1 | #!/bin/ksh |
---|
2 | |
---|
3 | #************************************************************** |
---|
4 | # Author: Patrick Brockmann, Martial Mancip |
---|
5 | # Contact: Patrick.Brockmann__at__cea.fr Martial.Mancip__at__ipsl.jussieu.fr |
---|
6 | # $Revision:: $ Revision of last commit |
---|
7 | # $Author:: $ Author of last commit |
---|
8 | # $Date:: $ Date of last commit |
---|
9 | # IPSL (2006) |
---|
10 | # This software is governed by the CeCILL licence see libIGCM/libIGCM_CeCILL.LIC |
---|
11 | # |
---|
12 | #************************************************************** |
---|
13 | |
---|
14 | #================================================== |
---|
15 | # The documentation of this file can be automatically generated |
---|
16 | # if you use the prefix #D- for comments to be extracted. |
---|
17 | # Extract with command: cat lib* | grep "^#D-" | cut -c "4-" |
---|
18 | #================================================== |
---|
19 | |
---|
20 | #================================================== |
---|
21 | # Add high level verbosity |
---|
22 | typeset -i Verbosity=${Verbosity:=3} |
---|
23 | |
---|
24 | #================================================== |
---|
25 | # DEBUG_debug |
---|
26 | # Add low level verbosity |
---|
27 | DEBUG_debug=${DEBUG_debug:=false} |
---|
28 | |
---|
29 | #D- postProcessingStopLevel (0,1,2,3) |
---|
30 | #D- 3 stop if any post-processing went wrong |
---|
31 | #D- 2 dont stop if atlas, monitoring or metrics failed |
---|
32 | #D- 1 dont stop if atlas, monitoring, metrics, create_ts or create_se failed |
---|
33 | #D- 0 dont stop if atlas, monitoring, metrics, create_ts, create_se, rebuild or pack_* failed |
---|
34 | postProcessingStopLevel=${postProcessingStopLevel:=0} |
---|
35 | |
---|
36 | #================================================== |
---|
37 | # GENERATE RANDOM ERROR ; only apply if ( ${DEBUG_debug} ) |
---|
38 | typeset -r RandomError=false |
---|
39 | |
---|
40 | #================================================== |
---|
41 | # NULL_STR |
---|
42 | # Default null string |
---|
43 | typeset -r NULL_STR="_0_" |
---|
44 | |
---|
45 | #================================================== |
---|
46 | # libIGCM_CurrentTag |
---|
47 | # Current libIGCM tag, check compatibilty with *.card |
---|
48 | typeset -r libIGCMVersion="2.8.1" |
---|
49 | |
---|
50 | #================================================== |
---|
51 | # Exit Flag (internal debug) |
---|
52 | # When true, end the master loop AFTER SAVES FILES |
---|
53 | ExitFlag=false |
---|
54 | |
---|
55 | #================================================== |
---|
56 | # When we start to run the simulation is not finished |
---|
57 | simulationIsOver=false |
---|
58 | |
---|
59 | #================================================== |
---|
60 | # When we start to run we dont flush AMQP messages |
---|
61 | FlushAMQP=false |
---|
62 | |
---|
63 | #================================================== |
---|
64 | # Declare a stack of functions calls |
---|
65 | unset IGCM_debug_Stack |
---|
66 | unset IGCM_debug_StackArgs |
---|
67 | unset IGCM_debug_StackTiming |
---|
68 | IGCM_debug_Stack[0]=${NULL_STR} |
---|
69 | IGCM_debug_StackArgs[0]=${NULL_STR} |
---|
70 | IGCM_debug_StackTiming[0]=${NULL_STR} |
---|
71 | IGCM_debug_LenStack=0 |
---|
72 | |
---|
73 | #D-#================================================================== |
---|
74 | #D-function IGCM_debug_getDate_ms |
---|
75 | #D- * Purpose: Give number of milliseconds since 01-jan-1970 |
---|
76 | function IGCM_debug_getDate_ms |
---|
77 | { |
---|
78 | typeset nanosecs ms |
---|
79 | # nano secondes since 01-jan-1970 |
---|
80 | nanosecs=$( date +%s%N ) |
---|
81 | |
---|
82 | # truncate the last 6 digits to get milliseconds since 01-jan-1970 |
---|
83 | ms=${nanosecs:0:${#nanosecs}-6} |
---|
84 | |
---|
85 | echo "$ms" |
---|
86 | } |
---|
87 | |
---|
88 | #D-#================================================================== |
---|
89 | #D-function IGCM_debug_sizeOfTabContent |
---|
90 | #D- * Purpose: Give sumed size of a list of files |
---|
91 | #D- * Usage: IGCM_debug_sizeOfTabContent entityList destination |
---|
92 | #D- * where entityList is a list of files or directory |
---|
93 | #D- * where dest is either a directory or a file name |
---|
94 | function IGCM_debug_sizeOfTabContent |
---|
95 | { |
---|
96 | typeset entityListe destination iEntity sizeKo sumSizeKo sumSizeMo |
---|
97 | |
---|
98 | eval set +A entityListe \${${1}} |
---|
99 | destination=${2} |
---|
100 | sumSizeKo=0 |
---|
101 | |
---|
102 | # Here we will try to compute size (file or directory size) from local path and not from archive. |
---|
103 | for ((i = 0; i < ${#entityListe[*]}; i += 1)) ; do |
---|
104 | if [ -f ${entityListe[$i]} ] ; then |
---|
105 | # One file or a bunch of files has been copied without renaming from a visible filesystem |
---|
106 | iEntity=${entityListe[$i]} |
---|
107 | elif [ -f ${entityListe[$i]##/*/} ] ; then |
---|
108 | # One file or a bunch of files has been copied without renaming from an non visible filesystem |
---|
109 | # remove path /home/login/../ from entityListe elements |
---|
110 | iEntity=${entityListe[$i]##/*/} |
---|
111 | elif [ -f ${destination} ] ; then |
---|
112 | # a file has been copied and renamed |
---|
113 | iEntity=${destination} |
---|
114 | elif [ -f ${destination}/${entityListe[$i]##/*/} ] ; then |
---|
115 | # a copy in a directory but not in ${PWD} |
---|
116 | iEntity=${destination}/${entityListe[$i]##/*/} |
---|
117 | elif [ -d ${entityListe[$i]} ] ; then |
---|
118 | # a directory has been copied from a non remote place |
---|
119 | iEntity=${entityListe[$i]} |
---|
120 | elif [ -d ${destination}/${entityListe[$i]##/*/} ] ; then |
---|
121 | # a directory has been copied from a remote archive and not renamed |
---|
122 | iEntity=${destination}/${entityListe[$i]##/*/} |
---|
123 | elif [ -d ${destination} ] ; then |
---|
124 | # a directory has been copied from a remote archive and renamed |
---|
125 | iEntity=${destination} |
---|
126 | fi |
---|
127 | sizeKo=$( du --apparent-size -skL ${iEntity} | gawk '{print $1}' ) |
---|
128 | sumSizeKo=$(( $sumSizeKo + $sizeKo )) |
---|
129 | done |
---|
130 | sumSizeMo=$( echo "scale=6;${sumSizeKo}/1024" | bc ) |
---|
131 | echo "${sumSizeKo}|${sumSizeMo}" |
---|
