1 | #!/bin/ksh |
---|
2 | |
---|
3 | #************************************************************** |
---|
4 | # Author: Sonia Labetoulle |
---|
5 | # Contact: sonia.labetoulle__at__ipsl.jussieu.fr |
---|
6 | # $Revision:: $ Revision of last commit |
---|
7 | # $Author:: $ Author of last commit |
---|
8 | # $Date:: $ Date of last commit |
---|
9 | # IPSL (2012) |
---|
10 | # This software is governed by the CeCILL licence see libIGCM/libIGCM_CeCILL.LIC |
---|
11 | # |
---|
12 | #************************************************************** |
---|
13 | |
---|
14 | #======================================================================# |
---|
15 | |
---|
16 | # Display a report on how a simulation is faring. |
---|
17 | |
---|
18 | function ChangeUsr { |
---|
19 | |
---|
20 | # echo ">$1<" |
---|
21 | |
---|
22 | echo $1 | sed -e "s/${CurrentUsr}/${TargetUsr}/" \ |
---|
23 | -e "s/${CurrentGrp}/${TargetGrp}/" |
---|
24 | |
---|
25 | } |
---|
26 | |
---|
27 | |
---|
28 | function SearchCatalog { |
---|
29 | |
---|
30 | typeset num |
---|
31 | unset SUBMIT_DIR |
---|
32 | |
---|
33 | NbOcc=$( awk -v JobName=${JobName} \ |
---|
34 | 'BEGIN {x=0} $1 ~ JobName {++x} END {print x}' ${SimuCatalog} ) |
---|
35 | |
---|
36 | if [ ${NbOcc} -eq 0 ] ; then |
---|
37 | if [ X${fg_new} == X.true. ] ; then |
---|
38 | echo "${JobName} not in Catalog, we'll try to find it in ${WORK_DIR}" |
---|
39 | |
---|
40 | set -A FileList |
---|
41 | FileList=$( find ${WORK_DIR}/ -name "Job_${JobName}" ) |
---|
42 | echo $FileList |
---|
43 | (( NbOcc = ${#FileList[@]} )) |
---|
44 | else |
---|
45 | echo "${JobName} not in Catalog." |
---|
46 | echo "Use option -n / --new to look for it and add it to the catalog." |
---|
47 | exit |
---|
48 | fi |
---|
49 | fi |
---|
50 | |
---|
51 | |
---|
52 | if [ ${NbOcc} -eq 0 ] ; then |
---|
53 | echo "${JobName} not found." |
---|
54 | exit |
---|
55 | else |
---|
56 | if [ ${NbOcc} -eq 1 ] ; then |
---|
57 | |
---|
58 | if ( [ ${TargetUsr} == $( awk -v JobName=${JobName} '$1 ~ JobName {print $2}' ${SimuCatalog} ) ] && [ ${HostName} == $( awk -v JobName=${JobName} '$1 ~ JobName {print $3}' ${SimuCatalog} ) ] ) ; then |
---|
59 | JobName=$( awk -v JobName=${JobName} '$1 ~ JobName {print $1}' ${SimuCatalog} ) |
---|
60 | TargetUsr=$( awk -v JobName=${JobName} '$1 ~ JobName {print $2}' ${SimuCatalog} ) |
---|
61 | HostName=$( awk -v JobName=${JobName} '$1 ~ JobName {print $3}' ${SimuCatalog} ) |
---|
62 | SUBMIT_DIR=$( awk -v JobName=${JobName} '$1 ~ JobName {print $4}' ${SimuCatalog} ) |
---|
63 | else |
---|
64 | echo "${JobName} not in Catalog." |
---|
65 | exit |
---|
66 | fi |
---|
67 | fi |
---|
68 | fi |
---|
69 | |
---|
70 | } |
---|
71 | |
---|
72 | |
---|
73 | function AffichResult { |
---|
74 | |
---|
75 | fg_first=".false." |
---|
76 | fg_last=".false." |
---|
77 | |
---|
78 | while [ $# -ne 0 ] ; do |
---|
79 | case ${1} in |
---|
80 | -f|--first) |
---|
81 | fg_first=".true." |
---|
82 | shift 1 ;; |
---|
83 | -l|--last) |
---|
84 | fg_last=".true." |
---|
85 | shift 1 ;; |
---|
86 | -*) |
---|
87 | echo "usage: ${0}" |
---|
88 | echo " options = -f; -l" |
---|
