[jira] Commented: (JCR-2861) Internal Timeout Handling in the TransactionContext is not XA Spec. conform

["Claus Köll (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/JCR-2861?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12986349#action_12986349 ] 

Claus Köll commented on JCR-2861:
---------------------------------

The internal TimerTask will only be scheduled if we are in a one phase transaction.

> Internal Timeout Handling in the TransactionContext is not XA Spec. conform
> ---------------------------------------------------------------------------
>
>                 Key: JCR-2861
>                 URL: https://issues.apache.org/jira/browse/JCR-2861
>             Project: Jackrabbit Content Repository
>          Issue Type: Improvement
>          Components: jackrabbit-core
>    Affects Versions: 2.2.0
>            Reporter: Claus Köll
>            Assignee: Claus Köll
>            Priority: Critical
>
> The problem here is that in a 2 phase transaction the xa spec does not  
> permit a RB* return code in response to xa_commit().  The xa spec says  
> the following about RB* return codes in the xa_commit() section:        
>                                                                         
> "The resource manager did not commit the work done on behalf of the     
> transaction branch.  Upon return, the resource manager has rolled back  
> the branch?s work and has released all held resources.  These values may
> be returned only if TMONEPHASE is set in flags"                         
>                                                                         
> Essentially, the only two return codes from xa_commit that J2EE Containers can     
> handle sensibly are XA_OK (normal case) and XA_RMFAIL.  RMFAIL will     
> cause the containers to retry to commit the  transaction.  Any other return code will result in a heuristic          
> transaction outcome (non-atomic).  
> In a xa environment the TMONEPHASE is not set on the flags and so XA_RBTIMEOUT is 
> not a permitted return code. A Container  transaction service cannot do anything to ensure an atomic     
> outcome if an XAResource fails to honour its promise to be able to commit it made when it answer XA_OK in response to xa_prepare(). 
> The internal timeout handling will rollback the Jackrabbit XAResource if the time exceeds between prepare and commit.
> and in the commit Method will always throw a XA_RBTIMEOUT.
> We should not handle the timeout internal because this should make the container in a 2 Phase transaction.

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