Yarn Containers Not Using Ram
The worker nodes on my cluster won’t use more than 11 of the 30 total (24 allocated) for mapreduce jobs running in Yarn, I’m hoping someone could help me figure out why.
I followed the steps listed here: http://docs.hortonworks.com/HDPDocuments/HDP2/HDP-188.8.131.52/bk_installing_manually_book/content/rpm-chap1-11.html, and http://hortonworks.com/blog/how-to-plan-and-configure-yarn-in-hdp-2-0/. to set various memory parameters, but no matter what I try, the nodes on the cluster don’t use more than 11GB of the allocated 26GB.
The yarn resource manager reports that it is using all of the allocated memory in the status across the top, but according to TOP and other such, it is not.
Using ps, I see org.apache.hadoop.mapred.YarnChild processes being created with -Xmx756m, but I can’t find this anywhere in mapreduce or yarn configurations. Does anyone have an idea what might be constraining the usage of Ram?
yarn.nodemanager.resource.memory-mb = 24576
yarn.scheduler.minimum-allocation-mb = 3072
yarn_heapsize=20000 (not really clear to me what this does…?)
mapreduce.map.memory.mb = 4096
mapreduce.reduce.memory.mb = 8192
mapreduce.map.java.opts = -Xmx3500
mapreduce.reduce.java.opts = -Xmx7000