[
https://issues.apache.org/jira/browse/HADOOP-2188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12580455#action_12580455
]
Hairong Kuang commented on HADOOP-2188:
---------------------------------------
Raghu, thanks for the review comments.
> Why does the patch remove purging the connections on server?
The purpose of this jira is not to throw away any calls. Otherwise the
applications will wait for the result for ever. I do not think we should
concern about the number of pending responses so much because hadoop-2910
throttle the client write rate and in general a client's read rate should not
be slower than the write rate. If this is really a concern, what we can do is
to limit the total number of calls in the server including unserved ones, being
served ones, and pending responding ones. HADOOP-2910 only limits the unserved
calls.
> RPC should send a ping rather than use client timeouts
> ------------------------------------------------------
>
> Key: HADOOP-2188
> URL: https://issues.apache.org/jira/browse/HADOOP-2188
> Project: Hadoop Core
> Issue Type: Improvement
> Components: dfs, ipc
> Reporter: Owen O'Malley
> Assignee: Hairong Kuang
> Attachments: ipc-timeout.patch, ipc-timeout1.patch,
> ipc-timeout2.patch, ipc-timeout3.patch, rpc-to.patch
>
>
> Current RPC (really IPC) relies on client side timeouts to find "dead"
> sockets. I propose that we have a thread that once a minute (if the
> connection has been idle) writes a "ping" message to the socket. The client
> can detect a dead socket by the resulting error on the write, so no client
> side timeout is required. Also note that the ipc server does not need to
> respond to the ping, just discard it.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
