[5268] | 1 | #!/bin/bash |
---|
| 2 | |
---|
[5383] | 3 | |
---|
[5438] | 4 | comments() { |
---|
[5509] | 5 | state=$1; LAST_TIME_STEP=0 |
---|
[5453] | 6 | |
---|
[5465] | 7 | if [ "$state" == 'E R R O R' ]; then |
---|
[5555] | 8 | [ -e time.step ] && LAST_TIME_STEP=$( cat time.step | tr -d ' ' ) |
---|
[5623] | 9 | export LAST_TIME_STEP |
---|
[5633] | 10 | [ -e ocean.output ] && line=$( grep -m1 -A4 "$state" ocean.output | tr -d '\n' ) |
---|
| 11 | else |
---|
| 12 | [ -e ocean.output ] && line=$( grep -m1 -A3 "^ $state" ocean.output | tr -d '\n' ) |
---|
[5455] | 13 | fi |
---|
[5453] | 14 | |
---|
[5560] | 15 | # if [[ -e ${REFE_DIR}/model.log && $( diff -q model.log ${REFE_DIR}/model.log ) ]]; then |
---|
| 16 | # line=$( diff model.log ${REFE_DIR}/model.log | grep '>' | sed '{N;s/\n/<BR>/}' ) |
---|
| 17 | # fi |
---|
[5455] | 18 | |
---|
[5465] | 19 | printf "Comments\n$line\n" | tee mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5438] | 20 | } |
---|
| 21 | |
---|
[5509] | 22 | |
---|
[5402] | 23 | diff_inputs() { |
---|
[5438] | 24 | mesg='Same' |
---|
[5453] | 25 | for file in inputs_list.txt $( ls namelist_* ) $( ls *.xml ); do |
---|
[5521] | 26 | diff -q $file ${REFE_DIR}/$file > /dev/null |
---|
[5438] | 27 | [ $? -ne 0 ] && mesg='Different' |
---|
[5402] | 28 | done |
---|
| 29 | |
---|
[5521] | 30 | printf "Input files\n$mesg\n" | tee mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5402] | 31 | } |
---|
| 32 | |
---|
[5509] | 33 | diff_results() { |
---|
| 34 | for file in ocean.output $( ls *.stat ); do |
---|
[5521] | 35 | [ ! -e ${REFE_DIR}/$file ] && export ST='FAILED' && get_out 7 |
---|
[5509] | 36 | diff -q $file ${REFE_DIR}/$file |
---|
| 37 | [ $? -ne 0 ] && export ST='FAILED' |
---|
| 38 | done |
---|
| 39 | } |
---|
| 40 | |
---|
[5383] | 41 | diff_restart() { |
---|
[5521] | 42 | [ ! -e ${REFE_DIR}/time.step ] && export ST='FAILED' && get_out 9 |
---|
[5485] | 43 | |
---|
[5509] | 44 | export LAST_TIME_STEP=$( cat ${REFE_DIR}/time.step | tr -d [:space:] ) |
---|
| 45 | echo 'Last time step of standard run: '${LAST_TIME_STEP} |
---|
[5485] | 46 | |
---|
[5528] | 47 | if [ $( find -regex ".*_0+${LAST_TIME_STEP}_restart.*\.nc" -print -quit ) ]; then |
---|
| 48 | base_name=$( find -regex ".*_0+${LAST_TIME_STEP}_restart.*\.nc" -print -quit | awk -F/ '{print $NF}' \ |
---|
| 49 | | sed "s/^\(.*\)$LAST_TIME_STEP\_restart.*/\1$LAST_TIME_STEP\_/" ) |
---|
[5383] | 50 | |
---|
[5432] | 51 | for comp in restart restart_ice restart_trc; do |
---|
[5485] | 52 | file=${base_name}${comp} && printf "$file.nc: " |
---|
[5424] | 53 | |
---|
[5483] | 54 | nb_dom=$( find -name "${file}_[0-9]*.nc" | wc -l | awk '{print $1}' ) |
---|
[5523] | 55 | if [ ${nb_dom} -gt 1 ]; then |
---|
[5528] | 56 | ${NEMO_TRBD}/rebuild_nemo -t $NPROC $file ${nb_dom} > /dev/null |
---|
| 57 | [ $? -eq 0 ] && rm -f ${base_name}${comp}_[0-9]*.nc > /dev/null |
---|
[5523] | 58 | elif [ ${nb_dom} -eq 0 ]; then |
---|
| 59 | export ST='FAILED' && get_out 10 |
---|
[5424] | 60 | fi |
---|
| 61 | |
---|
[5485] | 62 | if [ -e ${REFE_DIR}/$file.nc ]; then |
---|
[5629] | 63 | ## UNIX `cmp` not suitable (filename & timestamp in .nc file) |
---|
