slurm prefix scripts, possible bug with exit codes. added checking of _command file

anduril/slurm-maxfree

@@ -69,6 +69,20 @@ EXECPATH=$( pwd )
 # create the jobfile
 echo '#!/bin/bash' > "$JOBFILE"
 chmod 755 "$JOBFILE"
+# Find _command file
+for (( i=1; i<=$#; i++ ))
+do  if [[ "${!i}" == */_command ]]
+    then echo 'retrys=0' >> "$JOBFILE"
+         echo 'while :
+         do [ -f "'${!i}'" ] && {
+           break 
+         } || {
+           echo Waiting for _command file
+           sleep 1; [ "$retrys" -gt 15 ] && break;
+         }
+         done ' >> "$JOBFILE" 
+    fi 
+done
 
 echo 'echo Node: $HOSTNAME'" >> \"$STATFILE\" " >> "$JOBFILE"
 echo "pwd >> \"$STATFILE\" " >> "$JOBFILE"
@@ -81,8 +95,14 @@ done
 cat - >> "$STRMFILE"
 echo -n ' < "'$STRMFILE'"' >> "$JOBFILE"
 echo -e "\n" >> "$JOBFILE"
+echo 'EC=$?' >> "$JOBFILE"
 echo "date +'Stop: %s' >> \"$STATFILE\" " >> "$JOBFILE"
+echo 'exit $EC' >> "$JOBFILE"
 
 echo "The job file is in $JOBFILE"
 # send the job
 srun --nodelist=${NODELIST[$INDEX]} "$JOBFILE"
+
+
+
+

anduril/slurm-random

@@ -64,6 +64,20 @@ EXECPATH=$( pwd )
 # create the jobfile
 echo '#!/bin/bash' > "$JOBFILE"
 chmod 755 "$JOBFILE"
+# Find _command file
+for (( i=1; i<=$#; i++ ))
+do  if [[ "${!i}" == */_command ]]
+    then echo 'retrys=0' >> "$JOBFILE"
+         echo 'while :
+         do [ -f "'${!i}'" ] && {
+           break 
+         } || {
+           echo Waiting for _command file
+           sleep 1; [ "$retrys" -gt 15 ] && break;
+         }
+         done ' >> "$JOBFILE" 
+    fi 
+done
 
 echo 'echo Node: $HOSTNAME'" >> \"$STATFILE\" " >> "$JOBFILE"
 echo "pwd >> \"$STATFILE\" " >> "$JOBFILE"
@@ -76,7 +90,9 @@ done
 cat - >> "$STRMFILE"
 echo -n ' < "'$STRMFILE'"' >> "$JOBFILE"
 echo -e "\n" >> "$JOBFILE"
+echo 'EC=$?' >> "$JOBFILE"
 echo "date +'Stop: %s' >> \"$STATFILE\" " >> "$JOBFILE"
+echo 'exit $EC' >> "$JOBFILE"
 
 echo "The job file is in $JOBFILE"
 # send the job