Mailing-List: contact common-issues-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Date: Thu, 3 Aug 2017 16:59:00 +0000 (UTC)
From: "Daryn Sharp (JIRA)" <jira@apache.org>
To: common-issues@hadoop.apache.org
Message-ID: <JIRA.13082933.1498594466000.86077.1501779540720@Atlassian.JIRA>
In-Reply-To: <JIRA.13082933.1498594466000@Atlassian.JIRA>
References: <JIRA.13082933.1498594466000@Atlassian.JIRA> <JIRA.13082933.1498594466920@jira-lw-us.apache.org>
Subject: [jira] [Commented] (HADOOP-14599) RPC queue time metrics omit timed
 out clients
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Thu, 03 Aug 2017 16:59:06 -0000


    [ https://issues.apache.org/jira/browse/HADOOP-14599?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16113082#comment-16113082 ] 

Daryn Sharp commented on HADOOP-14599:
--------------------------------------

General implementation issues:
# No need to change UGI.  Revert them.
# Don't change {{RpcProtobufRequest#getRequestHeader}} to convert IOE to an illegal arg.
# In {{NamenodeWebHdfsMethods#doAsExternalCall}}, the changed indentation of methods like {{getHostInetAddress}} and {{getDeclaringClassProtocolName}} violate style guidelines.
# {{WritableRpcEngine#call}} doesn't appear to need the finally clause anymore?
# Is the change in {{Server}} to the deferred response handling is necessary?  It's subtlety changing the behavior.
# In the finally block that updates the metrics, please update _after_ clearing the call and closing the scope.  If for some reason the metrics update blow up, the handler will be left in an inconsistent state.

Most importantly: The queue time for skipped calls is recorded = great!; _but with a processing time of 0_ = bad.  As the call queue becomes congested with timing out clients, the average processing time will plummet and artificially make performance appear great when it's not.  The updates to queue time and processing time need to be independent.

> RPC queue time metrics omit timed out clients
> ---------------------------------------------
>
>                 Key: HADOOP-14599
>                 URL: https://issues.apache.org/jira/browse/HADOOP-14599
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: metrics, rpc-server
>    Affects Versions: 2.7.0
>            Reporter: Ashwin Ramesh
>            Assignee: Ashwin Ramesh
>         Attachments: HADOOP-14599.001.patch, HADOOP-14599-002.patch, HADOOP-14599-003.patch, HADOOP-14599-004.patch
>
>
> RPC average queue time metrics will now update even if the client who made the call timed out while the call was in the call queue.


--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

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