Linux Software Raid 1 Recovery Question

[kentborg at borg.org (Kent Borg)]

On Wed, Sep 25, 2002 at 08:35:46PM -0400, ron.peterson at yellowbank.com wrote:
> Ahh, I think I found that which you seek.
> 
> http://www.tldp.org/HOWTO/mini/Boot+Root+Raid+LILO-3.html
> 
> (Don't forget the mini-howto's...)
> 
> Basically, it looks like you need to make a pair of lilo.conf files.
> The boot= line should point to either of /dev/hda or /dev/hdc,
> respectively.  Well, nevermind my blathering on, just read the
> mini-howto.  It's a little out of date, but I think still relevant.

I'm back!  And I have various news.

Item: I was playing with Mandrake 9.0 over the weekend (on my "OK to
trash it" kitchen computer), and I like it.  It also does bootable
software raid 1 and it also only maked my /dev/hda bootable.  However,
I created an /etc/lilo.conf.hdc, changed the "boot=/dev/hda" to
"boot=/dev/hdc", powered down, pulled the first disk, powered up...and
it worked!

Item: My CD burner is on my basement ("not OK to trash it") server and
I used it to burn my Mandrake 9.0 CDs.  It worked.  The next morning,
however, I notice I have an e-mail from my raidcheck script telling me
that I lost my /dev/hda on my basement server.  The good news comes in
two parts: according to the logs it died during a DMA that never
completed yet the computer kept running just fine from the backup
disk, I was using it at the time and never knew anything had gone
wrong!  Cool!  Good news part 2: The fact that /dev/hda died right
when root was messing with /dev/hdb suggests to me that the disk
didn't really die, rather it got confused and will be fixed by power
cycling it.  

The obvious bad news is, of course, if I still don't know how to boot
a Red Hat raid 1 off of /dev/hdc and if /dev/hda really is broken,
power cycling will leave me dead in the water.  

I created an /etc/lilo.conf.hdc.  It is exactly like my original
/etc/lilo.conf, but the boot line is changed.  Where the original
said:

  boot=/dev/md4

The new one says:

  boot=/dev/hdc

I did a "lilo -C /etc/lilo.conf.hdc" and it both warned me that I was
installing it on something other than the first disk and it claimed to
be adding linux.  Cool.  But I have not tried rebooting for fear it
won't work.

Now at work today, a co-worker has a nearly matching machine (RH 7.3
instead of RH 7.2).  His lilo.conf also has an md value for the boot
line.  We created a new /etc/lilo.conf.hdc and put a "boot=/dev/hdc"
in there.  On running lilo we got the same output as I got at home.
But on attempting to boot from his /dev/hdc all we got was an "L".

So, I conclude my basement server will also not know how to boot off
of /dev/hdc and I hope it keeps running for the while.


New question: Does anyone know the difference between a
"boot=/dev/md4" and a "boot=/dev/hdc"?  And how does the md-version
ever work?


Thanks,


-kb, the Kent who is pleased that raid 1 really can keep running, that
hdc can boot in some circumstances, he just doesn't know how to get RH
to raid boot from /dev/hdc.