hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Yongjun Zhang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-6578) add toString method to DatanodeStorage etc for easier debugging
Date Sun, 22 Jun 2014 17:26:24 GMT

    [ https://issues.apache.org/jira/browse/HDFS-6578?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14040187#comment-14040187

Yongjun Zhang commented on HDFS-6578:

Hi [~arpit99],

Thanks for your comments. I didn't see it when I uploaded revision 002 for a minor change.

About your comments.  in patch 001, I wanted to give some more explanation why the last BlockManager.processReport
call is accurate, so I described it as "accumulate" based on my understanding. This seems
to be inaccurate. I wonder if you could  help saying a bit more on top of "The result of the
last BlockManager.processReport call is accurate"?

Say, since we are assigning noStaleStorages for each call, the patterns of noStaleStorage
value in the loop can be:

Iter             0      1        2        3        4 (end of loop)
pattern1.   false, false, false,  false, false
pattern2.   true,  true,   true,   true,  true
pattern3.   false, false, true,   true,   true
pattern4,   true,  true,  false,  false, false
pattern5.   false, true,  false,  true,  false

I didn't study the code hard enough, but my understanding (which seems to be wrong) was that,
pattern1 through pattern4 are valid, and pattern5 not.  Based on your description, it seems
actually any pattern of the above patterns could happen, each call get up-to-date return value
based on what processed so far. And the last iteration is most up-to-date. Is this understanding
correct? If so, can we say "The result of the last BlockManager.processReport call is most
up-to-date and accurate".

Thanks for your help toward my better understanding.

> add toString method to DatanodeStorage etc for easier debugging
> ---------------------------------------------------------------
>                 Key: HDFS-6578
>                 URL: https://issues.apache.org/jira/browse/HDFS-6578
>             Project: Hadoop HDFS
>          Issue Type: Bug
>            Reporter: Yongjun Zhang
>            Assignee: Yongjun Zhang
>         Attachments: HDFS-6578.001.patch, HDFS-6578.002.patch
> It seems to be nice to add a toString() method for DatanodeStorage class, so we can print
out its key info easier while doing debuging.
> Another thing is, in the end of BlockManager#processReport, there is the following message,
> {code}
>    blockLog.info("BLOCK* processReport: from storage " + storage.getStorageID()
>         + " node " + nodeID + ", blocks: " + newReport.getNumberOfBlocks()
>         + ", processing time: " + (endTime - startTime) + " msecs");
>     return !node.hasStaleStorages();
> {code}
> We could add node.hasStaleStorages() to the log, and possibly replace storage.getSorateID()
with the suggested storage.toString().
> Any comments? thanks.

This message was sent by Atlassian JIRA

View raw message