[jira] [Commented] (HBASE-7244) Provide a command or argument to startup, that formats znodes if provided

["stack (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-7244?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13646997#comment-13646997 ] 

stack commented on HBASE-7244:
------------------------------

Looks great.

One question.  The default, if you pass no arguments, is to clean everything?   Should instead the default be to do nothing but print out usage?   Make it so you have to pass in what you want cleaned?  Pass --cleanZk to clean zk, -- cleanHDFS to clean hdfs, and --cleanAll to clean all?

I'm upping priority on this issue because its critical to migration/prep-for-0.95 upgrade.
                
> Provide a command or argument to startup, that formats znodes if provided
> -------------------------------------------------------------------------
>
>                 Key: HBASE-7244
>                 URL: https://issues.apache.org/jira/browse/HBASE-7244
>             Project: HBase
>          Issue Type: New Feature
>          Components: Zookeeper
>    Affects Versions: 0.94.0
>            Reporter: Harsh J
>            Assignee: rajeshbabu
>            Priority: Minor
>             Fix For: 0.98.0, 0.94.8, 0.95.1
>
>         Attachments: HBASE-7244_2.patch, HBASE-7244_3.patch, HBASE-7244_4.patch, HBASE-7244_5.patch, HBASE-7244_6.patch, HBASE-7244.patch
>
>
> Many a times I've had to, and have seen instructions being thrown, to stop cluster, clear out ZK and restart.
> While this is only a quick (and painful to master) fix, it is certainly nifty to some smaller cluster users but the process is far too long, roughly:
> 1. Stop HBase
> 2. Start zkCli.sh and connect to the right quorum
> 3. Find and ensure the HBase parent znode from the configs (/hbase only by default)
> 4. Run an "rmr /hbase" in the zkCli.sh shell, or manually delete each znode if on a lower version of ZK.
> 5. Quit zkCli.sh and start HBase again
> Perhaps it may be useful, if the start-hbase.sh itself accepted a formatZK parameter. Such that, when you do a {{start-hbase.sh -formatZK}}, it does steps 2-4 automatically for you.
> For safety, we could make the formatter code ensure that no HBase instance is actually active, and skip the format process if it is. Similar to a HDFS NameNode's format, which would disallow if the name directories are locked.
> Would this be a useful addition for administrators? Bigtop too can provide a service subcommand that could do 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