[
https://issues.apache.org/jira/browse/HADOOP-12325?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14707488#comment-14707488
]
Anu Engineer commented on HADOOP-12325:
---------------------------------------
[~xyao] Thanks for the comments and the call graph to WritableRpcEngine. You
are absolutely right in the sense that it is possible to call into
WritableRpcEngine. But AFAIK no service in hadoop does it these days and all
services use protobuf RPC. Hence the logging code is contained in
ProtobufRPCEngine.
I would think that modifying WritableRPCEngine would not be very useful since
no services really is passing WritableRPCEngine as the RPC engine, and hence
logging slow calls is not very useful. Please let me know if you agree with
this analysis or if you really would like to move the code to {{server.java}}
and log slow calls for both ProtobufRpcEngine as well as WritableRpcEngine.
> RPC Metrics : Add the ability track and log slow RPCs
> -----------------------------------------------------
>
> Key: HADOOP-12325
> URL: https://issues.apache.org/jira/browse/HADOOP-12325
> Project: Hadoop Common
> Issue Type: Improvement
> Components: ipc, metrics
> Affects Versions: 2.7.1
> Reporter: Anu Engineer
> Assignee: Anu Engineer
> Attachments: Callers of WritableRpcEngine.call.png,
> HADOOP-12325.001.patch, HADOOP-12325.002.patch, HADOOP-12325.003.patch
>
>
> This JIRA proposes to add a counter called RpcSlowCalls and also a
> configuration setting that allows users to log really slow RPCs. Slow RPCs
> are RPCs that fall at 99th percentile. This is useful to troubleshoot why
> certain services like name node freezes under heavy load.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
