[ https://hibernate.onjira.com/browse/HHH-7265?page=com.atlassian.jira.plug... ] Steve Ebersole commented on HHH-7265: ------------------------------------- Well the problem I think is that it is cleared "mid-loop" by one of the synchronizations. We need to find out which so that I can plan how to best fix this. Was anyone able to reproduce this in an isolated test case? -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira

[ https://hibernate.onjira.com/browse/HHH-7265?page=com.atlassian.jira.plug... ] Strong Liu commented on HHH-7265: --------------------------------- https://issues.jboss.org/browse/JBPAPP-8796 -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira

[ https://hibernate.onjira.com/browse/HHH-7265?page=com.atlassian.jira.plug... ] Scott Marlow updated HHH-7265: ------------------------------ Attachment: hibernatesearchsync.jpg debugger session proving Gail's theory about there being another sync registered after the entitymanager closer. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira

[ https://hibernate.onjira.com/browse/HHH-7265?page=com.atlassian.jira.plug... ] Gail Badner commented on HHH-7265: ---------------------------------- IIUC, the session should not be auto-closed after the transaction completes for PersistenceContextType.EXTENDED unless the EntityManager is closed. I couldn't find a way to enable session auto-close when the EntityManager is closed without adding a setter to SessionImplementor. Maybe I'm missing something here? I created a pull request (https://github.com/hibernate/hibernate-orm/pull/320) that removes the synchronization that gets registered by EntityManagerImpl.close(). Instead the synchronization will be added by AbstractEntityManagerImpl$AfterCompletionActionImpl.doAction() if the session is still open, but the entity manager is closed. Adding the synchronization here should ensure that it's the last afterCompletion synchronization that closes the session. It would be tidier if AfterCompletionAction had something like "doAfterAfterCompletionAction" that simply closed the session (rather than relying on a synchronization to do it). See documentation in the pull request code for more info. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira

[ https://hibernate.onjira.com/browse/HHH-7265?page=com.atlassian.jira.plug... ] Scott Marlow commented on HHH-7265: ----------------------------------- I haven't reviewed the patch yet but wanted to share an even hackier way to possibly fix. ;) How about if we hacked the synchronizations so that the EntityManager.close is always last (via some hack logic that keeps the EM.close at the end of the list). Currently, the iteration order will be the order that the synchronizations were added. I think a cleaner fix would be to add a new method to SynchronizationRegistry: {code} /** * Register a system {@link Synchronization} level callback for this transaction. System level synchronizations will run user-system level synchronizations have run. * * @param synchronization The synchronization callback to register. * * @throws org.hibernate.HibernateException */ public void registerSystemSynchronization(Synchronization synchronization); {code} I assume we aren't changing the SPI now but this might be one way to address this (when the SPI can change). The current registerSynchronization(Synchronization) method could be hacked to put the passed Synchronization on a different list (or keep it at the end of an arraylist) until we make the SPI change. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira

[ https://hibernate.onjira.com/browse/HHH-7265?page=com.atlassian.jira.plug... ] Gail Badner commented on HHH-7265: ---------------------------------- I was also thinking that there should be a way delegate to the session owner (EntityManager) to determine if the session can be auto-closed. IMO, this is a better alternative than adding a "system" synchronization along with more assumptions about when it gets executed wrt observers, actions, and "normal" synchronizations. Steve, I'm not sure how properties can be used to determine if the session should be auto-closed when using PersistenceContextType.EXTENDED. Using PersistenceContextType.EXTENDED, IIUC, the session should only be auto-closed after a transaction if the EntityManager is closed, so it's not a simple property lookup. Am I missing something here (very possible)? There is a test in the pull request that illustrates this case: TransactionJoiningTest.testCloseAfterCommit(). -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira

[ https://hibernate.onjira.com/browse/HHH-7265?page=com.atlassian.jira.plug... ] Steve Ebersole updated HHH-7265: -------------------------------- Pull Requests: https://github.com/hibernate/hibernate-orm/pull/322 (was: https://github.com/hibernate/hibernate-orm/pull/322, https://github.com/hibernate/hibernate-orm/pull/320) -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira