[
https://issues.apache.org/jira/browse/MAPREDUCE-2386?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13043947#comment-13043947
]
Todd Lipcon commented on MAPREDUCE-2386:
----------------------------------------
Though on the surface they appear different, I think this is also caused by the
same underlying JVM bug as MAPREDUCE-2389. The JVM bug in question causes a
spurious dup2() call which overwrites an open fd. In this case, it's
overwriting the epoll fd. In the case of MAPREDUCE-2389, it's overwriting an fd
corresponding to a spill record
> TT jetty server stuck in tight loop around epoll_wait
> -----------------------------------------------------
>
> Key: MAPREDUCE-2386
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-2386
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: tasktracker
> Affects Versions: 0.23.0
> Environment: RHEL 6.0 "Santiago"
> Reporter: Todd Lipcon
>
> In some load testing, I got a TaskTracker into a state where its Jetty server
> is in a tight loop calling epoll_wait, which is returning EINVAL:
> [pid 19573] epoll_wait(157, 40829000, 8192, 0) = -1 EINVAL (Invalid argument)
> It's not responding to any HTTP connections - connections are accepted and
> then just hang.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
