Job record · NAS & RAID · MKD-2025-4341
A Four-Bay Array, Two Members Down.
A Bletchley staircase joinery kept its four-bay Synology on a shelf where sawdust filled the fan. A bay had flagged bad sectors since Christmas, the warnings ticked away to stop the emails. A job list would not open, so someone power-cycled it. Back it came: volume crashed
. The note with it — clone the bad disk in and rebuild
— was refused.
Same fault? Start here.
0800 6890668
Without the jargon.
RAID 5 carries one disk's worth of insurance, spread thinly across the whole set. With a single member gone, every stripe still has an answer. With two, the sums run short of knowns. Powering the box up to see whether it would run would have leaned hard on two tired survivors, and invited a crashed volume to lay new metadata over the very structures a reconstruction has to read. So the unit stayed off, and the clone-and-rebuild request was declined in writing. In work like this, imaging is not a precaution taken before the real plan. It is the plan.
The kit this job needed.
See how a job runs here →| Kit | Why it was used | What it gives us |
|---|---|---|
| DeepSpar Disk Imager 4 | New head stacks in both dead members, with the recent casualty read first | Maps the heads, then images each — resets, timeouts and power cycles controlled |
| Atola TaskForce 2 | Imaged the surviving pair side by side while the bench work went on | Copies several array members side by side, so a week of imaging finishes in days |
| UFS Explorer RAID Recovery | Parsed the mdadm superblocks and stood the volume up across the images | Unpicks a NAS layer by layer instead of treating it as one flat array |
What happened on the bench.
Treat the survivors as carefully as the casualties
Disks that carry a degraded array rarely come through unmarked, and both survivors held small crops of pending sectors. Both were put on the TaskForce and copied at the same time as the bench work proceeded on the failed pair. Working in parallel took days out of the schedule and left nothing in the set uncopied or taken on trust.
Two head transplants — and a date hiding in the SMART log
Both casualties accepted a matched donor stack and could read once more. Then the logs changed the picture. The disk flagged since Christmas had in fact dropped out in January. That left its contents nine months stale — the volume as it used to be, not as it finished. The recent failure was the copy that counted, its damage confined to defined bands.
Reassemble it on the bench, and stand the stale member down
Member order, stripe size and the parity rotation were taken from the array's own records rather than assumed. Three current images went into the reconstruction and the January dropout stayed out of it, because folding months-old blocks into a rebuild is how convincing corruption gets manufactured. The handful of sectors the recent casualty withheld sat in space the filesystem was not using.
The outcome.
The volume came up first time. Every drawing, cutting list and ledger was matched against the firm's own job numbers before anything shipped, on fresh disks. The NAS now lives in the office, where the fan can breathe and someone reads the warnings.
Other jobs from the casebook.
See also RAID & NAS.
Does that sound like your device?
One lesson runs through the lot: cut the power, let us look for free, then decide with the facts in hand.