Has anyone gotten the High, Medium, and Low example at http://slurm.schedmd.com/preempt.html working with slurm-16.05?
It looks pretty simple, there's been some changes from previous versions of slurm. Mainly that it's PriorityTier instead of Priority and OverSubscribe instead of SHARED. So their new example (on 3 lines, sorry for the warp): PartitionName=low Nodes=lnx Default=YES OverSubscribe=NO PriorityTier=10 \ PreemptMode=requeue PartitionName=med Nodes=lnx Default=NO OverSubscribe=FORCE:1 PriorityTier=20 \ PreemptMode=suspend PartitionName=hi Nodes=lnx Default=NO OverSubscribe=FORCE:1 PriorityTier=30 \ PreemptMode=off I poured over that page and came up with this: $ cat slurm.conf | egrep -v "^#" | egrep -i "selecttype|SelectTypeParameter|DefMemPerCPU|JobAcctGatherType|PreemptMode|PreemptType|PriorityTier|SchedulerType" DefMemPerCPU=2000 SchedulerType=sched/builtin SelectType=select/cons_res SelectTypeParameters=CR_CPU_Memory PreemptType=preempt/partition_prio PreemptMode=GANG,SUSPEND JobAcctGatherType=jobacct_gather/linux PartitionName=low Nodes=linux Default=NO OverSubscribe=NO \ PriorityTier=10 PreemptMode=requeue MaxTime=INFINITE State=UP DefMemPerCPU=2000 PartitionName=med Nodes=linux Default=NO OverSubscribe=FORCE:1 \ PriorityTier=20 PreemptMode=suspend MaxTime=INFINITE State=UP DefMemPerCPU=2000 PartitionName=high Nodes=linux Default=YES OverSubscribe=FORCE:1 \ PriorityTier=30 PreemptMode=off MaxTime=INFINITE State=UP DefMemPerCPU=2000 Each node has swap = twice ram. I noticed the pending job in high: $ squeue -p high --format="%.8i %.9P %.8T %.10M %.9l %.6D %12R %8Q" | head -10 JOBID PARTITION STATE TIME TIME_LIMI NODES NODELIST(REA PRIORITY 90204 high PENDING 0:00 UNLIMITED 1 (Resources) 2583 Has lower priority then the running jobs in med: $ squeue -p med --format="%.8i %.9P %.8T %.10M %.9l %.6D %12R %8Q" | grep RUNNING 89647 med RUNNING 2:34:26 UNLIMITED 1 c4-68 9945 89648 med RUNNING 2:30:15 UNLIMITED 1 c4-69 9926 Do I need to set Priority *and* PriorityTier maybe? Any other suggestions?
