[Mauiusers] running jobs & restarting maui

Thomas Dargel td at chemie.hu-berlin.de
Wed Nov 9 14:46:46 MST 2005


On Tue, Nov 08, 2005 at 04:25:28PM -0800, Garrick Staples wrote:
> On Tue, Nov 08, 2005 at 03:18:47PM -0800, Garrick Staples alleged:
> > On Wed, Nov 09, 2005 at 09:26:02AM +1100, Chris Samuel alleged:
> > > On Tue, 8 Nov 2005 08:20 pm, Thomas Dargel wrote:
> > > 
> > > > 11/08 09:20:41 ALERT: ? ?job '561' in state 'Running' has exceeded its
> > > > wallclock limit (0+S:0) by 16:43:00 (job will be cancelled)
> > > > 11/08 09:20:41 MSysRegEvent(JOBWCVIOLATION: ?job '561' in state 'Running'
> > > > has exceeded its wallclock limit (0) by 16:43:00 (job will be
> > > > cancelled) ?job start 
> > > 
> > > Bingo - the jobs are running with a walltime of 0 (i.e. not set) and for some 
> > > reason whilst Maui considers this to be infinite if the job is submitted 
> > > whilst it's running when it restarts it sees this as 0 and so kills the 
> > > job. :-(
> > 
> > Agreeing 100% with Chris, this would appear to be a definite bug in maui
> > that should be fixed.  But I'm unable to duplicate the problem with
> > maui-3.2.6p14-snap.1129921819.  I restart maui and it happily recreates
> > the infinite reservation.  Which version are you using?
> 
> Of course, plan B is to just set a really really long default walltime.

With  

Qmgr: set server resources_default.walltime=2400:00:00

checkjob -v 593
gives something like that:

Reservation '593' (00:00:00 ->   INFINITY  Duration:   INFINITY)

This seems to solve my problem without changing the maui-release.
But with qstat -f 593 I see the explicit walltikme setting

    Resource_List.walltime = 2400:00:00

Does this (server) overrule the scheduler setting 'INFINITY'???

Best regards,

 Thomas.


> 
> Qmgr: set server resources_default.walltime=9000:00:00
> 
> You might even become like me and consider the existance of infinite
> walltime jobs to be a bug!  If we allowed such a thing on our cluster,
> jobs would be forgotten and left to bitrot forever.
> 
> -- 
> Garrick Staples, Linux/HPCC Administrator
> University of Southern California




More information about the mauiusers mailing list