JPA - insert instead of update

Few days ago I was digging through the code in some project to find the reason why EclipseLink (2.1.1) is performing database insert instead of update. Let me say, that it was the hardest thing to find in my whole JPA-using developer career - so for the records, and to let you find it faster than me ;)

There was an association between entities Alpha and Beta, made in following manner:

public class Alpha ... {
    ...
    private Beta beta;

    @ManyToOne(..., optional = false)
    @JoinColumn(name = "beta_id")
    @JoinFetch(JoinFetchType.INNER)
    public Beta getBeta() { 
        return beta;
    }
    ...
}

I wrote about the @JoinFetch annotation in one of my previous posts already (see: JPA Demystified (episode 1) - @OneToMany and @ManyToOne mappings) - it is used here, to fetch the associated Beta along with the Alpha, in one query.

In the project analyzed by me, instance of Alpha was created at one point, edited in few steps, with persisting into database between the steps. Everything was fine after the first step. Alpha was written into database (first insert), but after the second step, primary key violation occurred each time I tested it (same entity was again inserted into DB, while it should be updated at this point).

My first hint was missing equals and hashcode methods on Alpha and Beta. I've added it, but it doesn't helped at all. So I meet with the best friend of all Developers - Debugger :) - and started to dig into the project and EclipseLink source code. At this point let me give you first suggestion - when you trace the JPA problems use the Debugger and verify performed database operations in parallel, because only both those things can give you full view of the problem lurking beneath the Application's surface ;)

Watching the SQL queries, I've found that EclipseLink is performing query fetching the Alpha along with the Beta (as directed by @JoinFetch annotation), right before the invalid insert. Matching this operation to the Java code, I've found that EclipseLink is trying to check what has changed in the Alpha entity, to perform only the required updates to the database. The SQL query was using inner join ...

Do you suppose what was wrong? :) - There was no Beta entity associated to the Alpha, and therefore this query - "select ... from Alpha inner join Beta ..." was returning no records at all. At this point EclipseLink was assuming that Alpha entity was not written into database yet, and tried to do it again - Alpha has already generated identifier (while first insert into database), and ... beautiful exception :)

What you should learn from the above situation is that optional attribute of association definition is informational only! - It's provider choice to throw an exception if the associated value will be null, or do something completely else, as you see EclipseLink developers are ignoring it silently :(

Last but not least - use @JoinFetch(JoinFetchType.INNER) if and only if you are absolutely sure, that associated entity will always exist - ex. when the database scheme doesn't allow other possibility. Don't trust JPA provider completely ;)