132 | } |
---|
133 | |
---|
134 | #D-#================================================================== |
---|
135 | #D-function IGCM_debug_send_AMQP_msg__MAILTUNNEL |
---|
136 | #D- * Purpose: Take over AMQP C client using mail as a message recipient |
---|
137 | #D- * One argument : base64 encoded message |
---|
138 | #D- * Attach encoded config.card when starting the simulation |
---|
139 | |
---|
140 | function IGCM_debug_send_AMQP_msg__MAILTUNNEL { |
---|
141 | |
---|
142 | typeset b64_encoded_msg mail_recipient |
---|
143 | typeset buffer send_messages mail_frequency |
---|
144 | typeset last_mail_date__file |
---|
145 | typeset secondsBetweenRefAndLastMail secondsSinceLastMail |
---|
146 | |
---|
147 | b64_encoded_msg=$1 |
---|
148 | |
---|
149 | mail_recipient="superviseur@ipsl.jussieu.fr" |
---|
150 | send_messages=0 |
---|
151 | mail_frequency=3600 # in seconds |
---|
152 | # use to keep track when was last mail sent (maybe to be replaced with global variable) |
---|
153 | last_mail_date__file=${R_BUF}/.stamp.${config_UserChoices_TagName}.${config_UserChoices_JobName} |
---|
154 | # use to accumulate messages before sending them |
---|
155 | buffer=${R_BUF}/.buffer.${config_UserChoices_TagName}.${config_UserChoices_JobName} |
---|
156 | |
---|
157 | # init |
---|
158 | if [ ! -f "${buffer}" ]; then |
---|
159 | touch ${buffer} |
---|
160 | fi |
---|
161 | |
---|
162 | if [ ! -f "${last_mail_date__file}" ]; then |
---|
163 | touch ${last_mail_date__file} |
---|
164 | else |
---|
165 | # compute last time the file was changed (in seconds) |
---|
166 | secondsBetweenRefAndLastMail=$(stat -c %Y ${last_mail_date__file}) |
---|
167 | status=$? |
---|
168 | # |
---|
169 | # Only execute this block when the stat command succeeded. |
---|
170 | # The stat command might fail in some circumstance but we consider it is ok to continue anyway. |
---|
171 | if [ ${status} -eq 0 ] ; then |
---|
172 | secondsSinceLastMail=$(( $(date +%s) - ${secondsBetweenRefAndLastMail} )) |
---|
173 | # send message when exceeding threshold |
---|
174 | [ ${secondsSinceLastMail} -gt ${mail_frequency} ] && send_messages=1 |
---|
175 | fi |
---|
176 | fi |
---|
177 | |
---|
178 | # queue messages in the buffer |
---|
179 | echo ${b64_encoded_msg} >> ${buffer} |
---|
180 | |
---|
181 | # send mail |
---|
182 | |
---|
183 | if [ X${initBigBro} = Xtrue ] ; then |
---|
184 | #echo $(date +"%Y-%m-%dT%H:%M:%S.%N%z") > ${SUBMIT_DIR}/mail.txt |
---|
185 | mailx -s "[TEMPORARY AMQP CHANNEL]" -a ${SUBMIT_DIR}/config.card.base64 ${mail_recipient} < ${buffer} # send buffer |
---|
186 | rm -f $buffer ; touch ${buffer} # clear buffer |
---|
187 | touch ${last_mail_date__file} # memorize last mail date |
---|
188 | rm -f ${SUBMIT_DIR}/config.card.base64 # remove temporary file |
---|
189 | initBigBro=false |
---|
190 | elif [ ${send_messages} -eq 1 ] ; then |
---|
191 | #echo $(date +"%Y-%m-%dT%H:%M:%S.%N%z") >> ${SUBMIT_DIR}/mail.txt |
---|
192 | mailx -s "[TEMPORARY AMQP CHANNEL]" ${mail_recipient} < ${buffer} # send buffer |
---|
193 | rm -f ${buffer} ; touch ${buffer} # flush the buffer |
---|
194 | touch ${last_mail_date__file} # memorize last mail date |
---|
195 | fi |
---|
196 | |
---|
197 | if ( ${FlushAMQP} ) ; then |
---|
198 | mailx -s "[TEMPORARY AMQP CHANNEL]" ${mail_recipient} < ${buffer} # send buffer |
---|
199 | rm -f ${buffer} # cleaning behind us |
---|
200 | rm -f ${last_mail_date__file} # cleaning behind us |
---|
201 | fi |
---|
202 | |
---|
203 | # Allways all good for now. |
---|
204 | return 0 |
---|
205 | } |
---|
206 | |
---|
207 | #D-#================================================================== |
---|
208 | #D-function IGCM_debug_sendAMQP_Purge |
---|
209 | #D- * Purpose: Take over AMQP C client using mail as a message recipient |
---|
210 | #D- * No argument : send a message indicating a simulation has been purged |
---|
211 | |
---|
212 | function IGCM_debug_sendAMQP_Purge { |
---|
213 | |
---|
214 | typeset mail_recipient encodedBody |
---|
215 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
216 | mail_recipient="superviseur@ipsl.jussieu.fr" |
---|
217 | # Clea nup tag on server side |
---|
218 | code=8888 |
---|
219 | # Usual AMQP message to route messages on server side |
---|
220 | encodedBody=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"msgUID\":\"$(uuidgen)\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" | base64 -w 0 ) |
---|
221 | # send mail |
---|
222 | |
---|
223 | IGCM_debug_Print 2 "IGCM_debug_sendAMQP_Purge" |
---|
224 | echo ${encodedBody}|mailx -s "[TEMPORARY AMQP CHANNEL]" ${mail_recipient} |
---|
225 | fi |
---|
226 | |
---|
227 | # Allways all good for now. |
---|
228 | return 0 |
---|
229 | } |
---|
230 | |
---|
231 | #D-#================================================================== |
---|
232 | #D-function IGCM_debug_sendAMQP_Metrics |
---|
233 | #D- * Purpose: Take over AMQP C client using mail as a message recipient |
---|
234 | #D- * Two arguments : - Directory where metrics.json files can be found |
---|
235 | #D- * - Metrics Group Name. metrics will be added to this group |
---|
236 | #D- * Attach encoded metrics.json files. |
---|
237 | |
---|
238 | function IGCM_debug_sendAMQP_Metrics { |
---|
239 | |
---|
240 | typeset mail_recipient encodedBody |
---|
241 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
242 | mail_recipient="superviseur@ipsl.jussieu.fr" |
---|
243 | # Metrics tag on server side |
---|
244 | code=7100 |
---|
245 | # Usual AMQP message to route messages on server side |
---|
246 | encodedBody=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"msgUID\":\"$(uuidgen)\",\"metricsGroupName\":\"${2}\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" | base64 -w 0 ) |
---|
247 | # send mail |
---|
248 | attachmentsOptions="" |
---|
249 | for metricsFile in $( ls $1/*json ) ; do |
---|
250 | attachmentsOptions="-a ${metricsFile} ${attachmentsOptions}" |
---|
251 | done |
---|
252 | IGCM_debug_Print 2 "IGCM_debug_sendAMQP_Metrics" |
---|
253 | echo ${encodedBody}|mailx -s "[TEMPORARY AMQP CHANNEL]" ${attachmentsOptions} ${mail_recipient} |
---|
254 | fi |
---|
255 | |
---|
256 | # Allways all good for now. |
---|
257 | return 0 |
---|
258 | } |
---|
259 | |
---|
260 | #D-#================================================================== |
