1 | #!/bin/bash |
---|
2 | |
---|
3 | |
---|
4 | comments() { |
---|
5 | state=$1 |
---|
6 | |
---|
7 | if [ "$state" == 'E R R O R' ]; then |
---|
8 | if [ -e ocean.output ]; then |
---|
9 | last_time_step=$( cat ${TEST_DIR}/time.step | tr -d [:space:] ) |
---|
10 | STEP='nemo.exe crashed at '${last_time_step} && export STEP |
---|
11 | line=$( grep -m1 -A5 "$state" ocean.output | tr -d '\n' ) |
---|
12 | else |
---|
13 | STEP='nemo.exe crashed at initialization' && export STEP |
---|
14 | return |
---|
15 | fi |
---|
16 | fi |
---|
17 | |
---|
18 | if [ $( diff -q model.log ${REFE_DIR}/model.log ) ]; then |
---|
19 | line=$( diff model.log ${REFE_DIR}/model.log | tr -d '\n' ) |
---|
20 | fi |
---|
21 | |
---|
22 | printf "Comments\n$line\n" | tee mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
23 | } |
---|
24 | |
---|
25 | diff_inputs() { |
---|
26 | mesg='Same' |
---|
27 | |
---|
28 | for file in inputs_list.txt $( ls namelist_* ) $( ls *.xml ); do |
---|
29 | diff $file ${REFE_DIR}/$file |
---|
30 | [ $? -ne 0 ] && mesg='Different' |
---|
31 | done |
---|
32 | |
---|
33 | printf "Input files\n$mesg\n" | tee mesg_09_inputfiles_${CFG_USER}_${CFG_ARCH}.txt |
---|
34 | } |
---|
35 | |
---|
36 | diff_restart() { |
---|
37 | last_time_step=$( cat ${REFE_DIR}/time.step | tr -d [:space:] ) |
---|
38 | echo 'Last time step of standard run: '${last_time_step} |
---|
39 | |
---|
40 | if [ $( find ${TEST_DIR} -name "*${last_time_step}_restart*.nc" -print -quit ) ]; then |
---|
41 | base_name=$( find ${TEST_DIR} -name "*${last_time_step}_restart*.nc" -print -quit \ |
---|
42 | | awk -F/ '{print $NF}' | sed "s/^\(.*\)$last_time_step.*/\1$last_time_step/" ) |
---|
43 | |
---|
44 | for comp in restart restart_ice restart_trc; do |
---|
45 | file=${base_name}'_'${comp} && printf "$file'.nc': " |
---|
46 | |
---|
47 | nb_dom=$( find ${TEST_DIR} -name "${file}_[0-9]*.nc" | wc -l | awk '{print $1}' ) |
---|
48 | if [ ${nb_dom} -gt 1 ]; then |
---|
49 | ${NEMO_TRBD}/rebuild_nemo -t $NPROC $file ${nb_dom} > /dev/null |
---|
50 | fi |
---|
51 | |
---|
52 | # UNIX `cmp` not suitable (filename & timestamp in .nc file) |
---|
53 | nc_diff=$( $CDO diffn ${TEST_DIR}/$file'.nc' ${REFE_DIR}/$file'.nc' 2> /dev/null | tail -1 ) |
---|
54 | if [ ! -z "${nc_diff}" ]; then |
---|
55 | export STATUS='FAILED' STEP='Restart files different' |
---|
56 | printf "$CDO diffn ${nc_diff}\n" |
---|
57 | else |
---|
58 | printf "identical\n" |
---|
59 | fi |
---|
60 | |
---|
61 | rm -f ${TEST_DIR}/${base_name}_${comp}_[0-9]*.nc &> /dev/null |
---|
62 | done |
---|
63 | |
---|
64 | else |
---|
65 | export STATUS='FAILED'; STEP='No restart files found at last time step' |
---|
66 | printf "no restart files found for comparison\n" |
---|
67 | fi |
---|
68 | |
---|
69 | } |
---|
70 | |
---|
71 | get_cpu_time() { |
---|
72 | real_cpu_time=$( eval ${JOB_TIME} ) |
---|
73 | printf "Real CPU time\n${real_cpu_time}\n" | tee mesg_10_realcputime_${CFG_USER}_${CFG_ARCH}.txt |
---|
74 | } |
---|
75 | |
---|
76 | get_inputs() { |
---|
77 | if [ ! -z "${FORC_TAR}" ]; then |
---|
78 | cmd_iol="tar -tvf ${NEMO_FORC}/${FORC_TAR}"; cmd_iof="tar -vxf ${NEMO_FORC}/${FORC_TAR}" |
---|
79 | else |
---|
80 | cmd_iol="ls ${NEMO_FORC}/*" ; cmd_iof="\cp ${NEMO_FORC}/* ." |
---|
81 | fi |
---|
