jobState

#!/bin/bash

# Gets state of slurm job given its id as an argument

job=$1

running=( CONFIGURING PENDING RUNNING RESIZING SUSPENDED PREEMPTED COMPLETING )
failed=( BOOT_FAIL CANCELLED DEADLINE FAILED NODE_FAIL TIMEOUT OUT_OF_MEMORY )
completed=COMPLETED

failedLabel=failed
runningLabel=running
completedLabel=success
dateS=$(date -d "-10 days" +'%Y-%m-%d')


# Gets state of job
let timeMultiplier=4
let sleepTime=1
while true
do
    sacctOut=`sacct -n -j $1 --format "JobID,State" -P -S $dateS`
    if [ "$?" == "0" ]
    then
        break
    fi
    
    sleep $sleepTime
    let sleepTime=$sleepTime*$timeMultiplier
    
    if [ "$sleepTime" -gt "120" ]
    then
        sleepTime=120
    fi
done

state=`echo "$sacctOut" | grep -v .extern | cut -f2 -d "|"` # Ignores extern process from SLURM



#Checking if completed
nlinesCompleted="$(echo "$state" | perl -ne "if (\$_ =~ /$completed/) {print}" | wc -l)"
nlines="$(echo "$state" | wc -l)"

if [ "$nlinesCompleted" == "$nlines" ]
then
    echo "$completedLabel"
    exit 0
fi


# Checking if running
for i in ${running[@]}
do
    
    # If running in anyway the number of lines will not be 0
    nlines="$(echo "$state" | perl -ne "if (\$_ =~ /$i/) {print}" | wc -l)"
    
    if [ "$nlines" != "0" ] 
    then
        echo "$runningLabel"
        exit 0
    fi
done


# Checking if failed
for i in ${failed[@]}
do
    
    # If failed in anyway the number of lines will not be 0
    nlines="$(echo "$state" | perl -ne "if (\$_ =~ /$i/) {print}" | wc -l)"
    
    if [ "$nlines" != "0" ] 
    then
        echo "$failedLabel"
        exit 0
    fi
done


echo "$runningLabel"
exit 0