---|
261 | #D-function IGCM_debug_sendAMQP_projectAccounting |
---|
262 | #D- * Purpose: Take over AMQP C client using mail as a message recipient |
---|
263 | #D- * One argument : - File name where project accounting details are stored |
---|
264 | #D- * Attach encoded accounting file. |
---|
265 | |
---|
266 | function IGCM_debug_sendAMQP_projectAccounting { |
---|
267 | |
---|
268 | typeset mail_recipient encodedBody |
---|
269 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
270 | mail_recipient="superviseur@ipsl.jussieu.fr" |
---|
271 | # Metrics tag on server side |
---|
272 | code=7010 |
---|
273 | # Usual AMQP message to route messages on server side |
---|
274 | encodedBody=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"msgUID\":\"$(uuidgen)\",\"accountingProject\":\"${PROJECT}\",\"centre\":\"${CENTER}\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" | base64 -w 0 ) |
---|
275 | # send mail |
---|
276 | attachmentsOptions="-a ${1}" |
---|
277 | IGCM_debug_Print 2 "IGCM_debug_sendAMQP_projectAccounting" |
---|
278 | echo ${encodedBody}|mailx -s "[TEMPORARY AMQP CHANNEL]" ${attachmentsOptions} ${mail_recipient} |
---|
279 | # clean behind |
---|
280 | rm -f $1 |
---|
281 | fi |
---|
282 | |
---|
283 | # Allways all good for now. |
---|
284 | return 0 |
---|
285 | } |
---|
286 | |
---|
287 | #D-#================================================================== |
---|
288 | #D-function IGCM_debug_SendAMQP |
---|
289 | #D- * Purpose: Send body; encoded body and config.card to rabbitMQ |
---|
290 | function IGCM_debug_sendAMQP { |
---|
291 | |
---|
292 | typeset decal first additionnalOption encodedBody |
---|
293 | |
---|
294 | # Encode message Body |
---|
295 | encodedBody=$( echo "${Body}" | base64 -w 0 ) |
---|
296 | |
---|
297 | # Send config.card ? |
---|
298 | if [ X${1} = Xactivate ] ; then |
---|
299 | # Encode config.card |
---|
300 | cat ${SUBMIT_DIR}/config.card | base64 -w 0 > ${SUBMIT_DIR}/config.card.base64 |
---|
301 | # Prepare additionnal option |
---|
302 | additionnalOption="-f ${SUBMIT_DIR}/config.card.base64" |
---|
303 | # |
---|
304 | initBigBro=true |
---|
305 | else |
---|
306 | additionnalOption= |
---|
307 | # |
---|
308 | initBigBro=false |
---|
309 | fi |
---|
310 | |
---|
311 | # Only cosmetics : stack file |
---|
312 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
313 | decal=0 |
---|
314 | while [ ${decal} -lt ${IGCM_debug_LenStack} ]; do |
---|
315 | printf ' ' >> ${StackFileLocation}/${StackFileName} |
---|
316 | (( decal = decal + 1 )) |
---|
317 | done |
---|
318 | # Log to stack file using human readable format |
---|
319 | echo "${Body}" >> ${StackFileLocation}/${StackFileName} |
---|
320 | fi |
---|
321 | |
---|
322 | # Log separately encoded AMQP message command for reuse in a mock up |
---|
323 | #echo sendAMQPMsg -h localhost -p 5672 ${additionnalOption} -b ${encodedBody} >> ${RUN_DIR_PATH}/send.AMQP.${config_UserChoices_JobName}.${config_UserChoices_ExperimentName}.${config_UserChoices_SpaceName}.${config_UserChoices_TagName}.${CumulPeriod}.history.txt |
---|
324 | |
---|
325 | # Send the message |
---|
326 | if [ X${BigBrotherChannel} = XMAIL ] ; then |
---|
327 | IGCM_debug_send_AMQP_msg__MAILTUNNEL "${encodedBody}" |
---|
328 | status=$? |
---|
329 | else |
---|
330 | sendAMQPMsg -h localhost -p 5672 ${additionnalOption} -b ${encodedBody} |
---|
331 | status=$? |
---|
332 | fi |
---|
333 | |
---|
334 | if [ ${status} -gt 0 ] ; then |
---|
335 | IGCM_debug_Print 2 "IGCM_debug_Push/PopStack/ActivateBigBro : command sendAMQPMsg failed error code ${status}" |
---|
336 | echo sendAMQPMsg -h localhost -p 5672 -b "${Body}" |
---|
337 | exit 1 |
---|
338 | fi |
---|
339 | } |
---|
340 | |
---|
341 | #D-#================================================================== |
---|
342 | #D-function IGCM_debug_CallStack |
---|
343 | #D-* Purpose: Print the call stack tree from the oldest to the youngest (opposite of the display standard) |
---|
344 | #D- |
---|
345 | function IGCM_debug_CallStack { |
---|
346 | if ( $DEBUG_debug ) ; then |
---|
347 | # Cosmetics |
---|
348 | typeset i decal |
---|
349 | i=0 |
---|
350 | until [ $i -eq ${IGCM_debug_LenStack} ]; do |
---|
351 | decal=0 |
---|
352 | until [ $decal -eq ${i} ]; do |
---|
353 | printf -- ' ' |
---|
354 | (( decal = decal + 1 )) |
---|
355 | done |
---|
356 | echo "$i - ${IGCM_debug_Stack[$(( $IGCM_debug_LenStack-$i-1 ))]}" "(${IGCM_debug_StackArgs[$(( $IGCM_debug_LenStack-$i-1 ))]})" |
---|
357 | ((i = i + 1)) |
---|
358 | done |
---|
359 | fi |
---|
360 | } |
---|
361 | |
---|
362 | #D-#================================================================== |
---|
363 | #D-function IGCM_debug_PushStack |
---|
364 | #D-* Purpose: Push a function name in the stack |
---|
365 | #D- |
---|
366 | function IGCM_debug_PushStack { |
---|
367 | if ( $DEBUG_debug ) ; then |
---|
368 | typeset decal inputs startTime_ms |
---|
369 | |
---|
370 | # Only cosmetics : stack file |
---|
371 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
372 | echo >> ${StackFileLocation}/${StackFileName} |
---|
373 | decal=0 |
---|
374 | while [ ${decal} -lt ${IGCM_debug_LenStack} ]; do |
---|
375 | printf ' ' >> ${StackFileLocation}/${StackFileName} |
---|
376 | (( decal = decal + 1 )) |
---|
377 | done |
---|
378 | |
---|
379 | # Fill the stack file |
---|
380 | echo "> ${IGCM_debug_LenStack} : ${@}" >> ${StackFileLocation}/${StackFileName} |
---|
381 | fi |
---|
382 | |
---|
383 | # Save input list in an indexed array |
---|
384 | INPUTS=( $@ ) |
---|
385 | |
---|
386 | # Get timing information |
---|
387 | startTime_ms=$( IGCM_debug_getDate_ms ) |
---|
388 | |
---|
389 | # We add function call name on beginning of the stack |
---|
390 | set +A IGCM_debug_Stack -- ${1} ${IGCM_debug_Stack[*]} |
---|
391 | |
---|
392 | # Save timing in milliseconds in an indexed array |
---|
393 | set +A IGCM_debug_StackTiming -- ${startTime_ms} ${IGCM_debug_StackTiming[*]} |
---|
394 | |
---|
395 | # We include the "null" Args in the beginning of the StackArgs |
---|
396 | set +A IGCM_debug_StackArgs ${NULL_STR} ${IGCM_debug_StackArgs[*]} |
---|
397 | |
---|
398 | # Then, we shift StackArgs tabular |
---|
399 | # Replacing blank separated list by comma separated list of quoted elements (except the first and last element) |
---|
