RAID10: How to mark spare as active, assume-clean documentation

RAID10: How to mark spare as active, assume-clean documentation

[Anshuman Aggarwal]

Hi all,
 I made a simple boo boo which I hope the pros here can quickly guide
me about. I have a near 2, 6 disk Raid 10 MD device (mdadm 3.1.4,
Ubuntu ocelot 3.0.0.12) in which I marked two consecutive devices as
failed and removed them (without bothering with that RAID10 can take
failure only of non consecutive devices in near 2 configuration). When
I re-added them they're showing as spares and the array is obviously
not assembling.....

I know the data is there and the disks are working reliably( I marked
them as failed coz I wanted these two drives reconstructed ...bad idea
...and there are other ways of resyncing that I now know about)

My questions:
1. I am guessing that I need to create the array with --assume-clean
the knowledge of the order of the drives (4 are known, only the order
of 2 spares is in question which I am also pretty confident about)??
2. Documentation of assume-clean is quite sparse. Does it not write
the superblock at all and only keep the details in memory? If so, how
would I write the superblocks after verifying (read-only) that the
array got recreated correctly? If it does write the superblock,
..wouldn't that destroy existing superblock in case I get the
order/some other parameter wrong on the first go?

I looked around on the list but couldn't get clear directions on the
correct use of assume-clean for such a situation. I'm hoping that a
thorough reply here could serve others looking for the same.

Thanks and regards,
Anshu

Re: RAID10: How to mark spare as active, assume-clean documentation

[Phil Turmel]

Good morning Anshu,

On 02/20/2012 01:35 PM, Anshuman Aggarwal wrote:
> Hi all,
>  I made a simple boo boo which I hope the pros here can quickly guide
> me about. I have a near 2, 6 disk Raid 10 MD device (mdadm 3.1.4,
> Ubuntu ocelot 3.0.0.12) in which I marked two consecutive devices as
> failed and removed them (without bothering with that RAID10 can take
> failure only of non consecutive devices in near 2 configuration). When
> I re-added them they're showing as spares and the array is obviously
> not assembling.....
> 
> I know the data is there and the disks are working reliably( I marked
> them as failed coz I wanted these two drives reconstructed ...bad idea
> ...and there are other ways of resyncing that I now know about)
> 
> My questions:
> 1. I am guessing that I need to create the array with --assume-clean
> the knowledge of the order of the drives (4 are known, only the order
> of 2 spares is in question which I am also pretty confident about)??

Yes, assume-clean is what you need here.  You may have to try it twice,
swapping the unknowns.  Read-only of course, until "fsck -n" gives you
the expected responses.

> 2. Documentation of assume-clean is quite sparse. Does it not write
> the superblock at all and only keep the details in memory? If so, how
> would I write the superblocks after verifying (read-only) that the
> array got recreated correctly? If it does write the superblock,
> ..wouldn't that destroy existing superblock in case I get the
> order/some other parameter wrong on the first go?

The superblocks will be written immediately.  Before using it, you must
save the output of "mdadm --detail /dev/mdX" for the array, and
"mdadm --examine /dev/sdX" for each member device.

You should also record the device vs. serial number map for your
system, in case you reboot at any point and device names change.  I'm
biased towards "lsdrv" [1], but you could also print the output of
"ls -l /dev/disk/by-id/"

> I looked around on the list but couldn't get clear directions on the
> correct use of assume-clean for such a situation. I'm hoping that a
> thorough reply here could serve others looking for the same.

It's a good practice to keep copies of the above diagnostics for a
properly running system to use when the $%^& hits the fan.

HTH,

Phil

[1] http://github.com/pturmel/lsdrv