Data Recovery Case File · NAS & RAID · One Failure Too Many

Built to lose one, lost two: a four-disk array's bad evening, a repair tool's visit, a dormant year — and the photographs reconstructed from what remained

The arithmetic of his enquiry was stated in its second sentence and haunts the whole genre. Four 1TB drives in a Buffalo LinkStation, "in a RAID config that allowed one disk to fail." Two failed the same evening — with different faults: "one is not even spinning." He ran a well-known disk-repair utility against them; "neither could be repaired, the one could not even be read." Then the detail that changes nothing and explains everything: the fault was over a year ago — the unit shelved since, its owner honest about the fog that settles on dormant arrays: "I do not even know what is on this... I am only after photos, as I have been unable to locate some, so assume my backup was on this." A parity array one failure past its budget, a repair tool's fingerprints, a year of dust, and a target defined by absence. All four threads, worked in order.

The decode: why "one failure" arrays lose two, what the repair tool risked, and what the year cost

The double failure first, because it isn't bad luck — it's the design's fine print, met elsewhere in this archive on mirrors and true here in parity form: four drives bought together, installed together, worked identically, age in lockstep — and a single-parity array runs degraded but silent-feeling after its first quiet loss, every read now leaning harder on the survivors until a same-evening second failure converts "protected" into "offline." Two members with different faults on one night is the correlated-ageing signature exactly. The repair utility next, graded honestly: disk-repair tools work by reading and rewriting — exercising a drive's surfaces at length to refresh or relocate marginal sectors — which is defensible maintenance on a healthy disk and precisely the wrong medicine for failed ones: hours of sustained load on a non-spinning mechanical casualty achieves nothing, and the same hours on a degrading member work the failing territory a recovery would rather have met fresh. That it "could not even read" the mechanical member was the mercy; the readable member took the treatment. Filed without scolding — the tool is famous, the instinct understandable — and with the doctrine attached: repair utilities are for drives you're keeping, never for drives you're recovering. The dormant year, lastly, priced fairly: the shelf itself cost nothing — unpowered drives hold their state, as this archive's long-wait cases keep proving — but memory decays faster than magnetism, and his honest "I don't even know what's on it" is the year's real bill: a recovery target defined by a hunch about missing photos. The parity mathematics, meanwhile, held one door open: an array that needs all-but-one member can be rebuilt if the recovery bench can raise either failed member far enough to read — because three readable disks plus parity reconstruct the fourth.

The reconstruction — three and a half witnesses

The work ran member by member. The two healthy survivors imaged routinely on the TaskForce 2 — write-blocked, complete. The degrading member — the repair tool's patient — was imaged under management on the PC-3000, its worked-over territory negotiated in bounded passes: coverage high, imperfect, honest. The non-spinning member went under laminar airflow, its mechanical fault addressed far enough to serve as the fourth witness where the third's gaps demanded — the parity set's beautiful arithmetic meaning every stripe needed only three of its four voices, and between the imperfect member and its mechanically-raised sibling, three voices were always available. The array's structure was identified and reassembled from the members' own metadata; the Linux volume above it parsed; and the unit's forgotten contents resolved the year's fog: a household backup exactly as he'd hoped — the missing family photographs present and central — verified by opening images across the set, delivered organised on new single-drive media, with the report's inventory finally answering "what was on it" in full.

Outcome

The photographs found, the array's contents inventoried, the parity maths cashed at three-of-four — and the case's filings for every owner of a "one failure allowed" box. Treat the first failure as the second's countdown: a degraded array is an unprotected array wearing yesterday's badge — replace the fallen member within days, and consider the survivors suspects, because they shared its whole life. Keep repair utilities away from recovery patients: the tools that maintain healthy disks spend failing ones — pull the drives and stop is the entire correct response to an offline array. And write down what your storage holds: a one-line inventory in a drawer converts "I assume my backup was on this" into a checklist, and makes any future recovery verifiable instead of hopeful. The array was built to survive one loss and met two. The mathematics — and a bench willing to raise half a dead drive — made three enough.

Our case files are drawn from genuine enquiries received by our laboratory over the past ten years, anonymised to protect client confidentiality. Each one describes the diagnostic and recovery procedure our engineers apply to that fault, using the equipment listed.