Notes
Short write-ups from day-to-day work. No newsletter, no tracking.
Why we still write restore tests by hand
A backup you have never restored is a hypothesis. Automation covers the copying, but the restore is where assumptions surface: the missing credential, the database that needs a different version, the firewall rule nobody remembered. We run one restore per client per quarter, on a scratch host, and write down the time it took.
Certificate expiry is a people problem
Every expired certificate we have seen this year had automation behind it. The renewal job failed quietly because a challenge endpoint moved, or because a DNS token expired. The fix is unglamorous: an external expiry check that alerts a human at thirty, fourteen and seven days, plus a monthly glance at the log.
Patching windows for teams without a night shift
Most of our clients run a business that stops at 18:00. Reboots therefore happen at 05:30, with a maintenance page on the load balancer and a rollback image ready. Two hosts at a time, never the whole fleet, and a written all-clear before the office opens.
What we read
Provider status pages, kernel security announcements, and the release notes nobody else reads. When something is worth summarising for a client, it ends up here.