[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [eclipselink-users] Order of operations in EclipseLink

Hi Patric,

While it's true that in simple projects without foreign key constraints, it would be possible to simply execute operations in the order they were called, solving that problem would require more complex way of detecting the changes you have made. As you changed things, we would have to build an ordered list of all the changes (including minor updates to fields).

EclipseLink's change detection is optimized to allow SQL to execute correctly in the general case. We do one of two things:

1. Accumulate an unordered list of changes as you make changes
2. Calculate a list of changes at flush time based on comparisons to clones

When we have the list of changes, we execute it against a pre-calculated commit order so as to avoid any issues with foreign keys or relationships.

Our goal is to write changes in the most efficient way possible while still ordering things correctly for the general case and avoiding foreign key issues. Maintaining an ordered list of changes would be inefficient and likely to cause issues in all but cases with very simple mappings.

  We provide the perform deletes flag to cover some additional cases.

-Tom

patric@xxxxxxxxxxx wrote:
Hi Tom,

I understand that any O/R mapper needs to track relationships/integrity of the entities that the
foreign key relations/constraints would not be broken or violated.


In my project we're only using disabled foreign key constraints in the oracle database to map the parent-child structure only but not to enforce the integrity.
So when looking at my code snippet below EclipseLink could do all operations in the correct order without any problems.


Are there other arguments beside enabled foreign key constraints which would prevent the use of the real operation order?

Although the use of a flush() would help me out, it's a little bit more difficult than my code example below.
Imagine that the first persist, the removal and the re-insertation are located in different code parts and some code outside their implementation control which parts are invoked.
To be safe, we have to add a flush() after each part. This would cause a significantly performance degrade because flush() is costly operation, especially in my project (we're using client session customizers to intercept changeset calculations to track and archive all changes in some database tables).


Thank you and best regards,
Patric



Zitat von Tom Ware <tom.ware@xxxxxxxxxx>:

As mentioned in the article listed below. EclipseLink (and many other Object-Relational providers) orders the deletes last because it facilitates maintenance of referential integrity. An O-R tool must do more than just send SQL to a database. It has to maintain the identity and the integrity of the relationships among the objects it is managing. (e.g. Imagine you delete an object that is part of a relationship. EclipseLink must be able to update any foreign keys that form the relationship before doing the delete, or you will get exceptions related to Foreign Key constraints)

You can control when different parts of you transaction are executed using the em.flush() API.

If you need to get at the performDeletesFirst API, you can use org.eclipse.persistence.jpa.JpaHelper.getEntityManager(entityManager) and call getUnitOfWork() on the returned JpaEntityManager.

-Tom

Rohit Banga wrote:
Hi Patric
See the following link
http://goo.gl/2oH4P <http://wiki.eclipse.org/Using_Advanced_Unit_of_Work_API_%28ELUG%29#How_to_Use_the_setShouldPerformDeletesFirst_Method_of_the_Unit_of_Work>



Michael
The above link shows how to configure it using UnitOfWork. Is there a way to do it using the EntityManager?


Thanks
Rohit

On 12/30/2010 4:56 PM, Michael Simons wrote:
Hi Patric,

AFAIK it is an intentional behaviour, EL just like any other JPA/JDO implementation that we've used (EL, OpenJPA, etc) do inserts before deletes.
I don't remember the reason. I just remember there was one. Maybe someone else can explain that to us.
To keep the AK you can either flush right after the delete or replace the delete/insert by an update.


-Michael

-----Ursprüngliche Nachricht-----
Von: eclipselink-users-bounces@xxxxxxxxxxx [mailto:eclipselink-users-bounces@xxxxxxxxxxx] Im Auftrag von patric@xxxxxxxxxxx
Gesendet: Mittwoch, 29. Dezember 2010 14:27
An: EclipseLink-users@xxxxxxxxxxx
Betreff: [eclipselink-users] Order of operations in EclipseLink


Hi,

imagine a simple table/entity TestTable with an AK on column status.

The following operations will fail using EclipseLink:

TestTable tt = new TestTable();
tt.setStatus("BLA");
em.persist(tt);
em.flush();

em.remove(tt);

TestTable tt2 = new TestTable();
tt2.setStatus("BLA");
em.persist(tt2);
em.flush(); //this will throw an exception due to AK violation


However, I would expect a successfull run because I previously removed the record which has the same AK.


It seems that the operation order is modified by EclipseLink -
inserts seems to be executed first which might change the outcome of a set of operations.
Is this an intentional behaviour?
If it is, is there a possibility to preserve the operation order?


Thank you and best regards,
Patric


_______________________________________________ eclipselink-users mailing list eclipselink-users@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/eclipselink-users _______________________________________________ eclipselink-users mailing list eclipselink-users@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/eclipselink-users

-- Thanks and Regards Rohit Banga Member Technical Staff Oracle Server Technologies


------------------------------------------------------------------------

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users