[5528] | 64 | nc_diff=$( $CDO diffn $file.nc ${REFE_DIR}/$file.nc 2> /dev/null \ |
---|
| 65 | | awk '/records/ {print $1}' ) |
---|
[5485] | 66 | |
---|
[5545] | 67 | if [[ ! -z ${nc_diff} && ${nc_diff} -ne 0 ]]; then |
---|
[5509] | 68 | export ST='FAILED' |
---|
[5485] | 69 | printf "$CDO ${nc_diff}\n" |
---|
| 70 | else |
---|
| 71 | printf "identical\n" |
---|
| 72 | fi |
---|
| 73 | |
---|
[5432] | 74 | else |
---|
[5523] | 75 | export ST='FAILED' && get_out 9 |
---|
[5432] | 76 | fi |
---|
[5438] | 77 | |
---|
[5424] | 78 | done |
---|
[5383] | 79 | |
---|
[5424] | 80 | else |
---|
[5509] | 81 | export ST='FAILED' |
---|
[5424] | 82 | fi |
---|
[5455] | 83 | |
---|
[5383] | 84 | } |
---|
| 85 | |
---|
[5472] | 86 | get_cpu_time() { |
---|
[5527] | 87 | real_cpu_time=$( eval ${JOB_TIME} ) |
---|
[5521] | 88 | printf "Real CPU time\n${real_cpu_time}\n" | tee mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5472] | 89 | } |
---|
| 90 | |
---|
| 91 | get_inputs() { |
---|
[5623] | 92 | if [ ! -z "${FORC_TARF}" ]; then |
---|
[5543] | 93 | cmd_iol="tar -tvf ${NEMO_FORC}/${FORC_TARF}"; cmd_iof="tar -vxf ${NEMO_FORC}/${FORC_TARF}" |
---|
[5472] | 94 | else |
---|
[5543] | 95 | cmd_iol="ls ${NEMO_FORC}/*" ; cmd_iof="\cp ${NEMO_FORC}/* ." |
---|
[5472] | 96 | fi |
---|
| 97 | |
---|
[5627] | 98 | ${cmd_iol} > inputs_list.txt && ${cmd_iof} > /dev/null |
---|
[5472] | 99 | } |
---|
| 100 | |
---|
[5424] | 101 | get_soft_rel() { |
---|
[5629] | 102 | for soft in $CDO $COMPILER $MPI $NETCDF; do |
---|
| 103 | soft_rel=$( echo $LOADEDMODULES | grep -o "$soft[^:]*" ) |
---|
| 104 | [ -z "${soft_rel}" ] && soft_rel=$( echo $PATH | grep -o "$soft/[^/]*" ) |
---|
| 105 | ## Not working inside CURIE cron (no modules command && cdo embedded in netcdf-*) |
---|
| 106 | #[ "$soft" == "$COMPILER" ] && soft_rel=$( $soft --version | grep -m1 -o ' [0-9.]* ' ) |
---|
| 107 | [[ "$soft" == "$CDO" && -z "${soft_rel}" ]] && soft_rel=$( echo $PATH | grep -o "netcdf-utils[^/]*" ) |
---|
| 108 | [[ "$soft" == "$NETCDF" && -z "${soft_rel}" ]] && soft_rel=$( echo $PATH | grep -o "netcdf[^/]*" | sort | head -1 ) |
---|
[5633] | 109 | echo ${soft_rel} >> model.log |
---|
[5383] | 110 | done |
---|
[5438] | 111 | |
---|
[5472] | 112 | sed -n 4p model.log | sed 'i\Fortran compiler' > mesg_06_compiler_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 113 | sed -n 5p model.log | sed 'i\MPI libs' > mesg_07_mpi_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 114 | sed -n 6p model.log | sed 'i\NetCDF libs' > mesg_08_netcdf_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5383] | 115 | } |
---|
| 116 | |
---|
[5424] | 117 | get_nemo_rev() { |
---|
[5444] | 118 | svn_opt='status' |
---|
[5383] | 119 | |
---|
[5629] | 120 | ## If -v|--version option has been set, modify svn command |
---|
[5607] | 121 | if [ $( echo ${NEMO_VERS} | grep "HEAD\|up\|update" ) ]; then |
---|
[5523] | 122 | svn_cmd='svn update -r HEAD' |
---|
[5607] | 123 | elif [ $( echo ${NEMO_VERS} | grep -o '{[0-9]\{4\}-[0-9]\{2\}-[0-9]\{2\}}' ) ]; then |
---|
| 124 | svn_cmd='svn update -r '$( echo ${NEMO_VERS} | grep -o '{[0-9]\{4\}-[0-9]\{2\}-[0-9]\{2\}}' ) |
---|
| 125 | elif [ $( echo ${NEMO_VERS} | grep -o '[0-9]*' ) ]; then |
---|
