Skip to content

Cost & efficiency

Cost Explorer

The Cost tab is a multi-cloud cost explorer. It computes Kubernetes cost client-side from your metrics datasource, with a 4-tier pricing engine that uses the best source available — and shows which tier priced each node, so the numbers stay honest:

TierSourceWhen
0 — OpenCostnode_total_hourly_costOpenCost installed (billing-accurate: spot, PV, LB, egress)
1 — Catalogbundled on-demand prices for AWS / GCP / Azure / DigitalOceannode instance-type label is exposed
2 — Cloud estimateper-cloud flat $/vCPU + $/GiBcloud known (from provider_id) but instance type isn't
3 — Flat estimateglobal flat rateon-prem / unknown

It surfaces estimated monthly + hourly cost, allocated vs idle (unrequested capacity), and breakdowns by namespace (top spenders + CPU/mem efficiency), by node (cloud, type, tier, $/hr·$/mo) and by cloud.

Cost Explorer

TIP

For catalog (Tier 1) pricing, expose the instance-type/region labels on kube-state-metrics: --metric-labels-allowlist=nodes=[node.kubernetes.io/instance-type,topology.kubernetes.io/region,topology.kubernetes.io/zone]. For billing-accurate cost, install OpenCost (Tier 0).

Cost & Efficiency dashboard

The bundled Cost & Efficiency dashboard gives you FinOps numbers in pure PromQL (portable) — it works with kube-state-metrics alone, and gets more precise with OpenCost.

Two modes

  • Estimated (default): cluster- and namespace-level monthly cost computed from resource requests × a configurable price list ($/vCPU-hour, $/GiB-hour, set per cluster). Works everywhere.
  • OpenCost: toggle OpenCost present in the cluster config to light up the real node-cost panel (node_total_hourly_cost). Empty if the exporter isn't installed.

Right-sizing

  • Efficiency gauges — usage as a share of requests (CPU & memory).
  • Waste — requested minus used, by namespace.
  • Over-provisioned — containers using < 50% of their CPU request.
  • No request set — containers consuming CPU with no request (scheduling risk).
  • Limit : request ratio — high ratios flag burst/eviction risk.

Use these to tighten requests/limits and cut spend without guesswork.

Apache-2.0 licensed.