Setting up a software RAID array
There are various forms of RAID: via a hardware RAID controller, via "fake RAID", and "software RAID" using
mdadm, which is Linux-only. These instructions only discuss the last form of RAID. Also, it only discusses how to setup a RAID array for arbitrary storage. It is possible to have one's system root /, or /var, or swap, or even one's /boot, on a RAID array. See Setting up disks manually for more details about doing any of that.
There are several "levels" of RAID to choose between:
- RAID0 essentially just glues two devices together, making a larger virtual drive. Reads and writes are "striped" between the drives for speed improvements. (That is, your hardware may read from, or write different data to, multiple devices in parallel.) A "device" here is usually a partition of a hard drive.
- RAID1 "mirrors" writes to two devices, for improved safety. Then if one of the devices fails, the data will still be available on the other.
- RAID5 is similar to RAID1, but it uses three devices and provides the space of two of them. The data will be preserved as long as any two of the three devices continue to work.
There are other RAID levels as well. Here is more explanation of their differences.
- Your /boot partition should either not be on RAID, or else be on a RAID1 array, with no further layers of encryption or LVM. (Alpine's default bootloader extlinux can't handle either. Grub2 can handle /boot being on LVM.) The usual practice is to create a small (32--100 MB) partition for /boot. That can be a mirrored (RAID1) volume, however this is just for post-init access. That way, when you write a new kernel or bootloader config file to /boot, it gets written to multiple physical partitions. During the pre-init, bootloader phase, only one of those partitions will be used (and it will be mounted read-only).
- You can put swap on a RAID0 volume, but there doesn't seem to be any good reason to do so. The Linux kernel already knows how to stripe several swap partitions. So you can just devote multiple ordinary (not-residing-on-RAID) partitions to swap, and get the same effect. The downside from doing either of these things is that when one of your disk fails, the system will go down. For better reliability, you can create a mirrored (RAID1) volume and put swap there. This will let your system keep running even when one of the disks fails.
- All partitions in a RAID array should be the same size.
- Don't ever mount just one of the devices in a RAID1 array, even though it "has the same data" as the other. If you mount it r/w, then---even if you don't explicitly write anything to the device---it may get out of sync with the unmounted device, for example because the journal on its filesystem has been updated. If you ever subsequently mount the other device, or the two of them together, your data will likely become corrupted. If you have to do this, make sure you mount your device r/o. Better yet, abandon the device you didn't mount. Zero out its RAID headers, and tell
mdadmthat that device has failed. Then you can if you like treat it as a new disk, which you can add as a replacement to your (now degraded) original RAID array.
- A mirrored RAID array (level 1 or 5) protects you against hardware failure. It doesn't protect against
rm -rf /, software errors, exploits, earthquakes, fire. Don't rely on RAID as a backup strategy.
- Running a mirrored RAID only provides one line of defense against drive failures. It doesn't license you to stop thinking about them. If a device in a RAID 1 starts failing and you aren't aware of it, your data will end up just as silently corrupted as it would be if you were running one drive. You have to watch your logs.
This document was updated for Alpine 2.4.6.
Loading needed modules
Start with loading the raid1 kernel module:
Add it to /etc/modules so it gets loaded during next reboot:
Creating the partitions
I will use /dev/sda and /dev/sdb in this document but your devices may be different. To find what disks you have available, look in /proc/partitions.
Create the partitions using fdisk.
I will create one single partition of type Linux raid autodetect. Use n in fdisk to create the partition and t to set type. Logical volumes will be created later. My partition table looks like this ('p' to print partition table):
Device Boot Start End Blocks Id System /dev/sda1 1 17753 8388261 fd Linux raid autodetect
Use w to write and quit. Do the same with your second disk.
Mine looks like this:
Device Boot Start End Blocks Id System /dev/sdb1 1 17753 8388261 fd Linux raid autodetect
Alternately, if your disks are the same size (as they should be, see above) you can copy the partition table from one to the other like this:
Setting up the RAID array
Install mdadm to set up the arrays.
Create the array.
Monitoring sync status
You should now be able to see the array syncronize by looking at the contents of /proc/mdstat.
~ # cat /proc/mdstat Personalities : [raid1] md0 : active raid1 sdb1 sda1 8388160 blocks [2/2] [UU] [=========>...........] resync = 45.3% (3800064/8388160) finish=0.3min speed=200003K/sec unused devices: <none>
You don't need to wait til it is fully syncronized to continue.
Create the /etc/mdadm.conf file so mdadm knows how your raid setup is:
To make sure the raid devices start during the next reboot run:If you're not running Alpine from a hard disk install, use as usual to save your configuration changes to your removable media.
The raid device /dev/md0 is now ready to be used with LVM or mkfs.
More Info on RAID
These resources may be helpful:
- Arch wiki page on RAID
- Arch wiki page on RAID and LVM
- Arch wiki page on Converting an existing system to RAID Gentoo wiki page on the same
- http://anonscm.debian.org/gitweb/?p=pkg-mdadm/mdadm.git;a=blob_plain;f=debian/FAQ;hb=HEAD Debian MDADM FAQ
- http://tldp.org/FAQ/Linux-RAID-FAQ/x37.html Linux RAID FAQ