Lustre discuss - Aug 2007 - Lustre keep crashing

Hi,

    If lustre state enter LBUG, are there any other way to fix the file 
system other than e2fsck + lfsck? Everytime we did it the whole file 
system crash (unable to mount) and need to be rebuilt. Or are there any 
"best practice" on managing Lustre to avoid the crash? Something like

1. fsck often. Especially after system crash.
2. lfsck once a month
3. If you see a lot of error in file system. Cancel fsck and re-building 
it immediately to save time.
4. blah blah..

    I ask this because re-building the whole things take sometime, 
especially for large partition size. Not to mention time consume by 
fsck. It took us whole day to fix.

PS: We''re using 1.6.1.

-- 

-----------------------------------------------------------------------------------
Somsak Sriprayoonsakul

Thai National Grid Center
Software Industry Promotion Agency
Ministry of ICT, Thailand
somsak_sr@thaigrid.or.th
-----------------------------------------------------------------------------------

On Aug 17, 2007  21:18 +0700, Somsak Sriprayoonsakul
wrote:
>    If lustre state enter LBUG, are there any other way to fix the file 
> system other than e2fsck + lfsck? Everytime we did it the whole file 
> system crash (unable to mount) and need to be rebuilt. Or are there any 
> "best practice" on managing Lustre to avoid the crash? Something
like
> 
> 1. fsck often. Especially after system crash.
> 2. lfsck once a month
> 3. If you see a lot of error in file system. Cancel fsck and re-building 
> it immediately to save time.
> 4. blah blah..
You shouldn''t have to run e2fsck or lfsck after an LBUG.  It is really
case-dependent on when you need to run e2fsck.  If the kernel reports
"mounting filesystem with errors" that is bad and you should unmount
and run e2fsck immediately.  I''d also suggest running e2fsck once
every month or two when there is a planned system outage, just to detect
any hardware or software errors that might not have been noticed by the
kernel.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.