i have simple 2 node cluster. one node with 2 GB and second with 1 GB RAM
on node with 1 GB is running application master - no memory left, nothing runs on node with 2 GB RAM
current application do not runs, it has only AM started and no resources for running mapper. If you take a look at screenshot from scheduler, interesting is that memory total is cluster metrics is just 1 GB.
http://postimage.org/image/t3pqmz6af/ its leak in cluster accounting?