400 | if [ $# -gt 1 ]; then |
---|
401 | IGCM_debug_StackArgs[0]=$(echo ${INPUTS[*]:1} | sed -e "s/\ /\",\"/g" ) |
---|
402 | fi |
---|
403 | |
---|
404 | # Increment LenStack |
---|
405 | (( IGCM_debug_LenStack = IGCM_debug_LenStack + 1 )) |
---|
406 | |
---|
407 | #IGCM_debug_CallStack |
---|
408 | fi |
---|
409 | } |
---|
410 | |
---|
411 | #D-#================================================================== |
---|
412 | #D-function IGCM_debug_PopStack |
---|
413 | #D-* Purpose: Pop a function name in the stack |
---|
414 | #D- |
---|
415 | function IGCM_debug_PopStack { |
---|
416 | if ( $DEBUG_debug ) ; then |
---|
417 | typeset i decal command arguments startTime_ms endTime_ms |
---|
418 | typeset instrumentation dest prefix |
---|
419 | # they are not typeset because they are send "by adress" to son functions |
---|
420 | # we unset them to avoid "memory effect" |
---|
421 | unset fileList source |
---|
422 | |
---|
423 | # INTRODUCE SIMPLE ERROR GENERATOR TO TEST SUPERVISOR |
---|
424 | # PROBABILITY ERROR IS 0.0001 PER COMMAND OR FUNCTION CALL |
---|
425 | # THERE ARE ~500 COMMAND OR FUNCTION CALL PER PERIOD |
---|
426 | # ONLY WHEN TaskType is "computing". |
---|
427 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
428 | if [ X${TaskType} = Xcomputing ]; then |
---|
429 | if ( ${RandomError} ) ; then |
---|
430 | if [ $((RANDOM%10000)) -le 10 ] ; then |
---|
431 | IGCM_debug_Print 1 "Random error has been triggered" |
---|
432 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
433 | echo "RANDOM ERROR" >> ${StackFileLocation}/${StackFileName} |
---|
434 | fi |
---|
435 | ExitFlag=true |
---|
436 | fi |
---|
437 | fi |
---|
438 | fi |
---|
439 | fi |
---|
440 | |
---|
441 | if [ "${IGCM_debug_Stack[0]}" = "${1}" ]; then |
---|
442 | # Everything is cool |
---|
443 | |
---|
444 | # Get timing information |
---|
445 | endTime_ms=$( IGCM_debug_getDate_ms ) |
---|
446 | |
---|
447 | # Save Stack information before poping the stack |
---|
448 | command=${IGCM_debug_Stack[0]} |
---|
449 | |
---|
450 | # Go from comma separated list of quoted elements (except the first and the last element) |
---|
451 | # to unquoted space separated elements in an array |
---|
452 | set -A arguments -- $( echo ${IGCM_debug_StackArgs[0]} | sed -e "s/\",\"/\ /g" ) |
---|
453 | |
---|
454 | # Save Stack information before poping the stack |
---|
455 | startTime_ms=${IGCM_debug_StackTiming[0]} |
---|
456 | |
---|
457 | # Pop the stack |
---|
458 | (( IGCM_debug_LenStack = IGCM_debug_LenStack - 1 )) |
---|
459 | set -A IGCM_debug_Stack -- ${IGCM_debug_Stack[*]:1} |
---|
460 | set -A IGCM_debug_StackArgs -- ${IGCM_debug_StackArgs[*]:1} |
---|
461 | set -A IGCM_debug_StackTiming -- ${IGCM_debug_StackTiming[*]:1} |
---|
462 | else |
---|
463 | echo 'IGCM_debug_Exit : stack is corrupted ! LenStack =' ${IGCM_debug_LenStack} |
---|
464 | IGCM_debug_Exit $@ |
---|
465 | fi |
---|
466 | |
---|
467 | # Special actions depending on command to prepare IGCM_debug_PrintInfosActions call |
---|
468 | # We are interested in: |
---|
469 | # 0. Which command performs the work |
---|
470 | # 1. Size of entity we are working with |
---|
471 | # 2. Where are we reading |
---|
472 | # 3. Where are we writing |
---|
473 | # 4. How long it took |
---|
474 | |
---|
475 | instrumentation=false |
---|
476 | |
---|
477 | case ${command} in |
---|
478 | # Classical copy (only files are given to IGCM_sys_Cp as options) |
---|
479 | IGCM_sys_Cp) |
---|
480 | instrumentation=true |
---|
481 | # All but the latest |
---|
482 | fileList=${arguments[*]:0:${#arguments[*]}-1} |
---|
483 | # just need the first file to get the directory |
---|
484 | source=${arguments[0]} |
---|
485 | # Nothing but the latest |
---|
486 | dest=${arguments[${#arguments[*]}-1]} |
---|
487 | # Size of file whose name are stored in a list |
---|
488 | entitySize=$( IGCM_debug_sizeOfTabContent fileList ${dest} ) |
---|
489 | ;; |
---|
490 | |
---|
491 | # Copy from archive machine or from buffer |
---|
492 | IGCM_sys_Get|IGCM_sys_GetBuffer) |
---|
493 | instrumentation=true |
---|
494 | if [ ${#arguments[*]} -eq 2 ] ; then |
---|
495 | source=${arguments[0]} |
---|
496 | dest=${arguments[1]} |
---|
497 | # Size of file whose name are stored in a variable |
---|
498 | entitySize=$( IGCM_debug_sizeOfTabContent source ${dest} ) |
---|
499 | elif ( [ ${#arguments[*]} -eq 3 ] && [ ${arguments[0]} = '/l' ] ) ; then |
---|
500 | # IGCM_sys_Get /l liste_file[*] /ccc/scratch/cont003/dsm/p86denv/RUN_DIR/985998_14754/ |
---|
501 | # Keep the array name hosting the all list |
---|
502 | eval set +A fileList \${${arguments[1]}} |
---|
503 | # just need the first file to get the directory |
---|
504 | source=${fileList[0]} |
---|
505 | dest=${arguments[2]} |
---|
506 | # Size of file whose name are stored in a list |
---|
507 | entitySize=$( IGCM_debug_sizeOfTabContent fileList[*] ${dest} ) |
---|
508 | elif [ [ ${#arguments[*]} -ge 3 ] ; then |
---|
509 | # All but the latest |
---|
510 | fileList=${arguments[*]:0:${#arguments[*]}-1} |
---|
511 | # just need the first file to get the directory |
---|
512 | source=${arguments[0]} |
---|
513 | # Nothing but the latest |
---|
514 | dest=${arguments[${#arguments[*]}-1]} |
---|
515 | # Size of file whose name are stored in a list |
---|
516 | entitySize=$( IGCM_debug_sizeOfTabContent fileList ${dest} ) |
---|
517 | fi |
---|
518 | ;; |
---|
519 | |
---|
520 | # Copy from compute node or copy to archive/buffer |
---|
521 | IGCM_sys_Get_Master|IGCM_sys_Get_Dir|IGCM_sys_Put_Out|IGCM_sys_PutBuffer_Out) |
---|
522 | instrumentation=true |
---|
523 | source=${arguments[0]} |
---|
524 | dest=${arguments[1]} |
---|
525 | # Size of file whose name are stored in a variable |
---|
526 | entitySize=$( IGCM_debug_sizeOfTabContent source ${dest} ) |
---|
527 | ;; |
---|
528 | |
---|
529 | # Rebuild command |
---|
530 | IGCM_sys_rebuild|IGCM_sys_rebuild_station) |
---|
531 | instrumentation=true |
---|
532 | # All but the first |
---|
533 | fileList=${arguments[*]:1:${#arguments[*]}-1} |
---|
534 | # just need a file to get the directory |
---|
535 | source=${arguments[1]} |
---|
536 | # Nothing but the first |
---|
537 | dest=${arguments[0]} |
---|
538 | # Size of file whose name are stored in a list |
---|
539 | entitySize=$( IGCM_debug_sizeOfTabContent fileList ${dest} ) |
