TopLink's Missing Grandchild

Question

This is quite a difficult problem to describe, so please bear with me.
We are using TopLink to map one-to-many relationships between three tables in a master->detail->subdetail configuration, with three corresponding Java domain classes. We are making use of back-references on the child and grandchild classes to facilitate population of foreign key references on the corresponding tables.
Unfortunately, we are observing some strange behaviour when the “batch reading” option is enabled on the one-to-many relationships. Say the data in our three tables expresses the following, with one row on the master table, two on the detail table, and four on the sub-detail table:
Parent
->Child1
->Grandchild1A
->Grandchild1B
->Child2
->Grandchild2A
->Grandchild2B
When a TopLink query is executed to retrieve the parent object, both of the child objects are correctly retrieved, but the first grandchild of the first child (Grandchild1A) is missing. All other grandchildren are correctly retrieved.
Analysing the SQL that TopLink has generated reveals the joins to be correct, and running this SQL against the database directly retrieves all of the expected data. Somewhere between TopLink executing the select and the population of the TopLink cache, the first grandchild goes missing. Interestingly, if an order-by is applied to the TopLink mapping such that GrandChild1B is retrieved before Grandchild1A, then it is 1B that goes missing, consistent with the theory that TopLink is somehow losing the first child of the first child.
Has anybody else observed this irregularity, and is there any work-around besides turning off batch reading?
Many thanks,
Tom

Answer 1

Has anyone else encountered this issue, where batch reading across 3 nested relationship, causes incorrect data in the lowest level relationship to be returned.
We are using Toplink 10.1.3.
This currently means that we are only able to specify batch reading within the one to many "Parent" to "Child" relationship.
As soon as batch reading is turned on for the one to many "Child" to "Grandchild" relationship, the data is incorrect.
As stated, all SQL appears to be correct, and removing batch reading means that the data returned by the Query is correct.
Please assist, as there are many areas in our system where we have these type of relationships, and currently we are unable to use the performance advantage of Batch Reading.
Many Thanks
Marc