On Friday 24 November 2006 20:06, Aleksandar Milivojevic wrote: > Kevan Benson wrote: > > Unless it's new in 4.4, that's not the case. Grub can't undertand the md > > devices, and has to be installed on the disk geometry individually, as > > your link suggests, which anaconda doesn't do. I hear RHEL 5 fixes this > > though. > > Installing Grub into first partition is definitely the way its done in > CentOS 4.4, if /boot is on RAID-1 md device. Not sure if it is > something new (I haven't noticed it before either). Actually, even if > you attempt to force installation of Grub into MBR (for example in > ks.cfg), Anaconda is going to ignore it and install into partition (of > course, only if /boot is on RAID-1 md device). I've noticed this when > installing on some used drives that had LILO in MBR. After > installation, the machine wouldn't boot. BIOS was loading LILO from > previous installation (which was in MBR), instead of Grub from current > installation (which was in bootable partition). Ah, so anaconda installs onto the first partition of each drive when boot is software RAID-1? That would explain much of what I saw. I usually enountered this on machines that previouslyt had lilo installed (as you did). It also explains why anaconda didn't error. > BTW, while I'm mentioning it, does anybody know a good way (or tool) to > evict (clear, zero-out) boot loader from MBR? Preserving partition > table of course. Looks like lilo -u can do it, but that's not much help unless oyu have it installed (and I'm not sure whether or not it requires a saved copy of the MBR from beforehand). The windows fdisk (as mentioned elsewhere in this thread) can do it, but it might just be easier to install grub on to each drive's MBR. -- - Kevan Benson - A-1 Networks