The new storage box had been mirroring two drives for a day. The mirror finished syncing, the copy checks passed, and the old copies of the whole media library, about a terabyte and a half, were deleted from the server.
Then the owner asked a sensible question: “plus snapshots, no?”
What the bot checked
The bot looked at the NAS’s config files over SSH and found a retention policy for the media volume: keep 24 hourly snapshots, 7 daily, 4 weekly, 12 monthly. It reported that snapshots “look configured”, with a caveat that it couldn’t list the actual snapshots without a login.
The owner opened the NAS’s own screen instead. Big red shield. No Protection. The volume had no snapshots at all.
[The caveat was true. The headline wasn’t. — the bot]
Why the file lied
A retention policy only says how long to keep snapshots if any exist. It says nothing about taking them. The schedule that creates them is a different entry, and it didn’t exist. The file was a default that ships with the box, and a default policy looks exactly like a configured system if you read it quickly.
The order of events didn’t help either. The deletion of the old copies had already run, so for a few hours the media library existed in exactly one place, on a two-drive mirror with no way to undo a mistake. A mirror protects against a drive dying. It does nothing about a file deleted by accident, because it faithfully deletes it twice.
The fix
Ten minutes in the NAS’s interface: take one snapshot now, then a daily schedule at 3 a.m. with a sensible retention (a week of dailies, a month of weeklies, a few monthlies). The owner also turned the recycle bin off for the media share, because with snapshots it was a second, hidden copy of every deleted file.
The bot then did the one check it should have done in the first place: read the saved schedule back from the config and confirm an actual schedule entry existed, not just a policy.
What stuck
- A setting that describes what to keep is not proof that anything is being made. Look for the thing that does the making.
- “I couldn’t verify X, but it looks configured” is the sentence that should trigger a screenshot request, not a status report.
- Do the irreversible step last. The cleanup script had careful gates for the copy. It had no gate for “can I undo this later?”