89 | exit ;; |
---|
90 | *) |
---|
91 | break ;; |
---|
92 | esac |
---|
93 | done |
---|
94 | |
---|
95 | |
---|
96 | # Define colors |
---|
97 | ColEsc="\033[" |
---|
98 | ColNon="${ColEsc}0m" # Return to normal |
---|
99 | ColExp="${ColEsc}1m" # Blanc - gras |
---|
100 | ColFat="${ColEsc}1;31m" # Fatal |
---|
101 | ColCpl="${ColEsc}1;32m" # Completed |
---|
102 | ColAtt="${ColEsc}1;30m" # Waiting |
---|
103 | ColDef="${ColEsc}1;34m" # Default |
---|
104 | ColRbl="${ColEsc}31m" # Rebuild |
---|
105 | |
---|
106 | |
---|
107 | |
---|
108 | if [ X${fg_first} == X.true. ] ; then |
---|
109 | Dum="" |
---|
110 | (( len = 67 - ${#JobName} )) |
---|
111 | echo "|===========================================================================================================|" |
---|
112 | printf "| JobName = ${ColExp}%-${#JobName}s${ColNon}" ${JobName} |
---|
113 | printf "%-${len}s" ${Dum} |
---|
114 | printf "run.card : ${ColExp}%-17s${ColNon}|\n" "${LastWrite} " |
---|
115 | echo "|-------------------------|-------------|-------------------------|-------------|-----:----------:----------|" |
---|
116 | echo "| | | | | Pending Rebuilds |" |
---|
117 | echo "| Date Begin - DateEnd | PeriodState | Current Period | CumulPeriod | Nb : from : to |" |
---|
118 | echo "|-------------------------|-------------|-------------------------|-------------|-----:----------:----------|" |
---|
119 | return |
---|
120 | fi |
---|
121 | |
---|
122 | if [ X${fg_last} == X.true. ] ; then |
---|
123 | printf "|===========================================================================================================|\n" |
---|
124 | date +"${DateFormat}" |
---|
125 | return |
---|
126 | fi |
---|
127 | |
---|
128 | printf "| %-10s - %-10s | " \ |
---|
129 | $DateBegin $DateEnd |
---|
130 | |
---|
131 | case $PeriodState in |
---|
132 | Fatal) |
---|
133 | Color=${ColFat} |
---|
134 | break ;; |
---|
135 | Completed) |
---|
136 | Color=${ColCpl} |
---|
137 | break ;; |
---|
138 | Waiting|OnQueue) |
---|
139 | Color=${ColAtt} |
---|
140 | break ;; |
---|
141 | *) |
---|
142 | Color=${ColDef} |
---|
143 | break ;; |
---|
144 | esac |
---|
145 | printf "${Color}%-11s${ColNon} | " $PeriodState |
---|
146 | |
---|
147 | printf "%-10s - %-10s | %11s | " \ |
---|
148 | $PeriodDateBegin $PeriodDateEnd $CumulPeriod |
---|
149 | |
---|
150 | if ( [ X${NbRebuild} != X. ] && [ X${NbRebuild} != X0 ] ) ; then |
---|
151 | printf "${ColRbl}%3s : %-8s : %-8s${ColNon} |\n" \ |
---|
152 | $NbRebuild $FirstRebuild $LastRebuild |
---|
153 | else |
---|
154 | printf "%3s : %-8s : %-8s |\n" \ |
---|
155 | $NbRebuild $FirstRebuild $LastRebuild |
---|
156 | fi |
---|
157 | |
---|
158 | printf "|-----------------------------------------------------------------------------------------------------------|\n" |
---|
159 | printf "| Last |\n" |
---|
160 | printf "| Rebuild | Pack_Output | Pack_Restart | Pack_Debug | Monitoring | Atlas |\n" |
---|
161 | printf "|------------------|------------------|------------------|------------------|--------------|----------------|\n" |
---|
162 | |
---|
163 | if [ X${Result[1]} == XOK ] ; then |
---|
