Skip to main content

Auto Scale

The Autoscale feature is enabled by default when users create a Kubernetes Cluster with D-FKE. D-FKE's Autoscale feature is built based on Kubernetes' cluster-autoscaler tool. The system will automatically adjust the number of worker nodes in the cluster when the following conditions are met:

  • Pods are in pending state due to lack of resources.
  • Node in the cluster uses less than 50% of resources (CPU & RAM) for a period of time (default is 15 minutes) and Pods on that node can be moved to other nodes

Reference: https://github.com/kubernetes/autoscaler/tree/master/cluster-autoscaler To use the cluster-autoscaler feature, users can use it combine with HPA (HorizontalPodAutoscaler).

  • Scale-up : HPA automatically creates additional Pods when resource usage exceeds a predefined threshold. When new Pods are created, if Worker Nodes have insufficient resources for that Pod request, cluster-autoscaler will automatically create more Worker Nodes to meet the number of newly created Pods.
  • Scale-down : HPA automatically reduces the number of Pods when resource usage is less than a set threshold. At that time, Nodes will reduce resource usage and are automatically reduced by cluster-autoscaler.

Reference: https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale- walkthrough/ Basic Use-cases :

  • To have the number of worker nodes fixed: user can set the value of Min nodes = Max nodes
  • To block Cluster Autoscaler from scaling down fixed node: Cluster Autoscaler would not scale down node that has following annotation: "cluster-autoscaler.kubernetes.io/scale-down-disabled": "true" Use this command to add annotation:
Copy  kubectl annotate node cluster-autoscaler.kubernetes.io/scale-down-disabled=true