diff --git a/anduril/srun-maxfree b/anduril/srun-maxfree new file mode 100755 index 0000000..c410adf --- /dev/null +++ b/anduril/srun-maxfree @@ -0,0 +1,49 @@ +#!/bin/bash + +if [ -z "${ANDURIL_NODELIST}" ] +then NODELIST=( vm3 vm4 vm5 vm6 vm7 vm8 vm9 ) +else read -a NODELIST <<< "$ANDURIL_NODELIST" +fi + +if [ "$1" == "-h" ] +then echo -ne ' +This tools selects the lowest cpu allocated node for slurm run +export ANDURIL_NODELIST="vm3 vm4 vm5" + ^ to have your own preferred list of nodes +Current nodelist: "'${NODELIST[@]}'"'"( ${#NODELIST[@]} nodes)\n" + exit +fi + + +NODERUNS=( ) +# find node with max free CPUs +for e in ${NODELIST[@]} +do NODERUNS+=( 0 ) +done +while [ true ] +do + MAX=0 + INDEX=0 + for (( i=0; i<${#NODERUNS[@]}; i++ )) + do NODERUNS[$i]=$(( $( scontrol -o show node ${NODELIST[$i]} | sed 's,.*CPUAlloc=\([0-9]\+\).*CPUTot=\([0-9]\+\).*,\2-\1 ,' ) )) + scontrol -o show node ${NODELIST[$i]} | grep State=DOWN > /dev/null && NODERUNS[$i]=-1 + echo -n "${NODELIST[$i]}:${NODERUNS[$i]} " + [ ${NODERUNS[$i]} -gt $MAX ] && { + MAX=${NODERUNS[$i]} + INDEX=$i + } + done + [ $MAX -gt 0 ] && { + break + } || { + SECONDS=$(( ( RANDOM % 30 ) )) + echo "waiting for free sockets for $SECONDS s." + sleep $SECONDS + } +done + +echo "srun: Node ${NODELIST[$INDEX]}, Free sockets: ${NODERUNS[$INDEX]}" >&2 + +srun --nodelist=${NODELIST[$INDEX]} "$@" + +