82 | |
---|
83 | ${cmd_iol} > inputs_list.txt |
---|
84 | ${cmd_iof} > /dev/null |
---|
85 | } |
---|
86 | |
---|
87 | diff_textfiles() { |
---|
88 | for file in ocean.output $( ls *.stat ); do |
---|
89 | diff $file ${REFE_DIR}/$file |
---|
90 | [ $? -ne 0 ] && export STATUS='FAILED' STEP='ASCII output file diff' |
---|
91 | done |
---|
92 | } |
---|
93 | |
---|
94 | get_soft_rel() { |
---|
95 | for rel in $CDO $COMPILER $MPI $NETCDF; do |
---|
96 | arch_rel=$( echo $LOADEDMODULES | sed "s#.*$rel/\([^:]*\).*#\1#" ) |
---|
97 | [ -z "${arch_rel}" ] && arch_rel=$( echo $PATH | sed "s#.*$rel/\([^/]*\).*#\1#" ) |
---|
98 | [ "$rel" == "$COMPILER" ] && arch_rel=$( $rel --version | grep -m1 -o ' [0-9.]* ' ) |
---|
99 | echo $rel ${arch_rel} >> model.log |
---|
100 | done |
---|
101 | |
---|
102 | sed -n 4p model.log | sed 'i\Fortran compiler' > mesg_06_compiler_${CFG_USER}_${CFG_ARCH}.txt |
---|
103 | sed -n 5p model.log | sed 'i\MPI libs' > mesg_07_mpi_${CFG_USER}_${CFG_ARCH}.txt |
---|
104 | sed -n 6p model.log | sed 'i\NetCDF libs' > mesg_08_netcdf_${CFG_USER}_${CFG_ARCH}.txt |
---|
105 | } |
---|
106 | |
---|
107 | get_nemo_rev() { |
---|
108 | svn_opt='status' |
---|
109 | |
---|
110 | # If -v|--version option has been set, modify svn command |
---|
111 | if [[ $( echo ${NEMO_VERS} | grep "HEAD\|up\|update" ) \ |
---|
112 | || $( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) ]]; then |
---|
113 | rev=$( echo ${NEMO_VERS} | tr -d '[:alpha:][:punct:]' ) || rev='HEAD' |
---|
114 | svn_cmd='svn update -r '$rev |
---|
115 | else |
---|
116 | svn_cmd='svn '${svn_opt} |
---|
117 | fi |
---|
118 | |
---|
119 | rev=0 |
---|
120 | for dir in ${NEMO_ARCH} ${NEMO_CONF} ${NEMO_ENGI} \ |
---|
121 | ${NEMO_EIOI} ${NEMO_EFCM} \ |
---|
122 | ${NEMO_TCMP} ${NEMO_TRBD} \ |
---|
123 | ${DIR_XIOS} ; do |
---|
124 | |
---|
125 | # For time being, just get revision number from XIOS |
---|
126 | if [ "$dir" == "${DIR_XIOS}" ]; then |
---|
127 | rev_loc=$( svn info $dir | awk '(NR == 5) {print $NF}' ) |
---|
128 | echo 'XIOS '${rev_loc} >> model.log |
---|
129 | printf "XIOS rev.\n" > mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt |
---|
130 | printf "<a href=\"https://forge.ipsl.jussieu.fr/ioserver/changeset/${rev_loc}\" target=\"_blank\">${rev_loc}</a>" \ |
---|
131 | >> mesg_05_xios_${CFG_USER}_${CFG_ARCH}.txt |
---|
132 | continue |
---|
133 | fi |
---|
134 | |
---|
135 | echo ${svn_cmd} $dir': '$( ${svn_cmd} $dir ) |
---|
136 | rev_loc=$( svn info $dir | awk '(NR == 5) {print $NF}' ) |
---|
137 | (( ${rev_loc} >= $rev )) && rev=${rev_loc} |
---|
138 | done |
---|
139 | |
---|
140 | echo 'NEMOGCM '$rev >> model.log |
---|
141 | # printf "NEMOGCM rev.\n$rev\n" > mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt |
---|
142 | printf "NEMOGCM rev.\n" > mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt |
---|
143 | printf "<a href=\"https://forge.ipsl.jussieu.fr/nemo/changeset/$rev\" target=\"_blank\">$rev</a>" \ |
---|
144 | >> mesg_04_nemogcm_${CFG_USER}_${CFG_ARCH}.txt |
---|
145 | } |
---|
146 | |
---|
147 | mesg_make() { |
---|
148 | paste -d ';' mesg_*_${CFG_USER}_${CFG_ARCH}.txt > trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt |
---|
149 | cat trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt |
---|
150 | } |
