Devices · RAID sets and servers
The first fault seldom finishes an array. What comes next usually does — a rebuild dragging a tired member through one unbroken read until it quits, an ejected disk pushed back into the set, drives moved between bays on the off-chance. A mirror buys less than people expect: both halves went in the same afternoon and have run the same hours since. Sets come in from Cardiff city-centre offices, from Newport, from the Vale. All of it happens on images, so nothing gets worse once they arrive.
Describe the symptoms to an engineer
0800 6890668
| What shows up | What has gone wrong | Do this now |
|---|---|---|
| Foreign Configuration Found (Dell PERC) | The disks describe a set the card does not recognise | Answer nothing; image the members |
| 1786 — Drive Array Recovery Needed (HP) | Redundancy is spent, and a rebuild is waiting or half done | Shut it down; copy the disks |
| 1784 — Drive Array Drive Failure (HP) | A member of the set has died | Fitting a spare will not fix it |
| 1788 — Improper Drive Replacement (HP) | Somebody has put the disks back in the wrong bays | Bay order decides this; stop now |
| Virtual Drive is Degraded / Offline | The volume has lost its parity, or dropped out of sight | Power the whole server off |
| 1720 — SMART drive detects imminent failure (HP) | A disk still in service is warning that it will fail | Get it imaged that day |
The diagnosis costs you nothing. Every item that arrives is logged under a case number of its own, and an engineer works out what has actually failed — then tells you plainly which files can come back and which cannot. The price follows that: one fixed figure, in writing, and no work begins until you have read it and said go ahead.
Each disk is copied on dedicated imaging hardware, including the ones the card had already written off. Everything after that happens on the copies, and nothing whatever is written back to the disks you sent us.
Stripe size, the sequence of the disks and the way parity rotates are read back out of the layout itself. The set is then built over the images on our own kit. Your card plays no part, and at no stage is anything asked to rebuild.
With the array standing, the file system is repaired, the virtual-machine containers next, the databases after those. Everything is checked against a complete listing before it leaves the lab.
Nothing is billed while you are still deciding. A complete listing of what came off the drive reaches you first, and the invoice only follows your yes. Files travel back on media bought new for the job, return postage ours, and we do not close it here until every one of them opens on the machine you will use.
Why single parity earned its reputation: makers publish consumer SATA disks at one sector they cannot read for every 1014 bits, which is about one in every 12.5TB it hands over. Rebuild a large single-parity set and the members left standing have to give up a good deal more than that without stumbling — which is where rebuilds tend to stop. The figure is the manufacturers' own, not ours, and how hard it bites in practice is still argued over.
The rebuild pulled a second member down on the Friday, and the live data was suddenly out of reach. The array was left switched off from that point. We imaged all four disks onto our own hardware, read the stripe back out of three clean copies and what little the fourth still gave, and the plant ran to time on the Monday.
Yes, send it in with the rest. A drive the controller has ejected often holds the most recent version of some stripes, and every block we use comes from the copy that surrendered it most cleanly.
Not at all. Stripe size, the sequence of the disks and the way parity steps round can all be worked out from the data on them. Routine analysis here, not trial and error.
Both. Once the array stands, the VMDK and VHDX files and any database stores come off first. Each one is then mounted and opened on the bench — a name on a file listing proves nothing by itself.
A set of several disks usually takes four to seven working days. If trading has genuinely stopped, say so when you ring: the case goes ahead of the queue and the intake is arranged around your deadline.
A failing drive has only so many starts left in it. Spend none of them. Keep the power off, and the free diagnosis will tell you what is still readable.