EclipseLink generates cartesian plan instead of (inner) joins in SQL. Why?

Question

In the projects where Hibernate is my persistence provider, I can issue queries with 'join fetch' expressions, and Hibernate will generate SQL that mirrors that: SQL containing join expressions using valid paths of comparison.

EclipseLink, however, issues SQL with ugly Cartesian plans, that hurt performance quite badly. While reading this article , it mentions that eagerly fetching might generate Cartesian plans, but it conveniently forgets that other providers (Hibernate) can optimize that.

So, is it possible to instruct EclipseLink to optimize these queries? I believe that few relationships can be optimized through the usage of the @FetchJoin annotation, but I'm hoping to find something that does not include spreading ORM-specific annotations on the domain model.

As an example, here's a (dynamic) query that I'm issuing as JPQL:

String query = "select w from WorkOrder w " +
            "inner join fetch w.type " +
            "inner join fetch w.details " +
            "inner join fetch w.project " +
            "inner join fetch w.priority " +
            "inner join fetch w.networkElement ";

And here's the EclipseLink output:

SELECT t1.ID, t1.CREATION, ... (the fetch fields here)
FROM swa_network_element t5, swa_priorities t4, swa_dispatch_project t3, swa_work_order_detail t2, swa_work_orders t1, swa_work_order_type t0
WHERE ((t1.alpha_id LIKE '%TSK%') AND (((((t0.ID = t1.TYPE_ID) AND (t2.worder_id = t1.ID)) AND (t3.id = t1.project_id)) AND (t4.ID = t1.priority_id)) AND (t5.ID = t1.ne_id))) ORDER BY t1.CREATION DESC

Best regards, Rodrigo Hartmann

Answer 1

Try using:

@org.eclipse.persistence.annotations.JoinFetch(JoinFetchType.INNER)

-- or --

@org.eclipse.persistence.annotations.JoinFetch(JoinFetchType.OUTER)

Outer joins are recommended for nullable columns.

I've found more information in this blog: http://vard-lokkur.blogspot.com/2010/09/jpa-demystified-episode-1-onetomany-and.html

Regards,

Victor Tortorello Neto

Answer 2

I'm not sure what you mean? The SQL seems correct, there is a join for each table, which table is missing a join?

What is the SQL you are expecting?

If you mean putting the join condition in the FROM clause instead of the WHERE clause, EclipseLink does this for outer joins as it is required, but does not for inner joins as it make no difference on any database that I am aware of. If you really want to, you can configure EclipseLink 2.4 to print the join condition for inner joins in the FROM clause using the DatabasePlatform setPrintInnerJoinInWhereClause(false) API, but it should have no impact on what the database does.

The blog post talking about join fetching cartesian issues is when you join fetch multiple 1-m relationships, this can cause a lot of data to be returned. There is no way to optimize this with join fetching, as you need all of the data. In EclipseLink you can use batch fetching, which is much more efficient than join fetching.

See, http://java-persistence-performance.blogspot.com/2010/08/batch-fetching-optimizing-object-graph.html