GRUB 1.99-21 can't boot from second drive of MD RAID

[Philip Pemberton]

Hi,

I have an Ubuntu 12.04 Server system set up with a combination of 
software RAID-1 (md) and LVM:

$ sudo parted /dev/sda print
Model: ATA WDC WD20EFRX-68E (scsi)
Disk /dev/sda: 2000GB
Sector size (logical/physical): 512B/4096B
Partition Table: gpt

Number  Start   End     Size    File system  Name         Flags
 1      1049kB  2097kB  1049kB               bios_grub_0  bios_grub
 2      2097kB  1076MB  1074MB               raid_boot_0
 3      1076MB  2000GB  1999GB               raid_lvm_0

The other drive is partitioned identically, except the names end "_1".

"raid_boot" is the /boot partition, an ext2fs RAID-1 created with "md". 
Ubuntu calls this "md0".
"raid_lvm" is another md RAID-1 which contains an LVM volume group.


I've just replaced both disks (one was faulty and the other wasn't far 
from it), cloned the partitions and installed GRUB. My problem is:

  - With both disks installed, the system boots fine
  - With only sda installed (simulating a failure of sdb), the system 
boots fine
  - With only sdb installed (simulating a failure of sda), the system 
boots to the GRUB menu, then tries to boot and reboots the system 
instead. If I exit the GRUB menu to a GRUB command line, any command 
(even 'help' or 'ls') causes GRUB to spontaneously reboot the PC without 
even attempting to boot Linux.

Curiously, if both drives are installed and I instruct the BIOS to boot 
from sdb, GRUB works fine.

I initially installed GRUB using 'grub-install /dev/sda' followed by 
'grub-install /dev/sdb'. I've also tried reinstalling GRUB (apt-get 
purge grub-common; apt-get install grub-pc) but this had no effect.

Can anyone tell me what I'm doing wrong here? All I want is for my 
server to boot even if one of the two RAID-1 mirrored drives has failed.

Thanks,
--
Phil.
address@hidden
http://www.philpem.me.uk/