[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 |
---|
[5509] | 8 | [ -e time.step ] && LAST_TIME_STEP=$( cat time.step | tr -d [:space:] ) |
---|
[5521] | 9 | export ${LAST_TIME_STEP} |
---|
[5455] | 10 | fi |
---|
[5453] | 11 | |
---|
[5485] | 12 | [ -e ocean.output ] && line=$( grep -m1 -A5 "$state" ocean.output | tr -d '\n' ) |
---|
[5521] | 13 | |
---|
[5485] | 14 | if [[ -e ${REFE_DIR}/model.log && $( diff -q model.log ${REFE_DIR}/model.log ) ]]; then |
---|
[5524] | 15 | line=$( diff model.log ${REFE_DIR}/model.log | grep '>' ) |
---|
[5455] | 16 | fi |
---|
| 17 | |
---|
[5465] | 18 | printf "Comments\n$line\n" | tee mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5438] | 19 | } |
---|
| 20 | |
---|
[5509] | 21 | |
---|
[5402] | 22 | diff_inputs() { |
---|
[5438] | 23 | mesg='Same' |
---|
[5453] | 24 | for file in inputs_list.txt $( ls namelist_* ) $( ls *.xml ); do |
---|
[5521] | 25 | diff -q $file ${REFE_DIR}/$file > /dev/null |
---|
[5438] | 26 | [ $? -ne 0 ] && mesg='Different' |
---|
[5402] | 27 | done |
---|
| 28 | |
---|
[5521] | 29 | printf "Input files\n$mesg\n" | tee mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5402] | 30 | } |
---|
| 31 | |
---|
[5509] | 32 | diff_results() { |
---|
| 33 | for file in ocean.output $( ls *.stat ); do |
---|
[5521] | 34 | [ ! -e ${REFE_DIR}/$file ] && export ST='FAILED' && get_out 7 |
---|
[5509] | 35 | diff -q $file ${REFE_DIR}/$file |
---|
| 36 | [ $? -ne 0 ] && export ST='FAILED' |
---|
| 37 | done |
---|
| 38 | } |
---|
| 39 | |
---|
[5383] | 40 | diff_restart() { |
---|
[5521] | 41 | [ ! -e ${REFE_DIR}/time.step ] && export ST='FAILED' && get_out 9 |
---|
[5485] | 42 | |
---|
[5509] | 43 | export LAST_TIME_STEP=$( cat ${REFE_DIR}/time.step | tr -d [:space:] ) |
---|
| 44 | echo 'Last time step of standard run: '${LAST_TIME_STEP} |
---|
[5485] | 45 | |
---|
[5528] | 46 | # if [ $( find -name "*${LAST_TIME_STEP}_restart*.nc" -print -quit ) ]; then |
---|
| 47 | if [ $( find -regex ".*_0+${LAST_TIME_STEP}_restart.*\.nc" -print -quit ) ]; then |
---|
| 48 | # base_name=$( find -name "*${LAST_TIME_STEP}_restart*.nc" -print -quit | awk -F/ '{print $NF}' \ |
---|
| 49 | # | sed "s/^\(.*\)$LAST_TIME_STEP\_restart.*/\1$LAST_TIME_STEP\_/" ) |
---|
| 50 | base_name=$( find -regex ".*_0+${LAST_TIME_STEP}_restart.*\.nc" -print -quit | awk -F/ '{print $NF}' \ |
---|
| 51 | | sed "s/^\(.*\)$LAST_TIME_STEP\_restart.*/\1$LAST_TIME_STEP\_/" ) |
---|
[5383] | 52 | |
---|
[5432] | 53 | for comp in restart restart_ice restart_trc; do |
---|
[5485] | 54 | file=${base_name}${comp} && printf "$file.nc: " |
---|
[5424] | 55 | |
---|
[5483] | 56 | nb_dom=$( find -name "${file}_[0-9]*.nc" | wc -l | awk '{print $1}' ) |
---|
[5523] | 57 | if [ ${nb_dom} -gt 1 ]; then |
---|
[5528] | 58 | ${NEMO_TRBD}/rebuild_nemo -t $NPROC $file ${nb_dom} > /dev/null |
---|
| 59 | [ $? -eq 0 ] && rm -f ${base_name}${comp}_[0-9]*.nc > /dev/null |
---|
[5523] | 60 | elif [ ${nb_dom} -eq 0 ]; then |
---|
| 61 | export ST='FAILED' && get_out 10 |
---|
[5424] | 62 | fi |
