Software RAID with mdadm

Look at what you have

cat /proc/mdstat

The first command for anything RAID-related. [UU] means both members are up; [U_] means degraded. A rebuild shows a progress bar and an estimated finish.

mdadm --detail /dev/md0 for the full picture, including the array UUID and each member's role.

Creating an array

mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sda1 /dev/sdb1

Then persist the configuration, or the array may come up with a different name after a reboot:

mdadm --detail --scan >> /etc/mdadm/mdadm.conf

update-initramfs -u

That second command is not optional on a root array. Without it the initramfs does not know how to assemble the array and the machine boots into a rescue prompt.

Set up the monitoring, today

A degraded array that nobody notices is a single disk pretending to be redundant. In /etc/mdadm/mdadm.conf:

MAILADDR you@example.com

and make sure mdmonitor is enabled. Better still, alert on it from your monitoring as well — see Prometheus and Grafana. Two independent warnings is the right number for this.

Scrub it monthly

echo check > /sys/block/md0/md/sync_action

This reads every block and finds latent bad sectors before a rebuild does. Most distributions ship a monthly timer for it; confirm yours is actually running. A mismatch count that climbs is a warning worth acting on.

Spares and growing

Add a hot spare so a failure starts rebuilding immediately instead of waiting for a human:

mdadm --manage /dev/md0 --add-spare /dev/sdc1

Growing an array is possible and slow. Expect hours, and take a backup first — it rewrites the layout of every disk:

mdadm --grow /dev/md0 --raid-devices=4

Then grow the filesystem on top, as LVM describes.

Choosing a level

RAID 1

Mirror. Simplest, fastest rebuild, best for anything with a database on it.

RAID 10

Mirrors striped together. The default for four or more disks where performance matters.

RAID 6

Two disks of parity. The sane choice for bulk storage with large drives.

RAID 5

One disk of parity, and a rebuild window long enough to be uncomfortable on modern capacities. Avoid.

It is not a backup

RAID survives a disk. It does not survive rm -rf, ransomware, a controller writing nonsense, or the building. See backups and testing a restore.

Tell us what layout you want before delivery and the machine arrives built that way, rather than rebuilt afterwards — see ordering a server.

---

Still stuck? Mail support@novogara.com — an engineer answers, at any hour. Back to the knowledge base