[913] | 1 | #!/bin/ksh |
---|
[2] | 2 | |
---|
| 3 | #************************************************************** |
---|
| 4 | # Author: Patrick Brockmann, Martial Mancip |
---|
[373] | 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 |
---|
[2] | 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 |
---|
[913] | 16 | # if you use the prefix #D- for comments to be extracted. |
---|
[2] | 17 | # Extract with command: cat lib* | grep "^#D-" | cut -c "4-" |
---|
| 18 | #================================================== |
---|
| 19 | |
---|
| 20 | #================================================== |
---|
| 21 | # Add high level verbosity |
---|
| 22 | typeset -i Verbosity=${Verbosity:=3} |
---|
| 23 | |
---|
| 24 | #================================================== |
---|
| 25 | # DEBUG_debug |
---|
| 26 | # Add low level verbosity |
---|
[1083] | 27 | DEBUG_debug=${DEBUG_debug:=false} |
---|
[2] | 28 | |
---|
[1290] | 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 | |
---|
[872] | 36 | #================================================== |
---|
| 37 | # GENERATE RANDOM ERROR ; only apply if ( ${DEBUG_debug} ) |
---|
[1244] | 38 | typeset -r RandomError=false |
---|
[872] | 39 | |
---|
[2] | 40 | #================================================== |
---|
| 41 | # NULL_STR |
---|
| 42 | # Default null string |
---|
[913] | 43 | typeset -r NULL_STR="_0_" |
---|
[2] | 44 | |
---|
| 45 | #================================================== |
---|
| 46 | # libIGCM_CurrentTag |
---|
| 47 | # Current libIGCM tag, check compatibilty with *.card |
---|
[1404] | 48 | typeset -r libIGCMVersion="2.8.3" |
---|
[2] | 49 | |
---|
| 50 | #================================================== |
---|
| 51 | # Exit Flag (internal debug) |
---|
| 52 | # When true, end the master loop AFTER SAVES FILES |
---|
| 53 | ExitFlag=false |
---|
| 54 | |
---|
| 55 | #================================================== |
---|
[1243] | 56 | # When we start to run the simulation is not finished |
---|
| 57 | simulationIsOver=false |
---|
| 58 | |
---|
| 59 | #================================================== |
---|
[1268] | 60 | # When we start to run we dont flush AMQP messages |
---|
| 61 | FlushAMQP=false |
---|
| 62 | |
---|
| 63 | #================================================== |
---|
[2] | 64 | # Declare a stack of functions calls |
---|
[59] | 65 | unset IGCM_debug_Stack |
---|
| 66 | unset IGCM_debug_StackArgs |
---|
[913] | 67 | unset IGCM_debug_StackTiming |
---|
[54] | 68 | IGCM_debug_Stack[0]=${NULL_STR} |
---|
| 69 | IGCM_debug_StackArgs[0]=${NULL_STR} |
---|
[913] | 70 | IGCM_debug_StackTiming[0]=${NULL_STR} |
---|
[2] | 71 | IGCM_debug_LenStack=0 |
---|
| 72 | |
---|
| 73 | #D-#================================================================== |
---|
[913] | 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 | { |
---|
[926] | 78 | typeset nanosecs ms |
---|
[913] | 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 |
---|
[924] | 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 |
---|
[913] | 94 | function IGCM_debug_sizeOfTabContent |
---|
| 95 | { |
---|
[924] | 96 | typeset entityListe destination iEntity sizeKo sumSizeKo sumSizeMo |
---|
| 97 | |
---|
| 98 | eval set +A entityListe \${${1}} |
---|
[941] | 99 | destination=${2} |
---|
[924] | 100 | sumSizeKo=0 |
---|
| 101 | |
---|
| 102 | # Here we will try to compute size (file or directory size) from local path and not from archive. |
---|
[941] | 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 |
