[5268] | 1 | #!/bin/bash |
---|
| 2 | |
---|
[5383] | 3 | |
---|
[5644] | 4 | ## Messenger filenames |
---|
| 5 | FILE_DATE=mesg_01_date_$PATTERNAME.txt ; FILE_STAT=mesg_02_status_$PATTERNAME.txt |
---|
| 6 | FILE_RESU=mesg_03_result_$PATTERNAME.txt ; FILE_NEMO=mesg_04_nemogcm_$PATTERNAME.txt |
---|
| 7 | FILE_XIOS=mesg_05_xios_$PATTERNAME.txt ; FILE_COMP=mesg_06_compiler_$PATTERNAME.txt |
---|
| 8 | FILE_MPIN=mesg_07_mpi_$PATTERNAME.txt ; FILE_NCDF=mesg_08_netcdf_$PATTERNAME.txt |
---|
| 9 | FILE_INPT=mesg_09_inputs_$PATTERNAME.txt ; FILE_CPUT=mesg_10_cputime_$PATTERNAME.txt |
---|
| 10 | FILE_COMM=mesg_11_comments_$PATTERNAME.txt; |
---|
[5453] | 11 | |
---|
[5667] | 12 | ## Trusting summary & archive timestamped filenames |
---|
[5644] | 13 | FILE_TRUS=trusting_${DATE}_$PATTERNAME.txt; FILE_ARCH=trusting_${DATE}_$PATTERNAME.tar.gz |
---|
| 14 | |
---|
[5681] | 15 | |
---|
[5644] | 16 | print_step() { printf "\nStep.....\n$1\n"; } |
---|
| 17 | |
---|
[5681] | 18 | |
---|
[5644] | 19 | init_files() { |
---|
[5664] | 20 | echo 'Date' > ${FILE_DATE}; echo 'Status' > ${FILE_STAT} |
---|
| 21 | echo 'Result' > ${FILE_RESU}; echo 'NEMOGCM rev.' > ${FILE_NEMO} |
---|
| 22 | echo 'XIOS rev.' > ${FILE_XIOS}; echo 'Fortran compiler' > ${FILE_COMP} |
---|
| 23 | echo 'MPI libs' > ${FILE_MPIN}; echo 'NetCDF libs' > ${FILE_NCDF} |
---|
| 24 | echo 'Input files' > ${FILE_INPT}; echo 'Real CPU time' > ${FILE_CPUT} |
---|
| 25 | echo 'Comments' > ${FILE_COMM} |
---|
| 26 | |
---|
[5667] | 27 | # 'Failed' status with 'Unknown error' by default |
---|
[5664] | 28 | echo $ST >> ${FILE_STAT}; echo 'Unknown error' >> ${FILE_RESU} |
---|
[5644] | 29 | } |
---|
| 30 | |
---|
[5681] | 31 | |
---|
[5644] | 32 | get_date() { |
---|
| 33 | local dat=$( date -ud ${DATE} +"%F %R %Z" ) |
---|
| 34 | |
---|
| 35 | echo $dat >> ${FILE_DATE} |
---|
| 36 | } |
---|
| 37 | |
---|
[5681] | 38 | |
---|
[5644] | 39 | get_nemo_rev() { |
---|
| 40 | local rev_loc svn_cmd |
---|
| 41 | local rev=0 svn_opt='status' |
---|
| 42 | |
---|
| 43 | ## If -v|--version option has been set, modify svn command |
---|
| 44 | if [ $( echo ${NEMO_VERS} | grep "HEAD\|up\|update" ) ]; then |
---|
| 45 | svn_cmd='svn update -r HEAD' |
---|
| 46 | elif [ $( echo ${NEMO_VERS} | grep -o '{[0-9]\{4\}-[0-9]\{2\}-[0-9]\{2\}}' ) ]; then |
---|
| 47 | svn_cmd='svn update -r '$( echo ${NEMO_VERS} | grep -o '{[0-9]\{4\}-[0-9]\{2\}-[0-9]\{2\}}' ) |
---|
| 48 | elif [ $( echo ${NEMO_VERS} | grep -o '[0-9]*' ) ]; then |
---|
