Each of the RAID levels represents some type of compromise among three things you'd like from a RAID (capacity, speed and survival) and therefore none of the RAID levels deliver all three. The understanding of what your particular RAID gave up by compromising those aspects will tell you exactly how your RAID will die. Below are all of the RAID levels explained as simply as possible along with their respective deaths.
A RAID array is primarily designed to ensure that your services are up and running after a disk fails (not that your data is safe). It rapidly replicates deletions, ransomware encryptions and accidental overwrites across all disks as fast as the hardware allows. Redundancy is essentially "uptime insurance", but the single costliest misconception in storage is to confuse uptime and safety.
There is a key concept to know before explaining RAID levels: parity. A parity value does not represent a copy of your data; it represents the mathematical representation of your data so the array can rebuild a lost or damaged piece of data if a disk goes down. It's very elegant. However, it is the source of the most common disaster in this field, because rebuilding must read every remaining disk, completely, at the worst time.
RAID 0 (striping). Splits data into stripes & distributes each stripe to all disks. Very fast, uses every byte that you paid for & offers no redundancy. If just one disk fails then the entire array will fail since every file is split and half of it existed on that failed disk. RAID 0 is not a compromise on safety but the presence of no safety, and yet there are many dual-bay products being shipped with RAID 0 as the "default" configuration and referred to as "backup drives".
RAID 1 (mirroring). Two disks with identical content. It's simple to set up. If a single drive dies, you still have access to your data. The drawback of this RAID type is its low capacity (you only get half the total space), and because mirroring creates an exact replica of all data, most users will rarely ever view their mirrored data. Because a user has two copies of every file, they may be less likely to check them frequently. A user may allow one copy to degrade over time. This means that if both drives are allowed to fail in succession, before the second disk fails, the first disk could have been dead for months.
RAID 5 (striped with single parity). The workhorse: capacity of all disks minus one. Survives 1 disk failure. The cost of this configuration is the harshest of all, and we discuss why below.
RAID 6 (double parity). A variation of RAID 5 that can survive up to two simultaneous failures by using an additional disk for redundancy. This comes with a price, however; you will lose one more drive's worth of usable space. Given current technology in larger disks this is your best option. The biggest threat to RAID 6 is complacency rather than mathematics.
RAID 10 (a mirror of stripes). Mirrored disk pairs are striped together. This offers both the speed of RAID 0 and the fault tolerance of RAID 1 but cuts your raw storage down by 50%. RAID 10 can survive the failure of one disk from each mirrored pair. However, should both sides of a pair fail it will instantly cease to function. When all four of the hard drives used for a RAID 10 are identical and come from one production batch, each has likely been spinning in the same warm enclosure for as long as the other three drives. Thus, the risk of two disk failures occurring in the same mirrored pair is real.
The RAID 5 rebuild. A disk within a RAID 5 configuration has failed and been replaced. The process then begins to rebuild the missing disk by reading through every sector of every remaining active disk within the array. When using large, multi-terabyte drives, such a rebuild is substantial. Furthermore, if those drives have sat idle in their enclosures for years, it is entirely possible that some portions of them quietly become "bad" or "unreadable". Once a "bad" sector is encountered during this rebuild process and another disk in the array goes bad while attempting to continue the rebuild, the entire array becomes unusable during the repair. A Paisley practice lost exactly this way, and this is the single most common array disaster there is.
Pulling the wrong disk under pressure. With the amber lights flashing and pressure to get things done, a good hard drive is pulled out instead of the dead one, and the array starts rebuilding nonsense on top of other good data. We've rebuilt an array after an error occurred because the rebuild started onto the wrong member, with the virtual machines still inside.
The power cut and the firmware update. RAID 0 in a NAS during a write operation (mid-write), and suddenly there's a power failure: some of the metadata has been written, but the array can't be mounted; and we have no parity to work with. A ReadyNAS has done exactly this. It is just as common for the controller to lose memory due to either a card replacement or a firmware update that will overwrite the original configuration, resulting in the array returning and claiming each of its members as "foreign".
Deletion, which redundancy cannot save you from. A RAID 5 volume is removed from all disks at once if it is removed on a working RAID 5, as happened to a Dell PowerEdge. There isn't enough parity in the world to reconstruct a deletion.
It is natural to want to get an array operating again after it has failed, whether by rebuilding, resynchronising, re-initialising, "repair volume" from within the NAS interface, etc. All of these options will write to the disks that hold your only copy of the data. It is much harder to reverse the effects of a failed rebuild than those of a failed hard drive. Therefore, the professional recovery process reverses this order: first, image every single member disk; secondly, construct the virtual array using those images; thirdly, recover the data from the virtual array; fourthly, worry about what to do with the physical hardware.
Start with order. Shut down the array. Don't let it run the rebuild again. Label the drives with which bay they originally came from: stripe order is important, and a group of four identical-looking disks without labels will be a problem we can normally solve, but prefer to avoid. Get a price quote before transport. If possible, your NAS should arrive with us as the individual drives, labelled by bay; if not, please send the entire unit. Drive(s) from a server or array will need to be removed from their respective chassis or enclosures, also labelled. The cost of our array recovery services starts from £500 + VAT, the diagnostic is free, and this price is fixed in writing prior to our work commencing on the array.
You can recover data from a degraded array. Recovering an array that was rebuilt "over the top" is a much longer process. Stop the resync, assign bay labels to your drives, and then use the free assessment to map out the remaining reality. RAID, NAS & server work from £500 + VAT, fixed in writing.