[ 
https://issues.apache.org/jira/browse/HBASE-6775?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13487378#comment-13487378
 ] 

Gregory Chanan commented on HBASE-6775:
---------------------------------------

I think config is the best option.  It should only be temporary (gone in 0.96), 
so it's not making things any more cryptic long term.

The basic problem is ZOOKEEPER-1495; you can't issue a command and figure out 
quickly whether the cluster supports multi or not.  You can try to figure this 
out yourself, using "srvr", but it's tricky.  You need to check that every node 
on the cluster supports multi.  Apparently just figuring out what servers are 
in the cluster is complicated (I was told jmx is the best way); the zk 
configuration in hbase-land is not sufficient because it may not list every 
node.  It is not sufficient to just check for a quorum of servers that support 
it, because the servers that don't support it will be kicked (remove 
themselves?) from the quorum and you'll lose your fault tolerance.

Even if you know every server in the cluster supports multi at some point in 
time, that may change by the time you actually issue your command.  A new 
server may join or a server version may be downgraded by the time you actually 
send your command.  So there's no good way to do this.

Any objection to going with the Configuration parameter, Stack or others? 
(config parameter would be called hbase.zookeeper.useMulti).
                
> Use ZK.multi when available for HBASE-6710 0.92/0.94 compatibility fix
> ----------------------------------------------------------------------
>
>                 Key: HBASE-6775
>                 URL: https://issues.apache.org/jira/browse/HBASE-6775
>             Project: HBase
>          Issue Type: Improvement
>          Components: Zookeeper
>    Affects Versions: 0.94.2
>            Reporter: Gregory Chanan
>            Assignee: Gregory Chanan
>            Priority: Minor
>             Fix For: 0.94.4
>
>
> HBASE-6710 fixed a 0.92/0.94 compatibility issue by writing two znodes in 
> different formats.
> If a ZK failure occurs between the writing of the two znodes, strange 
> behavior can result.
> This issue is a reminder to change these two ZK writes to use ZK.multi when 
> we require ZK 3.4+. 

--
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

Reply via email to