Linucate
~ Linucate_

204.1 Configuring RAID

All Levels

Introduction

RAID combines devices for performance, availability, or both. RAID is not a backup: deletion, corruption, compromise, and array loss can affect every member.

What you should be able to do after this lesson:

  • Compare RAID 0, 1, and 5.
  • Create and inspect an md array.
  • Monitor synchronization and degraded state.
  • Fail, remove, and replace a member safely.
  • Assemble an existing array and understand persistent metadata.

Big Idea: Separate the Array from the Data Above It

Linux software RAID creates one md block device from several members. Partition tables, LVM, and filesystems normally sit above that md device:

member devices -> /dev/md0 -> optional LVM -> filesystem -> mount point

Replacing a member repairs redundancy at the array layer. It does not repair corrupted files inside the filesystem and does not restore deleted data.

RAID Levels

  • RAID 0: striping, no redundancy, capacity is the sum of members.
  • RAID 1: mirroring, survives member failure while a good copy remains.
  • RAID 5: distributed parity, requires at least three members and normally survives one member failure.

RAID 5 rebuilds are I/O intensive and expose the remaining members to additional risk. Capacity, fault tolerance, and write performance must all be considered.

Inspect Devices First

lsblk -o NAME,SIZE,TYPE,FSTYPE,MOUNTPOINTS
wipefs /dev/<member>

Use equally sized partitions or devices when practical. Older partitioning conventions used type 0xFD for Linux RAID autodetection; modern md metadata carries assembly information.

Create an Array

Example RAID 1:

sudo mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sdb1 /dev/sdc1
cat /proc/mdstat
sudo mdadm --detail /dev/md0

After creation, build the filesystem on /dev/md0, not on an individual member.

Persistent Assembly

Scan array metadata and store the result in the distribution's mdadm.conf location:

sudo mdadm --detail --scan

The configuration may be /etc/mdadm.conf or /etc/mdadm/mdadm.conf. Rebuild the initramfs if the root or boot path depends on the array.

Replace a Failed Member

sudo mdadm /dev/md0 --fail /dev/sdb1
sudo mdadm /dev/md0 --remove /dev/sdb1
sudo mdadm /dev/md0 --add /dev/sdd1
watch cat /proc/mdstat

Confirm the correct failed member before removal. Do not remove a healthy member from an already degraded array.

Stop and Assemble

sudo umount /mountpoint
sudo mdadm --stop /dev/md0
sudo mdadm --assemble --scan

Stop users and unmount filesystems before stopping an array. --force can make a damaged situation worse and should follow a documented recovery decision.

Monitoring

Monitor /proc/mdstat, mdadm --detail, system logs, and device health. Configure notifications so degraded arrays are not discovered only after another failure.

Guided Practice: Design and Observe a Lab Array

Use disposable virtual disks or loop devices, never production storage. Plan a RAID 1 array and write down the expected state transitions:

sudo mdadm --create /dev/md/lpic-lab --level=1 --raid-devices=2 \
  /dev/<lab-member-a> /dev/<lab-member-b>
watch cat /proc/mdstat
sudo mdadm --detail /dev/md/lpic-lab

After synchronization, identify:

  • array UUID and metadata version
  • active and working device counts
  • member role numbers
  • whether a bitmap is present

Simulate a member failure only in the disposable array, remove it, add a replacement, and watch recovery. Verify that the filesystem remains mounted and readable during degraded operation, then confirm that redundancy returns.

Troubleshooting Scenario

After reboot, both member partitions exist but /dev/md0 does not. Their md metadata is intact, while the initramfs lacks the array configuration required for early assembly.

Use mdadm --examine to confirm member identity, assemble by metadata, generate the correct mdadm.conf entry, and rebuild the initramfs if the root or boot path depends on the array. Creating a new array with --create risks overwriting metadata and is not the first recovery step.

Exam Focus

  • RAID 0 provides no redundancy; RAID 1 mirrors; RAID 5 distributes parity.
  • Know mdadm, mdadm.conf, /proc/mdstat, and historical partition type 0xFD.
  • Distinguish create, assemble, fail, remove, add, stop, and monitor operations.
  • RAID availability does not replace independent, tested backups.

Recap

  • RAID availability and backup solve different risks.
  • Create filesystems on the md device.
  • Persist assembly metadata and include it in early boot when required.
  • A degraded array should be repaired promptly and carefully.
🎯

Test Your Knowledge

Complete the quiz to assess your understanding of this course's concepts.

Back home