---|
| 63 | |
---|
[5485] | 64 | if [ -e ${REFE_DIR}/$file.nc ]; then |
---|
[5464] | 65 | # UNIX `cmp` not suitable (filename & timestamp in .nc file) |
---|
[5528] | 66 | nc_diff=$( $CDO diffn $file.nc ${REFE_DIR}/$file.nc 2> /dev/null \ |
---|
| 67 | | awk '/records/ {print $1}' ) |
---|
[5485] | 68 | |
---|
[5528] | 69 | if [ ${nc_diff} -ne 0 ]; then |
---|
[5509] | 70 | export ST='FAILED' |
---|
[5485] | 71 | printf "$CDO ${nc_diff}\n" |
---|
| 72 | else |
---|
| 73 | printf "identical\n" |
---|
| 74 | fi |
---|
| 75 | |
---|
[5432] | 76 | else |
---|
[5523] | 77 | export ST='FAILED' && get_out 9 |
---|
[5432] | 78 | fi |
---|
[5438] | 79 | |
---|
[5424] | 80 | done |
---|
[5383] | 81 | |
---|
[5424] | 82 | else |
---|
[5509] | 83 | export ST='FAILED' |
---|
[5424] | 84 | fi |
---|
[5455] | 85 | |
---|
[5383] | 86 | } |
---|
| 87 | |
---|
[5472] | 88 | get_cpu_time() { |
---|
[5527] | 89 | real_cpu_time=$( eval ${JOB_TIME} ) |
---|
[5521] | 90 | printf "Real CPU time\n${real_cpu_time}\n" | tee mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5472] | 91 | } |
---|
| 92 | |
---|
| 93 | get_inputs() { |
---|
| 94 | if [ ! -z "${FORC_TAR}" ]; then |
---|
| 95 | cmd_iol="tar -tvf ${NEMO_FORC}/${FORC_TAR}"; cmd_iof="tar -vxf ${NEMO_FORC}/${FORC_TAR}" |
---|
| 96 | else |
---|
| 97 | cmd_iol="ls ${NEMO_FORC}/*" ; cmd_iof="\cp ${NEMO_FORC}/* ." |
---|
| 98 | fi |
---|
| 99 | |
---|
| 100 | ${cmd_iol} > inputs_list.txt |
---|
| 101 | ${cmd_iof} > /dev/null |
---|
| 102 | } |
---|
| 103 | |
---|
[5424] | 104 | get_soft_rel() { |
---|
[5383] | 105 | for rel in $CDO $COMPILER $MPI $NETCDF; do |
---|
[5509] | 106 | arch_rel=$( echo $LOADEDMODULES | sed "s#.*$rel/\([^:]*\).*#\1#" ) |
---|
| 107 | [ -z "${arch_rel}" ] && arch_rel=$( echo $PATH | sed "s#.*$rel/\([^/]*\).*#\1#" ) |
---|
[5465] | 108 | [ "$rel" == "$COMPILER" ] && arch_rel=$( $rel --version | grep -m1 -o ' [0-9.]* ' ) |
---|
[5464] | 109 | echo $rel ${arch_rel} >> model.log |
---|
[5383] | 110 | done |
---|
[5438] | 111 | |
---|
[5472] | 112 | sed -n 4p model.log | sed 'i\Fortran compiler' > mesg_06_compiler_${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 | |
---|
[5438] | 120 | # If -v|--version option has been set, modify svn command |
---|
[5523] | 121 | if [ $( echo ${NEMO_VERS} | grep "HEAD\|up\|update" ) ]; then |
---|
| 122 | svn_cmd='svn update -r HEAD' |
---|
| 123 | elif [ $( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) ]; then |
---|
| 124 | svn_cmd='svn update -r '$( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) |
---|
[5444] | 125 | else |
---|
| 126 | svn_cmd='svn '${svn_opt} |
---|
[5438] | 127 | fi |
---|
| 128 | |
---|
[5444] | 129 | rev=0 |
---|
[5464] | 130 | for dir in ${NEMO_ARCH} ${NEMO_CONF} ${NEMO_ENGI} \ |
---|
| 131 | ${NEMO_EIOI} ${NEMO_EFCM} \ |
---|
| 132 | ${NEMO_TCMP} ${NEMO_TRBD} \ |
---|
| 133 | ${DIR_XIOS} ; do |
---|
[5438] | 134 | |
---|
[5383] | 135 | # For time being, just get revision number from XIOS |
---|
[5465] | 136 | if [ "$dir" == "${DIR_XIOS}" ]; then |
