hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-2862) Name DFSClient for Improved Debugging
Date Thu, 22 Jul 2010 14:44:50 GMT

    [ https://issues.apache.org/jira/browse/HBASE-2862?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12891195#action_12891195
] 

stack commented on HBASE-2862:
------------------------------

I like this patch.

May I suggest the following N:

In HRS edit you do this, 

{code}
+        this.conf.set("mapred.task.id", 
+            "HRegionServer_" + this.serverInfo.getServerAddress().toString());
{code}

On commit, mind if I change the String to instead be HServerAddress.getServerName?  We're
trying to standardize on what we call the HRS.  getServerAddress is kinda ugly -- its <hostname>
',' <port> ',' <startcode> -- but it has all we'd ever need in it and its being
used in our logs, both local to RS and its how the HRS shows in the master log.  I see an
advantage to having the HRS name shown in NN logs be same as is shown in ours (getServerAddress
also adds startcode so we'll be able to distingush file actions either side of a crash).

Master name would remain as you have it -- we don't have an equivalent getServerAddress for
master (though maybe we should since there can be different masters over time).

If amenable to the above, no need to make a new patch.  I'll just make change on commit.

Thanks.

> Name DFSClient for Improved Debugging
> -------------------------------------
>
>                 Key: HBASE-2862
>                 URL: https://issues.apache.org/jira/browse/HBASE-2862
>             Project: HBase
>          Issue Type: Improvement
>          Components: master, regionserver
>            Reporter: Nicolas Spiegelberg
>            Assignee: Nicolas Spiegelberg
>            Priority: Trivial
>             Fix For: 0.90.0
>
>         Attachments: HBASE-2862.patch
>
>
> Reported by Hairong.  We had an HDFS error on our test cluster.  It was hard to debug
the HDFS NameNode logs because there was no way to map DFClient => RegionServer.  Hadoop
guys added a hacky config value, "mapred.task.id", which allows you to add a suffix to the
DFSClient ID for logging.  We should piggyback upon that for our HLog/HFile code 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message