164 | Color=${ColCpl} |
---|
165 | else |
---|
166 | Color=${ColFat} |
---|
167 | fi |
---|
168 | printf "| ${Color}%-8s${ColNon} |" ${LastDate[1]} |
---|
169 | |
---|
170 | if [ X${Result[2]} == XOK ] ; then |
---|
171 | Color=${ColCpl} |
---|
172 | else |
---|
173 | Color=${ColFat} |
---|
174 | fi |
---|
175 | printf " ${Color}%-8s${ColNon} |" ${LastDate[2]} |
---|
176 | |
---|
177 | if [ X${Result[3]} == XOK ] ; then |
---|
178 | Color=${ColCpl} |
---|
179 | else |
---|
180 | Color=${ColFat} |
---|
181 | fi |
---|
182 | printf " ${Color}%-8s${ColNon} |" ${LastDate[3]} |
---|
183 | |
---|
184 | if [ X${Result[4]} == XOK ] ; then |
---|
185 | Color=${ColCpl} |
---|
186 | else |
---|
187 | Color=${ColFat} |
---|
188 | fi |
---|
189 | printf " ${Color}%-8s${ColNon} |" ${LastDate[4]} |
---|
190 | |
---|
191 | Color=${ColExp} |
---|
192 | printf " ${Color}%-4s${ColNon} | ${Color}%-12s${ColNon} |\n" \ |
---|
193 | ${LastDate[5]} ${LastDate[6]} |
---|
194 | } |
---|
195 | |
---|
196 | #======================================================================# |
---|
197 | |
---|
198 | #D- Task type (computing or post-processing) |
---|
199 | TaskType=post-processing |
---|
200 | typeset -i Verbosity=0 |
---|
201 | |
---|
202 | CurrentUsr=$( whoami ) |
---|
203 | CurrentGrp=$( groups $CurrentUsr | awk '{print $3}' ) |
---|
204 | |
---|
205 | libIGCM=${libIGCM:=::modipsl::/libIGCM} |
---|
206 | |
---|
207 | . ${libIGCM}/libIGCM_debug/libIGCM_debug.ksh |
---|
208 | ( ${DEBUG_debug} ) && IGCM_debug_Check |
---|
209 | . ${libIGCM}/libIGCM_card/libIGCM_card.ksh |
---|
210 | ( ${DEBUG_debug} ) && IGCM_card_Check |
---|
211 | . ${libIGCM}/libIGCM_date/libIGCM_date.ksh |
---|
212 | ( ${DEBUG_debug} ) && IGCM_date_Check |
---|
213 | #------- |
---|
214 | . ${libIGCM}/libIGCM_sys/libIGCM_sys.ksh |
---|
215 | . ${libIGCM}/libIGCM_config/libIGCM_config.ksh |
---|
216 | |
---|
217 | if ( [ $# -eq 0 ] ) ; then |
---|
218 | $0 -h |
---|
219 | exit |
---|
220 | fi |
---|
221 | |
---|
222 | # Default argument values |
---|
223 | TargetUsr=${CurrentUsr} |
---|
224 | HostName=$( hostname | sed -e "s/[0-9].*//" ) |
---|
225 | |
---|
226 | fg_color=".true." |
---|
227 | fg_new=".false." |
---|
228 | fg_quiet=".false." |
---|
229 | (( NbHisto = 1 )) |
---|
230 | |
---|
231 | # Get arguments from command line |
---|
232 | while [ $# -ne 0 ] ; do |
---|
233 | case $1 in |
---|
234 | -h|--help|-help) |
---|
235 | echo "usage: $0 [-u TargetUsr] JobName" |
---|
236 | exit ;; |
---|
237 | -b|--nocolor) |
---|
238 | fg_color=".false." |
---|
239 | shift 1 ;; |
---|
240 | -n|--new) |
---|
241 | fg_new=".true." |
---|
242 | shift 1 ;; |
---|
243 | -q|--quiet) |
---|
244 | fg_quiet=".true." |
---|
245 | shift 1 ;; |
---|
246 | -p|--config-path) |
---|
247 | ConfigPath="$2" |
---|
248 | shift 2 ;; |
---|
249 | -u|--user) |
---|
250 | TargetUsr="$2" |
---|
251 | shift 2 ;; |
---|
252 | -*) |
---|
253 | $0 -h |
---|
254 | exit ;; |
---|
255 | *) |
---|
256 | break ;; |
---|
257 | esac |
---|
258 | done |
---|
259 | |
---|
260 | JobName=$1 |
---|
261 | |
---|
262 | echo "TargetUsr = ${TargetUsr}" |
---|
263 | echo "HostName = ${HostName}" |
