On Mon, 2006-08-28 at 19:12 -0500, Jeff Vian wrote: > On Mon, 2006-08-28 at 13:20 -0700, Timothy Alberts wrote: > > I am trying to use the Installation program with FC5 to setup a new > > system with 2 SATA drives. The motherboard has a 'hardware raid' but > > after a few years of monkeying with this system and drivers, software > > raid is the correct way to go. However, I am installing FC5 on new hard > > drives. The installation finishes, I reboot and the boot loader > > sometimes isn't found at all, or is found, but doesn't have a kernel to > > boot? The times that the installer puts a boot loader in without a > > kernel, I type in the parameters: > > > > root(hd0,0) > > kernel... > > initrd... > > > > and the system boots for the first time and I get through the 'first > > boot stuff (setting soundcard etc)', then I boot again (if I'm lucky) > > and the system boots. However, at this point is the real symptom of the > > problem. The raid never has both drives in each partition. The system > > is configured to have /boot as /dev/md0 (/dev/sda1 + /dev/sdb1)... / > It appears you have used raid 1. > > You need to make sure grub was installed on the MBR of the drive that is > booting, and it needs to be on the boot sector of both drives. > One way would be to boot to rescue mode and then do a grub-install on > the both drives. (or do the same once booted to normal mode.) > > After that regardless of which drive is booted it should work. > Jeff, Thank you for responding... Yes I'm setting up RAID1. It is apparent that I need to make sure that GRUB is installed in the MBR, I guess I'm curious why the installer (anaconda) isn't doing this? What is the point of an installer that installs an OS that can't boot... It says in the anaconda gui during boot loader configuration that the boot loader will be installed on /dev/md0. So I guess from that I should have expected the boot loader isn't in the MBR. Wouldn't it be better if anaconda took care of installing GRUB in the MBR of all drives during installation or are the variables of what the user is installing too much to 'hash' out? (ha! that's funny) > > > as /dev/md1 (/dev/sda2 + /dev/sdb2)... swap as /dev/md2 (/dev/sda3 > > + /dev/sdb3) except the system is only using sda in partion md0 and sdb > > in partion md1 and sda in partion md2. Never are both drives in their > > assigned partitions. > > > How do you know this? mdadm will tell you what the config is, and if > the raid is working correctly it will not matter which partition is > shown as the active one, the other will have a mirrored copy. > I know this because when I type 'mdadm --detail /dev/md0' (or 1 or 2), the report tells me that only 1 partition is being used. Also 'cat /proc/mdstat' tells me there is only 1 partition in each of the raid devices. Additionally, when I type 'mdadm /dev/md1 -a /dev/sda2' the missing partition is added to mdadm and it tells me it is re-building the raid. I'm certain that mdadm says more than just the 'config', it tells the current status because it gets regular updates when a raid is being rebuilt. > > So my first question, I've looked through the mailing list archives back > > a few months and I see no related issues. Is anyone aware of bugs with > > the software raid installation in FC5? > > > > I guess my second question is a general 'help'? > > > > Thank you in advance. > > > > On top of all this, I submitted a bug report to bugzilla #204420 which is the result of what happened when I tried making an IDE hard drive A the main disk in the raid and have /dev/sda as the mirror and /dev/sdb as the backup.