| 126 | svn_cmd='svn update -r '$( echo ${NEMO_VERS} | grep -o '[0-9]*' ) |
---|
[5444] | 127 | else |
---|
| 128 | svn_cmd='svn '${svn_opt} |
---|
[5438] | 129 | fi |
---|
| 130 | |
---|
[5444] | 131 | rev=0 |
---|
[5464] | 132 | for dir in ${NEMO_ARCH} ${NEMO_CONF} ${NEMO_ENGI} \ |
---|
| 133 | ${NEMO_EIOI} ${NEMO_EFCM} \ |
---|
| 134 | ${NEMO_TCMP} ${NEMO_TRBD} \ |
---|
| 135 | ${DIR_XIOS} ; do |
---|
[5438] | 136 | |
---|
[5629] | 137 | ## For time being, just get revision number from XIOS |
---|
[5465] | 138 | if [ "$dir" == "${DIR_XIOS}" ]; then |
---|
[5607] | 139 | rev_loc=$( svn info $dir | awk '(NR == 9) {print $NF}' ) |
---|
[5472] | 140 | echo 'XIOS '${rev_loc} >> model.log |
---|
| 141 | printf "XIOS rev.\n" > mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 142 | printf "<a href=\"https://forge.ipsl.jussieu.fr/ioserver/changeset/${rev_loc}\" target=\"_blank\">${rev_loc}</a>" \ |
---|
| 143 | >> mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5438] | 144 | continue |
---|
| 145 | fi |
---|
| 146 | |
---|
[5465] | 147 | echo ${svn_cmd} $dir': '$( ${svn_cmd} $dir ) |
---|
[5607] | 148 | rev_loc=$( svn info $dir | awk '(NR == 9) {print $NF}' ) |
---|
[5383] | 149 | (( ${rev_loc} >= $rev )) && rev=${rev_loc} |
---|
| 150 | done |
---|
| 151 | |
---|
[5472] | 152 | echo 'NEMOGCM '$rev >> model.log |
---|
| 153 | printf "NEMOGCM rev.\n" > mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 154 | printf "<a href=\"https://forge.ipsl.jussieu.fr/nemo/changeset/$rev\" target=\"_blank\">$rev</a>" \ |
---|
| 155 | >> mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5383] | 156 | } |
---|
| 157 | |
---|
| 158 | mesg_make() { |
---|
[5464] | 159 | paste -d ';' mesg_*_${CFG_USER}_${CFG_ARCH}.txt > trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 160 | cat trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5268] | 161 | } |
---|
| 162 | |
---|
[5383] | 163 | mesg_publish() { |
---|
[5268] | 164 | |
---|
| 165 | if [ $PUBLISH -eq 1 ]; then |
---|
| 166 | |
---|
[5472] | 167 | if [ -f ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt ]; then cmd='tail -1'; else cmd='cat'; fi |
---|
[5268] | 168 | |
---|
[5464] | 169 | $cmd trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt >> ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5424] | 170 | |
---|
[5629] | 171 | ## Send mail only when FAILED |
---|
[5509] | 172 | if [[ ! -z $EMAIL && "$ST" == 'FAILED' ]]; then |
---|
[5424] | 173 | cat << END_MAIL > trusting.mail |
---|
[5292] | 174 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
[5472] | 175 | |
---|
[5268] | 176 | Dear $USER, |
---|
| 177 | |
---|
[5472] | 178 | The trusting sequence for ${REFE_CONF} has failed. |
---|
| 179 | Directory: ${TEST_DIR} |
---|
[5622] | 180 | Archive: trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz |
---|
[5268] | 181 | |
---|
[5292] | 182 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
[5268] | 183 | |
---|
| 184 | END_MAIL |
---|
[5521] | 185 | |
---|
[5464] | 186 | if [ -e ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt ]; then |
---|
| 187 | cat ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt >> trusting.mail |
---|
| 188 | fi |
---|
[5607] | 189 | mail -s "[trusting ${REFE_CONF} $BRANCH] $ST $ERR" $EMAIL < trusting.mail |
