Rolling upgrade minimizes service disruption and degradation during the upgrade process. You can upgrade cluster components and restart upgraded services without taking the entire cluster down.

The rolling upgrade process follows a specific sequence based on component interactions and dependencies. Upgrade ZooKeeper first, followed by Ranger (if enabled on your cluster), core master processes, core slave processes, and onward through the stack, as shown in the following diagram.

You will usually upgrade each component across all of its deployed nodes before upgrading the next component in the sequence. For example, first you'll upgrade ZooKeeper across all nodes. Next you'll upgrade Ranger across all nodes:

The exception to this rule is the set of core slave processes. DataNodes, NodeManagers, and RegionServers must be updated on each node before proceeding to the next node: