hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Maysam Yabandeh (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-7426) Change nntop JMX format to be a JSON blob
Date Thu, 04 Dec 2014 06:54:12 GMT

    [ https://issues.apache.org/jira/browse/HDFS-7426?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14233960#comment-14233960
] 

Maysam Yabandeh commented on HDFS-7426:
---------------------------------------

Thanks for the taking care of this [~andrew.wang]. The patch looks good to me.

I have a comment about the omitted TopConf.ALL_USERS. The rational behind it was not to lose
the account for the entire volume of traffic. e.g., using that we can tell whether the top
10 users account for 60% of the traffic or 99%. I understand that this number is redundantly
present in NameNodeMetrics, but since over there it is not mapped to the time periods covered
by the TopMetics rolling windows, recomputing it would add more complexity than redundantly
maintaining it in TopMetrics. Later in html view we would use that to also compute the percentage
of the traffic that each user is responsible for, a number that we found visually quite effective
in the html view.

That being said, I have second thoughts about the necessity of the html view. The small cluster
deployments who are not yet equipped with an external monitoring tool can simply look at the
jmx output. Admitted that it is not as visually effective as a table in a html view, but does
not sound that important. Our monitoring tool also makes use of only 1-minute-long rolling
windows for which the metrics reported in NameNodeMetrics are good enough. I can cancel HDFS-7465
and then we can leave ALL_USERS omitted.

> Change nntop JMX format to be a JSON blob
> -----------------------------------------
>
>                 Key: HDFS-7426
>                 URL: https://issues.apache.org/jira/browse/HDFS-7426
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: namenode
>    Affects Versions: 2.7.0
>            Reporter: Andrew Wang
>            Assignee: Andrew Wang
>         Attachments: hdfs-7426.001.patch, hdfs-7426.002.patch
>
>
> After discussion with [~maysamyabandeh], we think we can adjust the JMX output to instead
be a richer JSON blob. This should be easier to parse and also be more informative.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message