Logged job · NAS & RAID · HUL-2025-0642
The Power Cut Cost Nothing. The Rebuild Did.
A press-tool subcontractor on the Bessingby estate in Bridlington came back after the bank holiday to a six-bay shelf two members short; the power had been off since the Saturday. There'd been an amber light on one of them since Friday
, and nobody had acted on it. A second disk went by the Tuesday. The firm that looks after their computers put a spare in and set a rebuild going
, and by Wednesday it had stuck at about sixty per cent
. Losing the power cost them nothing. The rebuild cost them two weeks.
Sounds like yours? Ring us.
0800 6890668
What causes it.
Two dead disks could have been worked with. The rebuild could not. RAID 5 keeps one member spare, and that spare was used up the morning the first bay went amber; when the second disk left, the controller was short two values and had no way of deriving either. Started in that state, a rebuild restores nothing and damages plenty, laying new parity through stripes a recovery needs to find in the state the failure left behind. Server shelves bring their own problems. SAS disks are frequently formatted to sector sizes ordinary desktop hardware cannot address, and vendor structures live in places consumer tools either corrupt or ignore entirely. Underneath all of it, an amber light nobody had answered, with parity holding the set up on its own.
The tools this one needed.
The steps we follow →| The gear | What this one did | Why we have it |
|---|---|---|
| PC-3000 SAS/SCSI | Ran all six of the server disks on their native SAS interface | Desktop hardware has no way of addressing a SAS or SCSI disk pulled from a server |
| Atola TaskForce 2 | Copied all six at once rather than one at a time, saving several days | Takes every disk in the set together, instead of one imager working down the row |
| UFS Explorer RAID Recovery | Worked out the array geometry and mounted the volume across the six images | Puts the array back together, then unwinds the volume layers a NAS lays over it |
The stages.
Copy every disk, the healthy ones included
The shelf was left powered down. All six disks came out, each marked with its bay number, and went onto imagers — the four sound ones first, so the failed pair could be assessed properly before anything was asked of them. Choose the wrong interface or the wrong sector size here and every copy taken afterwards is quietly worthless.
Getting sectors off the two failed disks
One had a failing head; the other grew new bad sectors while it was read. Neither was anywhere near empty. Both were taken in short passes with long rests, and the bulk of both surfaces came off into images. Two copies of each mattered here: where the reads disagreed on a sector, the cleaner one won, because parity could no longer decide anything.
Take the geometry from the metadata, not habit
The member order, the stripe width, the rotation of parity and its delay were all read out of the array's own structures, never inferred from how a given controller usually behaves. With those settled, a virtual volume was assembled over the six images and the filesystem read from that, the physical disks staying switched off throughout.
How it finished.
The assembled volume mounted, was compared against the directory tree inside it, and went back on new media with the return post paid by us. One caveat sits on the file: in the region Wednesday's rebuild had already overwritten, a live project folder returned incomplete. The engineer working on it still had a copy from two weeks earlier on his laptop.
What people read after this.
Other RAID & NAS cases.
Is yours doing the same?
Turn it off. Send it to us. The diagnosis comes first. It says which files are still readable and which are not.