[5268] | 1 | #!/bin/bash |
---|
[5286] | 2 | # |
---|
| 3 | # Author: Patrick Brockmann |
---|
| 4 | # Contact: Patrick.Brockmann@cea.fr |
---|
| 5 | # History: |
---|
| 6 | # Modification: Nicolas-thierry.Martin@locean-ipsl.upmc.fr |
---|
| 7 | # |
---|
[5268] | 8 | |
---|
| 9 | |
---|
[5276] | 10 | # Set defaults |
---|
[5355] | 11 | #--------------------------------------------------- |
---|
| 12 | NEMO_TRUS=$( pwd $( dirname $0 ) ) |
---|
[5383] | 13 | SVN_CMD='svn status' |
---|
[5276] | 14 | DEBUG=0; PUBLISH=0; status='OK' |
---|
[5268] | 15 | |
---|
[5355] | 16 | # Get options for replacing some initials settings |
---|
| 17 | #--------------------------------------------------- |
---|
[5268] | 18 | while [ $# -ne 0 ]; do |
---|
| 19 | case $1 in |
---|
[5286] | 20 | -b|--branch ) NEMO_BRAN=$2 ; shift 2;; -d|--debug ) set -vx; DEBUG=1 ; shift ;; |
---|
| 21 | -e|--email ) MAIL=$2 ; shift 2;; -j|--job ) NPROC=$2 ; shift 2;; |
---|
[5355] | 22 | -h|--help ) cat trusting_help.txt; exit 1;; -m|--machine) CFG_ARCH=$2 ; shift 2;; |
---|
| 23 | -n|--newconf) TEST_CONF=$2 ; shift 2;; -r|--refconf) REFE_CONF=$2 ; shift 2;; |
---|
[5286] | 24 | -t|--time ) TIME_LIMI=$2 ; shift 2;; -p|--publish) PUBLISH=1 ; shift ;; |
---|
[5383] | 25 | -u|--user ) CFG_USER=$2 ; shift 2;; -v|--version) NEMO_VERS=$2 ; shift 2;; |
---|
[5355] | 26 | *) cat trusting_help.txt; exit 1;; |
---|
[5268] | 27 | esac |
---|
| 28 | done |
---|
[5276] | 29 | |
---|
[5355] | 30 | #export DEBUG PUBLISH |
---|
[5268] | 31 | |
---|
| 32 | |
---|
[5355] | 33 | # Initialization (super-computer & user environment) |
---|
| 34 | #--------------------------------------------------- |
---|
| 35 | if [[ ! -z "${CFG_ARCH}" && ! -z "${CFG_USER}" ]]; then |
---|
[5268] | 36 | |
---|
[5355] | 37 | if [ ! $( find ${NEMO_TRUS}/cfg -name ${CFG_USER}.cfg ) ]; then |
---|
| 38 | echo 'No configuration file in '${NEMO_TRUS}'/cfg for '${CFG_USER}; exit 1 |
---|
| 39 | elif [ ! $( find ${NEMO_TRUS}/cfg -name ${CFG_ARCH}.cfg ) ]; then |
---|
| 40 | echo 'No configuration file in '${NEMO_TRUS}'/cfg for '${CFG_ARCH}; exit 1 |
---|
| 41 | else |
---|
| 42 | . ${NEMO_TRUS}/trusting.env |
---|
| 43 | fi |
---|
[5268] | 44 | |
---|
[5355] | 45 | else |
---|
| 46 | cat ${NEMO_TRUS}/trusting_help.txt; exit 1 |
---|
| 47 | fi |
---|
[5268] | 48 | |
---|
[5383] | 49 | # If -v|--version option has been set, modify svn command |
---|
| 50 | [ $( echo ${NEMO_VERS} | grep "up\|update" ) ] && SVN_CMD='svn update' |
---|
| 51 | [ $( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) ] && SVN_CMD='svn update -r '$( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) |
---|
| 52 | |
---|
[5355] | 53 | datestr1=$( LC_ALL=C date -d"${DATE_STR}" +%Y-%m-%dT%H:%M ) |
---|
| 54 | . ./trusting_func.sh |
---|
[5268] | 55 | |
---|
[5355] | 56 | mkdir -p ${TEST_DIR} ${REFE_DIR} |
---|
| 57 | echo ${TEST_DIR} && cd ${TEST_DIR} |
---|
| 58 | printf "Date\n%s\n" $datestr1 > mesg_01_date_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 59 | # Only at end status will change to OK |
---|
[5383] | 60 | printf "Status\nFAILED\n" > mesg_03_status_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 61 | |
---|
| 62 | |
---|
| 63 | # Check softwares versions |
---|
| 64 | #--------------------------------------------------- |
---|
[5383] | 65 | printf "Step......................\nGet softwares release\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 66 | . ${ARCH_ENV} > /dev/null # Sourcing architecture environment |
