Software RAID 1 with multiple arrays: mounts not honoured after reboot

Gerard Lally · 02-17-2012, 07:20 PM

I have two disks - dev/sda and /dev/sdb.

I follow the RAID readme and create a number of RAID 1 arrays as follows:

Primary - sda1+sdb1 = /dev/md0 - / (ext4)
Logical - sda5+sdb5 = /dev/md1 - swap
Logical - sda6+sdb6 = /dev/md2 - /tmp (ext2)
Logical - sda7+sdb7 = /dev/md3 - /usr (jfs)
Logical - sda8+sdb8 = /dev/md4 - /var (jfs)
Logical - sda9+sdb9 = /dev/md5 - /home (jfs)
Logical - sda10+sdb10 = /dev/md6 - not mounted - to be used at a later stage for logical volumes

I follow the instructions to the letter but after rebooting only / is mounted.

In addition, the device numbering for the RAID arrays is now /dev/md0, /dev/md122, /dev/md123, /dev/md124, /dev/md125, /dev/md126, and /dev/md127. Needless to say only / is mounted, because /etc/fstab is set up for /tmp on /dev/md2, /usr on /dev/md3, etc.

What am I doing wrong?

wildwizard · 02-18-2012, 01:32 AM

Actually the README_RAID.TXT is out of date.

Some important notes on how raid really works with the more recent kernels (like last 3 years)

1. Partition type fd is obsolete don't bother with it and it can cause problems later. Use partition type da instead.
2. You must use the initrd to boot or the raid devices wont be detected, see the section under "Using the generic kernel" for the details.

https://raid.wiki.kernel.org/

EDIT: Your problem may also be related to the use of logical drives, not sure if RAID ever worked with those.

Richard Cranium · 02-18-2012, 02:36 AM

Plus it would be a lot easier to make one large RAID1 partition and use it as a physical volume for LVM.

Well, you could make a small RAID1 partition for /boot and put the rest of the drive on another larger RAID1 partition. For example...

Code:

bash-4.1$ cat /proc/mdstat
Personalities : [linear] [raid0] [raid1] [raid10] [raid6] [raid5] [raid4] [multipath] 
md1 : active raid1 sdd3[0] sde3[2]
      142128606 blocks super 1.2 [2/2] [UU]
      
md2 : active raid1 sdd1[0] sde1[1]
      248964 blocks super 1.2 [2/2] [UU]
      
md3 : active raid1 sdc2[0] sda2[1]
      624880192 blocks [2/2] [UU]
      
unused devices: <none>
bash-4.1$ sudo /sbin/pvs
  PV         VG      Fmt  Attr PSize   PFree  
  /dev/md1   mdgroup lvm2 a-   135.53g  79.50g
  /dev/md3   mdgroup lvm2 a-   595.91g 403.66g
bash-4.1$

(sdc and sda also have a small partition set aside to be used as a RAID1 /boot partition in case sdd and sde fail.)

/usr/share/mkinitrd/mkinitrd_command_generator.sh will be your undocumented friend if you go this route.

Gerard Lally · 02-19-2012, 08:57 AM

Quote:

Originally Posted by wildwizard

Actually the README_RAID.TXT is out of date.

Some important notes on how raid really works with the more recent kernels (like last 3 years)

1. Partition type fd is obsolete don't bother with it and it can cause problems later. Use partition type da instead.
2. You must use the initrd to boot or the raid devices wont be detected, see the section under "Using the generic kernel" for the details.

https://raid.wiki.kernel.org/

EDIT: Your problem may also be related to the use of logical drives, not sure if RAID ever worked with those.

Thanks for the updated info and the link. I think the README should mention
that partition type fd is deprecated.

I have a feeling logical drives are indeed the problem. For the moment I have
done what Richard Cranium suggested and gone with a single RAID partition and
multiple Logical Volumes. Perhaps this is the best way after all, although in
the future I would like to try "disposable" mounts like /usr on RAID0 and
/home on RAID1.

Gerard Lally · 02-19-2012, 09:01 AM

Quote:

Originally Posted by Richard Cranium

Plus it would be a lot easier to make one large RAID1 partition and use it as a physical volume for LVM.

Thanks. I have gone down this route and it's working fine. There seems to be no performance penalty attached to LVM on top of software RAID.