source: branches/2015/dev_r5092_CNRS_SETTE/NEMOGCM/TRUST/trusting_func.sh @ 5528

#!/bin/bash


comments() {
    state=$1; LAST_TIME_STEP=0

    if [ "$state" == 'E R R O R' ]; then
   [ -e time.step ] && LAST_TIME_STEP=$( cat time.step | tr -d [:space:] )
   export ${LAST_TIME_STEP}
    fi

    [ -e ocean.output ] && line=$( grep -m1 -A5 "$state" ocean.output | tr -d '\n' )

    if [[ -e ${REFE_DIR}/model.log && $( diff -q model.log ${REFE_DIR}/model.log ) ]]; then
   line=$( diff model.log ${REFE_DIR}/model.log  | grep '>' )
    fi

    printf "Comments\n$line\n" | tee mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt
}


diff_inputs() {
    mesg='Same'
    for file in inputs_list.txt $( ls namelist_* ) $( ls *.xml ); do
   diff -q $file ${REFE_DIR}/$file > /dev/null
   [ $? -ne 0 ] && mesg='Different'
    done

    printf "Input files\n$mesg\n" | tee mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt
}

diff_results() {
    for file in ocean.output $( ls *.stat ); do
   [ ! -e ${REFE_DIR}/$file ] && export ST='FAILED' && get_out 7
   diff -q $file ${REFE_DIR}/$file
   [ $? -ne 0 ] && export ST='FAILED'
    done
}

diff_restart() {
    [ ! -e  ${REFE_DIR}/time.step ] && export ST='FAILED' && get_out 9

    export LAST_TIME_STEP=$( cat ${REFE_DIR}/time.step | tr -d [:space:] )
    echo 'Last time step of standard run: '${LAST_TIME_STEP}

#    if [ $( find -name "*${LAST_TIME_STEP}_restart*.nc" -print -quit ) ]; then
    if [ $( find -regex ".*_0+${LAST_TIME_STEP}_restart.*\.nc" -print -quit ) ]; then
#  base_name=$( find -name "*${LAST_TIME_STEP}_restart*.nc" -print -quit | awk -F/ '{print $NF}' \
#               | sed "s/^\(.*\)$LAST_TIME_STEP\_restart.*/\1$LAST_TIME_STEP\_/"                  )
   base_name=$( find -regex ".*_0+${LAST_TIME_STEP}_restart.*\.nc" -print -quit | awk -F/ '{print $NF}' \
                | sed "s/^\(.*\)$LAST_TIME_STEP\_restart.*/\1$LAST_TIME_STEP\_/"                         )

   for comp in restart restart_ice restart_trc; do
       file=${base_name}${comp} && printf "$file.nc: "

       nb_dom=$( find -name "${file}_[0-9]*.nc" | wc -l | awk '{print $1}' )
       if   [ ${nb_dom} -gt 1 ]; then
      ${NEMO_TRBD}/rebuild_nemo -t $NPROC $file ${nb_dom} > /dev/null
      [ $? -eq 0 ] && rm -f ${base_name}${comp}_[0-9]*.nc > /dev/null
       elif [ ${nb_dom} -eq 0 ]; then
      export ST='FAILED' && get_out 10
       fi

       if [ -e ${REFE_DIR}/$file.nc ]; then
       # UNIX `cmp` not suitable (filename & timestamp in .nc file)
      nc_diff=$( $CDO diffn $file.nc ${REFE_DIR}/$file.nc 2> /dev/null \
                 | awk '/records/ {print $1}'                           )

      if [ ${nc_diff} -ne 0 ]; then
          export ST='FAILED'
          printf "$CDO ${nc_diff}\n"
      else
          printf "identical\n"
      fi

       else
      export ST='FAILED' && get_out 9
       fi

   done

    else
   export ST='FAILED'
    fi

}

get_cpu_time() {
    real_cpu_time=$( eval ${JOB_TIME} )
    printf "Real CPU time\n${real_cpu_time}\n" | tee mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt
}

get_inputs() {
    if [ ! -z "${FORC_TAR}" ]; then
   cmd_iol="tar -tvf ${NEMO_FORC}/${FORC_TAR}"; cmd_iof="tar -vxf ${NEMO_FORC}/${FORC_TAR}"
    else
   cmd_iol="ls ${NEMO_FORC}/*"                ; cmd_iof="\cp ${NEMO_FORC}/* ."
    fi

    ${cmd_iol} > inputs_list.txt
    ${cmd_iof} > /dev/null
}

get_soft_rel() {
    for rel in $CDO $COMPILER $MPI $NETCDF; do
                                     arch_rel=$( echo $LOADEDMODULES | sed  "s#.*$rel/\([^:]*\).*#\1#" )
   [ -z "${arch_rel}"  ]     && arch_rel=$( echo $PATH          | sed  "s#.*$rel/\([^/]*\).*#\1#" )
   [ "$rel" == "$COMPILER" ] && arch_rel=$( $rel --version      | grep -m1 -o ' [0-9.]* '         )
   echo $rel ${arch_rel} >> model.log
    done

    sed -n 4p model.log | sed 'i\Fortran compiler' > mesg_06_compiler_${CFG_USER}_${CFG_ARCH}.txt
    sed -n 5p model.log | sed 'i\MPI libs'         > mesg_07_mpi_${CFG_USER}_${CFG_ARCH}.txt
    sed -n 6p model.log | sed 'i\NetCDF libs'      > mesg_08_netcdf_${CFG_USER}_${CFG_ARCH}.txt
}

get_nemo_rev() {
    svn_opt='status'

