Suspending and Resuming Instances ​

To manage cost for cloud-based hardware resources, it is a best practice to suspend a data warehouse instance when it is not in use for some period of time. Suspending an instance implies the suspension of all of its compute clusters, regardless of the suspension and timeout policy set for those clusters and their current state. How long you want to allow an unused instance to remain up and running is your choice, as long as you monitor the cost. In addition to considering the cost, keep in mind that it takes a little while to resume an instance once it has been suspended.

On the Yellowbrick Manager Dashboard, you can easily see how long each instance has been up and running:

You can click the pause button next to the run time to suspend an instance. The run time is not cumulative; it restarts from 0 when you suspend and resume an instance.

Important: When you create an instance, the default behavior is that an idle instance is not automatically suspended. After creating an instance, you can modify this behavior by selecting Actions > Change for the instance. On the Change Instance screen, select Automatically suspend on idle, then enter an Idle Timeout value and an Idle Grace Period value.

The Idle Timeout value is subject to an additional grace period. For example, if you set Idle Timeout to 30 minutes and Idle Grace Period to 1 hour, the instance will not automatically suspend until an idle time of 90 minutes has passed. The idle timeout is not enforced until the grace period has elapsed.

An instance is considered idle when no user-submitted operations (create and alter cluster operations, backend queries, bulk loads, backups, and so on) are running on clusters that belong to the instance. Front-end database queries and system work that runs in the background do not apply to the idle timeout thresholds.

A cluster that is running but idle does not prevent an instance from auto-suspending. The instance suspends its clusters when it auto-suspends.

Note that you cannot change the auto-suspend behavior of an instance if the instance is currently suspended.