Notes

Short write-ups from recent projects. No vendor pitches.

A backup you have not restored is a hope, not a backup

Most teams we audit have backups configured, and most have never restored one end to end. Schedule a quarterly drill: restore the latest snapshot into an isolated environment, start the application against it and run a smoke test. Record how long it took. That number is your real recovery time, not the one in the policy document.

Keep at least one copy outside the provider that runs production. A billing dispute or an account lock should not be able to take both the service and its backups.

Certificate expiry is a monitoring problem

Automatic renewal works until a DNS change, a firewall rule or a moved webroot quietly breaks it. Monitor the expiry date of every public certificate from outside your network and alert three weeks before it lapses. Three weeks leaves time to fix the renewal path without an emergency.

Plan the rollback before the cutover

For each migration step, write down how you undo it and test that path first. Lower DNS TTLs a day in advance, keep the old environment read-only rather than deleted, and compare checksums of data after the copy. A cutover is finished when the rollback plan is no longer needed for a week, not when traffic moves.

Measure regions from where your users are

Provider latency maps are measured from their backbone. Your users sit behind home and mobile networks. Run repeated downloads from a few real locations at different hours and look at the spread, not the best result. Our test node is available for that.