RAID-6 problem

Hi all,

I've a RAID-6, 10 HDDs, which, due to some hardware issue,
has lost 4 HDDs.
The RAID cannot be assembled anymore, of course.

The missing HDDs are now available and sane, so I guess it
could be possible to force an assemble operation.

What would be the reccomended way to proceed?

mdadm -A /dev/md1xx --force -c mdadm.conf

and then re-add the missing HDDs (does the above works?)

Or to re-create the array with "assume-clean" and with
the proper device order? The device role seems to be
available using "mdadm -E /dev/sdXy", I assume this
gives the proper order.

Thanks a lot in advance,

bye,

-- 

piergiorgio

On Sat, 13 Oct 2012, Piergiorgio Sartor wrote:

I've a RAID-6, 10 HDDs, which, due to some hardware issue,
has lost 4 HDDs.
The RAID cannot be assembled anymore, of course.

The missing HDDs are now available and sane, so I guess it
could be possible to force an assemble operation.

What would be the reccomended way to proceed?

mdadm -A /dev/md1xx --force -c mdadm.conf

and then re-add the missing HDDs (does the above works?)

Just --assemble --force if they all disappeared at roughly the same time 
(so all drives have not too different event count). If one drive has a 
very different event count, leave it out of the assembly and add it later.

Or to re-create the array with "assume-clean" and with
the proper device order? The device role seems to be
available using "mdadm -E /dev/sdXy", I assume this
gives the proper order.

Re-creating a raid5/6 with --assume-clean is for me a disaster recovery, 
something that should be performed when everything else fails. It's 
extremely unfortunate that a lot of guides bring this in very early in the 
fault finding process. I've talked to several people who completely 
screwed up their arrays by getting it wrong here.

-- 
Mikael Abrahamsson    email: swmike@swm.pp.se

On 2012-10-13, Mikael Abrahamsson [off-list ref] wrote:

Re-creating a raid5/6 with --assume-clean is for me a disaster recovery, 
something that should be performed when everything else fails. It's 
extremely unfortunate that a lot of guides bring this in very early in the 
fault finding process.

Actually the RAID wiki suggests exactly this path, and suggests against
--force, contrary to the advice I've gotten here from Neil and others.

https://raid.wiki.kernel.org/index.php/RAID_Recovery#Restore_array_by_recreating_.28after_multiple_device_failure.29

I could update this, but not being expert all it'd say is "use
--assemble --force; if it fails ask on the mailing list."  :)  Perhaps
that's actually the best advice to give, but if there's something more
useful to put on that page maybe someone with more expertise could
update it?

--keith

-- 
kkeller@wombat.san-francisco.ca.us

On Sat, 13 Oct 2012 11:45:10 -0700 Keith Keller
[off-list ref] wrote:

On 2012-10-13, Mikael Abrahamsson [off-list ref] wrote:

quoted

Re-creating a raid5/6 with --assume-clean is for me a disaster recovery, 
something that should be performed when everything else fails. It's 
extremely unfortunate that a lot of guides bring this in very early in the 
fault finding process.

Actually the RAID wiki suggests exactly this path, and suggests against
--force, contrary to the advice I've gotten here from Neil and others.

https://raid.wiki.kernel.org/index.php/RAID_Recovery#Restore_array_by_recreating_.28after_multiple_device_failure.29

I could update this, but not being expert all it'd say is "use
--assemble --force; if it fails ask on the mailing list."  :)  Perhaps
that's actually the best advice to give, but if there's something more
useful to put on that page maybe someone with more expertise could
update it?

--keith

You should definitely try '--assemble --force' before trying to recreate the
array.  There is no question about that.

NeilBrown

On Sun, Oct 14, 2012 at 06:31:02AM +1100, NeilBrown wrote:

On Sat, 13 Oct 2012 11:45:10 -0700 Keith Keller
[off-list ref] wrote:

quoted

On 2012-10-13, Mikael Abrahamsson [off-list ref] wrote:

quoted

Re-creating a raid5/6 with --assume-clean is for me a disaster recovery, 
something that should be performed when everything else fails. It's 
extremely unfortunate that a lot of guides bring this in very early in the 
fault finding process.

Actually the RAID wiki suggests exactly this path, and suggests against
--force, contrary to the advice I've gotten here from Neil and others.

https://raid.wiki.kernel.org/index.php/RAID_Recovery#Restore_array_by_recreating_.28after_multiple_device_failure.29

I could update this, but not being expert all it'd say is "use
--assemble --force; if it fails ask on the mailing list."  :)  Perhaps
that's actually the best advice to give, but if there's something more
useful to put on that page maybe someone with more expertise could
update it?

--keith

You should definitely try '--assemble --force' before trying to recreate the
array.  There is no question about that.

Thanks to everybody,

The "-A --force" worked fine, the array had, in the
end, 0 mismatch count.

Thanks again,

bye,

pg

NeilBrown

-- 

piergiorgio