---|
[924] | 109 | # remove path /home/login/../ from entityListe elements |
---|
| 110 | iEntity=${entityListe[$i]##/*/} |
---|
| 111 | elif [ -f ${destination} ] ; then |
---|
[941] | 112 | # a file has been copied and renamed |
---|
[924] | 113 | iEntity=${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 |
---|
[941] | 118 | # a directory has been copied from a non remote place |
---|
[924] | 119 | iEntity=${entityListe[$i]} |
---|
| 120 | elif [ -d ${destination}/${entityListe[$i]##/*/} ] ; then |
---|
[941] | 121 | # a directory has been copied from a remote archive and not renamed |
---|
[924] | 122 | iEntity=${destination}/${entityListe[$i]##/*/} |
---|
| 123 | elif [ -d ${destination} ] ; then |
---|
[941] | 124 | # a directory has been copied from a remote archive and renamed |
---|
[924] | 125 | iEntity=${destination} |
---|
[917] | 126 | fi |
---|
[1083] | 127 | sizeKo=$( du --apparent-size -skL ${iEntity} | gawk '{print $1}' ) |
---|
[924] | 128 | sumSizeKo=$(( $sumSizeKo + $sizeKo )) |
---|
[913] | 129 | done |
---|
[924] | 130 | sumSizeMo=$( echo "scale=6;${sumSizeKo}/1024" | bc ) |
---|
| 131 | echo "${sumSizeKo}|${sumSizeMo}" |
---|
[913] | 132 | } |
---|
| 133 | |
---|
| 134 | #D-#================================================================== |
---|
[983] | 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 |
---|
[1051] | 138 | #D- * Attach encoded config.card when starting the simulation |
---|
| 139 | |
---|
[983] | 140 | function IGCM_debug_send_AMQP_msg__MAILTUNNEL { |
---|
| 141 | |
---|
[987] | 142 | typeset b64_encoded_msg mail_recipient |
---|
[1076] | 143 | typeset buffer send_messages mail_frequency |
---|
[987] | 144 | typeset last_mail_date__file |
---|
[1234] | 145 | typeset secondsBetweenRefAndLastMail secondsSinceLastMail |
---|
[987] | 146 | |
---|
[983] | 147 | b64_encoded_msg=$1 |
---|
| 148 | |
---|
[1398] | 149 | mail_recipient="superviseur@mail.ipsl.upmc.fr" |
---|
[983] | 150 | send_messages=0 |
---|
| 151 | mail_frequency=3600 # in seconds |
---|
| 152 | # use to keep track when was last mail sent (maybe to be replaced with global variable) |
---|
[1150] | 153 | last_mail_date__file=${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}.${config_UserChoices_JobName} |
---|
[983] | 156 | |
---|
| 157 | # init |
---|
| 158 | if [ ! -f "${buffer}" ]; then |
---|
[1150] | 159 | touch ${buffer} |
---|
[983] | 160 | fi |
---|
| 161 | |
---|
| 162 | if [ ! -f "${last_mail_date__file}" ]; then |
---|
[1150] | 163 | touch ${last_mail_date__file} |
---|
[983] | 164 | else |
---|
[1150] | 165 | # compute last time the file was changed (in seconds) |
---|
[1234] | 166 | secondsBetweenRefAndLastMail=$(stat -c %Y ${last_mail_date__file}) |
---|
| 167 | status=$? |
---|
| 168 | # |
---|
[1275] | 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 |
---|
[1234] | 175 | fi |
---|
[983] | 176 | fi |
---|
| 177 | |
---|
[997] | 178 | # queue messages in the buffer |
---|
| 179 | echo ${b64_encoded_msg} >> ${buffer} |
---|
| 180 | |
---|
| 181 | # send mail |
---|
[1051] | 182 | |
---|
[1053] | 183 | if [ X${initBigBro} = Xtrue ] ; then |
---|
[1087] | 184 | #echo $(date +"%Y-%m-%dT%H:%M:%S.%N%z") > ${SUBMIT_DIR}/mail.txt |
---|
[1051] | 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 |
---|
[1293] | 188 | rm -f ${SUBMIT_DIR}/config.card.base64 # remove temporary file |
---|
[1051] | 189 | initBigBro=false |
---|
[1150] | 190 | elif [ ${send_messages} -eq 1 ] ; then |
---|
[1087] | 191 | #echo $(date +"%Y-%m-%dT%H:%M:%S.%N%z") >> ${SUBMIT_DIR}/mail.txt |
---|
[1051] | 192 | mailx -s "[TEMPORARY AMQP CHANNEL]" ${mail_recipient} < ${buffer} # send buffer |
---|
[1150] | 193 | rm -f ${buffer} ; touch ${buffer} # flush the buffer |
---|
[1051] | 194 | touch ${last_mail_date__file} # memorize last mail date |
---|
[983] | 195 | fi |
---|
[987] | 196 | |
---|
[1189] | 197 | if ( ${FlushAMQP} ) ; then |
---|
[1150] | 198 | mailx -s "[TEMPORARY AMQP CHANNEL]" ${mail_recipient} < ${buffer} # send buffer |
---|
| 199 | rm -f ${buffer} # cleaning behind us |
---|
| 200 | rm -f ${last_mail_date__file} # cleaning behind us |
---|
| 201 | fi |
---|
| 202 | |
---|
[983] | 203 | # Allways all good for now. |
---|
| 204 | return 0 |
---|
| 205 | } |
---|
| 206 | |
---|
| 207 | #D-#================================================================== |
---|
[1355] | 208 | #D-function IGCM_debug_sendAMQP_Purge |
---|
| 209 | #D- * Purpose: Take over AMQP C client using mail as a message recipient |
---|
| 210 | #D- * No argument : send a message indicating a simulation has been purged |
---|
| 211 | |
---|
| 212 | function IGCM_debug_sendAMQP_Purge { |
---|
| 213 | |
---|
| 214 | typeset mail_recipient encodedBody |
---|
| 215 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
[1418] | 216 | mail_recipient="superviseur@mail.ipsl.upmc.fr" |
---|
[1355] | 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]" ${mail_recipient} |
---|
| 225 | fi |
---|
| 226 | |
---|
| 227 | # Allways all good for now. |
---|
| 228 | return 0 |
---|
| 229 | } |
---|
| 230 | |
---|
| 231 | #D-#================================================================== |
---|
[1162] | 232 | #D-function IGCM_debug_sendAMQP_Metrics |
---|
| 233 | #D- * Purpose: Take over AMQP C client using mail as a message recipient |
---|
[1202] | 234 | #D- * Two arguments : - Directory where metrics.json files can be found |
---|
| 235 | #D- * - Metrics Group Name. metrics will be added to this group |
---|
[1162] | 236 | #D- * Attach encoded metrics.json files. |
---|
| 237 | |
---|
| 238 | function IGCM_debug_sendAMQP_Metrics { |
---|
| 239 | |
---|
| 240 | typeset mail_recipient encodedBody |
---|
| 241 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
[1418] | 242 | mail_recipient="superviseur@mail.ipsl.upmc.fr" |
---|
[1162] | 243 | # Metrics tag on server side |
---|
| 244 | code=7100 |
---|
| 245 | # Usual AMQP message to route messages on server side |
---|
[1202] | 246 | encodedBody=$( echo "{${genericSimulationID},\"msgCode\":\"${code}\",\"msgUID\":\"$(uuidgen)\",\"metricsGroupName\":\"${2}\",\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" | base64 -w 0 ) |
---|
[1162] | 247 | # send mail |
---|
| 248 | attachmentsOptions="" |
---|
| 249 | for metricsFile in $( ls $1/*json ) ; do |
---|
| 250 | attachmentsOptions="-a ${metricsFile} ${attachmentsOptions}" |
---|
| 251 | done |
---|
[1355] | 252 | IGCM_debug_Print 2 "IGCM_debug_sendAMQP_Metrics" |
---|
[1162] | 253 | echo ${encodedBody}|mailx -s "[TEMPORARY AMQP CHANNEL]" ${attachmentsOptions} ${mail_recipient} |
---|
| 254 | fi |
---|
| 255 | |
---|
| 256 | # Allways all good for now. |
---|
| 257 | return 0 |
---|
| 258 | } |
---|
| 259 | |
---|
| 260 | #D-#================================================================== |
---|
[1285] | 261 | #D-function IGCM_debug_sendAMQP_projectAccounting |
---|
| 262 | #D- * Purpose: Take over AMQP C client using mail as a message recipient |
---|
| 263 | #D- * One argument : - File name where project accounting details are stored |