---|
[5268] | 190 | fi |
---|
| 191 | |
---|
| 192 | fi |
---|
| 193 | } |
---|
| 194 | |
---|
[5509] | 195 | print_step() { printf "Step.....\n$1\n"; } |
---|
[5438] | 196 | |
---|
[5383] | 197 | get_out() { |
---|
[5521] | 198 | ERR=$1 |
---|
| 199 | |
---|
[5542] | 200 | cd ${TEST_DIR} |
---|
[5509] | 201 | printf "Status\n$ST\n" > mesg_02_status_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 202 | |
---|
| 203 | if [ "$ST" == 'FAILED' ]; then |
---|
[5545] | 204 | file=mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 205 | [ ! -e $file ] && printf "Input files\n\n" > $file |
---|
| 206 | file=mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 207 | [ ! -e $file ] && printf "Real CPU time\n\n" > $file |
---|
| 208 | file=mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 209 | [ ! -e $file ] && printf "Comments\n\n" > $file |
---|
[5521] | 210 | |
---|
[5629] | 211 | ## Error identification |
---|
[5521] | 212 | case $ERR in |
---|
[5629] | 213 | ## Compilation |
---|
[5632] | 214 | 1) ERR='XIOS compilation failed' ;; 2) ERR='NEMO compilation failed' ;; |
---|
[5629] | 215 | ## Submission |
---|
[5632] | 216 | 3) ERR='Missing input files' ;; 4) ERR='Job submission error' ;; |
---|
[5629] | 217 | ## Running |
---|
[5632] | 218 | 5) ERR='Crashed at time step '${LAST_TIME_STEP};; 6) ERR='Exceeded time limit '${TIME_LIMI}'h';; |
---|
[5629] | 219 | ## Results |
---|
[5632] | 220 | 7) ERR='Missing previous outputs ' ;; 8) ERR='New outputs differ/missing' ;; |
---|
| 221 | 9) ERR='Missing previous restarts' ;; 10) ERR='New restarts differ/missing' ;; |
---|
[5629] | 222 | ## Undefined |
---|
[5632] | 223 | *) ERR='Unknown error' ;; |
---|
[5509] | 224 | esac |
---|
| 225 | |
---|
[5521] | 226 | export ERR |
---|
[5424] | 227 | fi |
---|
[5438] | 228 | |
---|
[5607] | 229 | printf "Result\n$ERR\n" > mesg_03_result_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5438] | 230 | |
---|
[5629] | 231 | ## Save tested configuration if trusting failed |
---|
[5622] | 232 | [[ "$ST" == 'FAILED' && $DEBUG -eq 0 ]] && tar -czf ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz * |
---|
[5521] | 233 | |
---|
[5509] | 234 | mesg_make |
---|
| 235 | mesg_publish |
---|
[5623] | 236 | |
---|
| 237 | exit 0 |
---|
[5268] | 238 | } |
---|
[5464] | 239 | |
---|
| 240 | job_pending() { |
---|
| 241 | time_elapsed=0; time_increment=30 |
---|
[5509] | 242 | |
---|
[5464] | 243 | sleep ${time_increment} |
---|
| 244 | while [[ $( ${JOB_LIST} | grep ${JOB_ID} ) && ${time_elapsed} -lt ${TIME_LIMI} ]]; do |
---|
[5527] | 245 | printf "\n####################################################\n" >> computation.log |
---|
| 246 | ${JOB_INFO} ${JOB_ID} >> computation.log |
---|
[5464] | 247 | sleep ${time_increment} |
---|
[5509] | 248 | time_elapsed=$(( ${time_elapsed} + ${time_increment} )) |
---|
[5464] | 249 | done |
---|
| 250 | sleep ${time_increment} |
---|
| 251 | |
---|
| 252 | if [ ${time_elapsed} -eq ${TIME_LIMI} ]; then |
---|
| 253 | ${JOB_DELE} ${JOB_ID} &> /dev/null |
---|
[5509] | 254 | TIME_LIMIT=$(( ${TIME_LIMI} / 3600 )) |
---|
| 255 | export ${TIME_LIMIT} |
---|
| 256 | get_out $ST ERR=6 |
---|
[5521] | 257 | fi |
---|
[5464] | 258 | } |
---|