How to Recover Data from RAID Drives
Our founder and lead technician, Joseph Montanti, handles cases personally.
To recover data from RAID drives, keep the original disks together. If the array went offline during a rebuild or a second disk dropped out, shut it down cleanly and leave it off so the controller cannot start another rebuild. Recovery of an offline volume starts with images of the member disks. Those images can be put back in the original order and read as one volume without creating a new array on the disks.
On this page
What the controller message means
The controller keeps track of which drives belong to the array and how the data is laid out across them. A computer may show one large disk even when there are four or eight physical drives in the enclosure. A controller replacement, missing member or unreadable sector can affect that one large volume in different ways.
Stop on a second fault. Shut the array down if a disk clicks, beeps or grinds, another disk drops out or the copy stalls. Do not scan the member disks with recovery software, reseat or reorder them, try rebuild combinations, initialize the set or force it online.
| What you see | What has happened | What to do |
|---|---|---|
| Degraded, folders still open | A redundant array has lost protection but can still supply the data | Stop if reads slow down or another disk drops offline. While the disks sound normal and the share reads steadily, copy folders you haven’t backed up to a separate disk now. |
| Offline or failed virtual disk | The controller has too few usable members, or it can't assemble the set it expects | Shut down cleanly and keep the set off to prevent an automatic rebuild. Leave the disks in place and check an independent backup. |
| Foreign configuration after moving disks or replacing a controller | The controller sees RAID information on the disks that differs from its current configuration | Use Preview to inspect the member list. Leave Import and Clear alone on the original disks. |
| Several bays show missing after a power failure | The controller has lost contact with several drives. The fault can be in something they share, such as the enclosure | If the array is running, shut it down cleanly. Leave it off when power returns to prevent an automatic rebuild. Keep the enclosure, controller and all disks together in their original positions. |
On a Dell PERC, Clear Foreign Configuration discards configuration information. Import makes the existing on-disk configuration available to the controller. A planned migration to a supported controller requires the complete disk group and its configuration. In a failure, keep every member in order with the original controller and any preserved cache. Importing a foreign member into an online or degraded array can corrupt the volume. Dell's foreign configuration instructions.
Use CHKDSK or fsck only on a disposable reconstructed copy after extracting the files. If the disks must come out, label each with its bay number first, including the disk the controller calls failed. Even a hard drive rejected by the controller can hold the only readable copy of part of a stripe.
The disks show up, but the virtual disk has disappeared
Look at the physical-disk list separately from the virtual-disk list. A Dell controller can identify every member by model and capacity yet have no active virtual disk for the operating system to use. Cancel any Windows prompt to initialize or format a member. What is missing may be the layout that tells the controller how those physical members fit together. Look for an existing configuration in the preview before assuming an empty list means empty disks.
After an outage a PERC can also report preserved cache. This is data the controller was holding for a virtual disk that went offline. It can include writes the operating system had already handed to the controller. Keep the original controller and its cache module with the drives. Clearing the cache to dismiss that message can discard writes that never reached the disks. Dell documents preserved cache and foreign configuration separately because the controller may need both pieces to complete the interrupted work. Dell's PowerEdge guidance on preserved cache.
RAID 0 data recovery
RAID 0 takes successive chunks of a volume and writes them across its members. On a two-disk stripe the first chunk goes to one disk, the second to the other, then back to the first. The disks share the work of a long read or write. Neither one holds a mirror or a parity block from which the other disk's missing chunks can be calculated.
Swapping in an empty drive leaves an empty space in every affected stripe. There is nothing for a rebuild to put there. A large video can have its opening on the good disk and the next part on the failed one. Copying the first disk or finding the video's name in a directory leaves those missing pieces exactly where they were.
Keep every original member for a full-volume recovery. Recreating or initializing an array with the old settings can write new configuration information to disks that still hold the original layout. Reconstruction on images lets a technician test the order and chunk size without committing a guess to the disk set. Intel's description of striped and mirrored volumes.
Recovering a RAID 1 mirror
A two-drive RAID 1 writes a copy to each member. If the first disk fails while the second is up to date and readable, the mirror can keep serving the files from the second disk. Both originals are worth keeping. One may have dropped out hours before the other, leaving two versions with the same label and quite different contents.
Start with the reason the files went missing. Deleting a folder while the mirror is working deletes it from both members. Neither a new disk nor a rebuild restores the folder to yesterday's state. A snapshot taken before deletion or a backup from that date is the place to look. With no such copy, preserve the members for deleted-file recovery before new writes reuse the space.
