When a RAID array drops offline, it is usually recoverable, because the data is still intact on the disks. Whether it is ever recovered depends on what you do next, and the single largest cause of permanent RAID loss is a DIY rebuild onto a RAID that is still degraded. Below is what actually fails at each level of RAID, the don’ts that matter most, and how to safely rebuild a RAID 0, RAID 5 or RAID 6 that has failed.
The data is still intact. A degraded array is capable of being reconstructed in nearly every case, but only if it is not written to first. Once a rebuild, re-initialisation or “repair” is performed, recoverable becomes unrecoverable.
The array stops assembling itself, but your data is still sitting on the platters of the disks.
A RAID distributes your data across multiple disks in fixed-size blocks, often with parity: additional maths that allows the array to keep functioning when a disk fails. The number of disks that can fail depends on the RAID level: RAID 0 cannot lose any, RAID 5 can lose one and RAID 6 can lose two. Any time you lose more than that (a second disk in a RAID 5, a dead controller, a corrupt configuration or enough bad sectors across multiple disks), the RAID cannot assemble itself and drops offline.
The key point: offline doesn’t mean gone. The blocks remain on the individual disks in exactly the same place they were written. Recovery means reading each disk, determining how the blocks were arranged, then reassembling them (away from the original hardware) so that nothing is ever written back to your data.
Failure and recovery differ slightly between the different types of RAID.
RAID 0 (striping). There is no redundancy at all (data is spread across all disks for speed), so a single failed drive will take down the entire array. Recovery requires all members: first we recover any physically failed drives, then we image all of them read-only and rebuild the stripe from the images by determining the block size and the order of the drives.
RAID 1 (mirroring). Every drive has its own complete copy, so losing one drive won’t lose the array, and even when the array can’t mount, we can still recover from either healthy member. Trouble usually means both drives have problems, or the file system on the mirror is corrupt.
RAID 5 (striping with parity). Survives the failure of one disk by rebuilding the lost disk using parity information. The classic disaster scenario is a second disk dropping (often while rebuilding), which takes the array offline. It is still generally possible to recover: we image all drives, including the failed ones as far as they will read, and use the parity information along with the remaining data to reconstruct the array.
RAID 6 (dual parity). Survives two failed disks, so three must fail before it goes offline, but the same rules and the same rebuild dangers apply. We follow an identical process to the one above: image all drives read-only, then reconstruct.
RAID 10 (mirrored stripes). A stripe of mirrors: it tolerates a disk failure in each mirror pair, but loses the entire array when both halves of one pair fail. Recovery combines the methods for RAID 1 and RAID 0: first recover the failed members, then rebuild the mirrors and the stripe.
Nearly all of the unrecoverable RAIDs we see were made worse once they had failed, not by the original fault.
Don’t rebuild onto a degraded array. If a disk drops in a RAID set, replacing it and letting the controller rebuild will stress every disk that remains in the set and write new parity information across all of them. If another disk is on the edge, the rebuild finishes it off, and your data goes with it.
Don’t re-initialise, recreate or “repair” the array. Recreating the RAID set or allowing a controller to initialise it can overwrite the configuration and the data layout, and once that is gone, it is much more difficult to reconstruct.
Don’t run chkdsk or fsck on the array, and do not continue to force the RAID set back online. Repair tools write to the volume, which can result in a partially rebuilt RAID set instead of a clean rebuild.
Don’t swap disks around guessing the order and, before you take anything apart, label each disk with its bay position. The physical order of the disks matters, and if they are not properly labelled, it will cost you time to get them sorted out.
Read everything, determine the geometry of the RAID set and reassemble a copy of it (never the originals).
The professional method of recovering a RAID set is the complete opposite of a rebuild. The first step is to image every member of the RAID set read-only, sector by sector, using write-blocked equipment. Any member disk that is physically failing is repaired or head-mapped first, then imaged around its weak areas, so no further degradation occurs. Once all the members have been imaged, the original disks are set aside and all further work happens on the images.
We then reverse-engineer the geometry of the array from the data itself: the stripe (block) size, the disk sequence, the parity rotation and whether it is left- or right-synchronous, and the start offset. This is why losing a controller or a configuration doesn’t have to mean the data is gone for good: the way the blocks are laid out is embedded within the data itself, so the layout can be worked out by analysing the images of the drives.
Once all the parameters are figured out, we rebuild the array virtually using those images, then recover what sits on top: the file system and its folders, or the virtual machines (VMDK, VHDX) and databases the array was hosting. For more on the specifics, please refer to our RAID 10 explainer and our guide on what to do when a RAID won’t rebuild.
If your rebuild has failed, a second disk has dropped or you have lost a business-critical array, power it off and let us create an image of it.
If a rebuild has been attempted and failed, a second disk has dropped, the controller no longer sees the array, or for whatever reason your system refuses to come back up and the data matters, then the safest option is to power it down and stop trying to bring it online. Every time you try again, you take a greater chance of writing over data that could otherwise be recovered.
Our RAID data recovery service handles arrays and NAS units in-house from a fixed £500 +VAT, with a written quote prior to any chargeable work based on a free diagnostic, and your drives remain within the UK. Simply remove the drives from your system, label each one with its corresponding bay number, pack them so they will not knock together in transit, and send them to us. Business-critical arrays receive priority treatment.
Typically, yes. RAID 5 arrays survive one failed disk. Once a second disk fails, the array is taken offline, but the data remains on the platters. We image all of the individual disks read-only (even if they are partially failing), determine the stripe size, the disk order and the parity rotation, and then rebuild the array offline from those images. How much of each disk is readable at the time of imaging determines how complete the result is, and the free diagnostic tells us that.
Stop and turn the system off. A rebuild requires every disk to be read in full and fresh parity to be written across the entire array, so if a second disk is already weak, the very data needed to reconstruct it can be overwritten; this is the most common cause of permanent RAID loss. Do not try to restart the rebuild or re-initialise the array. Label each disk with its bay order, then remove them and send them to us; we’ll image each one read-only, then rebuild the array offline.
Since RAID 0 does not contain redundancy, every disk is needed: the data is striped across all of them. If one disk has physically failed, we’ll recover that drive first, with work on the heads, board or firmware as required. Next we’ll create a read-only image of each member of the array, then rebuild the stripe using the images. To do this, we need to determine the block size and disk order used in the stripe. So long as the failed disk can be read, we can usually reassemble the array.
Generally no. We reconstruct the RAID parameters (stripe size, disk order, parity rotation and start offset) directly from the data on the imaged disks, so a dead controller or a lost configuration won’t prevent us from recovering your RAID. Having the original controller or a record of how things were set up can help, but it isn’t required.
Recovery services for RAID arrays, NAS units and servers begin at £500 +VAT. Prices vary based on the number of disks involved, the RAID level used and the capacity. All work begins with a free diagnostic followed by a fixed written quotation prior to any chargeable work, and all work is performed entirely in-house.
This will depend on both the number of disks and their condition, as imaging several drives read-only takes time, especially any that are failing. The free diagnostic is usually completed within 48 hours. Business-critical arrays receive priority, and we provide an honest turnaround time once we have had an opportunity to examine the disks.
Shut the system down, let us know the make, the RAID level and what happened, and send the disks in; we’ll give you a free diagnosis and a fixed written quote prior to any payment, with business-critical arrays prioritised.