Ask any recovery bench for its most repeated story and you’ll get this one: an array loses a drive, a replacement goes in, and during the rebuild a second ‘healthy’ drive fails — taking the redundancy’s last life with it. It happens so reliably it deserves an explanation, because the explanation changes what you should do in the window between failures.
Same batch, same hours, same wear — why a second drive so often follows the first.
Reliability maths assumes failures are independent events. Arrays violate that assumption from the day they’re built: the drives are typically the same model from the same production batch, installed the same hour, spun for the same tens of thousands of hours, warmed by the same enclosure and worked by the same workload. They aren’t four independent gambles — they’re quadruplets ageing in formation. Drive lifespans follow the ‘bathtub curve’: failures cluster early (defects) and late (wear-out). An array old enough for one wear-out failure has, by definition, siblings standing on the same cliff edge — the first failure isn’t bad luck, it’s the announcement that the batch has reached the falls.
Into that fragile moment arrives the most punishing week of an array’s life. A rebuild reads every sector of every surviving drive, flat out, for hours or days — drives that normally serve scattered light reads suddenly marathon at full tilt, often in a warmer-than-usual box because one bay is busy resilvering. Latent bad sectors that quietly existed for months get found, because a rebuild is the first process in years to read everything. The cruel arithmetic: the procedure that restores redundancy is precisely the stress test most likely to kill its remaining donors. This is why the bench’s rule for degraded arrays with irreplaceable data is heresy to intuition: don’t rebuild — image first. Copies can survive a rebuild; originals sometimes don’t.
The moment an array drops its first drive, you are inside the window, and three moves matter. Back up now — not after the rebuild; the degraded array is still serving data, and tonight may be its last night of doing so. Decide what the data is worth before touching anything: replaceable data can risk a rebuild; irreplaceable data justifies powering down and imaging every member first, rebuild afterwards from safety. And if a rebuild does fail partway — stop absolutely: the 62% case file shows how recoverable that exact disaster is when nobody tries a third time. Arrays forgive one emergency. They keep no second chances for confidence.
Degraded array humming behind you as you read this? That’s the window. The diagnostic is free, the quote is fixed in writing, and honest advice on 0131 202 0491 costs nothing at all.