A Buffalo LinkStation had been the office file server for years without complaint. It had even survived a disk swap two years earlier — replaced, rebuilt, back to work. Then one afternoon it simply stopped appearing on the reception PC: no shares, no folders, nothing but a users tab. The owner did what had worked last time and told it to rebuild. It failed at 2% — and that is where the real damage was done.
The unit came in as a four-bay RAID 5 array of 750 GB disks, and diagnosis found the situation was worse than the owner knew: two of the four had already failed SMART. That matters enormously, because RAID 5 survives the loss of exactly one disk. The array had been running degraded — quietly, for a long time, with nobody watching — and when the second disk let go, the volume dropped off the network.
Then came the rebuild. A rebuild reads every remaining disk end to end and writes reconstructed parity as it goes. Run that on an array already below its redundancy and you are not repairing anything: you are overwriting the very parity that made recovery possible, while hammering two disks that are already dying. It stopped at 2%, leaving the parity half-old, half-new and inconsistent throughout.
Nothing was done on the NAS itself. All four disks were pulled, write-blocked and imaged individually on the DeepSpar imager — including the two SMART had condemned, because a failing disk very often still holds the parity that makes the others readable. Weak sectors were read gently and in the right order, the imager working around bad regions rather than stalling on them.
The array was then reassembled virtually, from those images: disk order, stripe size, parity rotation and offset all worked out by hand from the data itself, rather than trusted to the controller that had just made everything worse. Because the failed rebuild had corrupted parity across part of the set, the affected stripes had to be recoded — parity recalculated from what survived, to reconstruct the blocks the rebuild had overwritten. Slow, careful work, and only possible because it happens on copies. Get it wrong on an image and you try again; get it wrong on the disks and there is nothing left to try.
The volume mounted from the reconstructed set and the office’s shares came back — the file server exactly as it had been before anyone pressed rebuild. Everything was returned on a fresh external drive, and the original disks went back precisely as they had arrived.
This is the most expensive mistake in RAID recovery, and it is made with the best of intentions every single week. If a NAS or server has fallen over: power it down. Do not rebuild, do not resync, do not swap disks in and hope. Label each disk with its bay position, send every one of them — including any the box has flagged as failed — and let them be imaged before anything is written back.
The free diagnostic costs nothing and tells you what is genuinely recoverable. Post the disks insured, or drop them at our Leeds address on Albion Street.