hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mit Desai (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-6597) New option for namenode upgrade
Date Tue, 24 Jun 2014 18:54:24 GMT

    [ https://issues.apache.org/jira/browse/HDFS-6597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14042518#comment-14042518
] 

Mit Desai commented on HDFS-6597:
---------------------------------

The idea seems to be good as it does not alter the way -upgrade currently works.
* I agree with [~cmccabe] and [~cnauroth] on the new name to be "force"
* Instead of -force, -halt or -upgradeOnly seems to be reasonable. But anything would be good
as long as it does nor imply we are forcing something to get done which it should not be doing.

Thanks,
Mit

> New option for namenode upgrade
> -------------------------------
>
>                 Key: HDFS-6597
>                 URL: https://issues.apache.org/jira/browse/HDFS-6597
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: namenode
>            Reporter: Danilo Vunjak
>         Attachments: JIRA-HDFS-30.patch
>
>
> Currently when namenode is started for upgrade (hadoop namenode -upgrade command), after
finishing upgrade of metadata, namenode starts working normally and wait for datanodes to
upgrade itself and connect to to NN. We need to have option for upgrading only NN metadata,
so after upgrade is finished on NN, process should terminate.
> I have tested it by changing in file: hdfs.server.namenode.NameNode.java, method: public
static NameNode createNameNode(String argv[], Configuration conf):
>  in switch added
>  case UPGRADE:
> case UPGRADE:
>       {
>     	  DefaultMetricsSystem.initialize("NameNode");
>           NameNode nameNode = new NameNode(conf);
>           if (startOpt.getForceUpgrade()) {
>         	  terminate(0);
>         	  return null;
>           }
>           
>           return nameNode;
>       }
> This did upgrade of metadata, closed process after finished, and later when all services
were started, upgrade of datanodes finished sucessfully and system run .
> What I'm suggesting right now is to add new startup parameter "-force", so namenode can
be started like this "hadoop namenode -upgrade -force", so we can indicate that we want to
terminate process after upgrade metadata on NN is finished. Old functionality should be preserved,
so users can run "hadoop namenode -upgrade" on same way and with same behaviour as it was
previous.
>  Thanks,
>  Danilo



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message