Latency is a distribution
A single ping is only one sample. A useful check records median latency, the slower tail, packet loss, and the time of day. Interactive work often feels worse because of jitter even when the average looks acceptable.
Keep the test path honest: use the same client, resolver, destination, and transport that the real workload uses. Changing several variables at once creates a prettier chart and a weaker conclusion.
Keep small services boring
A static site has fewer moving parts than a database-backed application. That matters on a small host: fewer background processes, less memory pressure, easier backups, and a smaller attack surface.
Routine maintenance should be reversible. Save the prior configuration, validate the new one before reload, and keep a second management session open during access-control changes.
Test the path users take
Process health and open ports are useful evidence, but they do not prove that a full request succeeds. A practical check starts from the client side, resolves the public name, completes TLS, fetches a real page, and records the returned status.
A dashboard is a map. A real request is the walk.
These notes are general operational reminders. They are not a service-level commitment.