---|
| 137 | rev_loc=$( svn info $dir | awk '(NR == 5) {print $NF}' ) |
---|
[5472] | 138 | echo 'XIOS '${rev_loc} >> model.log |
---|
| 139 | printf "XIOS rev.\n" > mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 140 | printf "<a href=\"https://forge.ipsl.jussieu.fr/ioserver/changeset/${rev_loc}\" target=\"_blank\">${rev_loc}</a>" \ |
---|
| 141 | >> mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5438] | 142 | continue |
---|
| 143 | fi |
---|
| 144 | |
---|
[5465] | 145 | echo ${svn_cmd} $dir': '$( ${svn_cmd} $dir ) |
---|
[5432] | 146 | rev_loc=$( svn info $dir | awk '(NR == 5) {print $NF}' ) |
---|
[5383] | 147 | (( ${rev_loc} >= $rev )) && rev=${rev_loc} |
---|
| 148 | done |
---|
| 149 | |
---|
[5472] | 150 | echo 'NEMOGCM '$rev >> model.log |
---|
| 151 | printf "NEMOGCM rev.\n" > mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 152 | printf "<a href=\"https://forge.ipsl.jussieu.fr/nemo/changeset/$rev\" target=\"_blank\">$rev</a>" \ |
---|
| 153 | >> mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5383] | 154 | } |
---|
| 155 | |
---|
| 156 | mesg_make() { |
---|
[5464] | 157 | paste -d ';' mesg_*_${CFG_USER}_${CFG_ARCH}.txt > trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 158 | cat trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5268] | 159 | } |
---|
| 160 | |
---|
[5383] | 161 | mesg_publish() { |
---|
[5268] | 162 | |
---|
| 163 | if [ $PUBLISH -eq 1 ]; then |
---|
| 164 | |
---|
[5472] | 165 | if [ -f ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt ]; then cmd='tail -1'; else cmd='cat'; fi |
---|
[5268] | 166 | |
---|
[5464] | 167 | $cmd trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt >> ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5424] | 168 | |
---|
[5268] | 169 | # Send mail only when FAILED |
---|
[5509] | 170 | if [[ ! -z $EMAIL && "$ST" == 'FAILED' ]]; then |
---|
[5424] | 171 | cat << END_MAIL > trusting.mail |
---|
[5292] | 172 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
[5472] | 173 | |
---|
[5268] | 174 | Dear $USER, |
---|
| 175 | |
---|
[5472] | 176 | The trusting sequence for ${REFE_CONF} has failed. |
---|
| 177 | Directory: ${TEST_DIR} |
---|
| 178 | Archive: ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz |
---|
[5268] | 179 | |
---|
[5292] | 180 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
[5268] | 181 | |
---|
| 182 | END_MAIL |
---|
[5521] | 183 | |
---|
[5464] | 184 | if [ -e ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt ]; then |
---|
| 185 | cat ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt >> trusting.mail |
---|
| 186 | fi |
---|
[5509] | 187 | mail -s "[trusting ${REFE_CONF}] $ST $ERR" $EMAIL < trusting.mail |
---|
[5268] | 188 | fi |
---|
| 189 | |
---|
| 190 | fi |
---|
| 191 | } |
---|
| 192 | |
---|
[5509] | 193 | print_step() { printf "Step.....\n$1\n"; } |
---|
[5438] | 194 | |
---|
[5383] | 195 | get_out() { |
---|
[5521] | 196 | ERR=$1 |
---|
| 197 | |
---|
[5509] | 198 | printf "Status\n$ST\n" > mesg_02_status_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 199 | |
