Small systems,
carefully run.
Notes from maintaining modest internet services: explicit changes, measured recovery, and documentation that remains useful months later.
Change control
Leave every change reversible
A deployment starts with the previous working artifact and a way back to it. We test configuration parsing first, switch one component, then compare the real service response before expanding the rollout.
Observability
Measure the request, not just the process
A running daemon may still be unable to serve users. Our checks follow the complete path: name resolution, TCP, certificate validation, and a small application request with a known answer.
Recovery
Practice before the incident
Recovery notes include the exact data snapshot, software version, expected row counts, and the person responsible for the final check. An untested archive is only a starting point.