External transaction control and ability to find new registered objects

Question

Hello, We are using Toplink with external transaction control and have a process inserting a complex hierarchy of objects. During the process we either do a registerObject or deepmergeClone depending on if the instance is already in the db. With external transaction control the registerObject does not actually do the commit to db until the global transaction (Container) issues the commit. Unfortunately we end up doing creating multiple instances of same objects ( because the assumption that registerObject would have written the row to the db ) with the same keys and when the container issues the commit we end up with duplicate key violation. Is there a way to find out if an object with a particular key is already registered?

Answer 1

This sounds like the kind of question that can only be answered with a whiteboard and a good review of your architecture.
In general, there should be no problem registering objects multiple times. I.e.,
x = some object
x1 = uow.registerObject(x);
Then x1==uow.registerObject(x), and x1==uow.registerObject(x1), etc. When you register an object with the UOW, based on PK it'll always return that same one.
Do you have multiple units of work on the go? (that may explain this behavior).
In any case, I think the real problem here is that you're somehow registering objects that are no longer cached. I.e., some object is serialized or rebuilt and then registered after it's gone from the cache. By default, TopLink determines if an object is new or existing (to determine INSERT vs UPDATE) by hitting the cache. You can change this default behavior in the Mapping Workbench, open the advanced property for "Identity" and change existence checking to "check database". Although, this can be a slow and tedious process to have to keep hitting the DB.
A little trick I use sometimes is to take advantage of the "readObject" API that will read the object from the databaes if it's not already in cache, and just return it from cache if it is in cache. Check out the UOW primer at http://otn.oracle.com/products/ias/toplink/index.html for more info, but the jist is that I would do this if I were you:
x = some object that you're not sure is cached and you want to register in UOW;
x' = uow.readObject(x);
IF the object was in cache, you'd get back a working copy, nice and fast. IF it's not in cache, you hit the database, it goes in cache, and you get your working copy. Now you don't have to change the existence checking option which could slow everything down.
- Don