The Synology that “crashed” in Hamilton.

The unit came in from Hamilton with both drives still seated and a confession: the web interface had been offering helpful-sounding buttons all morning and one of them had nearly been clicked. Good instinct not to — on a NAS, “reset”, “repair” and “create volume” are how recoverable situations become archaeology. The studio needed the client library back and, more urgently, the current fortnight’s unfinished jobs.

“Volume crashed” is a symptom, not a diagnosis. Out of the enclosure and onto the bench, disk 1 imaged cleanly; disk 2 was the culprit — a growing colony of unreadable sectors squarely inside the metadata region of the mirrored btrfs volume. DSM, unable to reconcile its own bookkeeping, had marked the whole volume dead. The data behind the bad bookkeeping was substantially intact.

Both members imaged — disk 2 slowly, weak zones last — then the SHR/mirror logic reassembled virtually from the copies and the filesystem repaired on the image, never on the originals. The quiet bonus: the studio had left Synology’s scheduled snapshots on without ever thinking about them, and the snapshot chain let us roll the handful of metadata-damaged current files back to versions hours old. The “lost fortnight” came back with it.

Full share tree verified and returned on an encrypted external within the week — plus the short sermon every NAS recovery ends with: a NAS is one copy in one box, RAID mirrors hardware failure but not deletion or corruption, and the snapshots that helped here were luck, not strategy. The studio now runs a real second copy. The DiskStation, on new disks, went back to being useful rather than trusted.