openjpa-1.2.1 partial commit problem

["Trenton D. Adams" <tr...@trentonadams.ca>]

Hi Guys,

I am not yet using the build time enhancer, as it currently blows up on one of my objects, as noted in another message to the list.  So, it's sticking to runtime subclassing.

I'm having the oddest problem.  I am using EJB3, and have a method that REQUIRES_NEW for the transaction.  It works fine like that, but as soon as I remove that attribute, to get the default transaction attribute, my records are only partially committed, literally.

So, what I'm trying to do, is 
1. put a journal entry in with the message "pre commit entry" in it's own transaction, to make sure something is entered in, in case something happens that blows up the rest of the commit process.  That way someone knows why the serial column has a missing journal number.
2. add some entity objects to the journal entry, as a collection
3. merge the journal entry again with objects in the collection

That's fine, but I decided to remove the REQUIRES_NEW, just to see if my unit tests would fail.  I found out really quickly that only one item of the collection is being persisted to the database.

I am effectively going like this (pseudo code)...

--- method with REQUIRES_NEW
            entityManager.persist(journalEntry);
            entityManager.flush();  // if this flush is removed, it works without REQUIRES_NEW, but not otherwise.

--- method that calls method with REQUIRES_NEW
            journalEntry = new JournalEntry(
                ledgerBroker.getJournalType(), "pre commit entry");
            sessionContext.getBusinessObject(
            RIJournalEntryManager.class).persistMethod(journalEntry);
            journalEntry.addItem(blah);
            journalEntry.addItem(blah);
            entityManager.merge(journalEntry);
            entityManager.flush();

It seems to me that this is a bug in openjpa, but I cannot be certain, as I'm VERY new to EJB/JPA.

RE: openjpa-1.2.1 partial commit problem

[C N Davies <cn...@cndavies.com>]

Use prepersist / preupate listeners, you can check the content of your objects inside there.

Trust me, you would be better to spend your time working out why build time enhancement blows up one of your objects, as I mentioned before and Pinaki re-iterated runtime enhancement will often cause issues that manifest themselves nowhere near the root cause of the issue. I would not be all surprise if your issue does related to run time enhancement. I went down the road of relying on run time enhancement for months so I din't have to worry about setting up ant, I saved time not having to work on ant but I lost 10 times the time trying to sort out strange issues that never made any sense but turned out to be the runtime enhancer. If it was up to me I'd remove it completely

Chris


-----Original Message-----
From: Trenton D. Adams [mailto:trent@trentonadams.ca] 
Sent: Thursday, 27 May 2010 2:41 PM
To: openjpa-users
Subject: openjpa-1.2.1 partial commit problem

Hi Guys,

I am not yet using the build time enhancer, as it currently blows up on one of my objects, as noted in another message to the list.  So, it's sticking to runtime subclassing.

I'm having the oddest problem.  I am using EJB3, and have a method that REQUIRES_NEW for the transaction.  It works fine like that, but as soon as I remove that attribute, to get the default transaction attribute, my records are only partially committed, literally.

So, what I'm trying to do, is 
1. put a journal entry in with the message "pre commit entry" in it's own transaction, to make sure something is entered in, in case something happens that blows up the rest of the commit process.  That way someone knows why the serial column has a missing journal number.
2. add some entity objects to the journal entry, as a collection
3. merge the journal entry again with objects in the collection

That's fine, but I decided to remove the REQUIRES_NEW, just to see if my unit tests would fail.  I found out really quickly that only one item of the collection is being persisted to the database.

I am effectively going like this (pseudo code)...

--- method with REQUIRES_NEW
            entityManager.persist(journalEntry);
            entityManager.flush();  // if this flush is removed, it works without REQUIRES_NEW, but not otherwise.

--- method that calls method with REQUIRES_NEW
            journalEntry = new JournalEntry(
                ledgerBroker.getJournalType(), "pre commit entry");
            sessionContext.getBusinessObject(
            RIJournalEntryManager.class).persistMethod(journalEntry);
            journalEntry.addItem(blah);
            journalEntry.addItem(blah);
            entityManager.merge(journalEntry);
            entityManager.flush();

It seems to me that this is a bug in openjpa, but I cannot be certain, as I'm VERY new to EJB/JPA.