[Wylug-discuss] Help needed with 'failed' Linux software RAID 10

[Tim Fletcher]

On Sat, 2010-05-29 at 19:43 +0100, Dave Fisher wrote:

> a) My, admittedly dodgy, understanding of RAID 10 is that if I can ID
> the two halves of a pair, I should be able to mount the pair like
> normal partitions or LVM volumes. Is this correct?

> b) Is it possible to readonly mount the md1 partitions in their
> current state - safely, without screwing-up further?

Having read over this I'm not sure I fully understand how things where
when they where working did you have:

md0 raid1 (DiskA+DiskB) 
md1 raid1 (DiskA+DiskB)
md2 raid0 (md0+md1)

or did you have

md0 raid0 (DiskA+DiskB) 
md1 raid0 (DiskA+DiskB)
md2 raid1 (md0+md1)

I think you had the former which means you need to tackle the problem in
2 layers, first of all you need to get at both parts of the RAID0 to
spin up, that means 1 half of each of the raid1's

You then need to get them to raid0 stripe together and that will give
you your LVM2 volume group.

Looking at the text attachment I think that sdf4+sdi4 and sdg4+sdh4 are
the 2 mirror sets, this is judging by the event counters on them. Also I
think that /dev/sdg4 has failed and maybe locked the sata bus.

I would suggest trying

Stopping the array altogether with:
mdadm -S /dev/md1

Telling mdadm to assemble the array by uuid but not start it if it's
degraded with this command:

mdadm -A /dev/md1 --no-degraded \
--uuid=f4ddbd55:206c7f81:b855f41b:37d33d37 

There is the --force option that will force mdadm to update the
superblocks and start the array that I have used to rescue me from a bad
power failure on a raid6 array.

-- 
Tim Fletcher
Learning Technologies Manager - Parrs Wood High School