---|
540 | ;; |
---|
541 | |
---|
542 | # NCO commands |
---|
543 | IGCM_sys_ncrcat|IGCM_sys_ncecat|IGCM_sys_ncra|IGCM_sys_ncks|IGCM_sys_cdo) |
---|
544 | # Example of what we want to catch : only filenames in those command lines |
---|
545 | # IGCM_sys_ncrcat -O -v ${list_var_final_ncrcat} ${OUT_SE[*]} ${RESULT_SE} |
---|
546 | # IGCM_sys_ncrcat --hst -v ${liste_coord}${var} ${file1} ${liste_file_tmp[*]} ${file_out} |
---|
547 | # IGCM_sys_ncrcat -p ${dir} ${liste_file_tmp} --output ${output} |
---|
548 | # IGCM_sys_ncrcat -x -v ${list_var} -p ${dir} ${liste_file_tmp} --output ${output} |
---|
549 | instrumentation=true |
---|
550 | keepGoing=true |
---|
551 | prefix=. |
---|
552 | i=0 |
---|
553 | while ( ${keepGoing} ) ; do |
---|
554 | # the last one is not interesting |
---|
555 | if [ ${i} -eq ${#arguments[*]}-1 ] ; then |
---|
556 | keepGoing=false |
---|
557 | # look after "-p" option. Path prefix is the following arguments |
---|
558 | elif [ ${arguments[${i}]} = "-p" ] ; then |
---|
559 | ((i = i + 1)) |
---|
560 | prefix=${arguments[${i}]} |
---|
561 | ((i = i + 1)) |
---|
562 | elif [ ${i} -eq ${#arguments[*]}-1 ] ; then |
---|
563 | keepGoing=false |
---|
564 | # looking for files |
---|
565 | elif [ -f ${prefix}/${arguments[${i}]} ] ; then |
---|
566 | fileList="${fileList} ${prefix}/${arguments[${i}]}" |
---|
567 | ((i = i + 1)) |
---|
568 | # other options are not interesting |
---|
569 | else |
---|
570 | ((i = i + 1)) |
---|
571 | fi |
---|
572 | done |
---|
573 | |
---|
574 | # i value is at least 1 |
---|
575 | # just need one file to get the directory |
---|
576 | source=$( echo ${fileList} | gawk '{print $1}' ) |
---|
577 | # Nothing but the latest |
---|
578 | dest=${arguments[${#arguments[*]}-1]} |
---|
579 | # Size of file whose name are stored in a list |
---|
580 | entitySize=$( IGCM_debug_sizeOfTabContent fileList ${dest} ) |
---|
581 | ;; |
---|
582 | esac |
---|
583 | |
---|
584 | # Print information related to instrumentation |
---|
585 | ( ${instrumentation} ) && IGCM_debug_PrintInfosActions ${command} ${entitySize} ${startTime_ms} ${endTime_ms} ${dest} ${source} |
---|
586 | |
---|
587 | # Only cosmetics : stack file |
---|
588 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
589 | decal=0 |
---|
590 | while [ ${decal} -lt ${IGCM_debug_LenStack} ]; do |
---|
591 | printf ' ' >> ${StackFileLocation}/${StackFileName} |
---|
592 | (( decal = decal + 1 )) |
---|
593 | done |
---|
594 | fi |
---|
595 | |
---|
596 | if ( ${ExitFlag} ) ; then |
---|
597 | # Inform the stack file |
---|
598 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
599 | echo '!!! ExitFlag has been activated !!!' >> ${StackFileLocation}/${StackFileName} |
---|
600 | fi |
---|
601 | |
---|
602 | # Unplugged message 4900 handling for now. To ease downstream treatment. |
---|
603 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
604 | if [ X${TaskType} = Xcomputing ]; then |
---|
605 | # RabbitMQ message code "COMPUTING JOBs COMMAND FAILURE" |
---|
606 | code=1900 |
---|
607 | elif [ X${TaskType} = Xpost-processing ]; then |
---|
608 | # RabbitMQ message code "POST-PROCESSING JOBs COMMAND FAILURE" |
---|
609 | code=2900 |
---|
610 | elif [ X${TaskType} = Xchecking ]; then |
---|
611 | # RabbitMQ message code "POST-PROCESSING FROM CHECKER JOBs COMMAND FAILURE" |
---|
612 | code=3900 |
---|
613 | fi |
---|
614 | # RabbitMQ message body |
---|
615 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"msgUID\":\"$(uuidgen)\",\"command\":\"${command}\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" ) |
---|
616 | |
---|
617 | # Fill the rabbitMQ queue |
---|
618 | IGCM_debug_sendAMQP |
---|
619 | fi |
---|
620 | else |
---|
621 | # Inform the stack file |
---|
622 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
623 | echo "< ${IGCM_debug_LenStack} : ${@}" >> ${StackFileLocation}/${StackFileName} |
---|
624 | fi |
---|
625 | fi |
---|
626 | |
---|
627 | # Reset array if necessary |
---|
628 | if [ ${IGCM_debug_LenStack} = 0 ]; then |
---|
629 | #echo |
---|
630 | #IGCM_debug_Print 3 "Clean stack array" |
---|
631 | #echo |
---|
632 | unset IGCM_debug_Stack |
---|
633 | unset IGCM_debug_StackArgs |
---|
634 | unset IGCM_debug_StackTiming |
---|
635 | IGCM_debug_Stack[0]=${NULL_STR} |
---|
636 | IGCM_debug_StackArgs[0]=${NULL_STR} |
---|
637 | IGCM_debug_StackTiming[0]=${NULL_STR} |
---|
638 | fi |
---|
639 | fi |
---|
640 | #IGCM_debug_CallStack |
---|
641 | } |
---|
642 | |
---|
643 | #D-#================================================================== |
---|
644 | #D-function IGCM_debug_BigBro_Initialize |
---|
645 | #D-* Purpose: switch rabbitMQ on |
---|
646 | #D- |
---|
647 | function IGCM_debug_BigBro_Initialize { |
---|
648 | IGCM_debug_PushStack "IGCM_debug_BigBro_Initialize" |
---|
649 | |
---|
650 | typeset postProcessingIDLength postProcessingName postProcessingDate postProcessingDimn postProcessingComp postProcessingFile accountingFile |
---|
651 | |
---|
652 | # Message type standard fields: |
---|
653 | # https://github.com/Prodiguer/prodiguer-docs/wiki/MQ-Standard-Message-Fields |
---|
654 | |
---|
655 | # Message type dictionnary and custom fields: |
---|
656 | # https://github.com/Prodiguer/prodiguer-docs/wiki/Monitoring-Message-Dictionary |
---|
657 | |
---|
658 | if [ X${BigBrother} = Xtrue ] ; then |
---|
659 | # create a unique ID for this specific job |
---|
660 | jobuid=$(uuidgen) |
---|
661 | |
---|
662 | # get the assigned id by the scheduler for that job |
---|
663 | IGCM_sys_getJobSchedulerID jobSchedulerID |
---|
664 | |
---|
665 | if [ X${TaskType} = Xcomputing ]; then |
---|
666 | if ( ${FirstInitialize} ) ; then |
---|
667 | # RabbitMQ message code "BEGIN A SIMULATION" |
---|
668 | code=0000 |
---|
669 | # create and persist a unique id for this simulation |
---|
670 | simuid=$(uuidgen) |
---|
671 | IGCM_card_WriteOption ${SUBMIT_DIR}/run.card Configuration simuid ${simuid} |
---|
672 | # Standard fields for the first message |
---|
673 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"activity\":\"IPSL\",\"name\":\"${config_UserChoices_JobName}\",\"experiment\":\"${config_UserChoices_ExperimentName}\",\"space\":\"${config_UserChoices_SpaceName}\",\"model\":\"${config_UserChoices_TagName}\",\"startDate\":\"${config_UserChoices_DateBegin}\",\"endDate\":\"${config_UserChoices_DateEnd}\",\"login\":\"${LOGIN}\",\"centre\":\"${CENTER}\",\"machine\":\"${MASTER}\",\"simuid\":\"${simuid}\",\"jobuid\":\"${jobuid}\"" ) |
---|
674 | # RabbitMQ message body with specific fields associated message codes treated here |
---|
675 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"accountingProject\":\"${PROJECT}\",\"jobWarningDelay\":\"${jobWarningDelay}\",\"jobSchedulerID\":\"${jobSchedulerID}\",\"jobSubmissionPath\":\"${SUBMIT_DIR}\",\"archivePath\":\"${R_SAVE}\",\"storagePath\":\"${R_BUFR}\",\"storageSmallPath\":\"${R_FIGR}\",\"msgUID\":\"$(uuidgen)\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" ) |
---|
676 | # Fill the rabbitMQ queue (the config.card in use will be sent) |
---|
677 | IGCM_debug_sendAMQP activate |
---|
678 | else |
---|
679 | # RabbitMQ message code "A NEW COMPUTING JOB IS RUNNING PART OF A SIMULATION" |
---|
680 | code=1000 |
---|
681 | # retrieve this simulation's unique id |
---|
682 | IGCM_card_DefineVariableFromOption ${SUBMIT_DIR}/run.card Configuration simuid |
---|
683 | simuid=${run_Configuration_simuid} |
---|
684 | # Using standard fields for message others than the first one. Still subject to change |
---|
685 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"simuid\":\"${simuid}\",\"jobuid\":\"${jobuid}\"" ) |
---|
686 | # RabbitMQ message body with specific fields associated message codes treated here |
---|
687 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"accountingProject\":\"${PROJECT}\",\"jobWarningDelay\":\"${jobWarningDelay}\",\"jobSchedulerID\":\"${jobSchedulerID}\",\"jobSubmissionPath\":\"${SUBMIT_DIR}\",\"archivePath\":\"${R_SAVE}\",\"storagePath\":\"${R_BUFR}\",\"storageSmallPath\":\"${R_FIGR}\",\"msgUID\":\"$(uuidgen)\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" ) |
---|
688 | # Fill the rabbitMQ queue |
---|
689 | IGCM_debug_sendAMQP |
---|
690 | fi |
---|
691 | |
---|
692 | # NOT VERY NICE BUT ... IT WORKS |
---|
693 | # Be sure that the genericSimulationID will be small from now on |
---|
694 | # Using standard fields for messages others than the first one. Still subject to change |
---|
695 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"simuid\":\"${simuid}\",\"jobuid\":\"${jobuid}\"" ) |
---|
696 | |
---|
697 | elif [ X${TaskType} = Xpost-processing ]; then |
---|
698 | # RabbitMQ message code "A NEW POST-PROCESSING JOB IS RUNNING PART OF A SIMULATION" |
---|
699 | code=2000 |
---|
700 | # retrieve this simulation's unique id |
---|
701 | IGCM_card_DefineVariableFromOption ${SUBMIT_DIR}/run.card Configuration simuid |
---|
702 | simuid=${run_Configuration_simuid} |
---|
703 | # Using standard fields for message others than the first one. Still subject to change |
---|
704 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"simuid\":\"${simuid}\",\"jobuid\":\"${jobuid}\"" ) |
---|
705 | |
---|
706 | # Specify the post-processing task we are dealing with |
---|
707 | postProcessingIDLength=$( echo "${Script_Post_Output}" | tr -d -c "\." | wc -c ) |
---|
708 | postProcessingName=$( echo "${Script_Post_Output}" | gawk -F. '{print $1}' ) |
---|
709 | postProcessingDate=$( echo "${Script_Post_Output}" | gawk -F. '{print $2}' ) |
---|
710 | postProcessingDimn="null" |
---|
711 | postProcessingComp="null" |
---|
712 | postProcessingFile="null" |
---|
713 | if [ ${postProcessingIDLength} -eq 2 ] ; then |
---|
714 | postProcessingDimn=$( echo "${Script_Post_Output}" | gawk -F. '{print $3}' ) |
---|
715 | elif [ ${postProcessingIDLength} -eq 4 ] ; then |
---|
716 | postProcessingComp=$( echo "${Script_Post_Output}" | gawk -F. '{print $4}' ) |
---|
717 | postProcessingFile=$( echo "${Script_Post_Output}" | gawk -F. '{print $5}' ) |
---|
718 | fi |
---|
719 | |
---|
720 | # RabbitMQ message body with specific fields associated message codes treated here |
---|
721 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"accountingProject\":\"${PROJECT}\",\"jobWarningDelay\":\"${jobWarningDelay}\",\"jobSchedulerID\":\"${jobSchedulerID}\",\"jobSubmissionPath\":\"${SUBMIT_DIR}\",\"msgUID\":\"$(uuidgen)\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\",\"postProcessingName\":\"${postProcessingName}\",\"postProcessingDate\":\"${postProcessingDate}\",\"postProcessingDimn\":\"${postProcessingDimn}\",\"postProcessingComp\":\"${postProcessingComp}\",\"postProcessingFile\":\"${postProcessingFile}\"}" ) |
---|
722 | # Fill the rabbitMQ queue |
---|
723 | IGCM_debug_sendAMQP |
---|
724 | |
---|
725 | elif [ X${TaskType} = Xchecking ]; then |
---|
726 | # retrieve this simulation's unique id |
---|
727 | IGCM_card_DefineVariableFromOption ${SUBMIT_DIR}/run.card Configuration simuid |
---|
728 | simuid=${run_Configuration_simuid} |
---|
729 | # Using standard fields for message others than the first one. Still subject to change |
---|
730 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"simuid\":\"${simuid}\",\"jobuid\":\"N\/A\"" ) |
---|
731 | fi |
---|
732 | |
---|
733 | # Turn the flag on |
---|
734 | ActivateBigBro=true |
---|
735 | |
---|
736 | if [ X${TaskType} = Xcomputing ]; then |
---|
737 | # Save project accounting details in a file |
---|
738 | accountingFile=cpt_${CENTER}_${PROJECT}_$( date +"%Y%m%d_%H%M" ).dat |
---|
739 | IGCM_sys_projectAccounting ${accountingFile} |
---|
740 | |
---|
741 | # And send it |
---|
742 | IGCM_debug_sendAMQP_projectAccounting ${accountingFile} |
---|
743 | fi |
---|
744 | fi |
---|
745 | IGCM_debug_PopStack "IGCM_debug_BigBro_Initialize" |
---|
746 | } |
---|
747 | |
---|
748 | #D-#================================================================== |
---|
749 | #D-function IGCM_debug_BigBro_Update |
---|
750 | #D-* Purpose: Update rabbitMQ messages exchanges during the run |
---|
751 | #D- |
---|
752 | function IGCM_debug_BigBro_Update { |
---|
753 | IGCM_debug_PushStack "IGCM_debug_BigBro_Update" |
---|
754 | |
---|
755 | # Message type standard fields: |
---|
756 | # https://github.com/Prodiguer/prodiguer-docs/wiki/MQ-Standard-Message-Fields |
---|
757 | |
---|
758 | # Message type dictionnary and custom fields: |
---|
759 | # https://github.com/Prodiguer/prodiguer-docs/wiki/Monitoring-Message-Dictionary |
---|
760 | |
---|
761 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
762 | if [ X${TaskType} = Xcomputing ]; then |
---|
