Re: Adding a new disk after disk failure on raid6 volume

[John Robinson <john.robinson@anonymous.org.uk>]

On 20/12/2011 09:21, Robin Hill wrote:
> On Tue Dec 20, 2011 at 09:46:13AM +0100, BERTRAND Joël wrote:
>
>> Hello,
>>
>> I use several softraid volumes for a very long time. Last week, a disk
>> has crashed on a raid6 volume and I have tried to replace faulty disk.
>> Today, when Linux boots, it only assembles this volume if the new disk
>> is marked as 'faulty' or 'removed', and I don't understand...
>>
>> System is a sparc64-smp server running linux debian/testing :
>>
>> Root rayleigh:[~]>  uname -a
>> Linux rayleigh 2.6.36.2 #1 SMP Sun Jan 2 11:50:13 CET 2011 sparc64
>> GNU/Linux
>> Root rayleigh:[~]>  dpkg-query -l | grep mdadm
>> ii  mdadm                                 3.2.2-1
>>
>> Faulty device is /dev/sde1 :
>>
>> Root rayleigh:[~]>  cat /proc/mdstat
>> Personalities : [raid1] [raid6] [raid5] [raid4]
>> md7 : active raid6 sdc1[0] sdi1[6] sdh1[5] sdg1[4] sdf1[3] sdd1[1]
>>         359011840 blocks level 6, 64k chunk, algorithm 2 [7/6] [UU_UUUU]
>>
>> All disks (/dev/sd[cdefghi]) are same model (Fujitsu SCA-2 73 GB) and
>> each disk only contains one partition (type FD, linux autodetect). If I
>> add /dev/sde1 to raid6 with mdadm -a /dev/md7 /dev/sde1, disk is added
>> and my raid6 runs with all disks. But I obtain the same superblock on
>> /dev/sde1 and /dev/sde ! If I remove /dev/sde superblock, /dev/sde1 one
>> disappears also (i think that both superblocks are the same).
>>
> <- SNIP info ->
>>
>> All disks return same information except /dev/sde when it is running
>> (mdadm --examine /dev/sde and mdadm --examine /dev/sde1 return the same
>> information). What is my mistake ? Is this a known issue ?
>>
> It's a known issue with 0.9 superblocks, yes. There's no information in
> the superblock with allows md to tell whether it's on the partition or
> the disk, so for full-disk partitions the same superblock could be valid
> for both. 0.1 superblocks contain extra information which can be used to
> differentiate between these. I'm a little surprised that the other
> drives don't get detected in the same way though.

I think the above issue only occurs on partitions with particular 
alignments, iirc starting at multiples of 8 sectors. Old fdisk would 
always create the first partition starting at sector 63, and that was 
the case with the output we saw for /dev/sdc, but a new fdisk will 
likely create the partition starting at sector 2048.

Alternatively, or additionally, the problem may be that very old fdisk 
had a bug where it miscounted and didn't create partitions right up to 
the last "cylinder" of the disc, so the md metadata on the last 
partition wasn't in the the same place as it would have been if it was 
for the whole disc.

Either way, I would recommend that the OP --fail, --remove and 
--zero-superblock his /dev/sde1, then copy a working partition table 
from sdc with `dd if=/dev/sdc of=/dev/sde bs=512 count=1`, then 
`blockdev --rereadpt /dev/sde`, then `fdisk -lu /dev/sde` just to make 
sure that there is now an sde1 that's identical to sdc1, then --add the 
new /dev/sde1.

Hope this helps!

Cheers,

John.

--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html