[5664] | 49 | svn_cmd='svn update -r '$( echo ${NEMO_VERS} | grep -o '[0-9]*' ) |
---|
[5633] | 50 | else |
---|
[5644] | 51 | svn_cmd='svn '${svn_opt} |
---|
[5455] | 52 | fi |
---|
[5453] | 53 | |
---|
[5644] | 54 | for dir in ${NEMO_ARCH} ${NEMO_CONF} ${NEMO_ENGI} \ |
---|
[5683] | 55 | ${NEMO_EAGR} ${NEMO_EIOI} ${NEMO_EFCM} \ |
---|
[5644] | 56 | ${NEMO_TCMP} ${NEMO_TRBD} \ |
---|
| 57 | ${DIR_XIOS} ; do |
---|
[5455] | 58 | |
---|
[5644] | 59 | ## For time being, just get revision number from XIOS |
---|
[5681] | 60 | if [ $dir == ${DIR_XIOS} ]; then |
---|
[5644] | 61 | rev_loc=$( svn info $dir | awk '(NR == 9) {print $NF}' ) |
---|
| 62 | echo 'XIOS '${rev_loc} >> model.log |
---|
| 63 | echo "<a href=\"https://forge.ipsl.jussieu.fr/ioserver/changeset/${rev_loc}\" target=\"_blank\">${rev_loc}</a>" \ |
---|
| 64 | >> ${FILE_XIOS} |
---|
| 65 | continue |
---|
| 66 | fi |
---|
| 67 | |
---|
| 68 | echo $dir |
---|
| 69 | ${svn_cmd} $dir |
---|
| 70 | rev_loc=$( svn info $dir | awk '(NR == 9) {print $NF}' ) |
---|
| 71 | (( ${rev_loc} >= $rev )) && rev=${rev_loc} |
---|
| 72 | done |
---|
| 73 | |
---|
| 74 | echo 'NEMOGCM '$rev >> model.log |
---|
| 75 | echo "<a href=\"https://forge.ipsl.jussieu.fr/nemo/changeset/$rev\" target=\"_blank\">$rev</a>" \ |
---|
| 76 | >> ${FILE_NEMO} |
---|
[5438] | 77 | } |
---|
| 78 | |
---|
[5681] | 79 | |
---|
[5644] | 80 | get_soft_rel() { |
---|
| 81 | local soft_rel |
---|
[5509] | 82 | |
---|
[5644] | 83 | for soft in $CDO $COMPILER $MPI $NETCDF; do |
---|
| 84 | soft_rel='' |
---|
[5664] | 85 | |
---|
| 86 | if (( ${I_MODULE} == 1 )); then |
---|
| 87 | soft_rel=$( echo $LOADEDMODULES | sed "s/.*$soft\/\([0-9.a-z_]*\).*/\1/i" ) |
---|
| 88 | else |
---|
| 89 | soft_rel=$( echo $PATH | sed "s/.*$soft\([0-9.a-z_]*\).*/\1/i" ) |
---|
| 90 | fi |
---|
| 91 | |
---|
[5681] | 92 | [ $soft == $COMPILER ] && soft_rel=$( $soft --version | grep -m1 -oe '\<[0-9. ]*\>' ) |
---|
[5672] | 93 | ## Cleaning characters string to display proper soft name |
---|
| 94 | soft=$( echo $soft | sed 's#\\##g; s#[/-]$##' ) |
---|
[5644] | 95 | echo $soft ${soft_rel} >> model.log |
---|
| 96 | done |
---|
| 97 | |
---|
[5672] | 98 | sed -n 4p model.log >> ${FILE_COMP}; sed -n 5p model.log >> ${FILE_MPIN} |
---|
| 99 | sed -n 6p model.log >> ${FILE_NCDF} |
---|
[5644] | 100 | } |
---|
| 101 | |
---|
[5681] | 102 | |
---|
[5644] | 103 | get_inputs() { |
---|
[5664] | 104 | local cmd_iol="tar -tvf ${NEMO_FORC}/${NEMO_TARF}" cmd_iof="tar -vxf ${NEMO_FORC}/${NEMO_TARF}" |
