Dogfooding the stack
Everything on this page is the real infrastructure behind cohns.net, observed in real time. Synthetic probes hit the site and API from five continents; a CloudWatch data source reads the ALB, ECS, and Aurora metrics directly. Every dashboard, probe, alarm, and the read-only IAM role Grafana assumes are Terraform in this repo — no dashboard was built by clicking.
- Platform Grafana Cloud (free tier) — $0/mo
- Synthetics Global HTTP probes, per-endpoint SLOs
- Metrics CloudWatch via keyless role-assumption
- Alerting CloudWatch alarms → SNS; per-account budget
Cost as a first-class metric
The comments API runs on Aurora Serverless v2 with a floor of zero. Under no load the database scales to nothing and the compute bill goes with it. The board below reads capacity in ACU and estimated $/hour straight from CloudWatch — through a keyless assumed role, no access keys — and sits at $0 while idle. Cost control isn't a spreadsheet exercise here; it's a metric with a current value.
How it's wired
Blackbox synthetics need no account access, so one set of probes covers the public sites and the API alike. The internals — request rate, 5xx, p95 latency, Aurora capacity — come from CloudWatch, read by a role Grafana Cloud assumes with an external id and read-only permissions. Paging stays on AWS-native alarms into SNS, so the alert path doesn't depend on the dashboard vendor being up. The write-up lives in docs/observability.md.