Case file · NAS & RAID · BDR-2025-0341
Two Drives Down in a Four-Drive Array.
A QNAP's four-drive RAID 5 lost two members, and both told the same story — a mechanical failure in both cases, with a clicking sound
. The owners' plan was pragmatic: fix one of the failed drives so we can rebuild our RAID5
. Reasonable-sounding, and exactly the move to talk them out of.
Same symptoms on your desk?
01273 964902
The decode.
RAID 5 reconstructs one missing member from parity. It cannot reconstruct two, because the arithmetic has nothing left to work with. Worse, the proposed plan — repair one drive, rebuild live — would have forced a full-surface read across two ageing survivors while writing new parity over the very stripes that reconstruction needs to read exactly as the failure left them. The array had to be rebuilt from copies, or not at all.
Equipment on this case.
How a case runs →| Platform | What it did here | Why this tool |
|---|---|---|
| DeepSpar Disk Imager 4 | Head-mapped imaging of both clicking members after donor head work | Head-map-aware imaging with per-head timeout, reset and power-cycle control |
| Atola TaskForce 2 | Imaged the two healthy members in parallel while the mechanical work proceeded | Images many drives at once — the difference between days and a week on an array |
| UFS Explorer RAID Recovery | Derived the array geometry and assembled the volume from the four images | Reads NAS volume managers as they actually are, not as a flat array |
On the bench.
Four images, not two — including the healthy pair
Every member was imaged, not just the failed ones. Reconstruction chooses each block from whichever copy holds it best, and the 'healthy' drives in a degraded array are frequently carrying their own quiet bad sectors. Imaging the good pair in parallel while the mechanical work went on also compressed the timeline considerably.
Donor heads on both clickers, weakest zones deferred
Both failed members received matched donor head stacks and were imaged head-map-first. Neither was uniformly damaged — the usual pattern held, with most of each surface reading cleanly and the difficulty confined to specific zones. Those zones were left until last on both drives.
Rebuild the array in software, from the copies
Stripe size, member order and parity rotation were derived from the on-disk metadata rather than assumed, then the volume assembled virtually across the four images. Where a block was weak on one image it was taken from another, which is precisely the flexibility a live rebuild does not give you.
The outcome.
The reconstruction mounted, the shares verified against the owners' own file list, and the QNAP was repopulated with new drives afterwards — rebuilt from safety rather than from hope.
Related on the index.
More from NAS & RAID.
Recognise your own drive in this story?
Same rule as every case above: power it down, and let the diagnosis be free before any decision has to be.