Task process exit with nonzero status of 134.

[Kiss Tibor <ki...@gmail.com>]

Hi,

I create a cluster on Amazon EC2 m1.large instances and I try to run a
simple distcp from hdfs to hdfs using a relatively small file and I end up
in a jvm crash and the tasktracker reports the "Task process exit with
nonzero status of 134."<https://issues.apache.org/jira/browse/MAPREDUCE-398>
I use a whirr setup which installs the jdk from
http://whirr.s3.amazonaws.com/0.1.0%2B23/sun/java/install
Then I upgraded jdk to the newest version, the jvm still crashes.
I attached the crash logs corresponding to the two different versions of
64bit jvm.

If I create a cluster of type m1.small instances, then everything it works
perfectly.

It seems that there is a problem related to 64bit JVM only.

I tried to set up
export HADOOP_TASKTRACKER_OPTS="-XX:+UseCompressedOops"
in /etc/hadoop/conf.dist/hadoop-env.sh (then restarted the nodes) but no
luck!

Then I was removing the 64bit jdk and installed the latest 32bit
jdk-6u22-linux-i586-rpm.bin which installs jdk-1.6.0_22-fcs.i586. With this
32bit jvm setup on m1.large instance my discp (and all my jobs) is working.

I would like to know if there are somebody who successfully run CDH
distribution of Hadoop (CDH3b3) on m1.large instances with 64bit jvm. I also
would like to know if that cluster has been created by whirr or something
else.

Thanks in advance.

Tibor