RAID 5 · one disk of parity
These land most weeks. A member leaves the set, the alert goes unread, a spare is fitted months afterwards, and the rebuild halts on a second disk. Sets arrive from across Plymouth — from engineering firms working into the dockyard, boatyards along the Cattewater and holiday-let offices out across the South Hams. The position is usually better than the front panel suggests, because failed is a controller's verdict rather than a diagnosis. What goes home to you is read off copies, and the disks you sent are left as they came.
You will be talking to an engineer
0800 6890668
| What shows up | What that points to | What to do |
|---|---|---|
| The controller says a member is missing | The parity is used up; a second fault ends the volume | Power it down before a second fails |
| A rebuild ran half the night and halted | The full read found a soft patch on another disk of the same vintage. That is how these usually end | Leave it powered off. No second attempt |
| Two members read as failed | Nearly always one finished disk beside one holding a few sectors it will not give up, rather than two alike | Send them all; the images will separate them |
| A rejected member was forced back online | Stale blocks now sit among the current stripes | Touch nothing more |
| A new controller cannot find the set | The argument is over metadata; your files have not moved an inch | An ordinary reassembly |
| Recovery software has already been pointed at the members | Something has written to your original disks | Send them regardless, and say which tool ran |
Whatever arrives is booked in under a case number on the day it reaches the bench. An engineer then works out the actual fault, and that part is free. You get a plain answer on what can come off the drive and what cannot, followed by one price in writing. Nothing further happens until you have read it and agreed.
All members go onto our own storage in full, including the ones the card wrote off. Slow reads are logged as they occur and the worst regions are saved for the end. Not a byte returns to the disks you posted.
Three things must be settled: which bay each member sat in, how large a stripe is, and the way parity walks round the set. They come from what each disk wrote about itself and from the pattern of your own data. None of it is estimated.
The array goes together above the images. Every block is lifted out of whatever copy gave the cleanest read, and where none of them gave one, parity supplies it — that is how files which lived on a written-off member still come home. Then the file system is repaired, and the volume mounted and checked.
Thinking it over costs you nothing. Everything the drive gave up is listed for you before any invoice exists, and the bill only follows your go-ahead. Files travel back on media bought in for your job, with return carriage paid at this end, and the case stays on the bench until you confirm they open on your own computer.
Do the arithmetic once and put it down: makers rate a consumer disk at 1014 bits between unrecoverable read errors — call it one awkward sector per 12.5TB — and an enterprise disk at 1015. Four 16TB drives in a chassis means a rebuild wants some 48TB of faultless reading out of the three survivors, so the exposure is real and not academic. Against that, disks routinely outlive their rating. Both are true, and both argue the same way: copies first, and the rebuild afterwards costs you nothing.
Two disks looked to have died at the same moment. The truth was slower than that: disk two had left the set in June and gone unremarked, and disk four then stopped part way through the rebuild in August. Nineteen bad sectors were all that imaging found wrong with the fourth, so three sound members and one nearly sound were enough between them to put the stripe back. The year-end filing went in on its due date.
Usually not. Controllers drop disks for their own reasons, and a handful of sectors answering slowly is reason enough. A rejected disk and a finished disk are different animals. With every member copied onto our own storage, whatever one image will not surrender comes back out of the parity the others are holding. Two disks ejected by a card is a normal week here. Two disks genuinely beyond reading is rare.
Not until copies exist. A rebuild orders every surviving member to read itself right through, which is the exact demand a tired disk fails. It also lays new parity over the parity that would otherwise have put your volume back. Once the images are taken, rebuild the set as many times over as suits you, because the copies stay as they were.
A URE is a sector the drive's own error correction could not repair, so the drive gives up on it. Makers publish roughly one per 12.5TB read on a consumer disk and one per 125TB on an enterprise one. Take those as the floor a manufacturer will stand behind rather than a clock running down. Modern firmware meets a URE by marking that stripe and pressing on.
No, the members are sufficient. Every disk in a set carries the description of that set, and where the description has been damaged your own data reveals the same thing. Assembly happens in software above the images. What you post is read and nothing more.
The diagnosis is free, and it comes back as a list: which files read, which do not, and what getting them off would take. Until then, leave the drive unplugged.