Jimson, On Mon, Sep 12, 2011 at 3:51 PM, Jimson K. James <[email protected]> wrote: > No, the code already does point to jobtracker using > hConfig.set("mapred.job.tracker", "master:8021"); > It seems weird to me, when running from cygwin, the job status is > correctly reported in jobtracker. > But when tested inside the cluster environment, the job fails to report > to jobtracker even though the job run to completion.
Perhaps you aren't submitting the job with a proper configuration directory on the classpath and that's causing this? With no cluster configuration on classpath, the submitter code loads defaults and runs against that (which uses local job-runner, a single-JVM instance to run your MR jobs over the local filesystem). Try making your job launcher read the proper configuration files? For instance, if I want my 'hadoop' exec to read my configs at a non standard location like /hadoop_home/conf, I'd run it as: $ export HADOOP_CONF_DIR=/hadoop_home/conf $ hadoop jar <myjar.jar …> -- Harsh J
