I will not pretend to be an expert, but.,.. have you tried rebuilding the initrd? It could be that mdadm.conf gets added to the initrd and is read during boot. so that would explain how to read mdadm.conf without having access to it :)
if memory servers, update-initramfs will do everything you need -nick On Thu, Apr 29, 2010 at 8:18 PM, [email protected] <[email protected]> wrote: > Good evening all, > Long time list member, but quasi first-time poster. > > I believe I screwed up my file server, it's non-operational at the moment, > but 100% recoverable, I just don't know how. > > Here's my setup: > > * Debian 5 on amd64 > * 5x 1TB drives (mostly 7200.12's) > ** all partitioned as > *** 1GB (sd*1) > *** 999GB (sd*2) > ** sda1, sdb1 = software raid 1 for /boot (others sd*1 are free or swap) (md0) > ** sd*2 = software raid 5 for LVM2 (md1) > > I recently grew the array from 4 to 5 disks, resized my volume group, resized > my logical volumes and accompanying FS's. Last time I had done it (3 to 4 > disks), it went fine. It worked great post reboot. However, this time, I > can't get past an initramfs shell. > > This time, I forgot to edit /etc/mdadm/mdadm.conf and tell it md1 has 5 > devices, and not 4. I believe I hadn't rebooted since (a while back), and got > caught last night when I finally did. > > So, when I boot, I get stuck in an initramfs shell. The previous console > messages are about software raid of course: > > mdadm: /dev/md0 was started with 2 drives > mdadm: no devices found for /dev/md1 > > Wait, what? > > Here's the thing, I can do this little mdadm/lvm dance leading to me being > able to mount all my logical volumes. If memory serves me right, it went > something like this: > > mdadm --assemble --scan /dev/md1 /dev/sda2 /dev/sdb2 /dev/sdc2 /dev/sdd2 > /dev/sde2 > lvm vgchange FirstGroup -a y > > At this point, /proc/mdstat says /dev/md1 is fine with all its 5 disks, > lvscan will tell me my volumes are all active. I can mount in r/w. I took the > liberty of editing /etc/mdadm/mdadm.conf to reflect 5 instead of 4 drives, > but that doesn't make it boot any better. Nor should it, really. The system > can't even read mdadm.conf because it's not even mounted, so I don't see how > that file has any effect on the issue. But perhaps there's a side effect, > some background process that does I don't know what. > > I re-read the documentation I used to grow the array > (http://scotgate.org/2006/07/03/growing-a-raid5-array-mdadm/), and there are > no last critical steps that sort of commit the changes survive a reboot, my > process was the same as last time, except editing mdadm.conf. > > So, there's something wrong with initrd or something of the sort. > > I'm not pulling my hair out like I did last night, I know it's 100% > recoverable, but it's just kind of annoying to have a system sit there not > knowing what to do. > > Does anyone have a clue? Reinstalling *is* a possibility, given my partition > scheme, but would prefer not to at this point in time. I'll be googling in > the mean time. > > Thanks, > Lawrence > > Sorry if this post is a little verbose, figured i'd be as clear as possible. > _______________________________________________ > mlug mailing list > [email protected] > https://listes.koumbit.net/cgi-bin/mailman/listinfo/mlug-listserv.mlug.ca > _______________________________________________ mlug mailing list [email protected] https://listes.koumbit.net/cgi-bin/mailman/listinfo/mlug-listserv.mlug.ca
