93 lines
2.5 KiB
Bash
Executable File
93 lines
2.5 KiB
Bash
Executable File
#!/bin/bash
|
|
|
|
if [ -z "$1" ]
|
|
then echo provide the script to run
|
|
exit 1
|
|
fi
|
|
echo CPUS $ANDURIL_PREFIX_CPU
|
|
if [ -z "${ANDURIL_NODELIST}" ]
|
|
then NODELIST=( vm3 vm4 vm5 vm6 vm7 vm8 )
|
|
else read -a NODELIST <<< "$ANDURIL_NODELIST"
|
|
fi
|
|
|
|
if [ "$1" == "-h" ]
|
|
then echo -ne '
|
|
This prefix selects the lowest cpu allocated node for slurm run.
|
|
Use it with anduril: --exec-mode prefix --prefix '$( basename $0 )'
|
|
To change the list of nodes (preferred order):
|
|
export ANDURIL_NODELIST="vm3 vm4 vm5"
|
|
|
|
Current nodelist: "'${NODELIST[@]}'"'"( ${#NODELIST[@]} nodes)\n"
|
|
exit
|
|
fi
|
|
|
|
NODERUNS=( )
|
|
|
|
# find node with least allocated CPUs
|
|
|
|
for e in ${NODELIST[@]}
|
|
#do NODERUNS+=( $( squeue | grep $e | wc -l ) )
|
|
do NODERUNS+=( $( scontrol show node $e | grep CPUAlloc | sed 's,.*CPUAlloc=\([0-9]\+\).*,\1,' ) )
|
|
done
|
|
MIN=${NODERUNS[0]}
|
|
INDEX=0
|
|
for (( i=0; i<${#NODERUNS[@]}; i++ ))
|
|
do (( ${NODERUNS[$i]} < MIN )) && {
|
|
MIN=${NODERUNS[$i]}
|
|
INDEX=$i
|
|
}
|
|
echo -n "${NODELIST[$i]}:${NODERUNS[$i]} "
|
|
done
|
|
echo "srun: Node ${NODELIST[$INDEX]}, occupied sockets: ${NODERUNS[$INDEX]}" >&2
|
|
|
|
|
|
JOBROOT="$HOME/.srun"
|
|
mkdir -p "$JOBROOT"
|
|
JOBNAME=job_$( date +"%y%m%d_%H%M%S" )_$( echo $@ | md5sum | cut -f1 -d" " )
|
|
JOBPATH="$JOBROOT/$JOBNAME"
|
|
while [ -d "$JOBPATH" ]
|
|
do echo Jobpath "$JOBPATH" exists
|
|
JOBNAME=job_$( date +"%y%m%d_%H%M%S" )_$( echo $@ | md5sum | cut -f1 -d" " )
|
|
JOBPATH="$JOBROOT/$JOBNAME"
|
|
done
|
|
mkdir -p "$JOBPATH"
|
|
JOBFILE="$JOBPATH/job"
|
|
STATFILE="$JOBPATH/statistics"
|
|
STRMFILE="$JOBPATH/stream"
|
|
EXECPATH=$( pwd )
|
|
|
|
# create the jobfile
|
|
echo '#!/bin/bash' > "$JOBFILE"
|
|
chmod 755 "$JOBFILE"
|
|
|
|
echo 'echo Node: $HOSTNAME'" >> \"$STATFILE\" " >> "$JOBFILE"
|
|
echo "pwd >> \"$STATFILE\" " >> "$JOBFILE"
|
|
echo "uptime >> \"$STATFILE\" " >> "$JOBFILE"
|
|
echo "date +'Start: %s' >> \"$STATFILE\" " >> "$JOBFILE"
|
|
echo -n "/usr/bin/time -o \"$STATFILE\" --append " >> "$JOBFILE"
|
|
for (( i=1; i<=$#; i++ ))
|
|
do echo -n "\"${!i}\" " >> "$JOBFILE"
|
|
done
|
|
#Catch the input stream (for R launcher)
|
|
cat - >> "$STRMFILE"
|
|
echo -n ' < "'$STRMFILE'"' >> "$JOBFILE"
|
|
echo -e "\n" >> "$JOBFILE"
|
|
echo "date +'Stop: %s' >> \"$STATFILE\" " >> "$JOBFILE"
|
|
|
|
echo "The job file is in $JOBFILE"
|
|
minutes_now=$( date "+%M" )
|
|
|
|
# send the job
|
|
srun --nodelist=${NODELIST[$INDEX]} "$JOBFILE"
|
|
exitcode=$?
|
|
# clear out old job definitions (semirandomly)
|
|
[[ -e "$JOBROOT"/.lastdel ]] || touch "$JOBROOT"/.lastdel
|
|
if test "$( find $JOBROOT/.lastdel -mmin +30 )"
|
|
then touch "$JOBROOT"/.lastdel
|
|
find "$JOBROOT" -maxdepth 2 -mindepth 2 -type f -mtime +10 -delete
|
|
find "$JOBROOT" -type d -depth -empty -delete
|
|
fi
|
|
|
|
exit $exitcode
|
|
|