Edge Observability with Prometheus

The problem

Edge traffic sits outside the usual Kubernetes metrics path. Teams were flying blind on Cloudflare behavior — or worse, getting alerts that didn’t correlate with actual user impact.

What I built

I developed a Cloudflare Prometheus exporter to pull meaningful signals into the same observability stack we already used for internal services. That made edge behavior visible alongside application and cluster metrics instead of living in a separate silo.

Impact

False-positive alert volume dropped by roughly 40%, and on-call engineers could trust that a page was worth investigating. Combined with Grafana dashboards for storage and capacity metrics, forecasting and troubleshooting got materially easier.

What I took away

Exporters are only useful when they measure the right things at the right cardinality. This project reinforced how much alert quality matters — noisy monitoring is almost as bad as no monitoring — and how much value there is in meeting teams where their infrastructure actually lives, not just where it’s convenient to scrape.

The public Observability Toolkit repo explores the same themes in a portable, chaos-validated stack you can run locally.