---|
264 | |
---|
265 | TargetGrp=$( groups $TargetUsr | awk '{print $3}' ) |
---|
266 | |
---|
267 | # Define the catalog in which the known simulations are stored |
---|
268 | SimuCatalog="$( ccc_home )/.simucatalog.dat" |
---|
269 | if [ ! -s ${SimuCatalog} ] ; then |
---|
270 | touch ${SimuCatalog} |
---|
271 | fi |
---|
272 | |
---|
273 | # Date format |
---|
274 | DateFormat="%d/%m/%y %R:%S" |
---|
275 | |
---|
276 | # Find SUBMIT_DIR from catalog |
---|
277 | |
---|
278 | SearchCatalog |
---|
279 | |
---|
280 | # Go to SUBMIT_DIR |
---|
281 | if [ ! X${SUBMIT_DIR} == X ] ; then |
---|
282 | |
---|
283 | echo "Submit: >${SUBMIT_DIR}<" |
---|
284 | cd $SUBMIT_DIR |
---|
285 | |
---|
286 | |
---|
287 | # Extract usefull information from run.card and config.card |
---|
288 | |
---|
289 | RunFile="${SUBMIT_DIR}/run.card" |
---|
290 | ConfFile="${SUBMIT_DIR}/config.card" |
---|
291 | |
---|
292 | IGCM_config_CommonConfiguration ${SUBMIT_DIR}/config.card |
---|
293 | |
---|
294 | |
---|
295 | if [ -s ${RunFile} ] ; then |
---|
296 | IGCM_card_DefineVariableFromOption ${RunFile} Configuration PeriodState |
---|
297 | PeriodState=${run_Configuration_PeriodState} |
---|
298 | else |
---|
299 | PeriodState="Waiting" |
---|
300 | fi |
---|
301 | |
---|
302 | if ( [ X${PeriodState} == XRunning ] || [ X${PeriodState} == XOnQueue ] ) ; then |
---|
303 | NbRun=$( ccc_mstat -f | grep -c ${JobName} ) |
---|
304 | |
---|
305 | if [ ${NbRun} -eq 0 ] ; then |
---|
306 | PeriodState="Fatal" |
---|
307 | fi |
---|
308 | fi |
---|
309 | |
---|
310 | DateBegin=${config_UserChoices_DateBegin} |
---|
311 | DateEnd=${config_UserChoices_DateEnd} |
---|
312 | TagName=${config_UserChoices_TagName} |
---|
313 | ExperimentName=${config_UserChoices_ExperimentName} |
---|
314 | SpaceName=${config_UserChoices_SpaceName} |
---|
315 | |
---|
316 | IGCM_card_DefineVariableFromOption ${RunFile} Configuration CumulPeriod |
---|
317 | CumulPeriod=${run_Configuration_CumulPeriod} |
---|
318 | |
---|
319 | if ( [ ! X${PeriodState} == XWaiting ] && [ ! X${PeriodState} == XCompleted ] ) ; then |
---|
320 | IGCM_card_DefineVariableFromOption ${RunFile} Configuration PeriodDateBegin |
---|
321 | IGCM_card_DefineVariableFromOption ${RunFile} Configuration PeriodDateEnd |
---|
322 | PeriodDateBegin=${run_Configuration_PeriodDateBegin} |
---|
323 | PeriodDateEnd=${run_Configuration_PeriodDateEnd} |
---|
324 | else |
---|
325 | PeriodDateBegin="." |
---|
326 | PeriodDateEnd="." |
---|
327 | fi |
---|
328 | |
---|
329 | DATA_DIR=${R_SAVE} |
---|
330 | POST_DIR=${R_BUFR}/Out |
---|
331 | CWORK_DIR=${R_FIGR} |
---|
332 | if [ X${config_Post_RebuildFromArchive} = Xtrue ] ; then |
---|
333 | REBUILD_DIR=${R_SAVE}/TMP |
---|
334 | else |
---|
335 | REBUILD_DIR=${BIG_DIR}/${config_UserChoices_TagName}/${config_UserChoices_JobName} |
---|
336 | fi |
---|
337 | |
---|
338 | if [ ${TargetUsr} != ${CurrentUsr} ] ; then |
---|
339 | DATA_DIR=$( ChangeUsr ${DATA_DIR}) |
---|
340 | POST_DIR=$( ChangeUsr ${POST_DIR} ) |
---|
341 | CWORK_DIR=$( ChangeUsr ${CWORK_DIR} ) |
---|
342 | REBUILD_DIR=$( ChangeUsr ${REBUILD_DIR} ) |
---|
343 | fi |
