On Sat, Oct 23, 2004 at 07:00:47PM +0300, Fabian Fagerholm wrote: > On Sat, 2004-10-09 at 21:16 +0300, Fabian Fagerholm wrote: > > I tried the following (3 x HDD): > > > > In partman, set up the three disks to have identical partition tables: > > > > * A small primary partition with boot flag set. > > * A large logical partition. > > * A small logical partition. > > > > I then took all the small primary partitions and created a RAID1 out of > > them (/dev/md/0). I put /boot on this (ReiserFS). > > > > Then, I took the large logical partitions and created a RAID5 out of > > them (/dev/md/1). The small logical partitions were marked as swap. > > > > I manually ran pvcreate and vgcreate on /dev/md/1. Then, using partman, > > I added logical volumes for /, /usr, /var, /home and /tmp. Still using > > partman, I assigned the mount points and chose to format all logical > > volumes as ReiserFS. Partman finished its work. > > > > However, as a result of this, GRUB install was disabled, and LILO was to > > be used instead. LILO install didn't work; I tried specifying the first > > disk's MBR manually but LILO failed whatever I tried. > > > > Am I right in assuming that both LILO and GRUB should work, since they > > can read the kernel image and initrd from one of the disks in the RAID1 > > array that comprises /boot, and the initrd contains all neccessary > > things to access / on RAID5 and LVM? Ie. is it enough that /boot is on > > RAID1 or plain partition with no LVM? > > I just tried with a cd image downloaded from the following url: > http://cdimage.debian.org/pub/cdimage-testing/daily/i386/current/sarge-i386-netinst.iso > The build date was 22-Oct-2004. > > It's now possible to mark a RAID array for use with LVM. However, if I > do the above, the installer disables GRUB, and LILO refuses to install > to the default option ("/dev/md1: Master Boot record") which seems to be > wrong for this kind of setup. I expected to see /dev/md0 as the default. > > I tried installing on /dev/md0 > and /dev/ide/host0/bus0/target0/lun0/disc, but I just got the error > messages "The path /dev/md0 does not represent a partition of a hard > disk device" and "Running /sbin/lilo failed with error code "1"". > > However, it does seem that it's impossible to do what I wanted. When I > adjusted the scenario, I got everything to work. I split up the three > disks into three partitions: > > * A primary partition with the boot flag set, > * a large logical partition and > * a small logical partition. > > I then marked the primary partitions for RAID use, the large logical > partition for RAID use, and the small logical partition as swap. > > Then, I assembled a RAID1 with the three primary partitions, and a RAID5 > with the three large logical partitions. After that (and here's the key > difference), I chose to format the RAID1 as ReiserFS and mount it as /, > and created an LVM volume group on the RAID5, which I populated with > logical volumes for /usr, /var, /home and /tmp. All partitions were > ReiserFS. > > This worked like a charm, and GRUB was nicely installed. > > So: it should be clearly documented that it is not possible, in any way, > to put your / on RAID5, not with GRUB and not with LILO. My assumption > that it would be enough for the boot loader to access the initrd from > one of the disks in the RAID1 array (with /boot on it) to be able to > access / on RAID5 was apparently not correct -- or at least it doesn't > work in d-i when installing the boot loader. Perhaps it would work if > booting off another medium and skipping the installation of a boot > loader in d-i.
I think you struct the same lilo-installer bug that I did yesterday, when I was doing some RAID1 experiments. I told it to use /dev/discs/disc0/disc and it installed fine, iirc. Joey said it's a known bug. -- linux.conf.au 2005 - http://lca2005.linux.org.au/ - Birthplace of Tux April 18th to 23rd - http://lca2005.linux.org.au/ - LINUX Canberra, Australia - http://lca2005.linux.org.au/ - Get bitten! -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]