[jbosscache-issues] [JBoss JIRA] Commented: (JBCACHE-1530) Stale copies of gravitated data left in "xxx:DEAD trees

[Brian Stansberry (JIRA)]

    [ https://jira.jboss.org/jira/browse/JBCACHE-1530?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12481010#action_12481010 ] 

Brian Stansberry commented on JBCACHE-1530:
-------------------------------------------

To get around the last twist above, having the DataGravitationCleanup command clean the defunct data trees looks like the only way to go. I've implemented this with a fairly kludgey mechanism to limit it to about a minute or so after the defunct data tree is created.  After that much time has elapsed any GravitateResult should have come from the defunct data tree.

> Stale copies of gravitated data left in "xxx:DEAD trees
> -------------------------------------------------------
>
>                 Key: JBCACHE-1530
>                 URL: https://jira.jboss.org/jira/browse/JBCACHE-1530
>             Project: JBoss Cache
>          Issue Type: Bug
>      Security Level: Public(Everyone can see) 
>    Affects Versions: 3.2.0.BETA1
>            Reporter: Brian Stansberry
>            Assignee: Manik Surtani
>             Fix For: 3.2.0.GA
>
>
> There's a race that can result in a stale copy of gravitated data being left in a "/BUDDY_BACKUP/xxx:DEAD" tree:
> Scenario:
> 1) Data is store in /B_B/xxx/foo
> 2) xxx is leaving group
> 3) Gravitate data command for /B_B/xxx/foo comes in, result is returned
> 4) Buddy group re-formation thread move /B_B/xxx/foo to /B_B/xxx:DEAD/1/foo
> 5) DataGravitationCleanupCommand comes in for /B_B/xxx/foo which results in nothing happening since the node is moved
> A fix might involve some analysis of the backup fqn in the DataGravitationCleanupCommand, try to detect this condition.  Or perhaps tracking successful GravitationResult responses, trying to match against the cleanup command.
> As a quick workaround I'm going to investigate an algorithm on the GravitateDataCommand sender side to not just accept the first successful result but rather to compare all positive results, giving preference to:
> 1) A result from the main tree
> 2) A result from a non :DEAD buddy backup tree
> 3) A result from xxx:DEAD/2
> 4) A result from xxx:DEAD/1

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: https://jira.jboss.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira