[ale] RAID mirror boot nightmare

[Phil Turmel]

Hi Bob!

On 07/09/2012 06:21 PM, Bob Toxen wrote:
> I'm having problems getting my RAID mirror booting again in
> CentOS 5.8 64-bit after GNU's gparted trashed my partition table when
> one of the new Seagate 1 TB SATA disks failed completely.
> 
> Booting causes the kernel to complain of an invalid root partition.
> I can't figure out what is invalid about it.  FSCK is happy.
> 
> I rebuilt the partition table with "CD Rescue" v2.7.1 bootable CD
> (wonderful stuff) and reinstalled my existing GRUB to the MBR.
> 
> 
> Unfortunately, CentOS 5.8 will no longer recognize my md3 root partition
> (still running in degrated mode since I've not yet replaced the failed
> disk).  The CentOS kernel does see the RAID partitions but just doesn't
> want to use one for /!!!
> 
> I CAN boot the "CD Rescue" disk and tell it to find and boot an
> existing OS on the hard disk and it will bring up my CentOS 5.8 on
> md3 (before I did mdadm -S /dev/md3 and used fdisk to change its
> partition type from RAID to Linux)
> or will boot up md4 (though apparently with a newer 3.x kernel from the CD).
> 
> However, no matter what I do I cannot get my CentOS 5.8 to boot the RAID
> mirror.  I also have the root and /boot file systems duplicated to
> md4 and md1 respectively.
> 
> I tried
> 
>   mdadm -S /dev/md3
>   mdadm -S /dev/md1
>   mdadm --create -n 2 /dev/md3 --level=mirror /dev/sda6 missing
>     "/dev/sd6 appears to be part of an existing RAID; do anyway" y
>   mdadm --create -n 2 /dev/md1 --level=mirror /dev/sda2 missing
>     "/dev/sd2 appears to be part of an existing RAID; do anyway" y
> 
> but when I try to boot CentOS 5.8 with "root-md3" it again fails.
> 
> However, booting the CD Rescue and selecting "find and boot from hard
> disk" happily boots either RAID partition.
> 
> 
> Please PLEASE give me some ideas to fix this.  $50 to the first right
> answer, REALLY!  I'm desperate.
> 
> Nothing on the file systems themselves, including  the /boot/initrd*
> file and kernels, appear to be damaged.  systems themselves)
> 
> 
> Since FSCK is happy all I can think of is damage to the md superblock at
> the end of the physical partitions.  I wonder if the CD Rescue kernel
> (being newer than the CentOS kernel) created to new a version of the
> mdsuperblock.  However, my recreating the RAID mirror under CentOS 5.8
> should have fixed it, shouldn't it?
> 
> HELP!  $50 prize.  Really!

Try adding "md-mod.start_dirty_degraded=1" to your kernel entry in GRUB.

> Bob Toxen
> bob at VerySecureLinux.com
> transam at VerySecureLinux.com
> 404-216-5100 Cell (except for the next half hour)