Uptime tools fail in two ways: they miss real outages, or they page you for nothing. Good monitoring is boring — clear targets, honest probes, and alerts that map to user pain.
What to probe
- A public health endpoint or homepage that proves the edge is alive.
- One authenticated or API path if downtime there is customer-visible.
- TLS expiry and DNS for the apex and www hosts.
- Critical third parties only if you have a real fallback plan.
Define “down” carefully
A single failed check isn’t an outage. Require consecutive failures, sensible timeouts, and expected status codes. Measure latency too — a site that responds in 12 seconds is “up” and still useless.
- Alert on multi-region failure, not one flaky network path.
- Separate warning (slow) from critical (unreachable).
- Include runbook links in alerts: where logs live, who owns the service.
- Mute deploys intentionally; don’t silence forever.
After the page
Every incident should end with a short note: detection time, user impact, root cause, and one prevention step. If the same alert fires monthly without action, either fix the system or delete the alert.
Health Mesh uptime checks give you a continuous pulse next to audits — so availability and quality live in the same operational picture.