[
https://issues.apache.org/jira/browse/HDFS-3207?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13442297#comment-13442297
]
liaowenrui commented on HDFS-3207:
----------------------------------
>In network down (or) machine down cases, we can never ensure with the
>available fence methods, whether the process is down or not.
we resolve this issue.hdfs is avaliable in this scenario.
>In other case, I don't see why fence can fail from fence methods configured
because the sharedstorage has strong writer level fencing (like how BKJM has
fencing) implemented.if active namenode is mutiple fencing fail,standby
namenode become active NameNode.
>where we can not ensure that older active is down from fence methods
>configured, in the case of machine is reachable?
if sharedstorage has strong writer level fencing (like how BKJM has fencing)
implemented,we cofigured that if active namenode is mutiple fencing
fail,standby namenode become active NameNode.
> Mechanism for HA failover to ignore fencing errors
> --------------------------------------------------
>
> Key: HDFS-3207
> URL: https://issues.apache.org/jira/browse/HDFS-3207
> Project: Hadoop HDFS
> Issue Type: New Feature
> Components: ha
> Affects Versions: HA branch (HDFS-1623)
> Reporter: Philip Zeyliger
> Assignee: Todd Lipcon
>
> If an administrator wants to ignore fencing (perhaps because he knows that
> the other namenode has been taken out of comission for sure, but the fencing
> is ssh-based), there should be a flag to the failover command to allow this.
--
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
