[infinispan-issues] [JBoss JIRA] (ISPN-4426) Transaction replayed but not committed

[Radim Vansa (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-4426?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12978441#comment-12978441 ] 

Radim Vansa commented on ISPN-4426:
-----------------------------------

Log is here: https://dl.dropboxusercontent.com/u/103079234/ispn4226.zip
The not committed transaction is {{GlobalTransaction:<edg-perf03-58446>:23655:remote}} executed by {{remote-thread-5}} at 10:07:31-10:07:32. Due to that, write to {{key_0000000000001519}} with value {{\[21 #17: 85, 300, 1263, 2398, 2909, 3485, 3684, 4037, 4473, 4673, 4754, 5156, 5429, 5444, 5595, 5641, 5767, \]}} was lost (notice the last 5767).

The test was executed against this JDG with some logging modifications [https://github.com/rvansa/jdg/tree/t_4426_logs] and with log filters as in [https://svn.devel.redhat.com/repos/jboss-qa/load-testing/etc/jdg-60/log4j/log4j-trace-fast.xml].


> Transaction replayed but not committed
> --------------------------------------
>
>                 Key: ISPN-4426
>                 URL: https://issues.jboss.org/browse/ISPN-4426
>             Project: Infinispan
>          Issue Type: Bug
>      Security Level: Public(Everyone can see) 
>          Components: State Transfer
>    Affects Versions: 7.0.0.Alpha4
>            Reporter: Radim Vansa
>            Assignee: Dan Berindei
>            Priority: Critical
>              Labels: 63gablocker
>
> Dist TX cache, node C is joining. In previous topology, entry is owned by A (primary) and B (backup). In new topology, primary ownership is transferred to C, B stays backup.
> 1. TX is prepared in old topology and is being committed, when topology changes
> 2. on C (the new owner), the TX info is received and later even the old entry
> 3. C receives the CommitCommand, therefore, it correctly replays the PrepareCommand.
> 4. When the entries are about to be committed, in TxInterceptor the transaction is found to be already completed as it has lower TxID.
> Result: the transaction is not being executed and stale data stay on the node (with my algortihm it eventually led to complete entry loss).



--
This message was sent by Atlassian JIRA
(v6.2.6#6264)