Clean up old snapshots on HDFS to reclaim as much space before the upgrade as you can.

Run Balancer to ensure that data distribution is done.

Confirm that the current state of NN does not indicate that there is a previous upgrade still in progress.

Confirm that all the services indicate a healthy state.

Confirm that there are no open Ambari alerts.

Take note of any configuration differences between nodes.

Confirm that the schemas for Apache Oozie, Ambari, Hive Metastore, Ranger and Hue have been backed up.

If you intend to perform a multi-step upgrade process, or if you know that you will reach points during the upgrade that require significant effort and time, be sure to have additional backups of the various RDBMS metastores.

Confirm that the entire cluster is either quiesced or brought to a steady state.