Linux Software RAID

June 16, 2022

Lang: cs en de es

Hardware RAID has many advantages. However, if you are not running a large infrastructure and don't really need the benefits of HW RAID for an enterprise solution, I prefer a software Linux RAID called mdadm because it offers many practical advantages. I'll describe the benefits and how to use mdadm in this tutorial.

You can find out what a RAID is at wikipedia. I'll just point out that people often confuse different terms and impressions, so it's a good idea to study the exact terms. RAID is not a backup. RAID is used to either consolidate disk space or to provide more service availability or better performance. Some motherboards are sold with RAID, but realistically, it's software RAID + Windows driver. Such a RAID is referred to as a fake RAID.

Which RAID to use? RAID 0, RAID 1, RAID 5, RAID 6, RAID 10. RAID 1 is the most typical choice due to data redundancy, ease of use and the need for only one extra disk. On the other hand, RAID 0 on its own is a completely inappropriate choice for server deployment because it reduces the security of data retention. Its use is only appropriate where we are willing to allow failure or loss of data - i.e. back up data.

Video RAID

You can find a verbal description of what RAID is and is not in this video:

Linux Software RAID

Linux software RAID, called mdadm, is a solution that is at the Linux kernel level.

    Its main advantages are:
  • stability
  • .
  • independence on HW
  • transparency
  • ability to migrate between different HW and Linux distributions
With Linux software RAID, you can run such a RAID without the need for special HW. So there is no need to maintain a spare piece of RAID controller. You can use the RAID on another Linux distribution. Or you can recover the data if you have only one disk of two left from RAID 1.

By the way, if you need RAID, you can use not only mdadm but also the BTRFS filesystem or LVM technology.

Disk partitioning

The size of the partition is individual to your needs. You can use the tools fdisk, cfdisk, parted or the graphical gparted to partition.

How to partition? We can safely derive the partitioning from the existing system, but with RAID in mind, it may be more appropriate to choose just one large partition across the entire disk. The reason for a single partition with RAID on each disk is performance. This is because the RAID driver tries to optimize reads from the array and reads either from the first disk that does nothing or from the disk from which it last requested access to the nearest partition. (The situation is described for RAID1) Since the last area accessed is stored in the information about each RAID array, this algorithm will not behave very efficiently with multiple arrays on a single disk.

You can create an identical layout on a second disk simply by using this command:

sfdisk -d /dev/sda | sfdisk /dev/sdb

Creating a RAID

Creating RAID 1 from two disks

mdadm --create /dev/md1 l 1 -n 2 /dev/sda1 /dev/sdb2

You can then create, for example, a JFS filesystem on this newly created block device, mount the device in a directory structure and start saving data.

Create a RAID 5 that is supposed to have three disks, but we only have two so far:

mdadm --create /dev/md/data -l5 -n3 /dev/sdb3 /dev/sdd3 missing

Adding a disk

Add an unused new disk to the RAID:

mdadm /dev/md0 --add /dev/sdc1

Disc Removal

The disk can be marked as defective by the system or we can mark it manually:

mdadm /dev/md0 --fail /dev/sdc1

Removing a faulty disk from the RAID:

mdadm /dev/md0 --remove /dev/sdb1 /dev/sdc1

If the disk is physically removed from the server it still needs to be removed from the RAID in the system. Because otherwise it will still be considered as used when connected to a PC.

Save configuration

It is a good idea to save the configuration of all RAIDs in a configuration file.

We can pro-scan the system and leave the output directly at the end of the configuration file:

mdadm --detail --scan >> /etc/mdadm/mdadm.conf
Or do the same for one specific disk array:
mdadm --detail /dev/md1 --brief >>/etc/mdadm.conf

It is also a good idea to have the mdadm service enabled:


 systemctl enable mdadm.service

Swap a RAID

How to swap and RAID?

    There are two approaches:
  • put swap on RAID
  • Put swap on RAID1

It is not advisable to put swap on a software RAID because of the possibility of deadlock that can occur when the system runs out of free memory. This has never happened to me personally.

If you are running a software RAID 1 to provide more availability and you don't put the swap on RAID1, then one disk dying will mean a failure. If you swap to RAID 1 it's fine.

I run swap on RAID1.

Checking RAID status

You can get a list and details of all RAIDs by using the following command:

cat /proc/mdstat

Start the data check on the RAID:

echo "check" > /sys/block/md1/md/sync_action

Command to start RAID synchronization:

echo repair >/sys/block/md<#>/md/sync_action

The number of errors can be traced in this file of the respective RAID:

cat /sys/block/md*/md/mismatch_cnt

If your system does not automatically fold the disk array on boot, for example you are booting from a rescue CD, you can manually fold it with this command:

mdadm --assemble --verbose /dev/md1 /dev/sdb1

Stopping the RAID array:

mdadm --stop /dev/mdXi

Remove the RAID information from the disk:

mdadm --zero-superblock /dev/X
Do this if you have invalidated the RAID on the disk. This will delete all RAID metadata and the system will no longer consider the disk to belong to a RAID.

If you are migrating the server to RAID you need to make sure you have updated the configuration in the boot loader, fstab file, and a new initramfs file generated, otherwise you will not be able to successfully boot.

Backup

Be sure to set up a boot loader (typically GRUB) and write it to all disks. Otherwise, you may have a problem booting the server if there is no bootloader anywhere.

Mdadm manual

Video tutorial on how to create a Linux software RAID1:

Other Resources and Topics

For more tutorials on linux servers, check out the ebook.
Next time I will elaborate on the information regarding Logical Volume Management.

Articles on a similar topic

VMware vs Proxmox: performance comparison
GitLab CI/CD: test automation and application deployment
Migrating VPS from VMware to Proxmox
VMware licensing change
Running Microsoft SQL Server on Linux
Backup: the Proxmox Backup Server
Linux as a router and firewall
How to upload a docker image to the Docker Registry
Linux: logical volume management
Running a web application behind a proxy
Mailbox migration
Docker multistage build
Backing up your data by turning on your computer
Podman
Importing Windows into Proxmox virtualization
Docker and PHP mail
Proxmox virtualization
Docker and Cron
Lenovo ThinkPad X1 Carbon: LTE modem EM7544 commissioning
Yocto Project: Build custom operating system for embedded devices
Preparing a Linux server to run a web application in Python
How to address poor file share performance in Docker
How to get started using Docker correctly
Installing Linux on a dedicated HPE ProLiant DL320e server
How to stress test a web application
Why use the JFS filesystem
How to boot from a 4TB drive with GTP using UEFI
Btrfs file system
Raspberry PI
WINE - running Windous programs under Linux
GNU/Linux operating system

Newsletter

If you are interested in receiving occasional news by email.
You can register by filling in your email news subscription.


+