nordugrid-arc-arex 5.3.0~rc1-1 / usr / share / arc / submit-lsf-job

#!/bin/sh
# set -xv
#
#  Submits job to LSF
#  Input: path to grami file (same as Globus).
#
# A temporary job script is created for the submission and then removed 
# at the end of this script. 

echo "----- starting submit_lsf_job -----" 1>&2
joboption_lrms=lsf

# ARC1 passes first the config file.
if [ "$1" = "--config" ]; then shift; ARC_CONFIG=$1; shift; fi

basedir=`dirname $0`
basedir=`cd $basedir > /dev/null && pwd` || exit $?

pkgdatadir="$basedir"

##############################################################
# Set LSF specific environment. # relies on $joboption_queue #
##############################################################

. ${pkgdatadir}/configure-lsf-env.sh || exit $?
. ${pkgdatadir}/submit_common.sh || exit $?

##############################################################
# Parse grami file, read arc config
##############################################################
perflogfilesub="${perflogdir}/submission.perflog"

if [ ! -z "$perflogdir" ]; then
   start_ts=`date +%s.%N`
fi

init $1

read_arc_conf

# enforce this 
#RUNTIME_FRONTEND_SEES_NODE='yes'
#RUNTIME_NODE_SEES_FRONTEND='yes'

#TODO should this be specified here?
joboption_localtransfer='no'

##############################################################
# Zero stage of runtime environments
##############################################################
RTE_stage0

##############################################################
# create temp job script
##############################################################
mktempscript

##############################################################
# Start job script
##############################################################

LSF_BSUB='bsub'
LSF_BPARAMS='bparams'
if [ ! -z "$LSF_BIN_PATH" ] ; then
  LSF_BSUB=${LSF_BIN_PATH}/${LSF_BSUB}
  LSF_BPARAMS=${LSF_BIN_PATH}/${LSF_BPARAMS}
fi

echo "#! /bin/bash" > $LRMS_JOB_SCRIPT
echo "#LSF batch job script built by grid-manager" >> $LRMS_JOB_SCRIPT
echo "#" >> $LRMS_JOB_SCRIPT

# Specify the bash shell as default
#echo "#BSUB -L /bin/bash" >> $LRMS_JOB_SCRIPT

# Write output to comment file:
echo "#BSUB -oo ${joboption_directory}.comment" >> $LRMS_JOB_SCRIPT

echo "" >> $LRMS_JOB_SCRIPT

# Choose queue(s).
if [ ! -z "${joboption_queue}" ] ; then
  echo "#BSUB -q $joboption_queue" >> $LRMS_JOB_SCRIPT
fi

if [ ! -z "${joboption_rsl_architecture}" ] ; then
     queuearch=`echo ${joboption_rsl_architecture}|sed 's/\"//g'`
     echo "#BSUB -R type=${queuearch}" >> $LRMS_JOB_SCRIPT
else
   if [ ! -z $CONFIG_lsf_architecture ] ; then
     echo "#BSUB -R type=$CONFIG_lsf_architecture" >> $LRMS_JOB_SCRIPT
   fi
fi

# Project name for accounting
if [ ! -z "${joboption_rsl_project}" ] ; then
  echo "#BSUB -P $joboption_rsl_project" >> $LRMS_JOB_SCRIPT
fi

# job name for convenience
if [ ! -z "${joboption_jobname}" ] ; then
  jobname=`echo "$joboption_jobname" | \
           sed 's/^\([^[:alpha:]]\)/N\1/' | \
           sed 's/[^[:alnum:]]/_/g' | \
	   sed 's/\(...............\).*/\1/'`
  echo "#BSUB -J $jobname" >> $LRMS_JOB_SCRIPT
fi
echo "LSF jobname: $jobname" 1>&2

##############################################################
# (non-)parallel jobs
##############################################################
set_count

##############################################################
# parallel jobs
##############################################################
echo "#BSUB -n $joboption_count" >> $LRMS_JOB_SCRIPT

# parallel structure
if [ ! -z $joboption_countpernode ] && [ $joboption_countpernode != '-1' ] ; then
 echo "#BSUB -R span[ptile=$joboption_countpernode]" >> $LRMS_JOB_SCRIPT
fi 
# exclusive execution 
if [ "$joboption_exclusivenode" = "true" ]; then
  echo "#BSUB -x" >> $LRMS_JOB_SCRIPT
fi

