hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Vinay (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-3854) Implement a fence method which should fence the BK shared storage.
Date Mon, 27 Aug 2012 05:33:07 GMT

    [ https://issues.apache.org/jira/browse/HDFS-3854?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13442280#comment-13442280
] 

Vinay commented on HDFS-3854:
-----------------------------

Hi Uma, Thanks for posting this Jira.

Its really a good idea to expose a fencing option for Shared storage to ZKFC.
In that case ZKFC itself will ensure complete fencing before going for failover.

How about making this as interface in JournalManager which will be called by the ZKFC for
all shared storages during fencing, if SSH fencing method fails...?
   If the shared storage doesnot have fencing implemented, it can return false. If this is
possible then we need not configure extra fencing method exlicitly for each shared storage.

If adding this interface is not possible, then we can think of implementing a fencing method
for BKJM separately.

{quote}only pain points what I can see is, we may have to put the BKJM jar in ZKFC lib for
running this fence method.{quote}
IMO, this might not be a big problem because ideally NN and ZKFC are started using same installations.
                
> Implement a fence method which should fence the BK shared storage.
> ------------------------------------------------------------------
>
>                 Key: HDFS-3854
>                 URL: https://issues.apache.org/jira/browse/HDFS-3854
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: name-node
>            Reporter: Uma Maheswara Rao G
>
> Currently when machine down or network down, SSHFence can not ensure that, other node
is completely down. So, fence will fail and switch will not happen.
> [ internally we did work around to return true when machine is not reachable, as BKJM
already has fencing]
> It may be good idea to implement a fence method, which should ensure shared storage fenced
propertly and return true.
> We can plug in this new method in ZKFC fence methods.
> only pain points what I can see is, we may have to put the BKJM jar in ZKFC lib for running
this fence method.
> thoughts?

--
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

Mime
View raw message