hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Vinayakumar B (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-13742) Expose "NumOpenConnectionsPerUser" as a metric
Date Wed, 16 Nov 2016 06:15:58 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-13742?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15669568#comment-15669568

Vinayakumar B commented on HADOOP-13742:

bq. First of all, the use of "Vs" is probably not the best choice. "Vs" is for two comparable
I think "userToConnectionsMap" might be better.
bq. I am wondering whether introducing a separate metrics class is better. (Something similar
to RpcDetailedMetrics) For typical metrics, a counter is simply incremented and the diff shows
the rate. If users are only interested in instantaneous counts and do not care at all about
cumulative counts, the approach in the patch might be okay. But I am thinking cumulative values
are useful too.
That's a good idea and may be required to for detailed analysis about each user's usage. In
simple terms there could be separate instances of RpcDetailedMetrics for each user, Just an
Idea. I feel this could be done in a follow-up jira and detailed discussions could be made.

bq. If we are to proceed with the proposed approach, I suggest simply synchronize on the lock
object inside incrUserConnections() and decrUserConnections() and lose the private methods.
Then no need to have AtomicInteger. The ideal way will be range-locking the map, but I don't
know whether there is any known good implementation of it. In any case, the number of rpc
reader threads is typically small, so this level of synchronization/monitor lock should be
In this jira only simple count of connections can be exposed as a metric, as title says.
 [~brahmareddy], please update the patch with simple synchronization, as suggested by [~kihwal].

> Expose "NumOpenConnectionsPerUser" as a metric
> ----------------------------------------------
>                 Key: HADOOP-13742
>                 URL: https://issues.apache.org/jira/browse/HADOOP-13742
>             Project: Hadoop Common
>          Issue Type: Improvement
>            Reporter: Brahma Reddy Battula
>            Assignee: Brahma Reddy Battula
>         Attachments: HADOOP-13742-002.patch, HADOOP-13742-003.patch, HADOOP-13742-004.patch,
> To track user level connections( How many connections for each user) in busy cluster
where so many connections to server.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: common-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-issues-help@hadoop.apache.org

View raw message