---|
344 | |
---|
345 | echo "Data: >${DATA_DIR}<" |
---|
346 | echo "Rebuild: >${REBUILD_DIR}<" |
---|
347 | echo "Post: >${POST_DIR}<" |
---|
348 | echo "Work: >${CWORK_DIR}<" |
---|
349 | |
---|
350 | if [ $PeriodState != "Waiting" ] ; then |
---|
351 | |
---|
352 | # Check missing rebuilds |
---|
353 | |
---|
354 | set -A RebuildList $( find ${REBUILD_DIR}/ -name "REBUILD_*" | sort ) |
---|
355 | if [ ${#RebuildList[*]} -gt 0 ] ; then |
---|
356 | NbRebuild=$( IGCM_sys_CountFileArchive ${REBUILD_DIR} ) |
---|
357 | |
---|
358 | FirstRebuild=$( basename ${RebuildList[0]} | cut -f2 -d\_ ) |
---|
359 | LastRebuild=$( basename ${RebuildList[ (( NbRebuild=${NbRebuild}-1 )) ]} | cut -f2 -d\_ ) |
---|
360 | else |
---|
361 | NbRebuild="." |
---|
362 | |
---|
363 | FirstRebuild="." |
---|
364 | LastRebuild="." |
---|
365 | fi |
---|
366 | else |
---|
367 | NbRebuild="." |
---|
368 | |
---|
369 | FirstRebuild="." |
---|
370 | LastRebuild="." |
---|
371 | fi |
---|
372 | |
---|
373 | typeset -A LastDate Result |
---|
374 | |
---|
375 | # Check last REBUILD |
---|
376 | FileType=rebuild_* |
---|
377 | (( ind = 1 )) |
---|
378 | |
---|
379 | LastDate[$ind]=$( ls ${POST_DIR}/${FileType}.*.out | tail -n ${NbHisto} | awk -F"." '{ print $2 }' ) |
---|
380 | |
---|
381 | dum=$( basename $( grep ^IGCM_sys_PutBuffer_Out ${POST_DIR}/${FileType}.${LastDate[$ind]}.out | tail -1 | awk '{ print $4 }' ) | awk -F"_" '{ print $3 }' ) |
---|
382 | |
---|
383 | if [[ ${dum} == ${LastDate[$ind]} ]] ; then |
---|
384 | Result[$ind]=OK |
---|
385 | else |
---|
386 | Result[$ind]=KO |
---|
387 | fi |
---|
388 | |
---|
389 | # Check last PACK jobs |
---|
390 | for FileType in pack_output pack_restart pack_debug ; do |
---|
391 | (( ind = ind + 1 )) |
---|
392 | |
---|
393 | LastDate[$ind]=$( ls ${POST_DIR}/${FileType}.*.out | tail -n ${NbHisto} | awk -F"." '{ print $2 }' ) |
---|
394 | |
---|
395 | dum=$( basename $( grep ^IGCM_sys_Put_Out ${POST_DIR}/${FileType}.${LastDate[$ind]}.out | tail -1 | awk '{ print $3 }' ) | awk -F"_" '{ print $3 }' ) |
---|
396 | |
---|
397 | if [[ ${dum} == ${LastDate[$ind]} ]] ; then |
---|
398 | Result[$ind]=OK |
---|
399 | else |
---|
400 | Result[$ind]=KO |
---|
401 | fi |
---|
402 | done |
---|
403 | |
---|
404 | |
---|
405 | # Check last MONITORING jobs |
---|
406 | |
---|
407 | #set -vx |
---|
408 | |
---|
409 | FileType=monitoring |
---|
410 | (( ind = ind + 1 )) |
---|
411 | |
---|
412 | LastDate[$ind]=$( cdo showyear ${CWORK_DIR}/MONITORING/files/ATM_bils_global_ave.nc 2> /dev/null | awk '{ print $NF }' ) |
---|
413 | |
---|
414 | # Check last ATLAS jobs |
---|
415 | |
---|
416 | FileType=atlas |
---|
417 | (( ind = ind + 1 )) |
---|
418 | |
---|
419 | LastDate[$ind]=$( ls ${CWORK_DIR}/ATLAS | tail -n ${NbHisto} ) |
---|
420 | |
---|
421 | |
---|
422 | # Time of last write on run.card |
---|
423 | LastWrite=$( ls -l --time-style=+"${DateFormat}" ${SUBMIT_DIR}/run.card | awk '{print $6 " " $7}' ) |
---|
424 | |
---|
425 | |
---|
426 | # Print results |
---|
427 | AffichResult -f |
---|
428 | AffichResult |
---|
429 | AffichResult -l |
---|
430 | |
---|
431 | fi |
---|
432 | |
---|