---|
| 200 | if [ "$ST" == 'FAILED' ]; then |
---|
[5455] | 201 | cd ${TEST_DIR} |
---|
[5485] | 202 | |
---|
[5521] | 203 | if [ ! -e mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt ]; then |
---|
| 204 | printf "Input files\n\n" > mesg_09_inputs_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 205 | fi |
---|
| 206 | |
---|
| 207 | if [ ! -e mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt ]; then |
---|
| 208 | printf "Real CPU time\n\n" > mesg_10_cputime_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 209 | fi |
---|
| 210 | |
---|
[5485] | 211 | if [ ! -e mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt ]; then |
---|
| 212 | printf "Comments\n\n" > mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 213 | fi |
---|
| 214 | |
---|
[5509] | 215 | # Error identification |
---|
[5521] | 216 | case $ERR in |
---|
[5509] | 217 | # Compilation |
---|
| 218 | 1) ERR='XIOS compilation failed' ;; 2) ERR='NEMO compilation failed' ;; |
---|
| 219 | # Submission |
---|
| 220 | 3) ERR='Missing input files' ;; 4) ERR='Job submission error' ;; |
---|
| 221 | # Running |
---|
| 222 | 5) ERR='nemo.exe crashed at '${LAST_TIME_ERR};; 6) ERR='Exceeded time limit '${TIME_LIMI}'h';; |
---|
| 223 | # Results |
---|
| 224 | 7) ERR='Missing previous outputs ' ;; 8) ERR='New outputs differ/missing' ;; |
---|
| 225 | 9) ERR='Missing previous restarts' ;; 10) ERR='New restarts differ/missing' ;; |
---|
| 226 | # Undefined |
---|
| 227 | *) ERR='Unknown error' ;; |
---|
| 228 | esac |
---|
| 229 | |
---|
[5521] | 230 | export ERR |
---|
[5424] | 231 | fi |
---|
[5438] | 232 | |
---|
[5509] | 233 | printf "Result.....\n$ERR\n" > mesg_03_result_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5438] | 234 | |
---|
[5521] | 235 | # Save tested configuration if trusting failed |
---|
| 236 | [ "$ST" == 'FAILED' ] && tar -czf ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz * |
---|
| 237 | |
---|
[5509] | 238 | mesg_make |
---|
| 239 | mesg_publish |
---|
[5465] | 240 | |
---|
[5424] | 241 | exit 1 |
---|
[5268] | 242 | } |
---|
[5464] | 243 | |
---|
| 244 | job_pending() { |
---|
| 245 | time_elapsed=0; time_increment=30 |
---|
[5509] | 246 | |
---|
[5464] | 247 | sleep ${time_increment} |
---|
| 248 | while [[ $( ${JOB_LIST} | grep ${JOB_ID} ) && ${time_elapsed} -lt ${TIME_LIMI} ]]; do |
---|
[5527] | 249 | printf "\n####################################################\n" >> computation.log |
---|
| 250 | ${JOB_INFO} ${JOB_ID} >> computation.log |
---|
[5464] | 251 | sleep ${time_increment} |
---|
[5509] | 252 | time_elapsed=$(( ${time_elapsed} + ${time_increment} )) |
---|
[5464] | 253 | done |
---|
| 254 | sleep ${time_increment} |
---|
| 255 | |
---|
| 256 | if [ ${time_elapsed} -eq ${TIME_LIMI} ]; then |
---|
| 257 | ${JOB_DELE} ${JOB_ID} &> /dev/null |
---|
[5509] | 258 | TIME_LIMIT=$(( ${TIME_LIMI} / 3600 )) |
---|
| 259 | export ${TIME_LIMIT} |
---|
| 260 | get_out $ST ERR=6 |
---|
[5521] | 261 | fi |
---|
[5464] | 262 | } |
---|