hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sanjay Radia (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HADOOP-2886) Track individual RPC metrics.
Date Thu, 13 Mar 2008 17:00:24 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-2886?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12578388#action_12578388
] 

Sanjay Radia commented on HADOOP-2886:
--------------------------------------

Looks good.
2 comments.
1. Hadoop brace style is std java brace style
    if ( ) {    <- brace is here, not on next line
        xxxx
    }
   synchronized ( xxx) {

   }



2. Since there are some static metrics such as RpcQueueTime  there is   a potential of 
   conflicts with rpc methods of the same name.
  Hence I suggest that you append the method metrics by "OP_" or something like that.
  (Which has the advantage that it obvious that the metrics is an rpc operation/method)

Alternatively you could prefix the static ones by a "_" or something that would not
conflic with method names (since java does not allow methods to start with that character).
This would be cheaper in cost but would break compatibility with the metrics xml file.

> Track individual RPC metrics.
> -----------------------------
>
>                 Key: HADOOP-2886
>                 URL: https://issues.apache.org/jira/browse/HADOOP-2886
>             Project: Hadoop Core
>          Issue Type: Improvement
>          Components: metrics
>            Reporter: girish vaitheeswaran
>         Attachments: rpc-metrics.patch, rpc-metrics.patch, rpc-metrics.path, rpcmetrics.patch
>
>
> There is currently no mechanism to track performance metrics at the granularity of a
specific RPC. So For e.g. if we wanted to capture average latency for the openFile RPC or
for the createFile RPC the current infrastructure does not support that.
> The implementation involves having a simple HashMap where every new Rpc metric being
added would be inserted into the HashMap. Since there is a mechanism to obtain RPC latencies
already (without the name of the specific RPC), the identification of what RPC is involved
would be done by doing a lookup on the HashMap.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message