get_number_cores_per_node

partition=$1
walltime=$2

scheduler=`cat $WORKDIR/scheduler`
version=`cat $WORKDIR/version`
job_submit=`cat $WORKDIR/job_submit`
job_cancel=`cat $WORKDIR/job_cancel`

rm -f $WORKDIR/cores.out

partition=`cat $WORKDIR/${partition}_partition`
excluded_nodes=`cat $WORKDIR/excluded_${partition}_nodes`

echo "#! /bin/bash"                           > $WORKDIR/get_number_of_cores
echo "#SBATCH --job-name=cores"              >> $WORKDIR/get_number_of_cores
echo "#SBATCH --output=$WORKDIR/cores.out"   >> $WORKDIR/get_number_of_cores
echo "#SBATCH --error=$WORKDIR/cores.err"    >> $WORKDIR/get_number_of_cores
echo "#SBATCH --time=00:00:"$walltime        >> $WORKDIR/get_number_of_cores
echo "#SBATCH --exclude=$excluded_nodes"     >> $WORKDIR/get_number_of_cores

if  [[ $partition == gpu ]] && [[ -e /etc/slurm/gres.conf ]]
then
    echo "#SBATCH --gres=gpu:0"              >> $WORKDIR/get_number_of_cores
fi

echo "#SBATCH --cpus-per-task=1"             >> $WORKDIR/get_number_of_cores
echo "#SBATCH --mem=2gb"                     >> $WORKDIR/get_number_of_cores
echo "#SBATCH --partition="$partition        >> $WORKDIR/get_number_of_cores
echo "WORKDIR=$WORKDIR"                      >> $WORKDIR/get_number_of_cores
echo "lscpu"                                 >> $WORKDIR/get_number_of_cores
chmod u+x                                       $WORKDIR/get_number_of_cores

echo "Trying to determine the number of CPU cores per ${partition} node. This may last up to $walltime seconds, please do not interrupt."

rm -f $WORKDIR/cores.*

$job_submit -Q $WORKDIR/get_number_of_cores
timeout ${walltime}s $vini_dir/wait_until_jobs_finish
if [ -e $WORKDIR/cores.out ]
then
    grep "CPU(s)" $WORKDIR/cores.out > tmp ; CPUs=`head -1 tmp` ; CPUs=`echo $CPUs | awk '{print $2}'`
    grep "Thread(s)" $WORKDIR/cores.out > tmp ; Threads=`head -1 tmp` ; Threads=`echo $Threads | awk '{print $4}'`
    cores=`echo $CPUs $Threads | awk '{print $1 / $2}'`
    echo "Found $cores CPU cores per ${partition} node."
    echo $cores > $WORKDIR/${partition}_cores
else
    ${job_cancel} -u $USER ; echo
    echo "Failed to access any ${partition} node. A probable reason is that all nodes are either busy or down."
    read -p "Write the number of CPU cores per ${partition} node here:" cores
    echo $cores > $WORKDIR/${partition}_cores
fi