#!/bin/bash

trap cleanup EXIT

cd `dirname $0`/..
CS_HOME=`pwd`
RUN_CASTEP="nice -15 $CS_HOME/bin/castep"

LOCK_FILE=$CS_HOME/cs_lock
Q_FILE=$CS_HOME/cs_queue
JOB_ID_FILE=$CS_HOME/cs_job_id

cleanup()
{
   # Wait till no-one else is writing
   while lsof | grep -q ~/etc/cs_active; do
       usleep $((RANDOM*100))
   done
   awk '!/'`hostname -s`'/{print}' ~/etc/cs_active > ~/etc/cs_active.tmp
   mv ~/etc/cs_active.tmp ~/etc/cs_active

   echo Terminating CASTEP server
   rm -f $LOCK_FILE

   exit 0
}

cont()
{
    echo Continuing. Looking for pending jobs
    RUNNABLE_JOBS=`awk -F'/' 'BEGIN {npend=0} ($2 == "PENDING") {npend++} END {print npend}' $Q_FILE`

    if [[ RUNNABLE_JOBS != 0 ]]; then
        JOB_ID=`awk -F'/' ' ($2 == "PENDING") { print $1; exit }' $Q_FILE`
        run_castep_job
    fi
    kill -TSTP $$
}

run_castep_job()
{
    cd $CS_HOME/$JOB_ID
    cp job.cell run.cell
    cp job.param run.param

    if [[ -e run.check ]]; then
        echo Resuming job $JOB_ID at `date`
        echo continuation: default >> run.param
        JOB_STATUS="RESUMING"
    else
        echo Starting job $JOB_ID  at `date`
        JOB_STATUS="RUNNING"
    fi

    # If species counts of this job match previous then copy .check file to speed up calculation
    if [[ -d ../$(($JOB_ID-1)) ]]; then
        if [[ (`$CS_HOME/bin/species_count job.cell` == \
            `$CS_HOME/bin/species_count ../$(($JOB_ID-1))/job.cell`) \
            && (-e ../$(($JOB_ID-1))/run.check) \
            && (! -e run.check ) ]]; then

            echo Reusing check file from previous job $(($JOB_ID-1)) for job $JOB_ID
            echo reuse: default >> run.param
            cp ../$((JOB_ID-1))/run.check run.check
        fi
    fi

    cd $CS_HOME
    awk -F'/' '{ if($1 == '$JOB_ID') { print '$JOB_ID'"/'$JOB_STATUS'/" $3} else {print}}' \
        $Q_FILE > cs_queue.tmp
    mv cs_queue.tmp $Q_FILE
    cd $CS_HOME/$JOB_ID

    rm -f run.*.err
    $RUN_CASTEP run
    echo Job $JOB_ID exited at `date`
    if [[ -s run.0001.err ]]; then
        echo Error encountered while running CASTEP job $JOB_ID
        JOB_STATUS="ERROR"
        cd ..
    else
        if [[ `tail -1 run.castep | cut -d' ' -f1-2` == "Total time" ]]; then
            echo Job $JOB_ID completed succesfully
            JOB_STATUS="SUCCESS"
        elif [[ -e killed ]]; then
            echo Job $JOB_ID killed.
            JOB_STATUS="KILLED"
        else
            echo Job $JOB_ID not completed, more time needed
            JOB_STATUS="TIMEOUT"
        fi
    fi
    cd $CS_HOME
    awk -F'/' '{ if($1 == '$JOB_ID') { print '$JOB_ID' "/'$JOB_STATUS'/" $3} else {print}}' \
        $Q_FILE > cs_queue.tmp
    mv cs_queue.tmp $Q_FILE
    # Notify qsubw if it's waiting for job to finish
    if [[ -e $JOB_ID/notify ]]; then
        kill -CONT `cat $JOB_ID/notify`
    fi

    if [[ $JOB_STATUS == "TIMEOUT" ]]; then
        echo Ran out of time running job, terminating server now
        exit 0
    fi
}

cd $CS_HOME

if [[ -f $LOCK_FILE ]]; then
    echo CASTEP server already running on this host, aborting
    exit 1
fi

# Wait till no-one else is writing
while lsof | grep -q ~/etc/cs_active; do
    usleep $((RANDOM*100))
done
echo `hostname -s` >> ~/etc/cs_active

echo CASTEP server started at `date`
echo $$ > $LOCK_FILE

[[ ! -f $Q_FILE ]] && touch $Q_FILE
[[ ! -f $JOB_ID_FILE ]] && echo 1 > $JOB_ID_FILE

echo Checking for new jobs or jobs to be resumed

RUNNABLE_JOBS=`awk -F'/' 'BEGIN {npend=0} ($2 == "PENDING" || $2 == "RUNNING" || $2 == "TIMEOUT") {npend++} END {print npend}' $Q_FILE`

if [[ $RUNNABLE_JOBS != 0 ]]; then
    JOB_ID=`awk -F'/' ' ($2 == "PENDING" || $2 == "RUNNING" || $2 == "TIMEOUT") { print $1; exit }' $Q_FILE`
    echo Resuming with job $JOB_ID
    run_castep_job
fi

while true; do

    # Suspend while we wait for a job
    echo Nothing to be done for now, suspending
    kill -STOP $$

    RUNNABLE_JOBS=`awk -F'/' 'BEGIN {npend=0} ($2 == "PENDING") {npend++} END {print npend}' $Q_FILE`
    while [[ $RUNNABLE_JOBS != 0 ]]; do
        JOB_ID=`awk -F'/' ' ($2 == "PENDING") { print $1; exit }' $Q_FILE`
        run_castep_job
        RUNNABLE_JOBS=`awk -F'/' 'BEGIN {npend=0} ($2 == "PENDING") {npend++} END {print npend}' $Q_FILE`
    done

done


