Thanks you Steven for your quick feedback. Setting yarn.app.mapreduce.am.command-opts works in a cluster, but it is not working in my local Hadoop setup. I am trying to find out the reason.
On Thu, Feb 9, 2017 at 2:00 PM, Steven Rand <[email protected]> wrote: > Have you tried setting yarn.app.mapreduce.am.command-opts? That should > allow you to set those Java options on only the Application Master and no > other processes. > > On Thu, Feb 9, 2017 at 11:52 AM, Tanvir Rahman <[email protected]> > wrote: > >> Hello everyone, >> I am currently working on a research project where i need to understand >> the yarn mapreduce Application Master code in hadoop 2.7.3. I have >> downloaded hadoop-2.7.3 source code, built it, I have a local Hadoop >> configuration on my machine. I can successfully run wordcount application >> in my hadoop setup using intelliJ IDE. >> >> However, I can not attach Intellij debugger to a running process to debug >> Application Master code. >> I can debug the wordcount application source code but I have trouble >> attaching the the application master (MRAppMaster.java). I tried to follow >> suggestions that I found on the web about setting HADOOP_OPTS and/or >> JAVA_OPTS to something like >> >> export >> JAVA_OPTS="-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=5005" >> >> but I ran into problems, since the same port can only be used once. >> Hadoop starts however multiple jvms, and by setting these options generally >> all jvm try to bind to that port. >> >> Does anybody have experience on how to specifically attach to the >> mapreduce application master ? >> >> Thanks in advance >> Tanvir >> > >
