Idle cluster cleanup: Decommission or consolidation

Before optimizing how your clusters run, eliminate what shouldn't be running at all. Idle clusters and abandoned workloads are pure waste as they consume resources 24/7 while delivering no business value. A cluster is considered idle when it has fewer than 5 application pods running (excluding kube-system, karpenter, monitoring namespaces) and node CPU utilization remains below 10% for 7+ consecutive days.

Identify idle clusters

Find clusters with little to no application workload running, these are candidates for decommissioning or consolidation. For sample commands, see this GitHub repository to identify idle clusters.

CloudWatch: Detect Idle Clusters via Metrics

For organizations managing multiple clusters across regions, CloudWatch metrics provide a centralized, automated way to flag idle clusters without requiring kubectl access to each one. Refer to this GitHub repository for sample commands to identify idle clusters.

Action

If a cluster shows <10% CPU utilization, zero application pods, or no deployments outside kube-system for 7+ days, escalate for decommissioning or consolidate workloads into another cluster. Every idle cluster costs you the full control plane fee plus node compute.

Before deleting an idle cluster, verify these items:

For the complete scripts and manifests, see the 01-idle-cluster-cleanup folder in the code repository.

The cheapest resource is the one you don't run. Eliminating idle clusters delivers immediate savings with zero risk to active applications.