[Lustre-discuss] Removing OST permantly

[Andreas Dilger]

On Jan 22, 2009  14:17 +0100, Heiko Schroeter wrote:
> i want to remove an OST permanently on the MDS:
> lctl --device 11 conf_param foo-OST0006.osc.active=0
> 
> Fine. Now lfs barks out on a 'lfs df -h /lustrefs':
> client # lfs df -h /lustrefs
> UUID                     bytes      Used Available  Use% Mounted on
> foo-MDT0000_UUID       154.8G    708.8M    145.3G    0% /misc/data[MDT:0]
> foo-OST0000_UUID         6.4T      5.1T   1002.9G   79% /misc/data[OST:0]
> foo-OST0001_UUID         6.4T      5.1T   1016.1G   79% /misc/data[OST:1]
> foo-OST0002_UUID         6.4T      5.0T      1.0T   79% /misc/data[OST:2]
> foo-OST0003_UUID         6.4T      5.0T      1.0T   78% /misc/data[OST:3]
> foo-OST0004_UUID         6.4T      5.6T    488.7G   87% /misc/data[OST:4]
> foo-OST0005_UUID         6.4T      5.6T    497.9G   87% /misc/data[OST:5]
> error: llapi_obd_statfs failed: Bad address (-14)
> 
> But there are OSTs numbered higher than 'OST-0006'.
> Getting the stripes located on those OSTs above is ok:
> 
> What to do to get lfs working again ?

Please file a bug on this.  It looks like there are two issues:
- lustre/utils/lfs.c:mntdf() should continue on this error instead
  of failing
- the ll_obd_statfs() function should return a better error

You can fix the first problem just by recompiling lfs, which will get
you a working lfs.  The second one is the "more correct" fix, but needs
more work and also recompiling/restarting Lustre.

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