A member disk of a RAID array shows predictive failure. The clone instinct says 'copy it before it dies' — but RAID was invented for exactly this moment, and using it wrong (cloning a degraded member, pulling the wrong disk) can take down data that redundancy was protecting.
A single failing drive and a failing RAID member look similar and are completely different jobs. A lone drive has no partner; a RAID member's data also lives, in shared form, on the other disks. That difference decides the method: mirrored and parity arrays are rebuilt; cloning is for the array as a whole, or for when there is no redundancy left to trust.
In a mirror (RAID 1) the partner disk holds everything; in parity arrays (RAID 5/6) the missing data is recalculated from the survivors. Either way, the procedure the hardware was built for is: confirm which physical disk is failing (slot lights and the controller utility agree), replace it with an equal or larger disk, and let the controller rebuild onto it. The server keeps running, users keep working, and at the end the array is whole again. No clone, no downtime window, no identity risk — those exist in the single-disk world, not here.
A rebuild stresses the surviving disks hard — they are read end to end. If a second disk was quietly weak, the rebuild is when it confesses, and a RAID 5 has no answer for that. So the order is: verify a current backup exists first (an actual restore-tested one, not a green icon from March), check the other members' SMART/health state in the controller utility, and only then pull the failing disk. Also mark the failing disk's slot physically — in a cage of identical drives, 'the one in bay 2' gets misremembered at exactly the wrong moment.
RAID 0 stripes data across disks with zero redundancy: every file lives in pieces on all of them. One member dying means the array's data is gone — not 'at risk', gone. There is nothing to rebuild onto a new disk and nothing meaningful to clone, because the surviving member holds stripes of files, not files. The rescue for RAID 0 is called restore from backup, and if there is no backup, the honest conversation is about professional recovery odds, not settings. Any server still running its system or data on RAID 0 should have that changed at the next maintenance window — this paragraph is why.
Cloning still has real RAID jobs. When the whole array is being retired — old small disks to a new larger array, or array to a single big SSD on a small server — a clone of the logical volume (the drive Windows sees) moves everything in one pass, booted from rescue media that can see the controller. When the array is degraded and a second member is suspect, imaging the logical volume to safe storage before attempting any rebuild is the cautious order: get a copy out while the array can still read, then rebuild. And when the 'RAID' is actually a motherboard fake-RAID or Windows dynamic disks, treat it with suspicion — read the controller documentation before pulling anything, because those setups vary the most.
A rebuild on large disks takes hours, sometimes a day, and the array is at its most fragile for every minute of it: a second failure during the rebuild is the classic data-loss story. So during the rebuild: no heavy maintenance, no firmware updates, no 'quick' reboots, and warn users the server may feel slower. Monitor progress in the controller utility rather than restarting to 'check'. When it completes, verify the array state reads Optimal, and only then return the replaced disk to the labelled-failed pile — never to the spare shelf.
The video handles a failing server drive with the method that fits it — the same judgement this page applies to arrays.
▶️ Watch Full VideoAct — on your schedule, not the disk's. Predictive failure means the disk's own counters crossed the line; arrays show Optimal until the disk actually drops out. Replace it by rebuild in the next maintenance window while you still get to choose the timing.
Yes, the extra space is simply unused by most controllers (some allow expanding the array afterwards — a separate, deliberate operation). What you cannot do is rebuild onto a smaller disk than the member it replaces.
Genuine hot-swap hardware on a hardware controller: yes, that is what it is for — the rebuild starts on insertion. But confirm the backplane and controller truly support it, and still do the backup check first. On cheap 'hot-swap style' bays without controller support, shut down instead.