[ 
https://issues.apache.org/jira/browse/ZOOKEEPER-597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12784344#action_12784344
 ] 

Flavio Paiva Junqueira commented on ZOOKEEPER-597:
--------------------------------------------------

It is ok to have cnxn = null in FinalRequestProcessor. For example, if a 
follower is forwarding a request, cnxn will be null for the leader. The 
problem, as Pat points out, seems to be that CommitProcessor is exiting at the 
follower that was supposed to finalize it.

Here is the stack trace from a faulty run:


{noformat}
java.nio.channels.CancelledKeyException
        at sun.nio.ch.SelectionKeyImpl.ensureValid(SelectionKeyImpl.java:55)
        at sun.nio.ch.SelectionKeyImpl.interestOps(SelectionKeyImpl.java:59)
        at 
org.apache.zookeeper.server.NIOServerCnxn.sendBuffer(NIOServerCnxn.java:350)
        at 
org.apache.zookeeper.server.NIOServerCnxn.sendResponse(NIOServerCnxn.java:1065)
        at 
org.apache.zookeeper.server.FinalRequestProcessor.processRequest(FinalRequestProcessor.java:360)
        at 
org.apache.zookeeper.server.quorum.CommitProcessor.run(CommitProcessor.java:73)
{noformat}

> ASyncHammerTest is failing intermittently on hudson trunk
> ---------------------------------------------------------
>
>                 Key: ZOOKEEPER-597
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-597
>             Project: Zookeeper
>          Issue Type: Bug
>          Components: tests
>            Reporter: Patrick Hunt
>            Assignee: Patrick Hunt
>            Priority: Critical
>             Fix For: 3.3.0
>
>         Attachments: ZOOKEEPER-597.patch
>
>
> ASyncHammerTest is failing intermittently on hudson trunk. There is no clear 
> reason why this is happening, but
> it seems from the logs that a session connection to a follower is failing 
> during session establishment - the
> failure seems to be a problem either on the follower or leader. The server 
> gets the session create request, but
> it stalls in the request processor pipeline. (we see it go in, but we do not 
> see it com eout)
> unfortunately all efforts to reproduce this on non-hudson trunk have failed. 
> Even trying to reproduce by
> running on hudson host itself (manually) has failed.
> We need to instrument the client session creation code in the test to dump 
> the thread stack if the
> session creation fails.

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

Reply via email to