[Lustre-discuss] meta freezing

[Andreas Dilger]

On Oct 14, 2008  13:07 +0200, Papp Tamas wrote:
> Since we switched from 1.6.4.3 to 1.6.5.1 on one of our cluster we have 
> a wierd problem.
> 
> One of the node of the cluster lock up and only reset can help, it's 
> usually the meta node. It's already not good, but there is also 
> something. When the node gets up again and the recovery is starting 
> agaian it locks up over and over again. It's counting back and sometimes 
> there is only a few clients, sometimes there is no more clients, but 
> it's always locks up.
> 
> So I mount the mdt, umount -f, mount again, recovery is in Sstatus 
> INACTIVE and the cluster is working.
> 
> Now I'm out of ideas.
> 
> The cluster was made with 1.6.5.1. Is it safe to move back to 1.6.4.3? I 
> mean just changing he utilities and the kernel and that's all, or do I 
> need the do further  steps?

Yes, it should always be possible to downgrade to the older version.
In some cases in the future (e.g. 2.0 -> 1.8.x downgrade) it will be
needed to remount the clients, but general consensus is that if you
are downgrading you already have major problems so a remount will not
contribute significantly to the problem.

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