[Lustre-discuss] ldlm_cli_cancel_req()) Got rc -108 from cancel RPC: canceling anyway

[Heiko Schröter]

Am Mittwoch 23 Dezember 2009 12:22:17 schrieb Christopher J. Walker:
> > 
> > Dec 22 17:18:49 proof kernelLustreError: 10917:0:
> > (ldlm_request.c:1030:ldlm_cli_cancel_req()) Got rc -108 from cancel RPC: 
> > canceling anyway
> > Dec 22 17:18:49 proof kernelLustreError: 10917:0:
> > (ldlm_request.c:1030:ldlm_cli_cancel_req()) Skipped 169 previous similar 
> > messages
> > Dec 22 17:18:49 proof kernelLustreError: 10917:0:
> > (ldlm_request.c:1533:ldlm_cli_cancel_list()) ldlm_cli_cancel_list: -108
> > Dec 22 17:18:49 proof kernelLustreError: 10917:0:
> > (ldlm_request.c:1533:ldlm_cli_cancel_list()) Skipped 169 previous similar 
> > messages
> > Dec 22 17:18:49 proof kernelLustre: client ffff81042fccf400 umount complete
> > Dec 22 17:19:02 proof kernelLustre: Client userdata-client has started
> > 
> > Is anybody else seeing these messages in this situation? Does anyboyd know for 
> > a workaround??
> 
> Like Ewan, our Lustre filesystem is automounted. Whilst I haven't done a 
> detailed study, it does look as though these messages occur immediately 
> before unmounting the filesystem.

Yes. These messages do occur before 'auto'-un-mounting. So nothing to worry about.
The above is the mount process.

Unmounting should look like this:
Jun 17 04:00:16 cluster1 LustreError: 6460:0:(ldlm_request.c:1043:ldlm_cli_cancel_req()) Got rc -108 from cancel RPC: canceling anyway
Jun 17 04:00:16 cluster1 LustreError: 6460:0:(ldlm_request.c:1632:ldlm_cli_cancel_list()) ldlm_cli_cancel_list: -108
Jun 17 04:00:16 cluster1 Lustre: client ffff8100c44d1000 umount complete

If you don't see the last line 'umount complete' automount + lustre will hang and there should be no further access to the lustre system.
Happend to us in our scenario.

> 
> Is automounting a bad idea?

It depends. We had some bad experiences with lustre-1.6.6 and automount. See the mail archive about it. Subject: 'Stalled autofs + lustre'
Our problem should be resolved with upgrading to 1.8.x.
We will test again in Jan/Feb 2010 when the upgrade is sheduled.

Regards
Heiko