##############################################################
# Execution times (obtained in seconds)
##############################################################

#OBS: Assuming here that LSB_JOB_CPULIMIT=y or is unset.
if [ -n "$joboption_cputime" ] && [ $joboption_cputime -gt 0 ]; then
    cputime=$(( ${joboption_cputime} / 60 ))
    echo "#BSUB -c ${cputime}" >> $LRMS_JOB_SCRIPT
fi

if [ -n "$joboption_walltime" ] && [ $joboption_walltime -gt 0 ] ; then
    walltime=$(( ${joboption_walltime} / 60 ))
    echo "#BSUB -W ${walltime}" >> $LRMS_JOB_SCRIPT	
fi	

##############################################################
# Requested memory (mb)
##############################################################

set_req_mem

#-M is memory limit per process in LSF, so no need to modify memory limit based on count.

if [ ! -z "$joboption_memory" ]; then
  memory=$(( ${joboption_memory} * 1024 ))
  echo "#BSUB -M ${memory}" >> $LRMS_JOB_SCRIPT
fi

##############################################################
# Start Time
##############################################################
if [ -n "$joboption_starttime" ] ; then
  echo "#BSUB -b ${joboption_starttime}" >> $LRMS_JOB_SCRIPT
fi

##############################################################
# priority
##############################################################
if [ ! -z "$joboption_priority" ]; then
  #first we must parse the max priority
  maxprio=`bparams -a| grep MAX_USER_PRIORITY | cut -f 2 -d '=' | cut -f 2 -d ' '`
  #scale priority LSF: 1 -> MAX_USER_PRIORITY ARC: 0-100
  if [ ! -z "$maxprio" ]; then
    if [ "$maxprio" -gt "0" ]; then
      priority=$((joboption_priority * ($maxprio - 1) / 100 +1))
      echo "#BSUB -sp ${priority}" >> $LRMS_JOB_SCRIPT
    fi
  fi
fi

##############################################################
# Override umask
##############################################################

echo "# Overide umask of execution node (sometime values are really strange)" >> $LRMS_JOB_SCRIPT
echo "umask 077" >> $LRMS_JOB_SCRIPT
echo " " >> $LRMS_JOB_SCRIPT

sourcewithargs_jobscript



##############################################################
# Add environment variables
##############################################################
add_user_env

##############################################################
# Check for existence of executable,
# there is no sense to check for executable if files are 
# downloaded directly to computing node
##############################################################
if [ -z "${joboption_arg_0}" ] ; then
  echo 'Executable is not specified' 1>&2
  rm -f "$LRMS_JOB_SCRIPT" "$LRMS_JOB_OUT" "$LRMS_JOB_ERR"
  echo "Submission: Job description error.">>"$failures_file"
  exit 1
fi
if [ ! "$joboption_localtransfer" = 'yes' ] ; then
  program_start=`echo ${joboption_arg_0} | head -c 1 2>&1`
  if [ "$program_start" != '$' ] && [ "$program_start" != '/' ] ; then
    if [ ! -f $joboption_directory/${joboption_arg_0} ] ; then 
      echo 'Executable does not exist, or permission denied.' 1>&2
      echo "   Executable $joboption_directory/${joboption_arg_0}" 1>&2
      echo "   whoami: "`whoami` 1>&2
      echo "   ls -l $joboption_directory/${joboption_arg_0}: "`ls -l $joboption_directory/${joboption_arg_0}`
      exit 1
    fi
    if [ ! -x $joboption_directory/${joboption_arg_0} ] ; then 
      echo 'Executable is not executable' 1>&2
      exit 1
    fi
  fi
fi
#######################################################################
# copy information useful for transfering files to/from node directly
#######################################################################
if [ "$joboption_localtransfer" = 'yes' ] ; then
  setup_local_transfer
fi