---|
| 67 | get_rel |
---|
| 68 | sed -n 2p model.log | sed 'i\Fortran compiler' > mesg_05_compiler_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 69 | sed -n 3p model.log | sed 'i\MPI libraries ' > mesg_06_mpi_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 70 | sed -n 4p model.log | sed 'i\NetCDF libraries' > mesg_07_netcdf_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5268] | 71 | |
---|
| 72 | |
---|
[5355] | 73 | # SVN action on XIOS & NEMO essentials directories |
---|
| 74 | #--------------------------------------------------- |
---|
[5383] | 75 | printf "Step......................\nGet NEMO SVN revision with: ${SVN_CMD}\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 76 | get_rev |
---|
| 77 | sed -n 6p model.log | sed 'i\NEMO revision' > mesg_02_nemo_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5292] | 78 | |
---|
[5355] | 79 | |
---|
| 80 | # NEMO config compiled from scratch |
---|
| 81 | #--------------------------------------------------- |
---|
[5383] | 82 | printf "Step......................\nCompiling ${CONF_REF} configuration\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5276] | 83 | cd ${NEMO_CONF} |
---|
[5355] | 84 | #[ -d ${TEST_CONF} ] && ./makenemo -n ${TEST_CONF} clean |
---|
| 85 | ./makenemo -n ${TEST_CONF} -r ${REFE_CONF} -m $ARCH -j $NPROC > /dev/null 2>&1 |
---|
| 86 | [ ! -e ${TEST_CONF}/BLD/bin/nemo.exe ] && get_out |
---|
[5268] | 87 | |
---|
| 88 | |
---|
[5355] | 89 | # Get namelists, xml & forced files for running |
---|
| 90 | #--------------------------------------------------- |
---|
[5383] | 91 | printf "Step......................\nPreparing job\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 92 | cd ${TEST_DIR} |
---|
| 93 | find ${NEMO_CONF}/${TEST_CONF}/EXP00 -regex '.*\(cfg\|opa\|ref\|xml\)' -exec cp {} . \; |
---|
[5268] | 94 | |
---|
[5286] | 95 | if [ ! -z "${NEMO_FORC}" ]; then |
---|
| 96 | cp ${NEMO_FORC}/${FORC_TAR} . |
---|
[5355] | 97 | tar -tvf ${FORC_TAR} > inputs_list.txt && tar -vxf ${FORC_TAR} > /dev/null |
---|
| 98 | [ $? -ne 0 ] && get_out |
---|
[5286] | 99 | gunzip *.gz |
---|
| 100 | else |
---|
| 101 | touch inputs_list.txt |
---|
| 102 | fi |
---|
[5268] | 103 | |
---|
[5286] | 104 | |
---|
[5355] | 105 | # Job submission & computing |
---|
| 106 | #--------------------------------------------------- |
---|
[5383] | 107 | printf "Step......................\nRunning\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt; echo ${JOB_SUBM} |
---|
[5355] | 108 | job_num=$( ${JOB_SUBM} | awk '{print $NF}' ) |
---|
| 109 | [ $? -ne 0 ] && get_out |
---|
[5268] | 110 | |
---|
[5355] | 111 | time_elapsed=0; time_increment=30 |
---|
[5268] | 112 | sleep ${time_increment} |
---|
[5355] | 113 | while [[ $( ${JOB_LIST} | grep $job_num ) && ${time_elapsed} -lt ${TIME_LIMI} ]]; do |
---|
[5286] | 114 | printf "\n####################################################" >> computation.log |
---|
| 115 | ${JOB_VIEW} $job_num >> computation.log |
---|
[5268] | 116 | sleep ${time_increment} |
---|
[5286] | 117 | let time_elapsed+=${time_increment} |
---|
[5268] | 118 | done |
---|
| 119 | |
---|
| 120 | |
---|
[5355] | 121 | # Eventual crash during computing |
---|
| 122 | #--------------------------------------------------- |
---|
[5383] | 123 | printf "Step......................\nTest if ASCII output files exist\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 124 | if [ ! -f ocean.output ]; then |
---|
