The fix I prefer is disabling qalter for users. Changing job characteristics after submit time is a horrible thing to do to any scheduler. It's about as easy and more fair overall to have the user submit a new job with the desired feature request and cancel the old job.
Kenneth On Fri, 23 Oct 2009, Jan Lindheim wrote: > Date: Fri, 23 Oct 2009 12:26:50 -0700 > From: Jan Lindheim <[email protected]> > To: [email protected] > Subject: [Mauiusers] maui bug triggered when changing node requirements > > We have run into what looks like a maui bug: > > We have used the following tagging for our two types of compute nodes: > core4: node with a total of 4 CPU cores > core8: node with a total of 8 CPU cores > > If a user submits a job like: > qsub -l nodes=64:core4+16:core8 -l walltime=4:00:00 jobscript > and later finds that many core8 nodes have become available, they > may want to alter the node requirements by doing something like: > qalter -l nodes=32:core4+32:core8 > to get the job to start sooner. > (Will give the user the same amount of CPU cores as the original job > submission.) > > Maui does not seem to update the requirements properly and the job just gets > stuck. A restart of maui will help maui get the information correctly, but > this is not a good work-around. > > Regards, > Jan Lindheim > _______________________________________________ > mauiusers mailing list > [email protected] > http://www.supercluster.org/mailman/listinfo/mauiusers > _______________________________________________ mauiusers mailing list [email protected] http://www.supercluster.org/mailman/listinfo/mauiusers
