Monitoring should answer a simple question: can people still use the service?

Measure the external path

CPU and memory are useful, but they do not replace an external availability check. A process can be running while DNS is wrong, a certificate has expired, or a network provider is filtering a port.

Use alerts sparingly

An alert needs a clear reason and a realistic response. Node unavailable, certificate renewal failed, and disk space nearly exhausted are useful examples. Frequent notifications without action quickly become background noise.

Review after an incident

A short note about the cause, the signal that revealed it, and the missing check is enough. The goal is to catch the same problem earlier next time, not to build a complicated reporting system.