The number of running nodes is a direct function of the number and size of your containers. The cluster autoscaler stops and and start servers based on how many containers are desired, which is in turn based on the Horizontal Pod Autoscaler inputs (e.g. incoming requests).