---|
[5644] | 105 | |
---|
| 106 | [ -z "${NEMO_TARF}" ] && cmd_iol="ls ${NEMO_FORC}/*" && cmd_iof="\cp ${NEMO_FORC}/* ." |
---|
| 107 | ${cmd_iol} > inputs_list.txt && ${cmd_iof} > /dev/null |
---|
| 108 | } |
---|
| 109 | |
---|
[5681] | 110 | |
---|
[5402] | 111 | diff_inputs() { |
---|
[5644] | 112 | local dif |
---|
[5664] | 113 | local files_list='' mesg='Same' |
---|
[5644] | 114 | |
---|
[5681] | 115 | for file in inputs_list.txt namelist_* *.xml cpp_*; do |
---|
[5644] | 116 | dif='' |
---|
[5664] | 117 | |
---|
[5681] | 118 | if [ -e ${REFE_DIR}/$file ]; then |
---|
| 119 | dif=$( diff -q $file ${REFE_DIR}/$file ) |
---|
| 120 | else dif=0; fi |
---|
| 121 | |
---|
[5664] | 122 | if [ -n "$dif" ]; then |
---|
| 123 | mesg='Different' |
---|
[5683] | 124 | [ "$dif" != '0' ] && echo $dif && files_list+=$file' ' |
---|
[5664] | 125 | fi |
---|
| 126 | |
---|
[5402] | 127 | done |
---|
| 128 | |
---|
[5685] | 129 | echo $mesg | tee -a ${FILE_INPT} |
---|
[5672] | 130 | [ -n "${files_list}" ] && echo 'Inputs : '${files_list}'differ<br>' \ |
---|
| 131 | >> temp_${FILE_COMM} |
---|
[5402] | 132 | } |
---|
| 133 | |
---|
[5681] | 134 | |
---|
[5644] | 135 | job_pending() { |
---|
[5672] | 136 | local outline=$( printf "%100s" ) time_elapsed=0 time_increment=30 |
---|
[5644] | 137 | |
---|
| 138 | sleep ${time_increment} |
---|
| 139 | |
---|
| 140 | while [[ $( ${JOB_LIST} | grep ${JOB_ID} ) && ${time_elapsed} -lt ${TIME_LIMI} ]]; do |
---|
[5672] | 141 | printf "\n%s\n" ${outline// /#} >> computation.log |
---|
| 142 | ${JOB_INFO} ${JOB_ID} >> computation.log |
---|
[5644] | 143 | sleep ${time_increment} |
---|
| 144 | time_elapsed=$(( ${time_elapsed} + ${time_increment} )) |
---|
| 145 | done |
---|
| 146 | |
---|
| 147 | sleep ${time_increment} |
---|
| 148 | |
---|
[5672] | 149 | if (( ${time_elapsed} == ${TIME_LIMI} )); then |
---|
[5644] | 150 | ${JOB_DELE} ${JOB_ID} &> /dev/null |
---|
[5664] | 151 | let TIME_LIMIT/=3600 |
---|
[5671] | 152 | export ${TIME_LIMIT} |
---|
[5644] | 153 | get_out 6 |
---|
| 154 | fi |
---|
| 155 | } |
---|
| 156 | |
---|
[5681] | 157 | |
---|
[5509] | 158 | diff_results() { |
---|
[5664] | 159 | local files_list='' |
---|
| 160 | |
---|
| 161 | for file in ocean.output *.stat; do |
---|
[5521] | 162 | [ ! -e ${REFE_DIR}/$file ] && export ST='FAILED' && get_out 7 |
---|
[5664] | 163 | diff -q $file ${REFE_DIR}/$file |
---|
| 164 | (( $? != 0 )) && export ST='FAILED' && files_list+=$file' ' |
---|
[5509] | 165 | done |
---|
