Last modified August 9, 2021

Advanced cluster autoscaler configuration

Your Giant Swarm installation comes with a default configuration for the Cluster Autoscaler addon

You can override these defaults in a ConfigMap named cluster-autoscaler-user-values.

Where is the user values ConfigMap

The following examples assume the cluster you are trying to configure has an id of 123ab

You will find the cluster-autoscaler-user-values ConfigMap on the Control Plane in the 123ab namespace:

$ kubectl -n 123ab get cm cluster-autoscaler-user-values --context=control-plane
NAME                                   DATA      AGE
cluster-autoscaler-user-values         0         11m


Please do not edit any other cluster autoscaler related ConfigMaps.

Only the user values ConfigMap is safe to edit.

On cluster creation the user values ConfigMap is empty (or might not exist yet) and the following defaults will be applied to the final Cluster Autoscaler deployment. To customize any of the configuration options, you just need to add the respective line(s) in the data field of the user ConfigMap.

How to set configuration options using the user values ConfigMap

On the Control Plane, create or edit a ConfigMap named cluster-autoscaler-user-values in the workload cluster namespace:

# On the Control Plane, in the abc12 namespace

apiVersion: v1
kind: ConfigMap
    app: cluster-autoscaler
  name: cluster-autoscaler-user-values
  namespace: abc12
  values: |
      scaleDownUtilizationThreshold: 0.30    

Configuration Reference

The following sections explain some of the configuration options and what their defaults are. They show only the ‘data’ field of the ConfigMap for brevity.

The most recent source of truth for these values can be found in the values.yaml file of the cluster-autoscaler-app

Scale down utilization threshold

The scaleDownUtilizationThreshold defines the proportion between requested resources and capacity, which under the value Cluster Autoscaler will trigger the scaling down action.

Our default value is 65%, which means in order to scale down, one of the nodes has to have less utilization (CPU/memory) than this threshold.

# 9.0.1 and greater
  values: |
      scaleDownUtilizationThreshold: 0.65    

# 9.0.0 and below
  scaleDownUtilizationThreshold: 0.65

Scan Interval

Define what interval is used to review the state for taking a decision to scale up/down. Our default value is 10 seconds.

  values: |
      scanInterval: "100s"    

Skip system pods

By default, the Cluster Autoscaler will never delete nodes which run pods of the kube-system namespace (except daemonset pods). It can be modified by setting following property to "false".

  values: |
      skipNodesWithSystemPods: "false"    

Skip pods with local storage

The Cluster Autoscaler configuration by default deletes nodes with pods using local storage (hostPath or emptyDir). In case you want to disable this action, you need to set the following property to "true".

  values: |
      skipNodesWithLocalStorage: "true"    

Further reading