Re: Can not start md0 after upgrade.

[Stan Hoeppner <stan@hardwarefreak.com>]

On 5/25/2011 10:06 AM, John McMonagle wrote:
> On Wednesday, May 25, 2011 09:54:32 am Stan Hoeppner wrote:
>> On 5/25/2011 8:19 AM, John McMonagle wrote:
>>> Just upgraded a poweredge 1850 server from Debian lenny to squeeze and
>>> can not boot with the new 2.6.32 kernel.
>>>
>>> From lspci  have this controller:
>>> SCSI storage controller: LSI Logic / Symbios Logic 53c1030 PCI-X
>>> Fusion-MPT Dual Ultra320 SCSI (rev 08)
>>>
>>>
>>> Running mdadm raid with root on md0.
>>>
>>> Normally run xen but all  info is for when running without xen.
>>>
>>> I can still boot with the 2.6.26 kernel but not with the new 2.6.32
>>> kernel. Under 2.6.32 it fails to start md0.
>>> in the busy box console
>>> Can see all the needed partitions.
>>> What was sda and sdb are now sdb and sdc that should not matter??
>>> mdadm.conf is:
>>> DEVICE partitions
>>> CREATE owner=root group=disk mode=0660 auto=yes
>>> HOMEHOST <system>
>>> MAILADDR xxxxx@advocap.org
>>> ARRAY /dev/md0 level=raid1 num-devices=2
>>> UUID=6f744c89:d2578f95:c150b018:d9f789b1
>>> ARRAY /dev/md1 level=raid1 num-devices=2
>>> UUID=7938d59c:28a69e5e:3facbdc2:12974557

>> This is probably due to udev changes.  What device is now sda?
>>
>> Using drive UUIDs instead of /dev/sdx in your arrays should fix this.
> 
> I think sda is a cd or virtual  cd now.
> 
> In the mdadm.conf it uses uuids and no /dev/sdx references or are you 
> referring to something else?

How is /dev/md0 assembled in your initramfs?  You said your root
filesystem is on /dev/md0.  Thus /dev/md0 must be assembled before
/etc/mdadm.conf can be read.

Another way around this problem is to create persistent udev rules.  But
since this requires created one-to-one mappings between
/dev/sdx<->drive_UUID mappings, it is easy to simply have mdraid use
drive UUIDs across the board, including within initramfs.

-- 
Stan