[
https://issues.apache.org/jira/browse/HBASE-2991?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12909949#action_12909949
]
Matthew Welty commented on HBASE-2991:
--------------------------------------
@stack: Ideal outcome of a -refreshNodes op would be to compare the contents of
an excludes configuration with the current in-memory state, and if there is a
conflict it should "decommission" a regionserver and no longer allow it to
participate in the cluster. In HDFS a datanode decommissioning amounts to a
period of block replication. For an HBase regionserver this may just amount to
closing itself down cleanly.
I don't think having an 'includes' list is necessary (it also breaks with the
convention of Hadoop Common), unless someone has a very odd transience issue
with rogue regionservers.
> Add support for refreshNodes in master
> --------------------------------------
>
> Key: HBASE-2991
> URL: https://issues.apache.org/jira/browse/HBASE-2991
> Project: HBase
> Issue Type: Wish
> Components: master
> Reporter: Matthew Welty
> Priority: Minor
>
> -refreshNodes capabilities should be added to bin/hbase (or to a new script
> called `hbaseadmin`, if we want to maintain command name consistency) with
> cluster exclusion functionality similar to the capabilities of dfsadmin and
> mradmin. For ease of administration, the behavior of this option should be
> procedurally identical to the functions of the Hadoop Common components,
> including modifying the exclusion list via a local 'excludes' configuration
> file and display of excluded nodes via the web UI.
> A command overview for the behavior of -refreshNodes in dfsadmin is here:
> http://hadoop.apache.org/common/docs/current/commands_manual.html#dfsadmin
> The command overview for -refreshNodes in mradmin is here:
> http://hadoop.apache.org/common/docs/current/commands_manual.html#mradmin
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
