Array work · RAID 5 recovery
RAID 5 keeps running on a missing member, and that is the entire promise. By the time an array reaches us the promise has been spent: a rebuild was tried, and a second disk fell out halfway. The good news: a controller calling a drive failed is not the same as a drive being dead, and everything we lift comes off sector-level clones.
Run the problem past an engineer
0800 6890668
| The problem | What's behind it | First move |
|---|---|---|
| One member has dropped and the array is running degraded | No redundancy left — leaving it running is a bet you can lose | Shut it down before the second goes |
| A rebuild began and stalled before finishing | Reading every sector turned up bad ground on a second, tired member — the usual ending | Leave it alone; no second attempt |
| The controller reports two failed members | Usually one genuinely dead disk plus one carrying a handful of bad sectors, not two write-offs | Send both in — the imaging decides |
| Members forced back online to revive the array | Old blocks are now mixed into current stripes | Stop there |
| New controller fitted, and the array vanished | Metadata disagreement only; the blocks below are intact | Standard rebuild job for us |
| DIY software has been run across the member disks | Writes have landed on your originals | Tell us what was used and send it in |
Every item gets its own case number on the day it lands here. An engineer traces the fault, gives you a plain assessment of what is realistically recoverable, and puts one fixed price in writing. The diagnosis is free, you are under no obligation, and chargeable work starts only when you say go.
Every disk is copied in full to our own storage, including whichever ones the controller condemned. Weak zones are mapped as we go, the worst surfaces are left until the final pass, and nothing is written to the disks you sent.
Two sources give us drive order, stripe width and parity rotation: controller metadata left on each member, plus the shape of the data itself. Nothing is guessed, and no working card is needed.
The volume goes back together across the images. Each block comes from whichever clone read it cleanest, and parity supplies anything no member returned properly — including stretches that only the condemned drives still hold. The file system above it is then repaired and tested.
You get the full list of recovered files first, and nothing is charged until you approve it. Data goes back on fresh media, return postage paid, and the job stays open on our side until you confirm every file opens at your end.
Do the arithmetic once, then stop worrying: the consumer specification allows one unrecoverable read error (URE) per 1014 bits, roughly a single bad sector every 12.5TB; enterprise disks are quoted at 1015. A four-bay set of 16TB drives asks the three survivors for about 48TB in one go, so on paper the exposure is genuine. Day to day, most drives beat their rating comfortably. Both things are true at once, which is why this page keeps saying one thing: image everything, then rebuild as much as you like.
Disk two had been down since June and nobody spotted it. In August disk four fell out mid-rebuild. Imaging showed drive four was not dead at all — nineteen bad sectors, no more. Three sound members and that almost-whole fourth put the stripe back together, and the filings went in to deadline.
Usually, yes. A 'failed' flag from the controller often comes down to a scatter of unreadable sectors, not a drive that has genuinely stopped. Because the reconstruction runs on clones rather than originals, any block one member cannot supply is read off another. Two members properly destroyed is rare; two thrown out of the set is the everyday case.
Not while a single member is still unimaged. Starting a rebuild means a full-surface read of each disk left standing — the exact job that finishes a tired one — and it lays new parity straight over data a reconstruction has to work from. With clones in hand first, that same rebuild is safe to run.
A URE is a sector the drive can no longer read back, even with its own error correction applied. Consumer disks are quoted at roughly one URE per 12.5TB read; enterprise drives at one per 125TB. Treat those as worst-case floors rather than a forecast, and remember that any recent controller drops a single stripe when it meets one, not the whole array.
Yes. The member disks carry controller metadata describing the geometry, and where that has been scrambled the data's own structure reveals it. Rebuilds are done from images on the bench — your original disks are never touched.
Switch a failing device on again and you lose a little more. Open a case before you do — the diagnosis is free, whatever it finds.