######################################################################
# Adjust working directory for tweaky nodes
# RUNTIME_GRIDAREA_DIR should be defined by external means on nodes
######################################################################
if [ ! -z "${RUNTIME_NODE_SEES_FRONTEND}" ] ; then
  setup_runtime_env
else
  echo "RUNTIME_JOB_DIR=$RUNTIME_LOCAL_SCRATCH_DIR/$joboption_gridid" >> $LRMS_JOB_SCRIPT
  echo "RUNTIME_JOB_DIAG=$RUNTIME_LOCAL_SCRATCH_DIR/${joboption_gridid}.diag" >> $LRMS_JOB_SCRIPT
  RUNTIME_STDIN_REL=`echo "${joboption_stdin}" | sed "s#^${joboption_directory}/*##"`
  RUNTIME_STDOUT_REL=`echo "${joboption_stdout}" | sed "s#^${joboption_directory}/*##"`
  RUNTIME_STDERR_REL=`echo "${joboption_stderr}" | sed "s#^${joboption_directory}/*##"`
  if [ "$RUNTIME_STDIN_REL" = "${joboption_stdin}" ] ; then
    echo "RUNTIME_JOB_STDIN=\"${joboption_stdin}\"" >> $LRMS_JOB_SCRIPT
  else
    echo "RUNTIME_JOB_STDIN=\"$RUNTIME_LOCAL_SCRATCH_DIR/$joboption_gridid/$RUNTIME_STDIN_REL\"" >> $LRMS_JOB_SCRIPT
  fi
  if [ "$RUNTIME_STDOUT_REL" = "${joboption_stdout}" ] ; then
    echo "RUNTIME_JOB_STDOUT=\"${joboption_stdout}\"" >> $LRMS_JOB_SCRIPT
  else
    echo "RUNTIME_JOB_STDOUT=\"$RUNTIME_LOCAL_SCRATCH_DIR/$joboption_gridid/$RUNTIME_STDOUT_REL\"" >> $LRMS_JOB_SCRIPT
  fi
  if [ "$RUNTIME_STDERR_REL" = "${joboption_stderr}" ] ; then
    echo "RUNTIME_JOB_STDERR=\"${joboption_stderr}\"" >> $LRMS_JOB_SCRIPT
  else
    echo "RUNTIME_JOB_STDERR=\"$RUNTIME_LOCAL_SCRATCH_DIR/$joboption_gridid/$RUNTIME_STDERR_REL\"" >> $LRMS_JOB_SCRIPT
  fi
fi

##############################################################
# Add std... to job arguments
##############################################################
include_std_streams

##############################################################
#  Move files to local working directory (job is done on node only)
#  RUNTIME_JOB_DIR -> RUNTIME_LOCAL_SCRATCH_DIR/job_id
##############################################################
move_files_to_node

echo "" >> $LRMS_JOB_SCRIPT
echo "RESULT=0" >> $LRMS_JOB_SCRIPT
echo "" >> $LRMS_JOB_SCRIPT


#####################################################
#  Download input files
####################################################
download_input_files

#####################################################
#  Go to working dir and start job
####################################################
echo "# Changing to session directory" >> $LRMS_JOB_SCRIPT
echo "cd \$RUNTIME_JOB_DIR" >> $LRMS_JOB_SCRIPT
echo "export HOME=\$RUNTIME_JOB_DIR" >> $LRMS_JOB_SCRIPT

##############################################################
#  Skip execution if something already failed
##############################################################
echo "" >> $LRMS_JOB_SCRIPT
echo "if [ \"\$RESULT\" = '0' ] ; then" >> $LRMS_JOB_SCRIPT
echo "echo \"runtimeenvironments=\$runtimeenvironments\" >> \"\$RUNTIME_JOB_DIAG\"" >> $LRMS_JOB_SCRIPT


##############################################################
#  Runtime configuration at computing node
##############################################################
RTE_stage1

#extra checks

if [ -z "$RUNTIME_NODE_SEES_FRONTEND" ] ; then
  echo "Nodes detached from gridarea are not supported when LSF is used. Aborting job submit" 1>&2
  rm -f "$LRMS_JOB_SCRIPT" "$LRMS_JOB_OUT" "$LRMS_JOB_ERR"
  exit 1
fi

gate_host=`uname -n`
if [ -z "$gate_host" ] ; then 
  echo "Can't get own hostname" 1>&2
  rm -f "$LRMS_JOB_SCRIPT" "$LRMS_JOB_OUT" "$LRMS_JOB_ERR"
  exit 1
fi