[5383] | 125 | printf "Step......................\nOpa crashed at initialization\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 126 | status='FAILED' |
---|
| 127 | elif [ $( grep 'E R R O R' ocean.output ) ]; then |
---|
[5383] | 128 | printf "Step......................\nOpa crashed\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 129 | grep 'E R R O R' ocean.output >> mesg_12_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 130 | sed 's/^/Time step :/' time.step >> mesg_12_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 131 | status='FAILED' |
---|
[5286] | 132 | fi |
---|
| 133 | |
---|
| 134 | |
---|
[5355] | 135 | # Inspect output text files |
---|
| 136 | #--------------------------------------------------- |
---|
[5383] | 137 | printf "Step......................\nTest ASCII output files difference\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5286] | 138 | for file in ocean.output solver.stat tracer.stat; do |
---|
[5355] | 139 | diff $file ${REFE_DIR}/$file |
---|
[5286] | 140 | [ $? -ne 0 ] && status='FAILED' |
---|
| 141 | done |
---|
[5268] | 142 | |
---|
| 143 | |
---|
[5355] | 144 | # Inspect output NetCDF files |
---|
| 145 | #--------------------------------------------------- |
---|
[5383] | 146 | printf "Step......................\nTest last restart NetCDF files difference\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 147 | diff_restart 2>/dev/null |
---|
[5286] | 148 | [ $? -ne 0 ] && status='FAILED' |
---|
[5268] | 149 | |
---|
| 150 | |
---|
[5355] | 151 | # Comments |
---|
| 152 | #--------------------------------------------------- |
---|
[5383] | 153 | echo 'Comments' > mesg_12_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 154 | grep 'W A R N I N G:' ocean.output >> mesg_12_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5268] | 155 | |
---|
| 156 | |
---|
[5355] | 157 | # Finals tests |
---|
| 158 | #--------------------------------------------------- |
---|
[5268] | 159 | if [ $status != 'FAILED' ]; then |
---|
[5383] | 160 | printf "Step......................\nGet real CPU time\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 161 | real_cpu_time=$( ${JOB_VIEW} $job_num | awk 'END{print $9}' ) |
---|
[5383] | 162 | printf "Real CPU time\n%s\n" ${real_cpu_time} > mesg_09_realcputime_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5268] | 163 | else |
---|
[5383] | 164 | printf "Step......................\nTest input files difference\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 165 | diff inputs_list.txt ${REFE_DIR}/inputs_list.txt |
---|
| 166 | |
---|
[5383] | 167 | if [ $? -ne 0 ]; then |
---|
| 168 | printf "Input files\nDifferent" > mesg_10_inputfiles_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 169 | else |
---|
[5383] | 170 | printf "Input files\nSame" > mesg_10_inputfiles_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5355] | 171 | fi |
---|
| 172 | |
---|
| 173 | get_out |
---|
[5268] | 174 | fi |
---|
| 175 | |
---|
| 176 | |
---|
[5355] | 177 | # So far, everything is good |
---|
| 178 | #--------------------------------------------------- |
---|
[5383] | 179 | printf "Status\nOK\n" > mesg_03_status_${CFG_USER}_${CFG_ARCH}.txt |
---|
| 180 | printf "Step......................\nCode is reliable\n" | tee mesg_04_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
[5268] | 181 | |
---|
[5355] | 182 | |
---|
| 183 | # End |
---|
| 184 | #--------------------------------------------------- |
---|
| 185 | get_out |
---|