The backup job has been green every night for a year, and the first real restore is usually the moment somebody finds out what green actually meant. Until a file comes back out of it, a backup is a hope with a schedule attached.
That sounds harsh until you have watched a restore fail for a reason that had nothing to do with the disk. The bytes were there. Everything around the bytes was not.
Green means the job finished, not that the data is there
A backup job reports on itself. It knows it opened the source, wrote something to the destination, and exited without an error. It does not know whether the file it wrote is usable, whether it caught the folder that was still being edited, or whether the destination quietly ran out of space halfway through and logged a warning nobody reads.
So the green tick is a statement about the process, not about the data. Those are different claims, and only one of them matters at the moment you need it.
What fails is usually the chain, not the file
When a restore does go wrong, it is rarely "the backup is corrupt". More often it is one of these:
- the service account that wrote the backup has no read permission on the repository it wrote to
- the encryption key lives on the machine that died
- the catalog knows about the files, but the catalog itself was never backed up
- the backup is on a drive that is currently disconnected, and nobody wrote down where it went
- the software version that reads this format is three upgrades behind, and the old installer was never kept
Every item on that list is invisible during the nightly run. All of them are fatal at 2am.
A restore you have never done is a promise somebody else will keep
The dangerous assumption is that restoring is a technical operation that will be performed by whoever is on shift. In a small company that person is one person, and that person is often on a plane, or has left, or is the one asking you for help.
If the only way to get data back is a sequence of steps that lives in one head, the backup has a single point of failure and it is human. Writing the steps down is not bureaucracy. It is the difference between a procedure and a hope.
Test the way you would have to do it on the worst day
Restoring a folder to the same machine, as the same user, with the original software still installed, proves almost nothing. The worst day is the one where the machine is gone.
So the useful test is deliberately inconvenient:
- Pick a file you did not choose when you set the backup up.
- Restore it to a different machine — one that never had the backup agent on it.
- Do it with the credentials a colleague would have, not yours.
If any of those steps needs you to log in as an administrator and "just" adjust something, you have found the part of your recovery plan that only works while you are alive and available.
A restore that has never been timed is not a plan
"How long will it take to get back?" is the question the business actually asks. It is also the question nobody can answer from a backup log.
Restore speed depends on where the copy is, how it is compressed, and how much of it you actually need. A full restore of everything is rarely what the business wants — it wants the accounting folder from last Tuesday, which is a completely different operation and often a different tool.
Time it once, write the number down, and then tell people the number. An honest two hours is manageable. A surprise of two days is a crisis.
Trust is built by repetition, not by documentation
One successful restore, done once, tells you the setup worked on that day with those files. It does not tell you it still works after an upgrade, a new disk, or a change of who runs the job.
The habit that actually holds up is small and boring: restore something on a schedule, keep it on the calendar, and write down what happened. Ten minutes a month buys you the right to say "yes" when someone asks whether the backups are good.
Three things worth starting this week:
- Restore one file you didn't pick, to a machine that isn't the original, using an account that isn't yours.
- Time it, and write the number where the business can see it.
- Keep the key, the catalog and the installer somewhere the dead machine cannot take with them.
And if the copy is on a drive that disconnects itself after every run, remember that this buys you distance from an attacker, not immunity from testing. Files sitting on that drive as plain copies are easy to check — plug it in, copy one back, look at it. Cheap test, real answer.
— Mario
I write about backups, self-hosting and keeping systems alive. I also build RansomShield (getransomshield.com), the air-gapped backup tool described above.

Top comments (0)