The Blaze, Spark, and Hive engines require access to the Hive staging directory. You can use the default directory, or you can create a directory on HDFS. For example, if you create a directory, you might run the following command:
hadoop fs –mkdir /staging
If you create a staging directory, update the yarn.app.mapreduce.am.staging-dir property in the mapred-site.xml file.
If you use the default directory or create a directory, you must grant execute permission to the Hadoop impersonation user and the mapping impersonation users.