Thursday, January 10, 2019

AKS Cluster Autoscaler is now on Preview

Cluster autoscaler (CA) is designed to manage the resource demand when the resource demand changes. CA does a good job on scaling up and down the cluster when in need as the resource demand increases due to any issues or actual demand as well as when they can be set to idle and save the resources.
  • CA will periodically check the pending pods or empty nodes for increasing the size of the AKS cluster. If possible, it will increase. 
  • This happens so fast, it will keep scanning the pending pods for every 10 seconds. 
  • It will remove the node if it is not needed for more than 10 minutes. 
  • Horizontal Pod Autoscaler (HPA) will update pod replicas and resources when needed. (Need to use HPA with CA)
  • If not enough or excess, CA will respond and act on it.

The Cluster Autoscaler also will show the state of cluster autoscaler on two levels.
  • Cluster-wide
  • Within each node
As it is in preview, currently it only supports single node clusters so will show the same information in both levels.

It provides a comprehensive set of statuses related to the node status.

They are,
  • Ready
  • Unready
  • NotStarted
  • LongNotStarted
  • Registered
  • Unregistered
  • Unhealthy
Read the full article on Microsoft Docs:

No comments:

Post a Comment