RAID backs up nothing: what it protects, and the six risks it ignores

RAID covers exactly one scenario: a disk failing. Deletion, corruption, ransomware, theft, mistyped commands — it does nothing. And rebuilds are where entire arrays are lost.

"My data is on RAID, it's protected." The shortcut is so widespread it sounds reasonable. It rests on confusing two properties that have almost nothing to do with each other.

What RAID actually does

RAID spreads data across several disks with redundancy, so that losing one — sometimes two, depending on the level — does not interrupt service. You replace the failed unit, the array rebuilds, nobody noticed anything.

That is availability, and it is valuable: it avoids downtime and a full restore every time a disk dies.

But note what it implies: there is only one version of your data, the current one. RAID keeps it reachable despite a hardware failure. It preserves no earlier state.

The six scenarios it does not cover

Deletion. You erase a directory by mistake. RAID performs the deletion across every disk, faithfully and instantly. Redundancy keeps no previous version.

Application corruption. A database writes inconsistent data, a failed update destroys a configuration: RAID faithfully stores what it is handed. It does not judge content.

Ransomware. Encryption is a legitimate write. RAID applies it across the whole array without slowing down. Here redundancy serves the attacker.

Theft, fire, water damage. Every disk is in the same enclosure, in the same place. One physical event takes them together.

The mistyped command. A format on the wrong volume, a partition table on the wrong device: applied to the entire array.

Silent corruption. A block degrades without anyone noticing. Classic RAID generally has no way to know which copy is correct — it can even propagate the bad one during a rebuild. Only filesystems with checksums genuinely detect and correct this, and that is a filesystem property, not a RAID one.

The rebuild: the most dangerous moment

Here is the point that discussions about RAID levels most often skip.

A disk dies, you replace it, the rebuild starts. To recompute the missing data, the system must read every surviving disk in full. It is the heaviest load they will ever experience, running for hours or days on large capacities.

Those disks are the same age, often from the same batch, and have carried exactly the same load since day one. They are in the same state of wear as the one that just failed.

If an unreadable sector is lurking anywhere, the rebuild will find it — and on an array tolerating a single failure, it fails. You then lose the whole array, not one disk.

A simple precaution: mix batches when buying, or even models. Disks that did not age together do not fail together. It is one of the rare free reliability measures.

The controller, the forgotten failure point

On hardware RAID, the controller card writes its metadata in a proprietary format. If it dies, you often need a compatible model to read the array back. On older hardware that can mean hunting for a part on the second-hand market — with perfectly intact and perfectly unusable disks.

Software RAID managed by the operating system does not have this flaw: the disks reassemble on any machine running the same system. For a homelab, where nobody keeps a spare card on the shelf, that is decisive.

So, should you use RAID?

Yes — for what it does. It avoids downtime on every disk failure, and on a machine providing a permanent service that is genuine comfort.

The question to ask is simply not "am I protected?" but "protected from what?". RAID answers one scenario out of seven. The other six require independent, versioned copies, at least one of them out of the machine's reach.

And if budget forces you to choose between the two, the choice is clear: a backup without RAID beats RAID without a backup. The first costs you downtime. The second costs you your data.

Frequently asked questions

What is the difference between availability and backup?
Availability answers "does the service keep running despite a failure?". Backup answers "can I recover my data after a disaster?". RAID addresses only the first: a disk dies, service continues, you replace it without downtime. It answers nothing for the second, because only one version of the data exists — the current one, mistakes included.
Why is a rebuild so risky?
Because it requires reading every remaining disk in full to recompute the missing data. It is the heaviest load they will ever face, applied to disks of the same age, the same batch, with the same wear. If an unreadable sector sits anywhere on one of them, it will be hit — and on an array tolerating a single failure, the rebuild fails. Statistically it is the most dangerous moment in an array's life.
Should I avoid buying all my disks at once?
It is a real and cheap precaution. Disks of the same model, from the same production batch, put into service on the same day under the same load share their ageing characteristics. A second failure shortly after the first is then not bad luck, it is the expected consequence. Mixing batches, or even models, desynchronises the failure curves.
Is software RAID less reliable than a controller card?
Generally no, and it has a decisive recovery advantage. A controller card writes metadata in its own proprietary format: if it dies and you cannot find a compatible model, the array can become unreadable while every disk is perfectly intact. Software RAID managed by the operating system reassembles on any machine running the same system. For a homelab, that is a serious argument.
Does RAID protect against silent data corruption?
Classic RAID does not: it faithfully stores what it is given, corrupted data included, and generally has no way to know which copy is correct. Filesystems with built-in checksums change that: they verify every block on read, detect alteration and repair it from redundancy. That is a property of the filesystem, not of RAID.

Did you enjoy this article?

Comments

Morgann Riu

Cybersecurity and Linux administration expert. I help companies secure and optimize their critical infrastructures.

Back to the blog

Checklist Sécurité Linux

30 points essentiels pour sécuriser un serveur Linux. Recevez aussi les nouveaux tutoriels par email.

Pas de spam. Désabonnement en 1 clic.