[5664] | 166 | |
---|
[5672] | 167 | [ -n "${files_list}" ] && echo 'Results : '${files_list}'differ<br>' \ |
---|
| 168 | >> temp_${FILE_COMM} |
---|
[5509] | 169 | } |
---|
| 170 | |
---|
[5681] | 171 | |
---|
[5383] | 172 | diff_restart() { |
---|
[5681] | 173 | local base_name dif file list_comp list_tmsp nb_dom |
---|
| 174 | local files_list='' dift=0 |
---|
[5644] | 175 | |
---|
[5681] | 176 | [ ! -e ${REFE_DIR}/time.step ] && export ST='FAILED' && get_out 7 |
---|
[5671] | 177 | export TIME_STEP=$( cat ${REFE_DIR}/time.step | tr -d [:space:] ) |
---|
[5672] | 178 | echo 'Last time step of benchmark run: '${TIME_STEP} |
---|
[5485] | 179 | |
---|
[5672] | 180 | ## Find all restart files to rebuild |
---|
| 181 | if [ $( find -regex ".*_restart.*[0-9]\.nc" -print -quit ) ]; then |
---|
| 182 | base_name=$( find -regex ".*_restart.*[0-9]\.nc" \ |
---|
| 183 | | sed "s#^\./\(.*\)_[0-9]*_restart.*#\1#" | sort -u ) |
---|
| 184 | list_comp=$( find -regex ".*_restart.*[0-9]\.nc" \ |
---|
| 185 | | sed "s#^.*\(restart[a-z_]*\)_[0-9].*\.nc#\1#" | sort -u ) |
---|
| 186 | list_tmsp=$( find -regex ".*_restart.*[0-9]\.nc" \ |
---|
| 187 | | sed "s#^.*\([0-9]\{8\}\)_restart.*#\1#" | sort -u ) |
---|
[5383] | 188 | |
---|
[5672] | 189 | for tmsp in ${list_tmsp}; do |
---|
| 190 | for comp in ${list_comp}; do |
---|
| 191 | file=${base_name}_${tmsp}_${comp} |
---|
| 192 | nb_dom=$( find -name "${file}_[0-9]*.nc" | wc -l | awk '{ print $1 }' ) |
---|
[5424] | 193 | |
---|
[5672] | 194 | if (( ${nb_dom} > 1 )); then |
---|
| 195 | ${NEMO_TRBD}/rebuild_nemo -t $NPROC $file ${nb_dom} > /dev/null |
---|
| 196 | [ $? -eq 0 ] && rm -f ${file}_[0-9]*.nc > /dev/null |
---|
| 197 | elif (( ${nb_dom} == 0 )); then |
---|
| 198 | export ST='FAILED' && get_out 8 |
---|
| 199 | fi |
---|
[5424] | 200 | |
---|
[5672] | 201 | ## Compare restart files at same time step |
---|
| 202 | if (( $tmsp == ${TIME_STEP} )); then |
---|
| 203 | if [ -e ${REFE_DIR}/$file.nc ]; then |
---|
| 204 | ## UNIX `cmp` not suitable (timestamp in .nc file) |
---|
| 205 | dif=$( $CDO diffn $file.nc ${REFE_DIR}/$file.nc 2> /dev/null \ |
---|
| 206 | | awk '/records/ {print $0}' | sed '2 s/^/,/' | tr -d '\n' ) |
---|
[5485] | 207 | |
---|
[5681] | 208 | if [[ -n "$dif" && (( $( echo $dif | awk '{print $1}' ) -ne 0 )) ]]; then |
---|
[5672] | 209 | export ST='FAILED' && files_list+=$file' ' |
---|
| 210 | let dift+=$( echo $dif | awk '{print $1}' ) |
---|
[5681] | 211 | echo $file.nc': '$dif |
---|
[5672] | 212 | fi |
