The Simplest Architecture for Kubernetes Cost Monitoring

Cost monitoring shouldn’t require its own platform. The leanest design is a single agent that lives next to your workloads and exports the few signals that matter.

Ingredients

• One Go agent with RBAC to read resources and metrics.
• Your price sheet (on-demand, spot, storage, egress) stored as ConfigMap/Secret.
• Optional CLI/UI to query and export reports; no extra DB.
• Admission webhook (optional) to block unlabeled or over-limit workloads.
• Tiny scratch storage or ConfigMap to cache reports; no PVC required.

Data flow

• Collect usage (requests, limits, p95 usage) from the metrics API.
• Apply price sheet to compute cost per namespace/service.
• Emit metrics and a daily report; optional admission hook for guardrails.
• Export CSV/JSON on demand; integrate with Grafana if teams want dashboards.

Why this wins

• Smaller blast radius and upgrades.
• Lower infra bill—no long-running Prometheus/Grafana just for cost.
• Faster to adopt; developers trust tools that stay out of the way.
• Easier compliance: data stays in-cluster; price sheets are versioned and auditable.

If your cost tool needs its own cluster, it’s solving the wrong problem.***