Use cases

Normal restart without --rolling-restart

Cloudera Manager (CM) stops and restarts every service on the cluster at once and redeploys client configuration. This is faster than a rolling restart but causes cluster downtime: while a service is being restarted, workloads dependent on it will fail. Use this only when you can tolerate a brief outage, for example, on a Cloudera Data Hub that is not currently serving traffic.

Rolling restart with --rolling-restart

CM restarts role instances in small batches so that a quorum of each service stays up for the whole duration. This is the recommended mode for production Cloudera Data Hub and Data Lake, because it minimizes the impact on running workloads.

A rolling restart minimizes downtime, but does not always eliminate it:

• Services that have only one role instance (for example, a single Hive Metastore) must still be stopped and started, and are unavailable for that window.
• Client sessions attached to the specific instances currently being recycled will see connection errors and need to reconnect.

Rolling restart is also significantly slower than a normal restart because role instances are cycled one batch at a time.

Rolling restart and –-only-restart-stale-configuration

This is the combination most often used to clear a "stale configurations" warning. CM only affects stale configuration services, and even those are restarted in batches. It is the least disruptive mode and should be your default when reacting to a stale configuration warning.