ChangeFields optimistic locking

Question

I am trying to use optmistic locking without changing my database schema by using the descriptor.useChangedFieldsLocking() on my EJB descriptor. Two clients try to do the following.
1) Query the shallow properties of the EJB (no getters called for unary or collection relationships).
2) Edit some of that simple data
3) Submit their changes
The optimistic locking works okay when just the shallow properties are being retrieved. If a client tries to submit when someone else has already done so, an OptimisticLockException is thrown which is the desired result.
However, if I wish to traverse one of the relationships on the EJB when it is first queried (by calling a getter for a collection for example) the following happens: When the second client attempts to load and edit this EJB the transaction times out with a deadlock exception.
I want to traverse relationships for display purposes in this case but in the furture will want to edit something in that relationship with ChangedFieldsLocking also set.
If anyone could tell me why the second client's transaction times out when it asks for the EJB I would be very grateful.
Regards
James.

Answer 1

Thanks for the reply.
After further investigation I have found that it's nothing to do with the locking (since I can switch it off and still have the problem) but may be a problem with OC4J (or how I am using it). The situation is as follows:
1) Client A starts a transaction
2) Client A asks for ejb1
3) Client A asks for ejb1.collection()
4) Client A asks for ejb1.collection(i).someMethod()
5) Client A uses the result of someMethod()
(transaction is still open)
6) Client B starts a transaction
7) Client B asks for ejb1
8) Client B asks for ejb1.collection()
9) Client B asks for ejb1.collection(i).someMethod()
10) Client B uses the result of someMethod()
The server hangs on step 9 and never enters someMethod() (the first line is debug and I never see it). Eventually it tells me that the transaction has timed out with a deadlock exception.
The following comes from the output of a ctrl-break at the point of hanging (someMethod() is getTitle() in this example) :
"HttpRequestHandler-22629000" prio=5 tid=0x434DE690 nid=0xab4 in Object.wait() [4400f000..4400fd8c]
at java.lang.Object.wait(Native Method)
- waiting on <08FC9AC8> (a com.evermind.server.ThreadState)
at com.evermind.server.ejb.AbstractEJBObject.startCall(AbstractEJBObject.java:226)
- locked <08FC9AC8> (a com.evermind.server.ThreadState)
at PersonNamesRemote_EntityBeanWrapper79.getTitle(PersonNamesRemote_EntityBeanWrapper79.java:3227)
at uk.co.corelogic.framework.common.ejb.entity.PeopleEJB.toBean(PeopleEJB.java:1800)
at PeopleRemote_EntityBeanWrapper60.toBean(PeopleRemote_EntityBeanWrapper60.java:1742)
at uk.co.corelogic.framework.common.ejb.session.FrameworkManager.retrieve(FrameworkManager.java:142)
I am using OC4J 9.0.4, Toplink 9.0.3.3. I have Toplink configured to generate BMP Entity beans.
My entity bean deployment descriptor has <reentrant>False</reentrant> but <reentrant>True</reentrant> makes no difference. It also has <trans-attribute>Required</trans-attribute> for everything which should be okay and trying <trans-attribute>Supports</trans-attribute> just causes null pointers to be thrown all over the place.
I will post this message on the OC4J forum too but would appreciate a reply if you have any ideas.
Thanks.
James.