Hibernate session.merge() is updating non-changed entities.

[jmoore]

It was found in r582 that calls to session.merge() were cascading to all instances so that versions in Experimenter, EventTypes, etc. are being incremented. This was hellish for clients trying to deal with optimistic locks.

This is a hibernate bug. See http://opensource.atlassian.com/projects/hibernate/browse/HHH-1401 (version 3.2 may help)

Revisions r590, r591, and r597 made temporary fixes so work could continue without versions.

Before we close this, we need to re-instate versions and finish the update tests (like DetachedPixelsGraphTest) which test the possible combinations of transient/detached graphs.

• Transient root with:
  • transient entity
  • "transient" collection
  • unloaded entity
  • filtered collection (ERROR)

• Detached root with:
  • transient entities/collections
  • detached entities/collections
  • filtered entities/collections

Also we need to make sure that the bug fix really worked, i.e.:

• re-merging unchanged detached --> no version increment
• re-merging changed --> version++ (also re: collections)

A possible Abstract test would be of the form:

  class {
    createGraph();
    saveGraph();
    checkIds();
    checkVersions();
    checkLogs();
  }

Originally http://bugs.openmicroscopy.org.uk/show_bug.cgi?id=646

[jmoore]

Need to try out newer (3.2.x) versions of Hibernate to see if this is fixed. Also important for security (#114) and the new model (#96). Moving to the next milestone:3.0-M3

[jmoore]

Not having versions is again causing massive troubles from the client side. (Just versions would cause an equal number of stale object exceptions, but in combination with the client session should be seamless.)

An example of a problem is the following:

    Experimenter user1 = iAdmin.lookupExperimenter("user1")";
    Experimenter newUser = iAdmin.createUser( userDTO );
    // ... create an object belonging to user1
    iUpdate.saveObject( object ); 
    // which removes newUser from the user group! 

This happens because the set of GroupExperimenterMaps attached to user1 doesn't contain the new map from newUser to the user group. Since there is no version to tell Hibernate that the sets from user are stale, it removes the link. Boom'''

[jmoore]

HHH-1564 appears to be related.

[jmoore]

http://forum.hibernate.org/viewtopic.php?t=950225&highlight=collectiontype+isdirty

[jmoore]

This is being fixed and is planned for Hibernate 3.2.1. There may be other issues: see HHH-85 (ref. the propery hibernate.batch_update_versioned)

[jmoore]

#81 might offer us another way out (update events)

[jmoore]

If we go the #81 route (using update events) then rather then throwing a SecurityViolation it needs to throw an OptimisticLockException.

[jmoore]

Note to previous comment: again if we got the #81 comment, we will have to prevent null update events for managed entities. I.e.

 * NullId/NullUpdateEvent    --> OK
 * NullId/NonNullUpdateEvent --> Not ok.
 * NonNullId/NullUpdateEvent --> Not ok.
 * Both non null             --> OK

[jmoore]

Still waiting on a response from the Hibernate team. Moving to milestone:Blocked

[jmoore]

They say they've fixed it as of hibernate:r10740 . We need to test that. Rescheduling.

See also: http://opensource.atlassian.com/projects/hibernate/browse/HHH-2217

[jmoore]

This just got missed in milestone:3.0-Beta2. Certainly time to upgrade Hibernate versions.

[jmoore]

We've upgraded, and this problem has not shown up again. However, versions still need to be reviewed. Pushing.