hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Purtell (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-7714) allow master to exclude nodes from running region servers
Date Wed, 30 Jan 2013 00:51:13 GMT

    [ https://issues.apache.org/jira/browse/HBASE-7714?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13566016#comment-13566016
] 

Andrew Purtell commented on HBASE-7714:
---------------------------------------

We could have something like the NameNode's dfs.hosts (if present, is the list of hosts who
are only allowed to connect) and dfs.exclude.hosts (if present, determines who should be drained
and shut down if already connected), reread by the master every 10 minutes if present. 
                
> allow master to exclude nodes from running region servers
> ---------------------------------------------------------
>
>                 Key: HBASE-7714
>                 URL: https://issues.apache.org/jira/browse/HBASE-7714
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: Ming Ma
>
> Want to bring up this requirement to get people's opinion. It can be done inside hbase
or at higher hadoop management layer.
> To support our clusters, we want to make sure RS doesn't run on certain nodes by accident:
a) either because it is a bad machine with slow NIC card, or b) we don't want HBase to take
away resource from other services. HDFS and MR have exclude files to support this.
> Currently master doesn't specify the RS membership; any machine as long as it has connectivity
and security credential can connect to the master.
> There are two ways to address this requirement.
> 1. Modify hbase to allow master have a say on the RS membership. If a machine is added
to hbase exclude file, master will try to migrate the regions from that machine and ask RS
to go away after it is done. 
> 2. Make no change to hbase. Have some external hadoop management layer handle that.
> We have been supporting our clusters using exclude files at HDFS and MR layer to handle
machine decomission and recommission. So I prefer #1.
> Any comments on 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

Mime
View raw message