[jira] Updated: (HBASE-3648) [replication] failover is sloppy with znodes

["Jean-Daniel Cryans (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HBASE-3648?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jean-Daniel Cryans updated HBASE-3648:
--------------------------------------

    Attachment: HBASE-3648.patch

This patch adds more targeted exception handling and forces the deletion of the lock znode at the end.

> [replication] failover is sloppy with znodes
> --------------------------------------------
>
>                 Key: HBASE-3648
>                 URL: https://issues.apache.org/jira/browse/HBASE-3648
>             Project: HBase
>          Issue Type: Bug
>    Affects Versions: 0.90.1
>            Reporter: Jean-Daniel Cryans
>            Assignee: Jean-Daniel Cryans
>            Priority: Critical
>             Fix For: 0.90.2
>
>         Attachments: HBASE-3648.patch
>
>
> ReplicationZookeeper is a bit sloppy in how it handles the znodes during failover:
> - when creating the lock, it doesn't cleanly handle the situation where the parent znode might already be deleted.
> - when deleting the znodes after a successful move, it doesn't make sure to delete the lock znode last.
> - after deleting the lock, there's a window where another region server could have already created another lock and deleted the znodes which would abort the first region server (saw it on one cluster).

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira