source: branches/2015/dev_r5092_CNRS_SETTE/NEMOGCM/TRUST/trusting_func.sh @ 5465

#!/bin/bash


comments() {
    state=$1

    if [ "$state" == 'E R R O R' ]; then
   if [ -e ocean.output ]; then
       last_time_step=$( cat ${TEST_DIR}/time.step | tr -d [:space:] )
       STEP='nemo.exe crashed at '${last_time_step} && export STEP
       line=$( grep -m1 -A5 "$state" ocean.output | tr -d '\n' )
   else
       STEP='nemo.exe crashed at initialization'    && export STEP
       return
   fi
    fi

    if [ $( diff -q model.log ${REFE_DIR}/model.log ) ]; then
   line=$( diff model.log ${REFE_DIR}/model.log  | tr -d '\n' )
    fi

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

get_cpu_time() {
    # Portability !
    real_cpu_time=$( ${JOB_VIEW} ${JOB_ID} | awk 'END{print $9}' )
    printf "Real CPU time\n${real_cpu_time}\n" | tee mesg_10_realcputime_${CFG_USER}_${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
}

diff_inputs() {
    mesg='Same'

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

    printf "Input files\n$mesg\n" | tee mesg_09_inputfiles_${CFG_USER}_${CFG_ARCH}.txt
}

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

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

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

       nb_dom=$( find ${TEST_DIR} -name "${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
       fi

       # UNIX `cmp` not suitable (filename & timestamp in .nc file)
       nc_diff=$( $CDO diffn ${TEST_DIR}/$file'.nc' ${REFE_DIR}/$file'.nc' 2> /dev/null | sed -n '$p' )
       if [ ! -z "${nc_diff}" ]; then
      export STATUS='FAILED' STEP='Restart files different'
      printf "$CDO diffn ${nc_diff}\n"
       else
      printf "identical\n"
       fi

       rm -f ${TEST_DIR}/${base_name}_${comp}_[0-9]*.nc &> /dev/null
   done

    else
   export STATUS='FAILED'; STEP='No restart files found at last time step'
   printf "no restart files found for comparison\n"
    fi

}

diff_textfiles() {
    for file in ocean.output $( ls *.stat ); do
   diff $file ${REFE_DIR}/$file
   [ $? -ne 0 ] && export STATUS='FAILED' STEP='ASCII output file diff'
    done
}

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\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"   ) \
     || $( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' )  ]]; then
   rev=$( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) || rev='HEAD'
   svn_cmd='svn update -r '$rev 
    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${rev_loc}\n" >  mesg_05_xios__${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 '$rev         >> model.log
#   printf "NEMOGCM rev.\n$rev\n" >  mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt
    printf "NEMOGCM rev.\n\
            <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='sed -n 2p'; 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 $EMAIL && "$STATUS" == 'FAILED' ]]; then
      cat << END_MAIL > trusting.mail
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
Dear $USER,

The trusting sequence for ${CONF_REF} has failed at step:
`tail -n 1 ${TEST_DIR}/mesg_03_step_${CFG_USER}_${CFG_ARCH}.txt`
Directory : ${TEST_DIR}

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 ${CONF_REF}] $STATUS $STEP" $EMAIL  <  trusting.mail
  fi

fi
}

print_step() {
    [ ! -z "$1" ] && STEP=$1
    export STEP && printf "Step.....\n$STEP\n"
}

get_out() {
    # Save tested configuration if trusting failed
    if [ "$STATUS" == 'FAILED' ]; then
   cd ${TEST_DIR}
   touch mesg_09_inputfiles_${CFG_USER}_${CFG_ARCH}.txt mesg_10_realcputime_${CFG_USER}_${CFG_ARCH}.txt \
         mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt
   #tar -czf ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz *
    fi

    printf "Status\n$STATUS\n"  > mesg_02_status_${CFG_USER}_${CFG_ARCH}.txt
    printf "Step.....\n$STEP\n" > mesg_03_step_${CFG_USER}_${CFG_ARCH}.txt

    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####################################################" >> computation.log
   ${JOB_VIEW} ${JOB_ID}                                           >> computation.log
   sleep ${time_increment}
   let time_elapsed+=${time_increment}
    done

    sleep ${time_increment}

    if [ ${time_elapsed} -eq ${TIME_LIMI} ]; then
   STEP='Exceeded time limit'

   if [ $( ${JOB_LIST} | grep ${JOB_ID} ) ]; then
       STEP='Job not finished on time: '$(( ${TIME_LIMI}/3600 ))'h'
   fi

   print_step
   ${JOB_DELE} ${JOB_ID} &> /dev/null
   get_out $STATUS $STEP
    fi
}