISTIC_WRITE);

// just some change to the row, it doesn't matter what
lock.setDateUpdated(new Timestamp(System.currentTimeMillis()));  
entityManager.merge(lock);
entityManager.flush();

// find your entity by unique constraint
// if it exists, update it
// if it doesn't, insert it