slurm-random

anduril/slurm-random

@@ -0,0 +1,85 @@
+#!/bin/bash
+
+if [ -z "$1" ]
+then echo provide the script to run
+     exit 1
+fi
+
+if [ -z "${ANDURIL_NODELIST}" ]
+then NODELIST=( vm3 vm4 vm5 vm6 vm7 vm8 vm9 )
+else read -a NODELIST <<< "$ANDURIL_NODELIST"
+fi
+JOBROOT="$HOME/.srun"
+
+if [ "$1" == "-h" ]
+then echo -ne '
+This tools selects the lowest cpu allocated node for slurm run.
+Use it with anduril:  --exec-mode prefix --prefix '$( 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=( )
+TRY=0
+for e in ${NODELIST[@]}
+do NODERUNS+=( 0 )
+done
+while [ true ]
+do
+    INDEX=$(( ( RANDOM % ${#NODELIST[@]} ) ))
+    NODERUNS=$(( $( scontrol -o show node ${NODELIST[$INDEX]} | sed 's,.*CPUAlloc=\([0-9]\+\).*CPUTot=\([0-9]\+\).*,\2-\1 ,' ) ))
+    scontrol -o show node ${NODELIST[$INDEX]} | grep State=DOWN > /dev/null && NODERUNS=-1
+    echo "srun: Node ${NODELIST[$INDEX]}, Free sockets: $NODERUNS" >&2
+    [ $NODERUNS -gt 0 ] && {
+        break
+    } || {
+        TRY=$(( $TRY + 1 ))
+        echo "try again $TRY"
+        [ $TRY -gt 3 ] && {
+            TRY=0
+            SECONDS=$(( ( RANDOM % 10 ) ))
+            echo "waiting for free sockets for $SECONDS s."
+            sleep $SECONDS
+        }
+    }
+done
+
+echo "srun: Node ${NODELIST[$INDEX]}, Free sockets: ${NODERUNS[$INDEX]}" >&2
+
+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\" " >> "$JOBFILE"
+echo "date +'Start: %s' >> \"$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 $JOBFILE"
+# send the job
+srun --nodelist=${NODELIST[$INDEX]} "$JOBFILE"