Parameters¶

Amazon EKS continues to run the control plane processes. You can set four fields. Each has a validated range. Amazon EKS applies the change with a rolling control plane update and records it in AWS CloudTrail.

Resource weights may include cpu, memory, nvidia.com/gpu, aws.amazon.com/neuron, and aws.amazon.com/neuroncore. If the resources list contains only cpu, memory is excluded from scoring. That is not the same as cpu: 100 and memory: 1. Accelerator weights apply only to pods that request those resources. Devices scheduled through Dynamic Resource Allocation are handled by a different plugin.

Settings apply to the whole cluster, not to a namespace. The minimum Kubernetes version is 1.31. Updates merge: omitted fields keep their current values. There is no reset operation; set a field back to its default explicitly. Use DescribeClusterVersions for the defaults and constraints of each Kubernetes version.

Event TTL¶

Kubernetes events such as Scheduled, Pulled, and FailedScheduling are stored in etcd. The default retention is 60 minutes. Reducing eventTtl does not change events that already exist. Expiry is set when the event is created.

If you rely on kubectl get events for investigation, export events elsewhere before shortening retention. Deleted events cannot be restored.

NodePort range¶

The lower bound of 10260 avoids kubelet port 10248 and kube-proxy health port 10256. The upper bound of 32767 stays below the typical Linux ephemeral port range. Narrowing the range does not change ports already assigned to Services. Recreating a Service allocates a port from the current range only.

HPA sync period¶

The minimum interval is 10s. The setting is available only on Provisioned Control Plane. Setting it on Standard fails. A non-default value prevents moving the control plane back to Standard until you restore 15s.

Amazon EKS accepts the API call even if the cluster has more HorizontalPodAutoscaler objects than a 10-second loop can reconcile. In that case some objects skip cycles and scale more slowly than intended.

On Provisioned Control Plane, Amazon EKS also increased HPA sync concurrency (up to 40 times the upstream default). A 10-second period is an additional adjustment, not a substitute for that change.

CLI¶

aws eks describe-cluster-versions --cluster-versions 1.36 \
  --query 'clusterVersions[0].controlPlaneComponentConfig'

aws eks describe-cluster --name "$CLUSTER" \
  --query 'cluster.{
      version: version,
      createdAt: createdAt,
      scheduler: kubeSchedulerConfig,
      controller: kubeControllerManagerConfig,
      apiserver: kubeApiServerConfig
    }'

aws eks update-cluster-config --name "$CLUSTER" \
  --kube-api-server-config '{"eventTtl":"45m"}'

aws eks wait cluster-active --name "$CLUSTER"

Live describe-cluster output, before and after Save, is on Existing clusters.