763 | # Send ${PeriodDateBegin}, ${PeriodDateEnd} and ${CumulPeriod} |
---|
764 | # RabbitMQ message body |
---|
765 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"1001\",\"PeriodDateBegin\":\"${PeriodDateBegin}\",\"PeriodDateEnd\":\"${PeriodDateEnd}\",\"CumulPeriod\":\"${CumulPeriod}\",\"msgUID\":\"$(uuidgen)\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" ) |
---|
766 | # Fill the rabbitMQ queue |
---|
767 | IGCM_debug_sendAMQP |
---|
768 | fi |
---|
769 | fi |
---|
770 | IGCM_debug_PopStack "IGCM_debug_BigBro_Update" |
---|
771 | } |
---|
772 | |
---|
773 | #D-#================================================================== |
---|
774 | #D-function IGCM_debug_BigBro_Finalize |
---|
775 | #D-* Purpose: Finalize rabbitMQ messages exchanges |
---|
776 | #D- |
---|
777 | function IGCM_debug_BigBro_Finalize { |
---|
778 | IGCM_debug_PushStack "IGCM_debug_BigBro_Finalize" |
---|
779 | |
---|
780 | # Message type standard fields: |
---|
781 | # https://github.com/Prodiguer/prodiguer-docs/wiki/MQ-Standard-Message-Fields |
---|
782 | |
---|
783 | # Message type dictionnary and custom fields: |
---|
784 | # https://github.com/Prodiguer/prodiguer-docs/wiki/Monitoring-Message-Dictionary |
---|
785 | |
---|
786 | if ( $DEBUG_debug ) ; then |
---|
787 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
788 | if [ X${TaskType} = Xcomputing ]; then |
---|
789 | if ( ${simulationIsOver} ) ; then |
---|
790 | # RabbitMQ message code "SIMULATION ENDS" |
---|
791 | code=0100 |
---|
792 | FlushAMQP=true |
---|
793 | elif ( ${ExitFlag} ) ; then |
---|
794 | # RabbitMQ message code "EXIT THE JOBS BECAUSE ERROR(S) HAS BEEN TRIGGERED" |
---|
795 | code=1999 |
---|
796 | FlushAMQP=true |
---|
797 | else |
---|
798 | # RabbitMQ message code "COMPUTING JOB ENDS" |
---|
799 | code=1100 |
---|
800 | fi |
---|
801 | elif [ X${TaskType} = Xpost-processing ]; then |
---|
802 | if ( ${ExitFlag} ) ; then |
---|
803 | # RabbitMQ message code "POST-PROCESSING JOB FAILS" |
---|
804 | code=2999 |
---|
805 | FlushAMQP=true |
---|
806 | else |
---|
807 | # RabbitMQ message code "POST-PROCESSING JOB ENDS" |
---|
808 | code=2100 |
---|
809 | FlushAMQP=true |
---|
810 | fi |
---|
811 | elif [ X${TaskType} = Xchecking ]; then |
---|
812 | if ( ${ExitFlag} ) ; then |
---|
813 | # RabbitMQ message code "POST-PROCESSING JOB FAILS" |
---|
814 | code=3999 |
---|
815 | FlushAMQP=true |
---|
816 | else |
---|
817 | # RabbitMQ message code "POST-PROCESSING JOB ENDS" |
---|
818 | code=3100 |
---|
819 | FlushAMQP=true |
---|
820 | fi |
---|
821 | fi |
---|
822 | # RabbitMQ message body |
---|
823 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"msgUID\":\"$(uuidgen)\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" ) |
---|
824 | # Fill the rabbitMQ queue |
---|
825 | IGCM_debug_sendAMQP |
---|
826 | fi |
---|
827 | fi |
---|
828 | |
---|
829 | IGCM_debug_PopStack "IGCM_debug_BigBro_Finalize" |
---|
830 | } |
---|
831 | |
---|
832 | #D-#================================================================== |
---|
833 | #D-function IGCM_debug_Exit |
---|
834 | #D-* Purpose: Print Call Stack and set ExitFlag to true |
---|
835 | #D- |
---|
836 | function IGCM_debug_Exit { |
---|
837 | IGCM_debug_PushStack "IGCM_debug_Exit" |
---|
838 | echo "IGCM_debug_Exit : " "${@}" |
---|
839 | echo |
---|
840 | echo "!!!!!!!!!!!!!!!!!!!!!!!!!!" |
---|
841 | echo "!! ERROR TRIGGERED !!" |
---|
842 | echo "!! EXIT FLAG SET !!" |
---|
843 | echo "!------------------------!" |
---|
844 | echo |
---|
845 | IGCM_debug_CallStack |
---|
846 | ExitFlag=true |
---|
847 | IGCM_debug_PopStack "IGCM_debug_Exit" |
---|
848 | } |
---|
849 | |
---|
850 | #D-#================================================== |
---|
851 | #D-function IGCM_debug_Verif_Exit |
---|
852 | #D-* Purpose: exit with number 1 if ExitFlag is true |
---|
853 | #D- |
---|
854 | function IGCM_debug_Verif_Exit { |
---|
855 | if ( ${ExitFlag} ) ; then |
---|
856 | echo "IGCM_debug_Verif_Exit : Something wrong happened previously." |
---|
857 | echo "IGCM_debug_Verif_Exit : ERROR and EXIT keyword will help find out where." |
---|
858 | # Only computing TaskType stops the job for now. |
---|
859 | if [ X${TaskType} = Xcomputing ] ; then |
---|
860 | IGCM_card_WriteOption ${SUBMIT_DIR}/run.card Configuration PeriodState "Fatal" |
---|
861 | echo " EXIT THE JOB." |
---|
862 | echo |
---|
863 | IGCM_debug_CallStack |
---|
864 | |
---|
865 | # Mail notification |
---|
866 | IGCM_sys_SendMail |
---|
867 | |
---|
868 | # Inform the rabbitMQ queue |
---|
869 | IGCM_debug_BigBro_Finalize |
---|
870 | |
---|
871 | # And Good Bye |
---|
872 | date |
---|
873 | exit 1 |
---|
874 | |
---|
875 | elif [ X${TaskType} = Xpost-processing ] ; then |
---|
876 | StopAll=false |
---|
877 | # Test if we need to stop the computing job |
---|
878 | case ${postProcessingName} in |
---|
879 | atlas*|monitoring*|metrics*) |
---|
880 | [ ${postProcessingStopLevel} -gt 2 ] && StopAll=true ;; |
---|
881 | create_*) |
---|
882 | [ ${postProcessingStopLevel} -gt 1 ] && StopAll=true ;; |
---|
883 | rebuild*|pack_*) |
---|
884 | [ ${postProcessingStopLevel} -gt 0 ] && StopAll=true ;; |
---|
885 | esac |
---|
886 | # Notify the computing job that something wrong happened and stop it. |
---|
887 | ( ${StopAll} ) && IGCM_card_WriteOption ${SUBMIT_DIR}/run.card Configuration PeriodState "Fatal.${Script_Post_Output}" |
---|
888 | |
---|
889 | # If SpaceName is PROD we stop when post_processing failed |
---|
890 | if [ X${config_UserChoices_SpaceName} = XPROD ] ; then |
---|
891 | echo " EXIT THE POST-PROCESSING JOB." |
---|
892 | echo |
---|
893 | IGCM_debug_CallStack |
---|
894 | |
---|
895 | # Inform the rabbitMQ queue |
---|
896 | IGCM_debug_BigBro_Finalize |
---|
897 | |
---|
898 | # And Good Bye |
---|
899 | date |
---|
900 | exit 1 |
---|
901 | else |
---|
902 | echo "In config.card the variable SpaceName is not in PROD" |
---|
903 | echo " SO WE DO NOT EXIT THE JOB." |
---|
904 | echo |
---|
905 | date |
---|
906 | fi |
---|
907 | elif [ X${TaskType} = Xchecking ] ; then |
---|
908 | echo "Nothing will happen for now" |
---|
909 | fi |
---|
910 | fi |
---|
911 | } |
---|
912 | |
---|
913 | #D-#================================================================== |
---|
914 | #D-function IGCM_debug_Print |
---|
