If the validation process fails during any step in the upgrade process and the issue cannot be resolved, downgrade the stack.
Important: Start the downgrade process with the component you most recently upgraded or attempted to upgrade. Continue through the stack in reverse order compared with the upgrade process:
Here are the downgrade starting points based on your most recently upgraded component:
Clients on cluster components (Flume, Sqoop, Pig, Tez, Phoenix, Mahout)
Components outside of the cluster (Hue, Kafka, Storm)
Non-core cluster components (Falcon, Oozie, Hive)
Core slave nodes (DataNodes, NodeManager, RegionServer)
Core master nodes (HDFS, YARN, HBase)