The Kubernetes integration tax: Prometheus, Cilium, and production reality

CNCF blog (2026-05-28) arguing that running production Kubernetes is dominated by an “integration tax” — the hidden labour of making ~20–30 standard CNCF tools (Prometheus, Cilium, cert-manager, Cluster API, GitOps) actually work together — not by the individual tools. A grounding source for platform-engineering and the operational face of observability. (Routed here 2026-06-05; previously parked in the hub _inbox under platform-ops-sre.)

The tax, concretely

• Cilium metrics invisible to Prometheus for want of a ServiceMonitor.
• cert-manager broken by ingress HTTP→HTTPS redirects.
• Generally: each tool works alone; the cost is the seams between them.

Recommended practices

Jsonnet-generated monitoring, NetworkPolicies shipped in Helm charts, two-repo GitOps, Sealed Secrets, and bootstrap-time DR.

Why it routed here

cloud-wiki’s domain is cloud-hosting providers (Hetzner/Oracle/AWS, pricing, regions/bandwidth), not the Kubernetes/CNCF platform-ops layer (orchestration, observability, networking, GitOps); force-fitting would dilute that spoke. This is core platform-engineering. Pairs with google-sre-agentic-ai and netflix-service-topology (production observability). See synthesis.