#!/bin/bash 
#
#
# Delft FEWS
# Copyright (c) 2003-2008 Deltares. All Rights Reserved.
#
# Developed by:
# Tessella Support Services plc
# Tauro Kantorencentrum
# President Kennedylaan 19
# 2517 JK  Den Haag
# The Netherlands
# email: info@tessella.com
# web:   www.tessella.com
#
# Project Ref: Tessella/NPD/5431
#
# File history
# Version           Date                           Author
# $Revision: 18767 $     $Date: 2008-05-07 14:09:48 +0200 (Wed, 07 May 2008) $                       $Author: broek_f $
#
# Description:
# Script to run ensembles for the Topkapi model on a Condor Grid Engine
# It is assumed that the condor bin directory is in the PATH
# 
# Return Values:
#  0  - Sucess
#  1  - Failure
#
# Command line options and arguments
#  Mandatory:
#  -o path  : The directory for the output log files (redirected stdout/stderr)
#  -n num   : Total number of ensemble members e.g. 16
#  -t num   : Number of seconds after which the condor cluster job should time out
#  -d path  : Root directory for the Topkapi module (local path) 
#              (where all the input/output files reside)
#  -r path  : Root directory for the Topkapi module (remote network path) 
#              This should be the same physical directory as the one
#              specified under -d
#              (where all the input/output files reside)
#
#  Optional:
#  -x path  : Full path to the Topkapi binary (Windows-style path)
#  -s path  : Full or relative path to the condor submit file
#  -w path  : Full or relative path to the wrapper script for Topkapi
#              Pre, Run and Post execution
#  -h       : Print help message
#

# Function to inform the user on usage
# Arguments : none
function print_usage() {
  echo ""
  echo "Script to run ensembles for the Topkapi model on a Condor Grid Engine."
  echo "It is assumed that the condor bin directory is in the PATH."
  echo ""
  echo "Usage: $0 options"
  echo "  options are:"
  echo "    Mandatory:"
  echo "    -o path  : The directory for the output log files (redirected stdout/stderr)"
  echo "    -n num   : Total number of ensemble members e.g. 16"
  echo "    -t num   : Number of seconds after which the condor cluster job should time out"
  echo "    -d path  : Root directory for the Topkapi module (remote network path)"
  echo "                (where all the input/output files reside)"
  echo ""  
  echo "    Optional:"
  echo "    -x path  : Full path to the Topkapi binary (Windows-style path)"
  echo "    -s path  : Full or relative path to the condor submit file"
  echo "    -w path  : Full or relative path to the wrapper script for Topkapi"
  echo "                Pre, Run and Post execution"
  echo "    -h       : Print help message"
  echo ""
}

# Function to create contents of the condor submit file
function create_submit() {
  echo "########################"
  echo "#"
  echo "# Submit description file for Topkapi run"
  echo "# Created: by $0 on $(date +"%F %T %Z") on $(hostname)" 
  echo "#"
  echo "########################"
  echo "Executable     = ${Topkapi_WRAPPER}"
  echo "Universe       = vanilla"
  echo 'Output         = $(Process)\log\Topkapi.out.$(Cluster).$(Process)'
  echo "Log            = ${CONDOR_Topkapi_LOG}"
  echo 'Error          = $(Process)\log\Topkapi.err.$(Cluster).$(Process)'
  echo "Notification   = Error"
  echo ""
  echo "Arguments      = \"'${Topkapi_BIN}' '${Topkapi_ROOT_DIR}' \$(Process)\""
  echo ""
  echo "run_as_owner = false"
  echo ""
  echo "Queue ${ENSEMBLE_SIZE}"
  echo ""
}

# Path to Topkapi executable can be overridden by command line option -x
Topkapi_BIN="E:\FEWS_PO_TestBench_01042008\Po_SA\Modules\TopkapiParallel\bin\topkapi.exe"

# Path to wrapper for Topkapi Pre, Run, and Post execution can be overridden by command line option -w
Topkapi_WRAPPER=Topkapi_pre_run_post.bat

