Operator note / October 7, 2026

The gap between uptime and availability

A process can stay alive for weeks while its useful work is interrupted. Uptime tells us when the process last started; availability asks whether a user could complete a task.

Probe the meaningful boundary

A successful TCP connection proves that something accepted a socket. An HTTP health response proves a little more. Neither guarantees that a dependency can answer or that a queued job will complete within its deadline.

Separate liveness from readiness. Keep lightweight liveness checks independent of remote dependencies, and make readiness reflect whether the service can accept useful work. Otherwise an upstream outage may cause unnecessary restart loops.

Do not hide the observation window

Availability without a stated interval is an ambiguous number. A dashboard should identify its window and explain exclusions such as scheduled maintenance. The percentages on our overview are synthetic examples, not an SLA.