[infinispan-issues] [JBoss JIRA] (ISPN-2137) Potential tx lock leaks when nodes are added to the cluster

[RH Bugzilla Integration (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-2137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12742457#comment-12742457 ] 

RH Bugzilla Integration commented on ISPN-2137:
-----------------------------------------------

Michal Linhard <mlinhard at redhat.com> made a comment on [bug 841891|https://bugzilla.redhat.com/show_bug.cgi?id=841891]

VERIFIED for 6.1.0.ER6

I've taken test: LockCleanupStateTransferTest (+added it to 5.1.5.FINAL)

and ran it for
5.1.5.FINAL - Failed
5.2.0.Beta6 - Passed

Note:
(the ER6 JAR was taken from MEAD repo - btw not md5sum-same as the JAR in jboss-datagrid-library-6.1.0.ER6.zip)

JAR in jboss-datagrid-library-6.1.0.ER6.zip: b8de37f0bd621246585a4de5939c66de
JAR in repo: a122144dac3b9d3e767389207bf6815a

(I tested both JARs)

The point of failure for 5.1.5.FINAL was here
https://github.com/infinispan/infinispan/blob/master/core/src/test/java/org/infinispan/tx/LockCleanupStateTransferTest.java#L152

Which IMO correspondes to the failing scenario described in ISPN-2137
                
> Potential tx lock leaks when nodes are added to the cluster
> -----------------------------------------------------------
>
>                 Key: ISPN-2137
>                 URL: https://issues.jboss.org/browse/ISPN-2137
>             Project: Infinispan
>          Issue Type: Bug
>          Components: State transfer, Transactions
>    Affects Versions: 5.1.5.FINAL
>            Reporter: Mircea Markus
>            Assignee: Mircea Markus
>            Priority: Critical
>              Labels: jdg6
>             Fix For: 5.2.0.Beta2, 5.2.0.Final
>
>
> Here's the sequence that might cause the leak:
> - a tx locking K is prepared and committed on node A
> - node B joins and becomes the primary owner of K
> - tx and the lock on key is moved from A to B as part of the state transfer
> - tx completion notification for tx is sent (async+oob) to A and not to B(old view)
> - there's no way for A to reply to the the originator telling it to re-send the tx completion notification to B as the call is async+oob
> - this would cause the lock transaction to leak on B
> Suggested solution:
> After a chat with Dan, following fixed seemed to be the most appropriate:
> when receiving a tx completion notification for a tx that's being migrated over as result of state transfer forward it to the new lock owner. With current blocking ST, this fwd call waits only after the tx state is transferred to the new owner. This would need some more thought for the new NBST code.
> Note: there's no issue in the case of nodes leaving the cluster, as the current logic of backup nodes would assure a proper cleanup.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira