Lustre discuss - Sep 2009 - recover OST data just using ./ROOT/ and ea.bak?

Ok, here''s a good one...

I have a lustre set up with version 1.8. Its has a combo mdt/mgs on one
machine and three OSSs with a couple of OSTs on each.

The array that the mdt/mgs was on died and decided to stay dead, taking its
data with it. But, I thought, no problem I have my file level backup that I
did a couple of days ago. But owing to a typo in the script I ran, I only
got the tar and getfattr of the ROOT directory, rather than the whole
volume.

My question to the lustre public is.. is it possible to recover the data on
the OSTs using this? If so, can anyone lend any pointers? (If, in theory its
not possible, I''ll stop banging my head against the wall and move on)

Any advice appreciated!
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090928/4a11b16d/attachment.html

Doh! Sorry to hear about your dead mdt. I''m no expert but something  
along the lines of the following may work-

1. mke2fs -j -J size=400 -I {inode_size} -i 4096 {new mdt dev}

where {inode_size} is at least 512 and possibly larger if the default  
stripe count is > 10 (inode_size = power_of_2_>=_than(384 +  
stripe_count * 24))

	tune2fs -O dir_index {new mdt dev}

2. Mount the new mdt as ldiskfs (mount -o ldiskfs) at say /mnt/mdt

3. Restore ROOT and your eas.  (unmount the mdt)

4. Run tunefs.lustre --fsname={fsname} --mdt --mgs --writeconf {new  
mdt dev } on the new mdt

5. Shutdown all your osts

6. Run tunefs.lustre --writeconf on your osts

7. Pray to the lustre gods for mercy

8. Mount your mdt as lustre (mount -o lustre)

9. Re-mount your osts and hope for the best

Hopefully one of the lustre experts on this list will comment on the  
above with a "yep that''ll work" or "you gotta be crazy
to do that"

-Aaron

On Sep 28, 2009, at 1:32 PM, Matt Franks wrote:
>
>
> Ok, here''s a good one...
>
> I have a lustre set up with version 1.8. Its has a combo mdt/mgs on  
> one machine and three OSSs with a couple of OSTs on each.
>
> The array that the mdt/mgs was on died and decided to stay dead,  
> taking its data with it. But, I thought, no problem I have my file  
> level backup that I did a couple of days ago. But owing to a typo in  
> the script I ran, I only got the tar and getfattr of the ROOT  
> directory, rather than the whole volume.
>
> My question to the lustre public is.. is it possible to recover the  
> data on the OSTs using this? If so, can anyone lend any pointers?  
> (If, in theory its not possible, I''ll stop banging my head against
> the wall and move on)
>
> Any advice appreciated!
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090928/ba3cf5c7/attachment.html

On Sep 28, 2009  21:58 -0400, Aaron Knister wrote:
> Doh! Sorry to hear about your dead mdt. I''m no expert but
something
> along the lines of the following may work-
>
> 1. mke2fs -j -J size=400 -I {inode_size} -i 4096 {new mdt dev}
>
> where {inode_size} is at least 512 and possibly larger if the default  
> stripe count is > 10 (inode_size = power_of_2_>=_than(384 +
stripe_count
> * 24))
>
> 	tune2fs -O dir_index {new mdt dev}
>
> 2. Mount the new mdt as ldiskfs (mount -o ldiskfs) at say /mnt/mdt
>
> 3. Restore ROOT and your eas.  (unmount the mdt)
>
> 4. Run tunefs.lustre --fsname={fsname} --mdt --mgs --writeconf {new mdt 
> dev } on the new mdt
>
> 5. Shutdown all your osts
>
> 6. Run tunefs.lustre --writeconf on your osts
>
> 7. Pray to the lustre gods for mercy
>
> 8. Mount your mdt as lustre (mount -o lustre)
>
> 9. Re-mount your osts and hope for the best
>
> Hopefully one of the lustre experts on this list will comment on the  
> above with a "yep that''ll work" or "you gotta be
crazy to do that"
It at least has a chance to work.  The majority of the Lustre-specific
files on the MDT will be (re-)created at "first" mount, and will fetch
the object IDs from the OSTs to recreate the entries in the lov_objids
file.
> On Sep 28, 2009, at 1:32 PM, Matt Franks wrote:
>> Ok, here''s a good one...
>>
>> I have a lustre set up with version 1.8. Its has a combo mdt/mgs on  
>> one machine and three OSSs with a couple of OSTs on each.
>>
>> The array that the mdt/mgs was on died and decided to stay dead,  
>> taking its data with it. But, I thought, no problem I have my file  
>> level backup that I did a couple of days ago. But owing to a typo in  
>> the script I ran, I only got the tar and getfattr of the ROOT  
>> directory, rather than the whole volume.
>>
>> My question to the lustre public is.. is it possible to recover the  
>> data on the OSTs using this? If so, can anyone lend any pointers? (If, 
>> in theory its not possible, I''ll stop banging my head against
the wall
>> and move on)
>>
>> Any advice appreciated!
>> _______________________________________________
>> Lustre-discuss mailing list
>> Lustre-discuss at lists.lustre.org
>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.