[Lustre-discuss] e2fsck --mdsdb segmentation fault

[Andreas Dilger]

On 2009-11-13, at 02:13, Heiko Schröter wrote:
> i think i've spotted the difference. Because it happend on _all_  
> OSTs i thought this cannot be an hardware issue.
>
>> That depends on what the errors on the OSTs are.
>
> What i've found out:
>
> When mounting/unmounting an OST partition with '-t ldiskfs' and do  
> some cp/rm jobs, the 'e2fsck -f -n' is fine with the mounted or  
> unmounted device.
>
> Now, mounting '-t lustre -o abort_recovery' (only a single OST in  
> the whole cluster to have a defined location to e2fsck) and do some  
> cp/rm Jobs to the lustre fs i get the inodes/blocks errors on the  
> mounted and unmounted OST.

If you are running "e2fsck -f -n" then it does not replay the journal,  
so
the filesystem will be inconsistent and may report some errors.  If you
did a real e2fsck (no "-n", but on an unmounted filesystem) then the  
journal
will be replayed and no errors should be seen.

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