CentOS - Dec 2010 - Fiddling with software RAID1 : continue working with one of two disks failing?

Hi,

I'm currently experimenting with software RAID1 on a spare PC with two 
40 GB hard disks. Normally, on a desktop PC with only one hard disk, I 
have a very simple partitioning scheme like this :

/dev/hda1  80 MB    /boot   ext2
/dev/hda2   1 GB    swap
/dev/hda3  39 GB    /       ext3

Here's what I'd like to do. Partition a second hard disk (say, /dev/hdb)
with three partitions. Setup RAID1 like this :

/dev/md0   80 MB    /boot   ext2
/dev/md1    1 GB    swap
/dev/md2   39 GB    /       ext3

I somehow managed to get this far. Here's what I have :

[root at raymonde ~]# fdisk -l /dev/hda

Disk /dev/hda: 41.1 GB, 41110142976 bytes
255 heads, 63 sectors/track, 4998 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes

    Device Boot   Start    End      Blocks   Id  System
/dev/hda1   *        1     11       88326   fd  Linux raid autodetect
/dev/hda2           12    134      987997+  fd  Linux raid autodetect
/dev/hda3          135   4998    39070080   fd  Linux raid autodetect

[root at raymonde ~]# fdisk -l /dev/hdb

Disk /dev/hdb: 41.1 GB, 41110142976 bytes
16 heads, 63 sectors/track, 79656 cylinders
Units = cylinders of 1008 * 512 = 516096 bytes

    Device Boot   Start    End      Blocks   Id  System
/dev/hdb1   *        1    156       78592+  fd  Linux raid autodetect
/dev/hdb2          157   2095      977256   fd  Linux raid autodetect
/dev/hdb3         2096  79656    39090744   fd  Linux raid autodetect

During install, my /dev/md1 and /dev/md2 got somehow mixed up, which 
doesn't really matter :

[root at raymonde ~]# cat /etc/fstab
/dev/md1        /               ext3    defaults        1 1
/dev/md0        /boot           ext2    defaults        1 2
tmpfs           /dev/shm        tmpfs   defaults        0 0
devpts          /dev/pts        devpts  gid=5,mode=620  0 0
sysfs           /sys            sysfs   defaults        0 0
proc            /proc           proc    defaults        0 0
/dev/md2        swap            swap    defaults        0 0

I wasn't sure where to install GRUB, so I chose /dev/md0.

I was wondering if this setup theoretically enabled me to continue 
working with one disk failure. So I tried unplugging the power cord of 
one of my hard disks... which resulted in a "GRUB Disk Error" on boot.

Question : is there a way to still run the system with either of the two 
disks "damaged" (in this case : unplugged)? And if so, how would I
have
to go about it in my setup?

Cheers from the freezing South of France,

Niki

On 12/4/10 11:34 AM, Niki Kovacs wrote:
>
> I wasn't sure where to install GRUB, so I chose /dev/md0.
Grub doesn't know anything about raid.  It only works because each component
of
a RAID1 looks just like a non-raid filesystem.  You should install grub on the 
master boot partition of both member disks.
> I was wondering if this setup theoretically enabled me to continue
> working with one disk failure. So I tried unplugging the power cord of
> one of my hard disks... which resulted in a "GRUB Disk Error" on
boot.
> Question : is there a way to still run the system with either of the two
> disks "damaged" (in this case : unplugged)? And if so, how would
I have
> to go about it in my setup?
Yes, raid1 isn't bothered at all by missing members.  You just have to
install
grub on the underlying disks as though you did not have raid. There can be some 
differences in the way ide/sata/scsi controllers handle a disk failure - and 
IDE's often hang the controller if one of two disks on the cable fails. 
Normally the first disk will always work if the 2nd fails, but if the first disk
fails both the bios and linux have to see the same shift in positions - so even 
though you are installing grub on what it calls hd1, if it boots from there it 
will see its root at hd0,1.   Worst case - just be prepared to boot a rescue 
disk and reinstall grub if your first disk fails.

When you replace the drive and want to rebuild the mirror, just:
mdadm --add /dev/md? /dev/sd??

