Cluster (Auto Mode)

Skip this page if you already have a suitable cluster. Jump to Argo CD capability or Kueue via GitOps.

Set variables

Reuse the exports from Prerequisites. Work under demo/:

export AWS_PROFILE=sandbox
export CLUSTER_NAME=kueue-lab
export AWS_REGION=ap-southeast-2
export AWS_DEFAULT_REGION="$AWS_REGION"
export IDC_REGION="$AWS_REGION"
export K8S_VERSION=1.36
cd "$LAB_DIR"

Create the cluster config

mkdir -p config

cat > config/cluster.yaml <<EOF
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

metadata:
  name: ${CLUSTER_NAME}
  region: ${AWS_REGION}
  version: "${K8S_VERSION}"

autoModeConfig:
  enabled: true
EOF

cat config/cluster.yaml

Auto Mode enables default system and general-purpose node pools. You do not manage managed node groups for this lab. Nodes appear when workloads need capacity — not necessarily right after create.

Create the cluster

eksctl create cluster -f config/cluster.yaml

Expect roughly 10–20 minutes for a new VPC and control plane. eksctl also installs the metrics-server addon and writes a kubeconfig entry when the cluster is ready.

Configure kubectl

aws eks update-kubeconfig --name "$CLUSTER_NAME" --region "$AWS_REGION"
kubectl get nodes

An empty node list (No resources found) is normal for a fresh Auto Mode cluster. Nodes scale in when Jobs (or other pods) are scheduled.

Note

Kueue requires Kubernetes 1.29+. This lab pins 1.36 by default (K8S_VERSION).

Verify Auto Mode

aws eks describe-cluster --name "$CLUSTER_NAME" --region "$AWS_REGION" \
  --query 'cluster.{version:version,status:status,compute:computeConfig}'

Expect status ACTIVE, version 1.36, and compute.enabled true with node pools such as general-purpose and system.

Next: Argo CD capability (optional) or Kueue via GitOps.