[
https://issues.apache.org/jira/browse/MAPREDUCE-4730?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13479257#comment-13479257
]
Jason Lowe commented on MAPREDUCE-4730:
---------------------------------------
A little more digging and I'm a bit more confident that this is a flow control
problem in the IPC layer. I think the scenario goes like this:
# 1000's of reducers start asking for map completion events about the same time
# IPC Server.Handler thread fields a call off the queue, makes the call and
gets 900K of data
# Handler thread queues up the response data to the connection, likely sees its
the only thing in the queue, and tries to push out the data
# It's too big to send it all without blocking so it pushes the remainder back
onto the response queue for the Responder thread to deal with and moves on to
another call from the call queue
# Lots of reducers are queueing up in the call queue to get their 900K of data,
and the handler threads are processing them and pushing that data on the
response queues as fast as they can
# Responder thread and/or socket I/O can't keep pace with the rate at which
handlers are generating 900K responses and we eventually exhaust memory
> AM crashes due to OOM while serving up map task completion events
> -----------------------------------------------------------------
>
> Key: MAPREDUCE-4730
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-4730
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: applicationmaster, mrv2
> Affects Versions: 0.23.3
> Reporter: Jason Lowe
> Priority: Blocker
>
> We're seeing a repeatable OOM crash in the AM for a task with around 30000
> maps and 3000 reducers. Details to follow.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
