Re: Use RAID1 mirror as backup during dist-upgrade?
Reiner Buehl <firstname.lastname@example.org> wrote:
> I have a Debian Etch system that runs on a RAID 1 software raid system.
> now I would like to upgrade it to Lenny by splitting the mirror off and
> keep one mirror as a backup. Alternatively I could add a third disk as a
> second mirror and split off this one.
> If the upgrade works, I then add in the old disk again. If it fails I'd
> like to be able to rebuild the array using the OLD mirror that I kept.
> Does anybody have a more detailed description on how to do this? I know
> I can fail a drive with mdadm, but my understanding is that the data on
> this this drive can't then be used again after that.
I don't have details to hand but the general principles may help. Be
aware that this is from memory, although I have successfully done it.
* When you fail the mirror, do it in single-user mode after a sync
* Note which physical disk is live (eg /dev/sdl) and failed (sdf)
* Boot your rescue CD and ensure you can differentiate sdl and sdf
* BACKUP your system. Or at the very least, the important bits
* Fail the mirror (/dev/sdl == live, sdf == failed)
* Use mdadm to zero the RAID superblock on sdf ONLY
* Trash/upgrade/reinstall using RAID on /dev/sdl. Ignore sdf
* When you're happy, bring in sdf (possibly with --force)
If it all goes badly...
* Boot from the rescue CD
* Use mdadm to zero the RAID superblocks on sdl AND sdf
* Use mdadm to create a RAID1 wrapper around sdf
* Bring in sdl as a new disk for the RAID1 array
* Update fstab, menu.lst, etc. to reference the new /dev/md* device