[ https://issues.apache.org/jira/browse/OPENJPA-1163?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12733960#action_12733960 ] 

Michael Dick commented on OPENJPA-1163:
---------------------------------------

Hi Ravi,

This patch demonstrates a race condition more than anything else. You have two transactions. Tran 1 gets a copy of the entity, removes everything from its collection, then adds ten more items. Tran 2 gets a copy of the entity and just adds ten items. To show this just print out the size of newA.getAItems() before committing tran 1.

In Tran 1 the entity has only 10 items , in tran 2 the entity has 20. The last one to commit wins. It's a unidirectional relationship - I'm guessing this is because it's a uni-directional relationship and A is the owner (even though the updates are in the AItem table) and therefore the state in A trumps the state in AItem. 

Interestingly enough if you updated any field in A you should see an OptimisticLockException which would explain the problem better. 

> Data consistency issues while modifying collections.
> ----------------------------------------------------
>
>                 Key: OPENJPA-1163
>                 URL: https://issues.apache.org/jira/browse/OPENJPA-1163
>             Project: OpenJPA
>          Issue Type: Bug
>          Components: kernel
>    Affects Versions: 2.0.0
>         Environment: openJPA trunk. Derby DB.
>            Reporter: Ravi P Palacherla
>            Assignee: Ravi P Palacherla
>             Fix For: 2.0.0
>
>         Attachments: OPENJPA-1163_trunk.patch
>
>
> There are data consistency issues when modifying more number of elements in a collection Vs less number of elements.
> Following is a detailed explanation about the issue with example:
>  
> - Entity A has a collection of Entities AItems with cascade ALL.
> - Test case :
>   Clear all the data inside tables representing Entity A and AItems.  
>   Create 3 entity managers em1,em2 and em3.
>    
>   em1.begin()
>       create A on em1 with id "1"
>       add 10 elements of AItems (id's from 0-9) to the created A(id 1).
>       persist A.
>   em1.commit()
>  
>   em1.begin()
>       merge A ( created in the previous step)
>       Remove 3 elements of AItems from the merged A.
>       Add 3 elements of AItems ( id's 10,11,12) to the merged A (id 1).
>    
> With out committing em1
>    
>   em2.begin()
>       query database to fetch A and construct object result2 of entity A.
>       Add 3 elements of AItems ( id's 13,14,15) to fetched A ( result2)      
>    em2.commit ()
>    em1.commit()
>    
>   em3.begin()
>      query database to check the size of AItems that are related to A ( id 1)
>   em3.commit()
>    
>   The result on em3's query for AItems related to A, returns 13 as expected.
>   13 ( Initial 10 - em1's 3 + em1's 3 + em2's 3).
>    
> When the same test case is repeated with removing and adding 10 elements instead of 3 as before then I get wrong results.
>    
>     Add initial 10 AItems (id's 0-9) for A.
>     commit()
>    
>     em1 will remove 10 AItems from the collection of A.
>     em1 will add 10 AItems (id's 10-19) to collection of A.
>    
>     em2 will add 10 AItems (id's 20-29) to collection of A.
>    
>     Commit em2.
>     Commit em1.
>    
>     Then instead of 20 elements ( Initial 10 - em1's 10 + em1's 10 + em2's 10), I see only 10 elements.
>    
>     The 10 elements that I see are from em1's added AItems ( id's 10-19).
> I think the cause of the issue is that, when more number of elements (compared to initial element count of collection) in a collection are modified then collection tracking is disabled and openJPA tries to do the following:
>  -- Delete every thing from the collection
>  -- Insert data back to collection.
> While Inserting the data back it does not consider adding the dirty records ( em2's 10 added elements ) because the collection tracking is disabled.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.