Start with a useful health check

A useful check describes what someone can actually do. A running process is one signal; a successful request through the complete path is another. Keep both, and record enough timing information to distinguish a slow response from a failed connection.

A backup is a recovery tool

Keep configuration and data backups separate from the service they protect. Record when each copy was made, protect access to it, and verify that the restoration procedure can read it before it is needed.

Make changes easy to understand

Change one behavior at a time. Preserve a known baseline, compare the outcome with the original goal, and keep a clear way back. Small changes are easier to review and easier to explain later.