Specialist · RAID 5 array recovery
RAID 5 tolerates the loss of one member and no more, and most arrays that get here have already spent that margin and had a rebuild thrown at them. What the panic hides: two disks marked down seldom means two disks destroyed, and whatever survives can be lifted from clones without one further write.
Speak to an engineer about it
0800 6890668
| The problem | What that tells you | First step |
|---|---|---|
| A member has dropped; the array is degraded | Redundancy is spent — from here on, every hour running is unprotected | Switch off before the second |
| A rebuild began and then stalled | The whole-surface read hit bad sectors on a marginal survivor — the usual ending | Halt there; no second attempt |
| Two members marked as failed | Usually a genuinely dead disk alongside one carrying a few unreadable sectors — not the same as losing both outright | Send both in; imaging settles it |
| Members forced back into the set | Out-of-date disks dropped into current stripes | Go no further |
| Array disappeared once the controller was changed | The metadata disagrees; the data below is fine | A job for reconstruction |
| 'RAID recovery' software was pointed at the live members | Writes landed on the evidence | Tell us plainly, then send it |
As soon as it lands with us, your device gets its own case number. An engineer works out the fault, says what can genuinely be pulled off, then puts one fixed price in writing — no charge for diagnosis, no obligation, and no chargeable work until you approve it.
Every disk — the two the controller condemned included — is cloned sector by sector, unreadable areas charted, fragile regions left till last. After that, nothing we do can worsen things.
Member order, stripe width and parity rotation all fall out of the data and the metadata the controller wrote to the disks — no live controller needed, nothing assumed.
We rebuild the set across the clones, each block taken from the member that reads it cleanest and recalculated from parity where none can — areas only the condemned disks still hold included. The file system above is then repaired and verified.
You sign off a complete list of the recovered files first; only then does the recovery fee fall due. Everything returns on fresh media, postage paid, and the job stays open until you confirm the files open at your end.
The figure that actually matters: a consumer disk is rated for one unrecoverable read error every 1014 bits — call it 12.5TB — and an enterprise disk 1015. Rebuilding a four-drive 16TB RAID 5 pulls around 48TB off the survivors, so on paper the exposure is genuine; in practice most disks run well beyond that floor clean. Neither claim is wrong — which is why the honest answer is not 'never rebuild' but 'never rebuild until you have images'.
The second disk failed in June and nobody spotted it until the fourth dropped out in August, mid-rebuild. Imaging revealed the truth: drive four held nineteen bad sectors, not a dead disk. The stripe came back from three members and a near-whole fourth, and the practice still filed everything on time.
Usually, yes. What a controller calls 'failed' is often just a handful of bad sectors rather than a dead disk — and rebuilding from images lets every block come from whichever member reads it cleanest. Two truly destroyed drives is rare; two dropped from the array is the norm.
Not until the members have been imaged. A rebuild touches every sector on every surviving disk — precisely the strain that kills marginal drives — and lays down parity as it runs, wiping what reconstruction would have leaned on. Image first and that same rebuild carries no risk at all.
An unrecoverable read error: a sector the drive's own correction cannot save, quoted at roughly one in every 12.5TB read on consumer disks, one in 125TB on enterprise. Take it seriously, but don't be frightened by it. The figure is a worst-case floor, and current controllers give up a stripe to a URE, not the entire array.
Yes. The array geometry sits on the disks themselves, held in controller metadata; if that is damaged, the patterns in the data reveal it. We rebuild arrays at the bench from images and nothing else.
Powering a damaged device up again costs you data. Start a case first; diagnosis is free whatever you decide.