Job on file · NAS & RAID · CDF-2025-0642
The Outage Was Free. The Rebuild Was Not.
A press-tool subcontractor on the Waterton estate in Bridgend found its six-bay shelf short-handed after the bank holiday, the power having been out since Saturday. One drive had shown amber since Friday
, and nobody made much of it. By Tuesday a second member had dropped out. The contractor who looks after their machines slotted a spare in and started a rebuild
, and on the Wednesday it stopped dead around sixty per cent
. The outage cost them nothing. The rebuild cost them a fortnight.
Does that match yours? Call us.
0800 6890668
Why that happens.
Two dead disks were survivable. What came next was not. RAID 5 holds a single member in reserve, and that reserve was spent the day the first bay turned amber; once a second disk left, the controller had two unknowns and no way to reach either. Set rebuilding in that condition it puts nothing back, and it does harm, writing fresh parity through stripes that a recovery needs to find as the failure left them. Server shelves add a difficulty of their own. SAS members are often formatted to sector sizes desktop hardware cannot address at all, and vendor structures sit in places consumer software either mangles or reads straight past. Under all of it is an amber light weeks old, with parity holding the set up single-handed.
What we used on this one.
Every step, in order →| The tools | What it did here | Why it helps |
|---|---|---|
| PC-3000 SAS/SCSI | Examined all six server disks on the interface they were designed for | Take a SAS or SCSI disk out of a server and ordinary desktop gear is deaf to it |
| Atola TaskForce 2 | Imaged all six together instead of one after another, which saved days | Images every member of the set at once, instead of working through them one by one |
| UFS Explorer RAID Recovery | Established the array layout and mounted the volume from the six images | Not just the RAID: it also unpicks the volume layers a NAS stacks on top |
How it went.
Image all six, including the ones still healthy
The shelf stayed off. All six disks were withdrawn, each labelled with the bay it had occupied, and put onto imagers — the healthy four first, so the failed pair could be looked at properly before anything was demanded of them. Get the interface or the sector format wrong at this point and every copy made afterwards is quietly ruined.
Pulling sectors off the two that had failed
One had a single failing head; the other picked up new bad sectors while we watched. Neither disk was close to empty. Each was read in short, patient passes, and most of both surfaces reached an image. Two copies mattered: where they disagreed on a sector, the cleaner read decided it, since parity could no longer settle anything.
Let the metadata say how the array was built
Member order, stripe width, which way parity rotated and how far it lagged all came off the array's own structures, not from the way a given controller tends to behave. Those fixed, a virtual volume was built across the six images and the filesystem read out of it, with the disks themselves left powered down.
Signing it off.
The assembled volume mounted, was checked against the directory tree it held, and went home on new media with the return postage ours. One qualification is on the file: where Wednesday's rebuild had already written, a live project folder came back incomplete. The engineer who had been working on it still had a fortnight-old copy on his laptop.
Pages people often read next.
The RAID & NAS jobs.
Is that what yours is doing?
Power it down, get it to us, and hold off on any decision until the diagnosis says what can still be read.