[Lustre-discuss] Lustre mount problem

[Bernd Schubert]

Hello Frank,

On Wednesday 30 April 2008 17:47:55 Frank Mietke wrote:
> Hi Andrew,
>
> On Wed, Apr 30, 2008 at 08:58:42AM -0600, Lundgren, Andrew wrote:
> > From what I have been able to gather, this is not possible at the moment.
> >
> > The dead OST will always be there.  There is no functioning way to
> > actually remove it at the moment.
> >
> > https://bugzilla.lustre.org/show_bug.cgi?id=15345
>
> thank you for pointing me to this bug report.
>
> > We are running in failout mode rather than failover.  When our new
> > clients tried to reconnect our test cluster, they appeared to block
> > forever.  I would need to recreate the situation to validate that is the
> > behavior, but I am dealing with another issue at the moment.
> >
> > If you are on a production cluster, you may be in a bad way.  The only
> > way I have found to recover this is to wipe the cluster and start fresh. 
> > (Not a good option.)
>
> I could live with a "dead" OST in the configuration but as I've written in
> the update, every call to a proc-entry of this OST on the clients hangs
> forever. Not really optimal.

you very first approach to set the re-created ost to the old index was 
actually the right way to go. In you present situation I would create another 
very very small ost and set it to the old index number. 

Lustre will again refuse to re-register this ost, but there is way to convice 
it not to complain. Here is what I already wrote to the list, when I run into 
the same problem as you:

<quote of myself from 2008-02-05 21:45 "Re: [Lustre-discuss] how to recreate 
an OST?">

>Now I mounted the mgs as ldiskfs, and in CONFIGS/ there is no file for the 
>missing ost. 
>But now I just found the reason - the failed OST was still activated on the 
>clients. After deleting CONFIGS/{fsname}-client and remounting as type lustre 
>again, registering the failed ost works!
>I guess one shouldn't do it this way if one still has important data on the 
>filesystem ;) 

</quote>

I never got an reply of sun/clusterfs developers if this is the right 
filesystem, but I have tested it several times and it seems to be the right 
way to go.


Hope it helps,
Bernd

-- 
Bernd Schubert
Q-Leap Networks GmbH