[SOLVED] FC 10: RAID failure test: grub cannot find boot device

jot · 03-25-2009, 05:29 AM

Trying to setup a software RAID on a system with 2 identical SCSI drives (contain /boot) and 2 identical IDE drives (for data). OS version: FC 10, which I can normally boot from harddisk via grub.

The installation was rather smooth, the RAID mirroring appears to work fine, which I have checked with

Code:

mdadm --detail /dev/md0

(there are two more RAID devices: md1 and md2, and all look good).

The Problem:
Now I tried to simulate a harddisk failure by unplugging one of the SCSI drives. The result, I cannot boot at all (with either SCSI unplugged, or even when either IDE drive was unplugged) and get the message "Error 21: Selected disk does not exist" instead. Plugging the disk back in solves the error.

After checking in a number of places, I discovered one odd thing: the mapping of devices from RAID device to actual drive/partion differs from the installation. During the installation things were setup this way:

Code:

md0 --> sdc1/sdd1 (/boot) (SCSI drives)
md1 --> sdc3/sdd3 (/)     (SCSI drives)
md2 --> sda1/sdb1 (/data) (IDE drives)

But after starting FC10 the decribed mapping seems to have changed! Now the above became this instead:

Code:

md0 --> sda1/sdb1 (/boot) (SCSI drives)
md1 --> sda3/sdb3 (/)     (SCSI drives)
md2 --> sdc1/sdd1 (/data) (IDE drives)

Question:
Could this "re-mapping" cause the described grub problem? Is this normal anyway? What could I try?

Any hint/help is appreciated! Thanks

mostlyharmless · 03-25-2009, 11:51 AM

You have to make sure that grub is installed on both mirrors. I don't know whether FC10 does that for you automatically.

In addition, your bios should be set up to boot from your one mirror as your first boot device and the other mirror as your second boot device. The fact that you saw a change in booting with the IDE drive unplugged indicates the IDE drive is probably higher in the bios boot order than the SATA, which is usually the default.

As far as the mapping changing after booting, that isn't necessarily an issue, but it does mean that when you install grub that you have to make sure you're installing to the correct device. I recommend you use a "native" install from something like SuperGrub rather than using grub-install from your installed system to avoid confusion. Tab completion at the grub> prompt is your friend.

jot · 03-26-2009, 05:51 AM

Hi Mostlyharmless

Summarizing the good news - RAID recovery after disk failure works now! Thank you!!

Answering your points:
1) Grub was installed on both mirrors. I also don't know if FC 10 does that automatically, I had already done it manually.

2) My system's BIOS could not be set up to boot from one mirror as the first boot device and the other mirror as the second boot device. This is because only the IDE drives were listed as boot order options. The SCSI drives were not offered. So I did not touch this.

3) Regarding the change in drive order after whichever disk was gone: the BIOS' drive order turned out to be an unexpected third version compared to the two reported in my first message:

Reported by Super-Grub:

Code:

hd0 --> sda (SCSI drive)
hd1 --> sdb (IDE drive)
hd2 --> sdc (IDE drive)
hd3 --> sdd (SCSI drive)

If you plugged out whichever SCSI drive it became:

Code:

hd0 --> sda (SCSI drive)
hd1 --> sdb (IDE drive)
hd2 --> sdc (IDE drive)

After you plugged out whichever IDE drive:

Code:

hd0 --> sda (SCSI drive)
hd1 --> sdb (IDE drive)
hd2 --> sdc (SCSI drive)

This revelation was the break-through. By default, the FC 10 installer had configured hd3 as the default boot device. So I had followed that. But obviously, this must fail if one disk is missing considering the findings above.

Solution: set up grub to use hd0 instead of hd3 as boot device (on both mirrors). This did the trick.

4) Installing grub to the correct device: even the commands of the "normal" grub (FC10 had grub 0.97) ensured you hit the right partition. It checked for the boot files and reported an error if one didn't point to the correct partition.

5) Super-Grub saved the day, mainly because it revealed the the "hdx --> sdxx" type of mapping. Good advice!

--
Additional information:
Booting after SCSI disk failure worked without human intervention. However, unplugging one of the IDE drives generated a BIOS error, forcing me to press F1 first in order to boot. But this might just be my particular BIOS.

Once again, thanks!

Jot

mostlyharmless · 03-26-2009, 10:11 AM

You're welcome, congratulations.