If an upgrade step fails, there are two options for reverting to your prior state:
Rolling downgrade restores software to the pre-upgrade version. The downgrade will preserve the user data that is generated during the upgrade process.
The downgrade process starts with the component you most recently worked with during the upgrade process, and proceeds back down the stack. For example, if you upgraded Zookeeper, Ranger, master and slave components, Hive, and Oozie, you would start the downgrade process with Oozie and then Hive. You would progress back through the stack, downgrading components in the reverse order, finishing with Ranger and then Zookeeper.
Rollback is an emergency procedure that restores your cluster services, components, metadata, and user data to a previously checkpointed state before rolling upgrade started. Rollback requires cluster downtime; it requires all services to be shut down before reverting to the previous version and restarting.
The upgrade process includes several preparatory steps for rollback.
Downgrade and rollback operations are only available during the rolling upgrade process, before the upgrade is terminated (either by finalizing the upgrade, by downgrading, or by rolling back).