---|
| 213 | |
---|
[5681] | 214 | else export ST='FAILED' && get_out 7; fi |
---|
[5672] | 215 | |
---|
[5681] | 216 | else continue; fi |
---|
[5485] | 217 | |
---|
[5672] | 218 | done |
---|
[5424] | 219 | done |
---|
[5383] | 220 | |
---|
[5681] | 221 | if (( $dift != 0 )); then |
---|
| 222 | echo 'Restarts: '${files_list}$dift' record(s) differ<br>' \ |
---|
| 223 | >> temp_${FILE_COMM} |
---|
| 224 | else echo 'No restart differ'; fi |
---|
[5455] | 225 | |
---|
[5681] | 226 | else export ST='FAILED'; fi |
---|
| 227 | |
---|
[5383] | 228 | } |
---|
| 229 | |
---|
[5681] | 230 | |
---|
[5472] | 231 | get_cpu_time() { |
---|
[5644] | 232 | local real_cpu_time=$( eval ${JOB_TIME} ) |
---|
[5472] | 233 | |
---|
[5644] | 234 | echo ${real_cpu_time} | tee -a ${FILE_CPUT} |
---|
[5472] | 235 | } |
---|
| 236 | |
---|
[5681] | 237 | |
---|
[5644] | 238 | comments() { |
---|
| 239 | local state=$1 |
---|
[5438] | 240 | |
---|
[5644] | 241 | if [ "$state" == 'E R R O R' ]; then |
---|
[5669] | 242 | [ -e time.step ] && export TIME_STEP=$( cat time.step | tr -d ' ' ) |
---|
[5644] | 243 | [ -e ocean.output ] && line=$( grep -m1 -A4 "$state" ocean.output | tr -d '\n' ) |
---|
[5444] | 244 | else |
---|
[5644] | 245 | [ -e ocean.output ] && line=$( grep -m1 -A2 "^ $state" ocean.output | tr -d '\n' ) |
---|
[5438] | 246 | fi |
---|
| 247 | |
---|
[5664] | 248 | echo $line |
---|
[5667] | 249 | printf "$line<br>" >> temp_${FILE_COMM} |
---|
[5383] | 250 | } |
---|
| 251 | |
---|
[5681] | 252 | |
---|
[5383] | 253 | mesg_make() { |
---|
[5669] | 254 | ## Format comments for web |
---|
[5672] | 255 | [ -e temp_${FILE_COMM} ] && cat temp_${FILE_COMM} | tr -d '\n' | sed 's/<br>$//' \ |
---|
| 256 | >> ${FILE_COMM} |
---|
[5669] | 257 | |
---|
[5664] | 258 | paste -d ';' mesg_*.txt | tee ${FILE_TRUS} |
---|
[5268] | 259 | } |
---|
| 260 | |
---|
[5681] | 261 | |
---|
[5383] | 262 | mesg_publish() { |
---|
[5644] | 263 | local cmd |
---|
[5268] | 264 | |
---|
[5644] | 265 | if [ $PUBLISH -eq 1 ]; then |
---|
[5268] | 266 | |
---|
[5644] | 267 | if [ -f ${REFE_DIR}/trusting_$PATTERNAME.txt ]; then cmd='tail -1'; else cmd='cat'; fi |
---|
[5424] | 268 | |
---|
[5644] | 269 | $cmd ${FILE_TRUS} >> ${REFE_DIR}/trusting_$PATTERNAME.txt |
---|
| 270 | |
---|
[5681] | 271 | ## Send mail only when FAILED |
---|
| 272 | if [[ ! -z "$EMAIL" && $ST == 'FAILED' ]]; then |
---|
| 273 | cat <<END_MAIL \ |
---|
| 274 | > trusting.mail |
---|
| 275 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