---|
151 | |
---|
152 | mesg_publish() { |
---|
153 | |
---|
154 | if [ $PUBLISH -eq 1 ]; then |
---|
155 | |
---|
156 | if [ -f ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt ]; then cmd='tail -1'; else cmd='cat'; fi |
---|
157 | |
---|
158 | $cmd trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt >> ${REFE_DIR}/trusting_${CFG_USER}_${CFG_ARCH}.txt |
---|
159 | |
---|
160 | # Send mail only when FAILED |
---|
161 | if [[ ! -z $EMAIL && "$STATUS" == 'FAILED' ]]; then |
---|
162 | cat << END_MAIL > trusting.mail |
---|
163 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
164 | |
---|
165 | Dear $USER, |
---|
166 | |
---|
167 | The trusting sequence for ${REFE_CONF} has failed. |
---|
168 | Directory: ${TEST_DIR} |
---|
169 | Archive: ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz |
---|
170 | |
---|
171 | XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX |
---|
172 | |
---|
173 | END_MAIL |
---|
174 | #`tail -n 1 ${TEST_DIR}/mesg_03_step_${CFG_USER}_${CFG_ARCH}.txt` |
---|
175 | if [ -e ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt ]; then |
---|
176 | cat ${TEST_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.txt >> trusting.mail |
---|
177 | fi |
---|
178 | mail -s "[trusting ${REFE_CONF}] $STATUS $STEP" $EMAIL < trusting.mail |
---|
179 | fi |
---|
180 | |
---|
181 | fi |
---|
182 | } |
---|
183 | |
---|
184 | print_step() { |
---|
185 | [ ! -z "$1" ] && STEP=$1 |
---|
186 | export STEP && printf "Step.....\n$STEP\n" |
---|
187 | } |
---|
188 | |
---|
189 | get_out() { |
---|
190 | # Save tested configuration if trusting failed |
---|
191 | if [ "$STATUS" == 'FAILED' ]; then |
---|
192 | cd ${TEST_DIR} |
---|
193 | printf "Input files\n\n" > mesg_09_inputfiles_${CFG_USER}_${CFG_ARCH}.txt |
---|
194 | printf "Real CPU time\n\n" > mesg_10_realcputime_${CFG_USER}_${CFG_ARCH}.txt |
---|
195 | printf "Comments\n\n" > mesg_11_comments_${CFG_USER}_${CFG_ARCH}.txt |
---|
196 | #tar -czf ${REFE_DIR}/trusting_${DATESTR}_${CFG_USER}_${CFG_ARCH}.tar.gz * |
---|
197 | fi |
---|
198 | |
---|
199 | printf "Status\n$STATUS\n" > mesg_02_status_${CFG_USER}_${CFG_ARCH}.txt |
---|
200 | printf "Step.....\n$STEP\n" > mesg_03_step_${CFG_USER}_${CFG_ARCH}.txt |
---|
201 | |
---|
202 | mesg_make; mesg_publish |
---|
203 | |
---|
204 | exit 1 |
---|
205 | } |
---|
206 | |
---|
207 | job_pending() { |
---|
208 | time_elapsed=0; time_increment=30 |
---|
209 | sleep ${time_increment} |
---|
210 | |
---|
211 | while [[ $( ${JOB_LIST} | grep ${JOB_ID} ) && ${time_elapsed} -lt ${TIME_LIMI} ]]; do |
---|
212 | printf "\n####################################################" >> computation.log |
---|
213 | ${JOB_INFO} ${JOB_ID} >> computation.log |
---|
214 | sleep ${time_increment} |
---|
215 | let time_elapsed+=${time_increment} |
---|
216 | done |
---|
217 | |
---|
218 | sleep ${time_increment} |
---|
219 | |
---|
220 | if [ ${time_elapsed} -eq ${TIME_LIMI} ]; then |
---|
221 | STEP='Exceeded time limit' |
---|
222 | |
---|
223 | if [ $( ${JOB_LIST} | grep ${JOB_ID} ) ]; then |
---|
224 | STEP='Job not finished on time: '$(( ${TIME_LIMI}/3600 ))'h' |
---|
225 | fi |
---|
226 | |
---|
227 | print_step |
---|
228 | ${JOB_DELE} ${JOB_ID} &> /dev/null |
---|
229 | get_out $STATUS $STEP |
---|
230 | fi |
---|
231 | } |
---|