Resubmit jobs on Sun Grid Engine

cascade_job_resubmit.sh

#!/bin/bash

# restart all currently running jobs which are hanging for some reason
if [ "$1" == "qstat" ]; then
    qstat=$(qstat -u $USER)
    jobIDs=$(echo "$qstat" | awk -v col=1 'NR > 2 {print $col}' | sort | uniq)
    taskIDs=$(echo "$qstat" | awk -v col=10 'NR > 2 {print $col}')

    # kill currently running jobs
    for id in $jobIDs; do
        echo "kill job $id"
        qdel $id
    done

    # restart the jobs
    for id in $taskIDs; do
        ./cascade_job_100k.sh $id
    done
    echo "$taskIDs" > cascade_job_resubmit_qstat.txt


elif [ "$1" == "find" ]; then
    find=$(find ../data/ana -name "*.root" -type f -size -2M)
    taskIDs=""
    for f in $find; do
        f2="${f#*cascade_}"
        id="${f2%%_*}"
        taskIDs=$(printf "$taskIDs\n$id")
    done
    taskIDs=$(echo "$taskIDs" | tr ' ' '\n' | sort -u)
    for id in $taskIDs; do
        ./cascade_job_100k.sh $id
    done
    echo "$taskIDs" > cascade_job_resubmit_find.txt


elif [ "$1" == "missing" ]; then
    # retrieve further information for the range
    maxID=100
    if test "$2" != ""; then
        if [[ $2 =~ ^[0-9]+$ ]]; then
            maxID=$2
        fi
    fi

    # get the list of IDs that ran through
    find=$(find ../data/ana -name "*.root" -type f)
    completedIDs=""
    for f in $find; do
        f2="${f#*cascade_}"
        id="${f2%%_*}"
        completedIDs=$(printf "$completedIDs\n$id")
    done
    completedIDs=$(echo "$completedIDs" | tr " " "\n" | sort -un)

    # find the gaps
    lastID=0
    gapIDs=""
    for id in $completedIDs; do
        if (( $id > $maxID )); then
            break
        fi

        if (( $id != $lastID + 1 )); then
            if (( $lastID + 2 == $id )); then
                gapIDs=$(printf "$gapIDs\n$(( $lastID + 1 ))")
            else
                gapIDs=$(printf "$gapIDs\n$(( $lastID + 1 ))-$(( $id - 1 ))")
            fi
        fi
        lastID=$id
    done

    # catch IDs in the gap between last and maximum
    if (( $lastID < $maxID )); then
        if (( $lastID + 1 == $maxID )); then
            gapIDs=$(printf "$gapIDs\n$(( $lastID + 1 ))")
        else
            gapIDs=$(printf "$gapIDs\n$(( $lastID + 1 ))-$(( $maxID ))")
        fi
    fi

    # call the job control
    for id in $gapIDs; do
        ./cascade_job_100k.sh $id
    done
    echo "$gapIDs" > cascade_job_resubmit_missing.txt
fi