[
https://issues.apache.org/jira/browse/HADOOP-2188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12541750
]
Devaraj Das commented on HADOOP-2188:
-------------------------------------
Owen, could you please clarify this. So, the way things work is that if a
client connection has been left idle too long (controlled via
ipc.client.connection.maxidletime), the client closes the connection. This is
to handle scalability issues to do with connection caching (as discussed in
HADOOP-312). What you are proposing will effectively enable connection caching.
Did I get you right? (BTW, for the task->tasktracker communication, the idle
timeout is set to a very high number since the tasktracker server has to deal
with only a very few number of clients).
> RPC should send a ping rather than use client timeouts
> ------------------------------------------------------
>
> Key: HADOOP-2188
> URL: https://issues.apache.org/jira/browse/HADOOP-2188
> Project: Hadoop
> Issue Type: Improvement
> Components: ipc
> Reporter: Owen O'Malley
> Fix For: 0.16.0
>
>
> 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.
