I am running Hadoop 0.19.1. One of my Mapper tasks was failing and the
problem that was reported was:
Task process exit with nonzero status of 1...
Looking through the mailing list archives, I got the impression that this
was only caused by a JVM crash.
After much hair pulling, I figured out that the problem was actually caused
by an exception thrown in my Mapper's close() method. I was catching the
initial exception, printing a message and a stack trace, and rethrowing the
exception as a runtime exception. But my printouts never showed up. Also,
when the task was restarted after the initial failure, the task would always
fail again but in seemingly random places. Sometimes not printing anything
at all in the logs.
I'm not sure if throwing runtime exceptions in the close() method is a
discouraged practice for Hadoop Mappers. In any case, I thought I'd report
my experience in case it helps anyone else.