915 | #D-* Purpose: Print arguments according to a level of verbosity. |
---|
916 | #D- |
---|
917 | function IGCM_debug_Print |
---|
918 | { |
---|
919 | typeset level=$1 |
---|
920 | shift |
---|
921 | |
---|
922 | if [ X"${1}" = X"-e" ]; then |
---|
923 | typeset cmd_echo="echo -e" |
---|
924 | shift |
---|
925 | else |
---|
926 | typeset cmd_echo="echo" |
---|
927 | fi |
---|
928 | |
---|
929 | if [ ${level} -le ${Verbosity} ] ; then |
---|
930 | typeset i |
---|
931 | case "${level}" in |
---|
932 | 1) for i in "$@" ; do |
---|
933 | ${cmd_echo} $(date +"%Y-%m-%d %T") "--Debug1-->" ${i} |
---|
934 | done ;; |
---|
935 | 2) for i in "$@" ; do |
---|
936 | ${cmd_echo} $(date +"%Y-%m-%d %T") "--------Debug2-->" ${i} |
---|
937 | done ;; |
---|
938 | 3) for i in "$@" ; do |
---|
939 | ${cmd_echo} $(date +"%Y-%m-%d %T") "--------------Debug3-->" ${i} |
---|
940 | done ;; |
---|
941 | esac |
---|
942 | fi |
---|
943 | } |
---|
944 | |
---|
945 | #D-#================================================================== |
---|
946 | #D-function IGCM_debug_PrintVariables |
---|
947 | #D-* Purpose: Print arguments when match a pattern |
---|
948 | #D- according to a level of verbosity. |
---|
949 | function IGCM_debug_PrintVariables |
---|
950 | { |
---|
951 | typeset level=$1 |
---|
952 | shift |
---|
953 | |
---|
954 | list=$( set | grep ^$1 | sed -e "s/'//g" ) |
---|
955 | |
---|
956 | if [ "X${list}" != X ] ; then |
---|
957 | IGCM_debug_Print ${level} ${list} |
---|
958 | fi |
---|
959 | } |
---|
960 | |
---|
961 | #D-#================================================================== |
---|
962 | #D-function IGCM_debug_PrintInfosActions |
---|
963 | #D-* Purpose: Print information related to instrumentation |
---|
964 | function IGCM_debug_PrintInfosActions |
---|
965 | { |
---|
966 | typeset actionType=$1 |
---|
967 | typeset entitySize=$2 |
---|
968 | typeset start_ms=$3 |
---|
969 | typeset end_ms=$4 |
---|
970 | |
---|
971 | typeset dest=$5 |
---|
972 | typeset source=$6 |
---|
973 | |
---|
974 | typeset diff_ms entitySizeKo entitySizeMo flux_Ko_ms flux_Ko_s flux_Mo_s |
---|
975 | typeset dirFrom dirTo |
---|
976 | |
---|
977 | diff_ms=$(( $end_ms - $start_ms )) |
---|
978 | # echo "diff_ms=$diff_ms" |
---|
979 | |
---|
980 | entitySizeKo=$( echo ${entitySize} | gawk -F"|" '{print $1}' ) |
---|
981 | # echo "entitySizeKo=$entitySizeKo" |
---|
982 | entitySizeMo=$( echo ${entitySize} | gawk -F"|" '{print $2}' ) |
---|
983 | |
---|
984 | # flux en Ko / ms |
---|
985 | flux_Ko_ms=$( echo "scale=6;${entitySizeKo}/${diff_ms}" | bc ) |
---|
986 | # echo "flux_Ko_ms=$flux_Ko_ms" |
---|
987 | |
---|
988 | # flux en Ko / s |
---|
989 | flux_Ko_s=$(( $flux_Ko_ms * 1000 )) |
---|
990 | # echo "flux_Ko_s=$flux_Ko_s" |
---|
991 | |
---|
992 | # flux en Mo / s |
---|
993 | flux_Mo_s=$( echo "scale=6;${flux_Ko_s}/1024" | bc ) |
---|
994 | # echo "flux_Mo_s=$flux_Mo_s" |
---|
995 | |
---|
996 | if [ -d $dest ] ; then |
---|
997 | dirTo=$( readlink -f ${dest} ) |
---|
998 | else |
---|
999 | dirTo=$( readlink -f $( dirname ${dest} ) ) |
---|
1000 | fi |
---|
1001 | |
---|
1002 | if [ -d $source ] ; then |
---|
1003 | dirFrom=$( readlink -f ${source} ) |
---|
1004 | else |
---|
1005 | dirFrom=$( readlink -f $( dirname ${source} ) ) |
---|
1006 | fi |
---|
1007 | |
---|
1008 | instrumentationContent=$( echo "\"actionName\":\"${actionType}\",\"size_Mo\":\"${entitySizeMo}\",\"duration_ms\":\"${diff_ms}\",\"throughput_Mo_s\":\"${flux_Mo_s}\",\"dirFrom\":\"${dirFrom}\",\"dirTo\":\"${dirTo}\"" ) |
---|
1009 | |
---|
1010 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
1011 | echo "{${instrumentationContent}}" >> ${StackFileLocation}/${StackFileName} |
---|
1012 | fi |
---|
1013 | |
---|
1014 | # Inform the rabbitMQ queue |
---|
1015 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
1016 | # RabbitMQ message body |
---|
1017 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"7000\",\"msgUID\":\"$(uuidgen)\",${instrumentationContent},\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" ) |
---|
1018 | # Fill the rabbitMQ queue |
---|
1019 | IGCM_debug_sendAMQP |
---|
1020 | fi |
---|
1021 | } |
---|
1022 | |
---|
1023 | #D-#================================================================== |
---|
1024 | #D-function IGCM_debug_Check |
---|
1025 | #D- * Purpose: Check the present file by comparison with a reference file |
---|
1026 | function IGCM_debug_Check |
---|
1027 | { |
---|
1028 | #--------------------- |
---|
1029 | if [ ! -n "${libIGCM}" ] ; then |
---|
1030 | echo "Check libIGCM_debug ..........................................[ FAILED ]" |
---|
1031 | echo "--Error--> libIGCM variable is not defined" |
---|
1032 | exit 2 |
---|
1033 | fi |
---|
1034 | |
---|
1035 | #--------------------- |
---|
1036 | if [ ! -n "${Verbosity}" ] ; then |
---|
1037 | echo "Check libIGCM_debug ..........................................[ FAILED ]" |
---|
1038 | echo "--Error--> Verbosity variable is not defined" |
---|
1039 | exit 3 |
---|
1040 | fi |
---|
1041 | |
---|
1042 | #--------------------- |
---|
1043 | # Need to remove timestamps here |
---|
1044 | diff ${libIGCM}/libIGCM_debug/IGCM_debug_Test.ref <(${libIGCM}/libIGCM_debug/IGCM_debug_Test.ksh | sed -e "s:[0-9][0-9][0-9][0-9]-[0-9][0-9]-[0-9][0-9] [0-9][0-9]\:[0-9][0-9]\:[0-9][0-9] ::g") > /dev/null 2>&1 |
---|
1045 | status=$? |
---|
1046 | |
---|
1047 | if [ ${status} -eq 0 ] ; then |
---|
1048 | echo "Check libIGCM_debug ..............................................[ OK ]" |
---|
1049 | else |
---|
1050 | echo "Check libIGCM_debug ..........................................[ FAILED ]" |
---|
1051 | echo "--Error--> Execution of ${libIGCM}/libIGCM_debug/IGCM_debug_Test.ksh" |
---|
1052 | echo " has produced the file IGCM_debug_Test.ref.failed" |
---|
1053 | echo " Please analyse differences with the reference file by typing:" |
---|
1054 | echo " diff IGCM_debug_Test.ref.failed ${libIGCM}/libIGCM_debug/IGCM_debug_Test.ref" |
---|
1055 | echo " Report errors to the author: Patrick.Brockmann@cea.fr" |
---|
1056 | diff ${libIGCM}/libIGCM_debug/IGCM_debug_Test.ref <(${libIGCM}/libIGCM_debug/IGCM_debug_Test.ksh | sed -e "s:[0-9][0-9][0-9][0-9]-[0-9][0-9]-[0-9][0-9] [0-9][0-9]\:[0-9][0-9]\:[0-9][0-9] ::g") |
---|
1057 | exit 4 |
---|
1058 | fi |
---|
1059 | #--------------------- |
---|
1060 | } |
---|