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] [Updated] (HDFS-6982) nntop: top­-like tool for name node users
Date Thu, 04 Sep 2014 22:25:24 GMT

     [ https://issues.apache.org/jira/browse/HDFS-6982?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Maysam Yabandeh updated HDFS-6982:
----------------------------------
    Attachment: HDFS-6982.patch

I am attaching the first patch that imports the main components from our standalone nntop
tool to hdfs repo. The patch is not complete but I am attaching it to get the preliminary
feedbacks, e.g., whether we can move forward with a single patch or we need to divide it into
multiple patches, etc.

A couple of things missing in the patch:
# I removed the simple web page that shows the top users since it was written using org.apache.hadoop.yarn.webapp.
I am thinking that we probably need to write a whole new page that does not depend on yarn.
I am not sure at this point which technology is recommended for that and to which page nntop
results could be added. I am wondering perhaps we can postpone this to another patch, which
would also make this patch simpler.
# The code for registering TopMetrics to metrics system is not present in the patch as I was
not sure where is the best place to put it.
# A user appearing in the current top list might not be present in the next generated top
list. The previously reported users should then be removed from the exported jmx metrics.
This desired behavior is not achieved with the current implementation of MetricsSourceAdapter
which resets its cache only if a new metrics is added:
{code}
      int oldCacheSize = attrCache.size();
      int newCacheSize = updateAttrCache();
      if (oldCacheSize < newCacheSize) {
        updateInfoCache();
      }
{code}
In our tool we had changed this behavior to always invoke updateInfoCache. I did not import
this change in the patch since I was not sure this change is desirable for all use cases.
I guess we can work on another patch that makes this optional in MetricsSourceAdapter?

Comments are highly appreciated.

> nntop: top­-like tool for name node users
> -----------------------------------------
>
>                 Key: HDFS-6982
>                 URL: https://issues.apache.org/jira/browse/HDFS-6982
>             Project: Hadoop HDFS
>          Issue Type: New Feature
>            Reporter: Maysam Yabandeh
>         Attachments: HDFS-6982.patch, nntop-design-v1.pdf
>
>
> In this jira we motivate the need for nntop, a tool that, similarly to what top does
in Linux, gives the list of top users of the HDFS name node and gives insight about which
users are sending majority of each traffic type to the name node. This information turns out
to be the most critical when the name node is under pressure and the HDFS admin needs to know
which user is hammering the name node and with what kind of requests. Here we present the
design of nntop which has been in production at Twitter in the past 10 months. nntop proved
to have low cpu overhead (< 2% in a cluster of 4K nodes), low memory footprint (less than
a few MB), and quite efficient for the write path (only two hash lookup for updating a metric).



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

Mime
View raw message