    # If -v|--version option has been set, modify svn command
    if   [ $( echo ${NEMO_VERS} | grep  "HEAD\|up\|update"   ) ]; then
   svn_cmd='svn update -r HEAD'
    elif [ $( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) ]; then
   svn_cmd='svn update -r '$( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' )
    else
   svn_cmd='svn '${svn_opt}
    fi

    rev=0
    for dir in ${NEMO_ARCH} ${NEMO_CONF} ${NEMO_ENGI} \
               ${NEMO_EIOI} ${NEMO_EFCM}              \
          ${NEMO_TCMP} ${NEMO_TRBD}              \
          ${DIR_XIOS}                             ; do

   # For time being, just get revision number from XIOS
   if [ "$dir" == "${DIR_XIOS}" ]; then
       rev_loc=$( svn info $dir | awk '(NR == 5) {print $NF}' )
       echo   'XIOS '${rev_loc} >> model.log
       printf "XIOS rev.\n"     >  mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt
       printf "<a href=\"https://forge.ipsl.jussieu.fr/ioserver/changeset/${rev_loc}\" target=\"_blank\">${rev_loc}</a>" \
                                >> mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt
       continue
   fi

   echo ${svn_cmd} $dir': '$( ${svn_cmd} $dir )
   rev_loc=$( svn info $dir | awk '(NR == 5) {print $NF}' )
   (( ${rev_loc} >= $rev )) && rev=${rev_loc}
    done

    echo   'NEMOGCM '$rev   >> model.log
    printf "NEMOGCM rev.\n" >  mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt
    printf "<a href=\"https://forge.ipsl.jussieu.fr/nemo/changeset/$rev\" target=\"_blank\">$rev</a>" \
                       >> mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt
}

mesg_make() {
    paste -d ';' mesg_*_${CFG_USER}_${CFG_ARCH}.txt > trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt
    cat                                               trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt
}

mesg_publish() {
  
if [ $PUBLISH -eq 1 ]; then

  if [ -f ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt ]; then cmd='tail -1'; else cmd='cat'; fi

  $cmd trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt >> ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt

  # Send mail only when FAILED
  if [[ ! -z $EMAIL && "$ST" == 'FAILED' ]]; then
      cat << END_MAIL > trusting.mail
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Dear $USER,

The trusting sequence for ${REFE_CONF} has failed.
Directory: ${TEST_DIR}
Archive: ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz

XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

END_MAIL

      if [ -e ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt ]; then
     cat ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt  >> trusting.mail
      fi
      mail -s "[trusting ${REFE_CONF}] $ST $ERR" $EMAIL  <  trusting.mail
  fi

fi
}

print_step() { printf "Step.....\n$1\n"; }

get_out() {
    ERR=$1

    printf "Status\n$ST\n"  > mesg_02_status_${CFG_USER}_${CFG_ARCH}.txt

    if [ "$ST" == 'FAILED' ]; then
   cd ${TEST_DIR}

   if [ ! -e mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt ]; then
       printf "Input files\n\n"   > mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt
   fi

   if [ ! -e mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt ]; then
       printf "Real CPU time\n\n" > mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt
   fi

   if [ ! -e mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt ]; then
       printf "Comments\n\n"      > mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt
   fi

        # Error identification
   case $ERR in
       # Compilation
       1) ERR='XIOS compilation failed'            ;;  2) ERR='NEMO compilation failed'            ;;
       # Submission
       3) ERR='Missing input files'                 ;;  4) ERR='Job submission error'               ;;
       # Running
       5) ERR='nemo.exe crashed at '${LAST_TIME_ERR};;  6) ERR='Exceeded time limit '${TIME_LIMI}'h';;
       # Results
       7) ERR='Missing previous outputs '           ;;  8) ERR='New outputs  differ/missing'        ;;
       9) ERR='Missing previous restarts'           ;; 10) ERR='New restarts differ/missing'        ;;
       # Undefined
       *) ERR='Unknown error'                       ;;
   esac

   export ERR
    fi

    printf "Result.....\n$ERR\n" > mesg_03_result_${CFG_USER}_${CFG_ARCH}.txt

    # Save tested configuration if trusting failed
    [ "$ST" == 'FAILED' ] && tar -czf ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz *

    mesg_make
    mesg_publish

    exit 1
}

job_pending() {
    time_elapsed=0; time_increment=30

    sleep ${time_increment}
    while [[ $( ${JOB_LIST} | grep ${JOB_ID} ) && ${time_elapsed} -lt ${TIME_LIMI} ]]; do
   printf "\n####################################################\n" >> computation.log
   ${JOB_INFO} ${JOB_ID}                                             >> computation.log
   sleep ${time_increment}
   time_elapsed=$(( ${time_elapsed} + ${time_increment} ))
    done
    sleep ${time_increment}

    if [ ${time_elapsed} -eq ${TIME_LIMI} ]; then
   ${JOB_DELE} ${JOB_ID} &> /dev/null
   TIME_LIMIT=$(( ${TIME_LIMI} / 3600 ))
   export ${TIME_LIMIT}
   get_out $ST ERR=6
    fi
}