If you have one good copy on a member, protect it before rebuilding. A rebuild takes a source and writes its contents to a destination. Choosing a stale source and a current destination copies the older state over your newer work. A disk that used to be the mirror partner is a poor choice of blank destination when you haven't yet checked what survives on it.
One failed disk in a four-drive RAID 5
RAID 5 distributes parity across the members. For each stripe, the surviving data and parity let the controller calculate the contents of one missing block. A four-drive RAID 5 with one failed disk can therefore remain online and readable. The remaining three disks now have to supply everything, including the information needed to reconstruct the missing disk.
With a current, checked backup, one positively identified failed drive and three healthy members, replacing that disk with a compatible drive and rebuilding is normal maintenance. Use the procedure for the exact controller or NAS and retain the original failed disk. For a Synology pool, the Repair option applies to a degraded pool with redundancy. Synology's storage pool repair instructions.
A rebuild reads across the surviving disks to fill the replacement. Until it completes, there is still only enough redundancy to cover the first loss. A bad sector encountered on another disk can leave two unknown blocks in the same stripe. Some controllers abort the rebuild, and a file using that stripe may have a hole even when files elsewhere on the volume are intact.
A RAID 6 has two independent sets of parity, so it can tolerate two missing members. A RAID 10 stripes across mirror pairs. Two missing disks from different pairs can leave a readable copy in each pair, while losing both members of a single pair takes part of the stripe away. The bay positions matter as much as the number of warning lights. Dell's RAID level comparison.
Where a hot spare fits in
A spare bay can change the timeline of a failure. An unused hot spare is a disk set aside to take over if a member fails. It carries no current member's share of the array before takeover. When the controller starts rebuilding automatically onto that spare, it becomes a partly written replacement. A chassis with four active RAID 5 members and a fifth drive in a spare bay still has one parity block per stripe. The fifth drive shortens the wait to start a rebuild. It doesn't turn that set into RAID 6. IBM's description of automatic spare takeover.
If the set fails while the hot spare is rebuilding, stop the attempt with the controller's supported stop procedure or a controlled shutdown. Leave the failed set offline. Preserve the drive that dropped out first and the one that took over for it. For a file written before takeover, the old member may still hold useful blocks. For a file written later, that member can be out of date, even though it was the original disk for that bay.
Windows Storage Spaces
Windows can combine several physical disks into a storage pool, then allocate space from that pool to one or more virtual disks. Simple, mirror and parity spaces distribute data differently. A drive letter in Explorer identifies a volume on a virtual disk. It doesn't tell you which pieces are on the drive in the first bay. Microsoft's Storage Spaces layout options.
Keep all the original pool members in order, even if you only need files from one drive letter. Leave the disks in the pool. Microsoft's Reset-PhysicalDisk command removes pool configuration and data from the selected member in Storage Spaces. Removing a disk from the pool or resetting it to clear a warning changes the material needed to reconstruct the space. Microsoft's Reset-PhysicalDisk reference.
How the files come off an offline array
Whether you bring the disks in for RAID recovery or use mail-in recovery, include every member, even the disk that dropped out first. A disk that stayed online longer might contain the newest directory or the only current version of a database. Imaging each readable member separately preserves those differences. An unreadable region remains a gap in that image.
Next comes the arrangement. Drive order tells you which disk supplies each successive chunk. Chunk size tells you when to switch to the next disk. A starting offset accounts for space reserved before the volume data, and the parity layout determines which member holds the calculated block for a given stripe. A recovery program can assemble the images into a virtual disk using those settings. Changing that virtual arrangement leaves the saved images available for another attempt. UFS Explorer's RAID Builder documentation.
If the array held a virtual machine, seeing its disk image in the recovered directory is only the first layer. The guest's folders sit inside that large file. Make a separate working copy of the recovered virtual disk before opening it with the appropriate virtualization software. Check a document or database from inside the guest.
For the array itself, test a large file that crosses several stripe boundaries. For business data, open a restored project, database backup or archive from the separate destination and check that its contents match the version you need.
Get help with data recovery
Bring your device to our Lincroft lab during opening hours, or mail it in.
How to mail your deviceThe $100 diagnostic is non-refundable once diagnostics start, and donor drives or replacement parts you approve are non-refundable once ordered, even if recovery fails.
Published by MDrepairs
MDrepairs is a data recovery and electronics repair lab in Lincroft, New Jersey. Our founder and lead technician Joseph Montanti films real cases for our YouTube channel, showing tools, repair steps and recovery outcomes.
Keep reading
- Windows Storage Spaces Failure: Preserve the Pool Before Resetting Disks
- Synology Volume Crashed: What to Do While Files Are Still Readable
- RAID 1 Recovery: Why a Mirror Does Not Save Every Deleted File
