HDR / The work we take on / RAID sets and servers

Servers · failed RAID sets

Hull server recovery. The box will not boot, but the disks inside are usually the healthy part.

The disks are usually the healthy part. What has gone is everything around them: a controller whose record of the set no longer matches the drives, a prompt where both answers do damage, a cache still holding writes the mains cut short. Every member is imaged first and the work happens on those images. Your card can stay in the server. What goes back is the whole volume — file shares, databases, virtual machines. An accountancy practice in the Old Town and a freight forwarder off Hedon Road send the same box, both of them in Hull.

Most jobs: nothing back, nothing to pay Free to look at, then one fixed price Parcels come in from Beverley, Grimsby and Skegness

An engineer will take the details
0800 6890668

Putting a name to the symptom.

Nothing similar? Go to the triage →

What you see	What the fault is	What to try
PERC card: Foreign Configuration Detected	The layout on the drives and the layout in the card's memory disagree. It stops rather than guess. Nothing has been lost	Give the screen no answer
1786 — Drive Array Recovery Needed (HP)	A rebuild started and may have stalled part-way	Shut it down. Image the drives first
Amber drive lights after a power cut	Writes stopped half-finished. What the cache was holding is unknown. The layout on the members is untouched	Do not power it up again
The server starts, but the data volume is gone	The fault is in the card's configuration or in the file system. What sits under both is still there	Usually comes back
The RAID card is dead	The member order, the stripe size and the parity rotation are all on the disks	No replacement card needed
The files copied, but no VM or database will start	A copy is not the same as a consistent copy, and only consistent data runs	Now it is a database recovery job

Sending it to us: pack it well, insure it for the full value and send it tracked to the Leeds intake lab. Return postage is ours. If you want an engineer to look over the packing before you tape the box shut, ring first. The detail is in the sending page.

The server disks already know the layout.

PERC & MegaRAID cardsDDF is a published format, and both of these cards write it into the last sectors of every member. Read it out of the images and the whole set goes back together with the controller left in the box.

HP Smart Array cardsHP does it differently. RIS structures sit at the front of every member. The parity block is offset by one stripe from the usual place. Aim a default preset at that and you get a volume that never existed on your server.

The prompt where both answers hurtThis is the one to leave alone. Import can lay out-of-date metadata over stripes that were completely sound. Clear takes the layout off every disk and nothing brings it back. Image the members first and the prompt stops being dangerous.

Writes left in the cacheA write counts as done as soon as battery-backed memory has it, which is well before it reaches a platter. Lose power mid-flush and the damage lands inside the file system, where nothing shows it. That on its own is the argument for imaging.

From the box arriving to the files going back.

Jobs we have logged →

A case number first, then a free diagnosis Free

Each item that arrives gets a case number the day it lands. An engineer looks at it and finds the real fault, and that costs you nothing. You are told in plain terms which files can come off the drive and which cannot. A price follows — one figure, written down. No work starts until you have seen it and agreed to it.

The diagnosis is freeOne price, put in writingNothing to pay yet

Leave the server exactly as it stopped

Power the server down. Pull the drives one at a time and mark each with the bay number it came from. Do not answer the console, whatever it asks. On the bench each member is imaged on its own, on hardware built for it — the ones the controller had already thrown out as well. No file is read until every image is finished.

Each disk imaged on its ownIncluding the write-offs

Put the set back together in software

The DDF or RIS records came across with the images. Reading them gives the RAID level, the stripe size, which bay each member occupied and — on RAID 5 and RAID 6 — the direction parity rotates. That is all the geometry needed to assemble the set in software. Once it is assembled, a dead card stops mattering.

Geometry read off the disksNo original controller needed

Copied is one thing. Opening is another.

Whatever sat on top — NTFS, ReFS, XFS, ext4, a VMFS datastore — is repaired on the rebuilt copy. Nothing is ever written back to your drives. Then it gets tested properly: guest machines boot, shares mount, databases attach, and the files themselves open.

Repairs done on images onlyProved by booting and mounting it

You sign it off, then it comes back

You are not billed while you are still thinking. The list of what came off the drive goes to you first, and the invoice waits on your word. Files go home on media bought in for the job, with the postage back on us. We keep the job open until you have opened them on your own machine.

You approve the listing firstCopied onto new mediaReturn postage at our cost

What turns up most often

The map is not at the same end on every card — HP writes RIS at the head of each member; PERC and MegaRAID put DDF in the last sectors. Look at the wrong end and the geometry is a guess.
The ejected drive is still evidence — a member the card dropped weeks back may hold the only current copy of some regions, so the images get compared to one another, byte for byte, and nothing here is taken on trust.
A generic preset gets Smart Array wrong — HP offsets the parity block by a stripe from where standard tools expect it, so the volume such a tool builds never existed. It has to be rebuilt to HP's scheme.
The cache battery is one of the first questions — whether the write-back data reached the platters at all decides which state the file system is in. That is settled early, before anybody is given an expectation.

What those two options really do: Import believes the foreign metadata and writes it over the configuration the card is currently running. Clear wipes the layout off every member, and there is no undo. Both sit one keypress away, with nothing on screen to warn you — though once full images exist, neither one can take anything from you. Imaging is not one step among several here. The whole job rests on it.

One case from the casebook.

HU · HUL-2026-0816LOGGED ✓

A flat cache battery, and the ProLiant asking questions

The power had been off long enough for the cache battery to go flat. When it came back up, the controller put a question to the practice that none of them could safely answer, so nobody answered it. Every member was copied before anything else was tried. The array's own metadata was read back from those images, the parity rotation worked out, and the set assembled on copies alone. The practice database opened in full, up to the last write the cache had managed.

Database came up complete5 days after it landed

Before you box it up.

Do these first

Photograph every screen the controller puts up
Mark every drive with the bay number it sat in
Send all the disks, including the ones that look dead
Tell us what it held — SQL, Exchange, VMs, file shares

What not to do

Pressing Import or Clear at that prompt
Fitting a spare disk and letting it rebuild
Flashing new firmware onto the controller now
Powering the server back up before the disks are imaged

Asked most weeks, answered here.

The server is dead. Is everything on it gone?

Rarely. In most cases the disks are undamaged and the trouble sits a layer above them — the controller's record, the map of the set, or a file system that has been mangled. Leave the power off, give the console no answer, and the odds stay on your side.

'Foreign Configuration Detected' — what does that mean?

That the layout on the disks is not the layout it was expecting. A swapped controller, a disk that went back into a different slot, or a card on its way out will each cause it. Stopping there is the card doing its job. Import and Clear both destroy data, so the answer is to image the lot first. After that, neither option can reach what matters.

Does the RAID controller need to come too?

No. Each member carries the layout on it — DDF, which is an open specification, on PERC and MegaRAID; HP writes its own RIS structures instead. That gets read off the images and the set is rebuilt in software. A matching card is never part of it.

Will the databases and virtual machines actually run?

Usually, but it is two pieces of work. The first is getting the VMDK or the MDF off the rebuilt volume intact. The second is making a hypervisor mount that VMDK, or SQL Server attach that MDF. The database recovery page deals with the second part.

What people read after this.

A RAID 5 rebuild → Wallet files → Email and databases → Everything we take in →

A drive left switched off loses nothing more.

Leave it switched off. A failing drive has a limited number of starts in it, and each one spends a little more. The free diagnosis will say which files can be read.

0800 6890668

Have it looked at How to post it to us

Devices and media we work on

Sorted by fault

Forensic and CCTV evidence

What to expect