[
https://issues.apache.org/jira/browse/GIRAPH-601?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13624978#comment-13624978
]
Eli Reisman commented on GIRAPH-601:
------------------------------------
there are several yarn-site resource settings. I had this type of problem when
I didn't know it but asked for too much and the cluster isn't always good at
telling you thats the deal. Two things:
1. each yarn task needs the heap you choose for it, plus "a little extra" for
the container itself. So keep that in mind.
2. you always have to "pay" for your app master too, which is not part of the
Giraph API so if you want "-w 5" you are getting:
one app master with some amount of YARN resources, one master task, and 5
worker tasks (with the master taking a share of heap equal to what each of the
5 workers gets)
point being its extremely easy to overpower a small cluster on local machine
without knowing it ;)
> Exception when running pagerank benchmark with 6 or more workers on a
> pseudodistributed setup: SendVertexRequest cannot be cast to MasterRequest
> ------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: GIRAPH-601
> URL: https://issues.apache.org/jira/browse/GIRAPH-601
> Project: Giraph
> Issue Type: Bug
> Reporter: Eugene Koontz
> Attachments: instrumentation.patch, print_addresses.patch
>
>
> Building Giraph with:
> {code}
> mvn -DskipTests -Phadoop_2.0.3 clean compile
> {code}
> Running pagerank like this:
> {code}
> $HADOOP_RUNTIME/bin/hadoop jar $JAR \
> org.apache.giraph.benchmark.PageRankBenchmark \
> -e 10 -s 10 -v -V 10 -w 6
> {code}
> I see this in
> /tmp/userlogs/application_1364578380737_0003/container_1364578380737_0003_01_000002/
> :
> {code}
> 2013-03-29 10:58:06,371 DEBUG [org.apache.giraph.master.MasterThread]
> org.apache.giraph.master.BspServiceMaster: barrierOnWorkerList: Got finished
> worker list = [Eugenes-MacBook-Pro.local_1, Eugenes-MacBook-Pro.local_3],
> size = 2, worker list = [Worker(hostname=Eugenes-MacBook-Pro.local,
> MRtaskID=2, port=30002), Worker(hostname=Eugenes-MacBook-Pro.local,
> MRtaskID=1, port=30001), Worker(hostname=Eugenes-MacBook-Pro.local,
> MRtaskID=4, port=30004), Worker(hostname=Eugenes-MacBook-Pro.local,
> MRtaskID=3, port=30003), Worker(hostname=Eugenes-MacBook-Pro.local,
> MRtaskID=5, port=30005), Worker(hostname=Eugenes-MacBook-Pro.local,
> MRtaskID=0, port=30010)], size = 6 from
> /_hadoopBsp/job_1364578380737_0003/_vertexInputSplitDoneDir
> 2013-03-29 10:58:06,373 WARN [netty-server-exec-3]
> org.apache.giraph.comm.netty.handler.RequestServerHandler: exceptionCaught:
> Channel failed with remote address /172.16.175.1:56236
> java.lang.ClassCastException:
> org.apache.giraph.comm.requests.SendVertexRequest cannot be cast to
> org.apache.giraph.comm.requests.MasterRequest
> at
> org.apache.giraph.comm.netty.handler.MasterRequestServerHandler.processRequest(MasterRequestServerHandler.java:27)
> at
> org.apache.giraph.comm.netty.handler.RequestServerHandler.messageReceived(RequestServerHandler.java:106)
> at
> org.jboss.netty.channel.Channels.fireMessageReceived(Channels.java:296)
> at
> org.jboss.netty.handler.codec.oneone.OneToOneDecoder.handleUpstream(OneToOneDecoder.java:71)
> at
> org.jboss.netty.handler.execution.ChannelUpstreamEventRunnable.doRun(ChannelUpstreamEventRunnable.java:45)
> at
> org.jboss.netty.handler.execution.ChannelEventRunnable.run(ChannelEventRunnable.java:69)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:895)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:918)
> at java.lang.Thread.run(Thread.java:680)
> {code}
--
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