-- 
   Les Mikesell
    lesmikesell at gmail.com

At Sat, 04 Dec 2010 18:34:26 +0100 CentOS mailing list <centos at
centos.org> wrote:
> 
> Hi,
> 
> I'm currently experimenting with software RAID1 on a spare PC with two 
> 40 GB hard disks. Normally, on a desktop PC with only one hard disk, I 
> have a very simple partitioning scheme like this :
> 
> /dev/hda1  80 MB    /boot   ext2
> /dev/hda2   1 GB    swap
> /dev/hda3  39 GB    /       ext3
> 
> Here's what I'd like to do. Partition a second hard disk (say,
/dev/hdb)
> with three partitions. Setup RAID1 like this :
> 
> /dev/md0   80 MB    /boot   ext2
> /dev/md1    1 GB    swap
> /dev/md2   39 GB    /       ext3
> 
> I somehow managed to get this far. Here's what I have :
> 
> [root at raymonde ~]# fdisk -l /dev/hda
> 
> Disk /dev/hda: 41.1 GB, 41110142976 bytes
> 255 heads, 63 sectors/track, 4998 cylinders
> Units = cylinders of 16065 * 512 = 8225280 bytes
> 
>     Device Boot   Start    End      Blocks   Id  System
> /dev/hda1   *        1     11       88326   fd  Linux raid autodetect
> /dev/hda2           12    134      987997+  fd  Linux raid autodetect
> /dev/hda3          135   4998    39070080   fd  Linux raid autodetect
> 
> [root at raymonde ~]# fdisk -l /dev/hdb
> 
> Disk /dev/hdb: 41.1 GB, 41110142976 bytes
> 16 heads, 63 sectors/track, 79656 cylinders
> Units = cylinders of 1008 * 512 = 516096 bytes
> 
>     Device Boot   Start    End      Blocks   Id  System
> /dev/hdb1   *        1    156       78592+  fd  Linux raid autodetect
> /dev/hdb2          157   2095      977256   fd  Linux raid autodetect
> /dev/hdb3         2096  79656    39090744   fd  Linux raid autodetect
> 
> During install, my /dev/md1 and /dev/md2 got somehow mixed up, which 
> doesn't really matter :
> 
> [root at raymonde ~]# cat /etc/fstab
> /dev/md1        /               ext3    defaults        1 1
> /dev/md0        /boot           ext2    defaults        1 2
> tmpfs           /dev/shm        tmpfs   defaults        0 0
> devpts          /dev/pts        devpts  gid=5,mode=620  0 0
> sysfs           /sys            sysfs   defaults        0 0
> proc            /proc           proc    defaults        0 0
> /dev/md2        swap            swap    defaults        0 0
> 
> I wasn't sure where to install GRUB, so I chose /dev/md0.
No, you install GRUB (or alternitively, lilo) on *both* /dev/hda AND
/dev/hdb, with your root /dev/hda1.  Neither grub (nor lilo) know about
RAID (ditto for the BIOS).  This is not a problem, since the *elements*
of a RAID1 set look like 'normal' partitions with normal file systems
on them.  You want grub to be in the MBR of /dev/hda -- duping it in
/dev/hdb's MBR allows you to boot (with degraded RAID sets) from
/dev/hdb (cabled & jumpered to be /dev/hda) in the event /dev/hda dies. 
> 
> I was wondering if this setup theoretically enabled me to continue 
> working with one disk failure. So I tried unplugging the power cord of 
> one of my hard disks... which resulted in a "GRUB Disk Error" on
boot.
> 
> Question : is there a way to still run the system with either of the two 
> disks "damaged" (in this case : unplugged)? And if so, how would
I have
> to go about it in my setup?
Yes, see above.

Minor performance nit: Doing RAID with two IDE disks on the *same*
controller is not going to buy you anything in terms of performance. I
suspect this is just experimental, mostly to get the feel for how to set
things up, so this is not a major issue.
> 
> Cheers from the freezing South of France,
> 
> Niki
> 
> 
> 
> 
> 
> _______________________________________________
> CentOS mailing list
> CentOS at centos.org
> http://lists.centos.org/mailman/listinfo/centos
> 
>
-- 
Robert Heller             -- 978-544-6933 / heller at deepsoft.com
Deepwoods Software        -- http://www.deepsoft.com/
()  ascii ribbon campaign -- against html e-mail
/\  www.asciiribbon.org   -- against proprietary attachments