RAID, in plain language

RAID means Redundant Array of Independent Disks. Linux treats several physical drives as one logical disk. Some RAID levels copy data for safety. Some stripe data for speed. RAID is not a backup.

Why dedicated servers use RAID

Hetzner often ships two same-size SSDs or NVMe drives. If you install the OS on only one disk, a single failure takes the server down. RAID1 writes the same data to both disks. One disk can die and the website stays up until you replace the drive.

Two disks holding the same A B C data blocks, illustrating RAID 1 mirroring — Picture: RAID 1 is a photocopy. Disk A and disk B store the same blocks. You get the capacity of one disk, and one disk may fail.

Important note

RAID survives a disk failure. It does not survive ransomware, rm -rf, or a bad installimage. Keep backups somewhere else (storage box, object storage, another machine).

Note

/dev/md2 is one layer. Root is often RAID → LVM (/dev/mapper/vg0-root), and can also sit on LUKS or a plain partition. The full “where is /” table: Rescue System.

The levels you will actually see

RAID 0 · stripe

Data is split across disks for speed and full capacity.

2×1 TB → 2 TB · one disk dies = all data gone

RAID 1 · mirror

Two (or more) copies. Hetzner default for two OS disks.

2×1 TB → 1 TB · one disk may fail

RAID 5 · parity

Needs 3+ disks. Capacity of N−1. Heavier writes.

3×1 TB → 2 TB · one disk may fail

RAID 6 · dual parity

Needs 4+ disks. Two disks may fail. Common on many-drive boxes.

4×1 TB → 2 TB · two disks may fail

RAID 10 · mirror + stripe

Best mix of speed and safety. Needs 4 disks (or more, even count).

4×1 TB → 2 TB · one disk per mirror may fail

/boot stays RAID1

Even if data is RAID 0 or 10, Hetzner keeps /boot as RAID1 so GRUB can boot.

md0 = boot mirror

installimage can build software RAID 0, 1, 5, 6, or 10. For two disks, use RAID1 unless you fully accept RAID0 data loss.

“Linux RAID” on the partition is only a label

MBR type fd and GPT “Linux RAID” tell other tools “this slice is meant for mdadm.” They do not build the array. You still run mdadm --create (or let installimage do it). The real RAID member shows up as linux_raid_member in lsblk -f. Partition labels vs filesystems: Partitioning.

Software RAID vs hardware RAID

	Software RAID (mdadm)	Hardware RAID
Who does the work	Linux CPU, tiny cost on modern hardware	A RAID card in the server
You see	`/dev/md0`, `md1`, `md2`	One fake disk, e.g. `/dev/sda`
Hetzner	Usual on auction / two-disk servers	Set up the array before installimage
Rescue	Linux can assemble the array	The card presents the volume already

Read RAID status

Is RAID healthy?

root@server ~ # cat /proc/mdstat

What you should see

Personalities : [raid1]md2 : active raid1 nvme0n1p3[0] nvme1n1p3[1]      498860032 blocks super 1.2 [2/2] [UU] md1 : active raid1 nvme0n1p2[0] nvme1n1p2[1]      1046528 blocks super 1.2 [2/2] [UU] md0 : active raid1 nvme0n1p1[0] nvme1n1p1[1]      262080 blocks super 1.2 [2/2] [UU] unused devices: <none>

Circle [UU] with your eyes: two disks, both Up. [_U] or [U_] means a disk is missing. A progress bar under the array is a rebuild — wait for it.

Same story as a tree

root@server ~ # lsblk

What you should see

NAME            MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINTnvme0n1         259:0    0 476.9G  0 disk└─nvme0n1p3     259:3    0 475.7G  0 part  └─md2           9:2    0 475.6G  0 raid1 /nvme1n1         259:4    0 476.9G  0 disk└─nvme1n1p3     259:7    0 475.7G  0 part  └─md2           9:2    0 475.6G  0 raid1 /

One array in detail

root@server ~ # mdadm --detail /dev/md2

What you should see

/dev/md2:           Version : 1.2     Creation Time : Sun Apr 12 09:11:02 2026        Raid Level : raid1        Array Size : 498860032 (475.56 GiB)     Used Dev Size : 498860032 (475.56 GiB)      Raid Devices : 2     Total Devices : 2             State : clean    Active Devices : 2   Working Devices : 2    Failed Devices : 0     Spare Devices : 0           UUID : 8f3a1c2d:9e0b4a17:c5d6e8f0:11223344         Devices : 2     Number   Major   Minor   RaidDevice State       0     259        3        0      active sync   /dev/nvme0n1p3       1     259        7        1      active sync   /dev/nvme1n1p3

What /proc/mdstat stands for

/proc = process filesystem (live kernel data). md = Multiple Device (software RAID). stat = status. Full list of short names: Abbreviations.

Create a data array (extra disks)

Only do this for spare drives that are not already in the OS array. Creating RAID wipes those devices.

New RAID1 on unused disks

root@server ~ # mdadm --create --verbose /dev/md3 --level=1 --raid-devices=2 /dev/sdc /dev/sddroot@server ~ # mkfs.ext4 /dev/md3root@server ~ # mkdir -p /dataroot@server ~ # mount /dev/md3 /data

What you should see

mdadm: size set to 976747520Kmdadm: array /dev/md3 started.mke2fs 1.47.0Creating filesystem with 244186880 4k blocksSuperblock backups stored on blocks: ...Writing superblocks and filesystem accounting information: done

Persist the mount in /etc/fstab using the UUID from blkid /dev/md3.

Email when a disk fails

On Debian and Ubuntu, set a mail address in /etc/mdadm/mdadm.conf and keep AUTOCHECK=true in /etc/default/mdadm. You need a working mail stack for this to leave the server.

Send a test mdadm mail

root@server ~ # mdadm --monitor --test --oneshot /dev/md0

What you should see

mdadm: Monitor using /proc/mdstat for arrays(a test mail is sent if a mail address is set in /etc/mdadm/mdadm.conf)

Stop RAID in Rescue before a clean install

Stop arrays

root@rescue ~ # mdadm --stop /dev/md0root@rescue ~ # mdadm --stop /dev/md1root@rescue ~ # mdadm --stop /dev/md2root@rescue ~ # mdadm --stop /dev/md*

What you should see

mdadm: stopped /dev/md0mdadm: stopped /dev/md1mdadm: stopped /dev/md2