Case record · NAS & RAID · MKD-2025-4642
One RAID 5 Array, Two Drives Dead.
A Northampton engineering consultancy, a six-bay shelf, and a bank holiday power cut. It came back degraded — one bay was flagged before the weekend
— and a second disk dropped out on Tuesday. Their IT contractor put a disk in from the spares drawer and pressed rebuild
; by Wednesday it had stopped at sixty per cent
. That was the expensive part.
Same fault? Start here.
0800 6890668
Without the jargon.
RAID 5 does one piece of arithmetic and there is no give in it: parity covers a single missing disk. Take a second away and the controller holds equations with more unknowns than knowns. Telling it to rebuild anyway is worse than doing nothing, because it starts writing new parity across stripes a recovery needs exactly as the failure left them. The second trap was the hardware. Enterprise shelves format their disks in ways desktop gear does not expect, and vendor metadata sits where ordinary tools either mangle it or refuse to look. Then the quiet part: that first flag had been showing for weeks, with parity quietly doing two jobs the whole time.
The kit this job needed.
See how a job runs here →| Kit | Why it was used | What it gives us |
|---|---|---|
| PC-3000 SAS/SCSI | Examined and read all six server-class disks over the interface they expect | Server-room SAS and SCSI disks do not answer to the ports on an ordinary bench |
| Atola TaskForce 2 | Copied members side by side rather than one after another, saving days | Copies several array members side by side, so a week of imaging finishes in days |
| UFS Explorer RAID Recovery | Worked out the layout and brought the volume up from the six images | Unpicks a NAS layer by layer instead of treating it as one flat array |
What happened on the bench.
Six members, six images — survivors as well
The shelf was not asked to do anything again. All six disks came out, were marked with their bay numbers, and went to the imagers — the healthy ones first, the casualties after they had been examined. Getting the interface and the sector formatting right at this point is the whole foundation; a mistake here quietly ruins everything built on top of it.
Persuade the two failures to read what they still hold
One casualty had a head reading weakly; the other was growing bad sectors by the hour. Neither was anything close to blank. Patient handling brought the great majority of both surfaces into images — and with both on file, a contested block could be settled by whichever image read it cleanest, rather than by parity alone.
Take the layout from the array, not from habit
The array itself supplied member order, stripe width, rotation and delay. Nothing was assumed and nothing guessed. The set was then assembled in software from the six images, and the filesystem walked out of that virtual volume while the original disks sat still.
The outcome.
The volume mounted, verified, and went back on the firm's spare disks — demoted to delivery duty, the only job left in them. The asterisk is owed: where Wednesday's rebuild had written over stripes, part of one live project folder could not be reassembled. A fortnight-old copy existed on an engineer's laptop.
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.