##############################################################
#  Execution
##############################################################
cd_and_run

##############################################################
#  End of RESULT checks
##############################################################
echo "fi" >> $LRMS_JOB_SCRIPT

##############################################################
#  Runtime (post)configuration at computing node
##############################################################
configure_runtime

#####################################################
#  Upload output files
####################################################
if [ "$joboption_localtransfer" = 'yes' ] ; then
echo "UPLOADER=\${UPLOADER:-$basedir/uploader}" >> $LRMS_JOB_SCRIPT
cat >> $LRMS_JOB_SCRIPT <<'EOSCR'
if [ "$RESULT" = '0' ] ; then
  $UPLOADER -p -c 'local' "$RUNTIME_CONTROL_DIR" "$RUNTIME_JOB_DIR" 2>>${RUNTIME_CONTROL_DIR}/job.local.errors
  if [ $? -ne '0' ] ; then
    echo 'ERROR: Uploader failed.' 1>&2
    if [ "$RESULT" = '0' ] ; then RESULT=1 ; fi
  fi
fi
rm -f "${RUNTIME_CONTROL_DIR}/job.local.proxy"
EOSCR
fi


##############################################################
#  Move files back to session directory (job is done on node only)
#  RUNTIME_JOB_DIR -> RUNTIME_LOCAL_SCRATCH_DIR/job_id
##############################################################
move_files_to_frontend

#######################################
#  Submit the job
#######################################
# Execute bsub command
cd "$joboption_directory"
#chmod 0755 $LRMS_JOB_SCRIPT

# We make the assumption that $joboption_directory is locally available according to the requirements of any arc installation

echo "----------------- BEGIN job script -----" 1>&2
cat $LRMS_JOB_SCRIPT 1>&2
echo "----------------- END job script -----" 1>&2

if [ ! -z "$perflogdir" ]; then
   stop_ts=`date +%s.%N`
   t=`awk "BEGIN { printf \"%.3f\", ${stop_ts}-${start_ts} }"`
   echo "[`date +%Y-%m-%d\ %T`] submit-lsf-job, JobScriptCreation: $t" >> $perflogfilesub
fi

if [ ! -z "$perflogdir" ]; then
   start_ts=`date +%s.%N`
fi

${LSF_BSUB} < $LRMS_JOB_SCRIPT 1>$LRMS_JOB_OUT 2>$LRMS_JOB_ERR
LSF_RESULT="$?"

if [ ! -z "$perflogdir" ]; then
   stop_ts=`date +%s.%N`
   t=`awk "BEGIN { printf \"%.3f\", ${stop_ts}-${start_ts} }"`
   echo "[`date +%Y-%m-%d\ %T`] submit-lsf-job, JobSubmission: $t" >> $perflogfilesub
fi

if [ $LSF_RESULT -eq '0' ] ; then
   job_id=`cat $LRMS_JOB_OUT | awk '{split($0,field," ");print field[2]}' | sed 's/[<>]//g'`

   if [ "${job_id}" = "" ] ; then
      echo "job *NOT* submitted successfully!" 1>&2
      echo "failed getting the jobid for the job!" 1>&2
   else
      echo "joboption_jobid=$job_id" >> $arg_file
      echo "job submitted successfully!" 1>&2
      echo "local job id: $job_id" 1>&2
      # Remove temporary job script file      
      rm -f $LRMS_JOB_SCRIPT $LRMS_JOB_OUT $LRMS_JOB_ERR
      echo "----- exiting submit_lsf_job -----" 1>&2
      echo "" 1>&2
      exit 0
   fi
else
  echo "job *NOT* submitted successfully!" 1>&2
  echo "got error code from qsub!" 1>&2
fi

echo "Output is:" 1>&2
cat $LRMS_JOB_OUT 1>&2
echo "Error output is:"
cat $LRMS_JOB_ERR 1>&2

rm -f $LRMS_JOB_SCRIPT $LRMS_JOB_OUT $LRMS_JOB_ERR

echo "----- exiting submit_lsf_job -----" 1>&2
echo "" 1>&2
exit 1