Cluster autoscaler

Why are there N nodes in my cluster now, but it's not getting downsized?

Autoscaling doesn't stop nodes with pods that can't be evicted. Scaling barriers:

• Pods whose eviction is limited with PodDisruptionBudget.
• Pods in the kube-system namespace:
  • That haven't been created under the DaemonSet controller.
  • That don't have PodDisruptionBudget or whose eviction is limited by PodDisruptionBudget.
• Pods that haven't been created under a replication controller (ReplicaSet, Deployment, or StatefulSet).
• Pods with a local storage.
• Pods that can't be evicted anywhere due to restrictions. For example, due to lack of resources or lack of nodes matching the affinity or anti-affinity selectors.
• Pods with an annotation that disables eviction: "cluster-autoscaler.kubernetes.io/safe-to-evict": "false".

kubectl annotate pod <pod name> cluster-autoscaler.kubernetes.io/safe-to-evict=true

Other possible causes:

• The node group has already reached its minimum size.

• The node is idle for less than 10 minutes.

• During the last 10 minutes, the node group has been scaled up.

• During the last 3 minutes, there was an unsuccessful attempt to scale down the node group.

• There was an unsuccessful attempt to stop a certain node. In this case, the next attempt occurs in 5 minutes.

• The node has an annotation that prohibits stopping it on scale-down: "cluster-autoscaler.kubernetes.io/scale-down-disabled": "true". You can add or remove annotation using kubectl.

  Check for annotation on the node:

  kubectl describe node <node name> | grep scale-down-disabled
  

  Command output:

  Annotations:        cluster-autoscaler.kubernetes.io/scale-down-disabled: true
  

  Set the annotation:

  kubectl annotate node <node name> cluster-autoscaler.kubernetes.io/scale-down-disabled=true
  

  Remove the annotation by running the kubectl command with -:

  kubectl annotate node <node name> cluster-autoscaler.kubernetes.io/scale-down-disabled-
  

Why has the pod been deleted, but the node group fails to downsize?

If the node is underloaded, it's removed in 10 minutes.

Why isn't autoscaling performed even when the number of nodes gets less than the minimum or greater than the maximum?

Autoscaling won't violate preset limits, but Managed Service for Kubernetes isn't explicitly controlling the limits. Scaling up will only trigger if there are pods in the unschedulable status.

To get answers to other questions about autoscaling, see the documentation Kubernetes.