Researching some information about tuning YARN memory to fit the cluster size, I found the following settings to work for my configuration:   yarn.nodemanager.resource.memory-mb  =  20GB yarn.scheduler.minimum-allocation-mb =   4GB yarn.scheduler.maximum-allocation-mb =  20GB mapreduce.map.memory.mb              =   4GB mapreduce.reduce.memory.mb           =   8GB mapreduce.map.java.opts              = 3.2GB mapreduce.reduce.java.opts           = 6.4GB yarn.app.mapreduce.am.resource.mb    =   8GB yarn.app.mapreduce.am.command-opts   = 6.4GB   That allowed my particular Hive query to execute on our 10 node cluster with 30GB physical RAM each.     ... View more