[5472] | 276 | |
---|
[5268] | 277 | Dear $USER, |
---|
| 278 | |
---|
[5472] | 279 | The trusting sequence for ${REFE_CONF} has failed. |
---|
| 280 | Directory: ${TEST_DIR} |
---|
[5644] | 281 | Archive created: ${FILE_ARCH} in ${REFE_DIR} |
---|
[5268] | 282 | |
---|
[5681] | 283 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
[5268] | 284 | |
---|
[5683] | 285 | `cat ${TEST_DIR}/${FILE_TRUS}` |
---|
[5268] | 286 | END_MAIL |
---|
[5683] | 287 | # cat ${TEST_DIR}/${FILE_TRUS} \ |
---|
| 288 | # >> trusting.mail |
---|
[5672] | 289 | mail -s "[NEMO Trusting ${NEMO_HPCC} ${REFE_CONF} ${NEMO_BRAN}] $ST $ERR" $EMAIL \ |
---|
| 290 | < trusting.mail |
---|
[5644] | 291 | fi |
---|
[5521] | 292 | |
---|
[5644] | 293 | fi |
---|
[5268] | 294 | } |
---|
| 295 | |
---|
[5681] | 296 | |
---|
[5383] | 297 | get_out() { |
---|
[5521] | 298 | ERR=$1 |
---|
| 299 | |
---|
[5542] | 300 | cd ${TEST_DIR} |
---|
[5509] | 301 | |
---|
[5681] | 302 | if [ $ST == 'FAILED' ]; then |
---|
[5521] | 303 | |
---|
[5629] | 304 | ## Error identification |
---|
[5521] | 305 | case $ERR in |
---|
[5629] | 306 | ## Compilation |
---|
[5667] | 307 | 1) ERR='XIOS compilation failed' ;; 2) ERR='NEMO compilation failed' ;; |
---|
[5629] | 308 | ## Submission |
---|
[5667] | 309 | 3) ERR='Missing input files' ;; 4) ERR='Job submission error' ;; |
---|
[5629] | 310 | ## Running |
---|
[5667] | 311 | 5) ERR='Crashed at time step '${TIME_STEP};; 6) ERR='Exceeded time limit '${TIME_LIMI}'h';; |
---|
[5629] | 312 | ## Results |
---|
[5667] | 313 | 7) ERR='Missing previous outputs' ;; 8) ERR='New outputs differ' ;; |
---|
[5629] | 314 | ## Undefined |
---|
[5664] | 315 | *) ERR='Unknown error' ;; |
---|
[5509] | 316 | esac |
---|
| 317 | |
---|
[5521] | 318 | export ERR |
---|
[5424] | 319 | fi |
---|
[5438] | 320 | |
---|
[5667] | 321 | sed -i "2 s/.*/$ST/" ${FILE_STAT} && sed -i "2 s/.*/$ERR/" ${FILE_RESU} |
---|
[5438] | 322 | |
---|
[5629] | 323 | ## Save tested configuration if trusting failed |
---|
[5681] | 324 | if [[ $ST == 'FAILED' && $PUBLISH -eq 1 && $DEBUG -eq 0 ]]; then |
---|
[5644] | 325 | echo 'Creating archive '${FILE_ARCH}' under '${REFE_DIR} |
---|
[5683] | 326 | tar -czf ${REFE_DIR}/${FILE_ARCH} * ${NEMO_HOME}/CONFIG/${TEST_CONF}/MY_SRC \ |
---|
[5681] | 327 | ${NEMO_HOME}/CONFIG/${TEST_CONF}/WORK |
---|
[5635] | 328 | fi |
---|
[5521] | 329 | |
---|
[5509] | 330 | mesg_make |
---|
| 331 | mesg_publish |
---|
[5623] | 332 | |
---|
| 333 | exit 0 |
---|
[5268] | 334 | } |
---|