# Path to condor submit file can be overridden by command line option -s
CONDOR_Topkapi_SUBMIT=Topkapi.submit

# Name of the condor log file
CONDOR_Topkapi_LOG=Topkapi.log

# Process command line options
while getopts "ho:x:w:s:d:t:n:" options; do
  case $options in 
   o ) LOGFILE_PATH=$OPTARG ;;
   x ) Topkapi_BIN=$OPTARG ;;
   w ) Topkapi_WRAPPER=$OPTARG ;;
   d ) Topkapi_ROOT_DIR=$OPTARG ;;
   t ) CONDOR_TIMEOUT=$OPTARG ;;
   n ) ENSEMBLE_SIZE=$OPTARG ;;
   h ) print_usage  # help option
          exit 0;;
   \? ) print_usage  # unknown option
          exit 1;;
   * )  print_usage  # any other value/option
          exit 1;;
  esac
done


# Check path to lisflood wrapper
if [ ! -e ${Topkapi_WRAPPER} ];
then
  echo "Error: Topkapi wrapper ${Topkapi_WRAPPER} does not exist."
  exit 1
fi

if [ ! -x ${Topkapi_WRAPPER} ];
then
  echo "Error: ${Topkapi_WRAPPER} is not an executable file."
  exit 1
fi


# Ensure that all options have been set correctly

# Test logfile path
if [ -z ${LOGFILE_PATH} ];
then
  echo "Error: output log file path not set (use option -o)."
  exit 1
fi

if [ ! -d ${LOGFILE_PATH} ];
then
  echo "Error: output log file path ${LOGFILE_PATH} "
  echo "  does not exist or is not a directory."
  exit 1
fi


# Test Topkapi root directory
if [ -z ${Topkapi_ROOT_DIR} ];
then
  echo "Error: root directory for Topkapi module not set"
  echo "  (use option -d)."
  exit 1
fi

if [ ! -d ${Topkapi_ROOT_DIR} ];
then
  echo "Error: root directory for Topkapi module ${Topkapi_ROOT_DIR} "
  echo "  does not exist or is not a directory."
  exit 1
fi

# Test number of ensemble members
if [ -z ${ENSEMBLE_SIZE} ];
then
  echo "Error: number of ensemble members not set (use option -n)."
  exit 1
fi

# Test time out 
if [ -z ${CONDOR_TIMEOUT} ];
then
  echo "Error: time out for entire enseble run (in seconds) not set (use option -t)."
  exit 1
fi

# Create the subdirectory for the condor log files
declare -i i=0
while [ $i -lt ${ENSEMBLE_SIZE} ];
do
  if [ ! -d ${Topkapi_ROOT_DIR}/$i/log ]; 
  then
    echo "Creating directory ${Topkapi_ROOT_DIR}/$i/log"
    mkdir -p ${Topkapi_ROOT_DIR}/$i/log
    if [ ! -d ${Topkapi_ROOT_DIR}/$i/log ];
    then
      echo "ERROR: Unable to create directory ${Topkapi_ROOT_DIR}/$i/log"
      exit 1
    fi
  fi
  i=$i+1
done

# Update the condor submit file
create_submit > ${CONDOR_Topkapi_SUBMIT}

# Submit the job and retrieve the cluster id
# condor_submit returns the following lines:
#  Submitting job(s).....
#  Logging submit event(s).....
#  5 job(s) submitted to cluster 86.
echo "Submitting condor job..."
CONDOR_SUBMIT_OUTPUT=$(condor_submit ${CONDOR_Topkapi_SUBMIT} 2>&1)
echo ${CONDOR_SUBMIT_OUTPUT}
CLUSTER_ID=$(echo ${CONDOR_SUBMIT_OUTPUT} | grep "to cluster" | grep -o -E "[0-9]+\.+"|grep -o -E "[0-9]+")

# Wait until the cluster job is done
echo
echo "Waiting for a maximum of ${CONDOR_TIMEOUT} second(s) for condor job with cluster ID ${CLUSTER_ID}"
echo " by watching condor log file $(pwd)/${CONDOR_Topkapi_LOG}"
condor_wait -wait ${CONDOR_TIMEOUT} ${CONDOR_Topkapi_LOG} ${CLUSTER_ID}
wait
retval_wait=$?
echo "condor_wait returned value: ${retval_wait}"

# Parse and combine the diagnostics files

# Clean up 

# Return the return value of condor_wait
exit ${retval_wait}