Thanks Paddy, Could you file a bug on this http://bugs.schedmd.com
Thanks, Danny On 11/06/2012 10:04 AM, Paddy Doyle wrote: > Hi all, > > I noticed something odd when I was testing some resource limit stuff. > > - AccountingStorageEnforce set to 'assocations', or turned off completely > (previously it had been 'limits') > - some associations previously had GrpCPUMins limits set, from earlier testing > - jobs were still being killed when their usage went over the limit, e.g. with > this message: > > "Job 759 timed out, assoc 452 is at or exceeds group max cpu minutes limit > 5940 with 5956 for account testing" > > Looking at src/slurmctld/job_mgr.c it seems that the job_time_limit() function > wasn't actually checking to see if limits were being enforced before killing > the > job. > > See attached a patch which checks to see if limits or qos are enforced before > killing the job. I've tested it with 2.4.3 and it does what I expect - haven't > tried 2.4.4, but the job_time_limit() logic seems to the same. > > Thanks, > Paddy >
