I started a new instance with a fresh install of hadoop/hive/tez (all the latest). Followed the steps above, and the hive queries still always default to using regular mapreduce.
If I follow this tutorial: http://docs.hortonworks.com/HDPDocuments/HDP2/HDP-18.104.22.168/bk_installing_manually_book/content/rpm-chap-tez-2.html. I get this exception: java.lang.ClassNotFoundException: org.apache.hadoop.yarn.api.records.DelegationToken
Do I need to add yarn jars specifically to the hive/hadoop classpath?