---|
| 264 | #D- * Attach encoded accounting file. |
---|
| 265 | |
---|
| 266 | function IGCM_debug_sendAMQP_projectAccounting { |
---|
| 267 | |
---|
| 268 | typeset mail_recipient encodedBody |
---|
| 269 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
[1418] | 270 | mail_recipient="superviseur@mail.ipsl.upmc.fr" |
---|
[1285] | 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} |
---|
[1290] | 279 | # clean behind |
---|
| 280 | rm -f $1 |
---|
[1285] | 281 | fi |
---|
| 282 | |
---|
| 283 | # Allways all good for now. |
---|
| 284 | return 0 |
---|
| 285 | } |
---|
| 286 | |
---|
| 287 | #D-#================================================================== |
---|
[913] | 288 | #D-function IGCM_debug_SendAMQP |
---|
| 289 | #D- * Purpose: Send body; encoded body and config.card to rabbitMQ |
---|
[1053] | 290 | function IGCM_debug_sendAMQP { |
---|
| 291 | |
---|
[913] | 292 | typeset decal first additionnalOption encodedBody |
---|
| 293 | |
---|
| 294 | # Encode message Body |
---|
| 295 | encodedBody=$( 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 ${SUBMIT_DIR}/config.card.base64" |
---|
[1076] | 303 | # |
---|
[1051] | 304 | initBigBro=true |
---|
[913] | 305 | else |
---|
| 306 | additionnalOption= |
---|
[1051] | 307 | # |
---|
| 308 | initBigBro=false |
---|
[913] | 309 | fi |
---|
| 310 | |
---|
| 311 | # Only cosmetics : stack file |
---|
[1115] | 312 | if [ X${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 |
---|
[913] | 321 | |
---|
[983] | 322 | # Log separately encoded AMQP message command for reuse in a mock up |
---|
[1120] | 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 |
---|
[913] | 324 | |
---|
| 325 | # Send the message |
---|
[983] | 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 | |
---|
[913] | 334 | if [ ${status} -gt 0 ] ; then |
---|
| 335 | IGCM_debug_Print 2 "IGCM_debug_Push/PopStack/ActivateBigBro : command sendAMQPMsg failed error code ${status}" |
---|
[1051] | 336 | echo sendAMQPMsg -h localhost -p 5672 -b "${Body}" |
---|
[1090] | 337 | exit 1 |
---|
[913] | 338 | fi |
---|
| 339 | } |
---|
| 340 | |
---|
| 341 | #D-#================================================================== |
---|
[2] | 342 | #D-function IGCM_debug_CallStack |
---|
[913] | 343 | #D-* Purpose: Print the call stack tree from the oldest to the youngest (opposite of the display standard) |
---|
[2] | 344 | #D- |
---|
| 345 | function IGCM_debug_CallStack { |
---|
[544] | 346 | if ( $DEBUG_debug ) ; then |
---|
[913] | 347 | # Cosmetics |
---|
[544] | 348 | typeset i decal |
---|
[823] | 349 | i=0 |
---|
[544] | 350 | until [ $i -eq ${IGCM_debug_LenStack} ]; do |
---|
| 351 | decal=0 |
---|
| 352 | until [ $decal -eq ${i} ]; do |
---|
[869] | 353 | printf -- ' ' |
---|
[823] | 354 | (( decal = decal + 1 )) |
---|
[544] | 355 | done |
---|
[869] | 356 | echo "$i - ${IGCM_debug_Stack[$(( $IGCM_debug_LenStack-$i-1 ))]}" "(${IGCM_debug_StackArgs[$(( $IGCM_debug_LenStack-$i-1 ))]})" |
---|
[823] | 357 | ((i = i + 1)) |
---|
[544] | 358 | done |
---|
| 359 | fi |
---|
[2] | 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 { |
---|
[544] | 367 | if ( $DEBUG_debug ) ; then |
---|
[913] | 368 | typeset decal inputs startTime_ms |
---|
| 369 | |
---|
| 370 | # Only cosmetics : stack file |
---|
[1115] | 371 | if [ X${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 |
---|
[2] | 378 | |
---|
[1115] | 379 | # Fill the stack file |
---|
| 380 | echo "> ${IGCM_debug_LenStack} : ${@}" >> ${StackFileLocation}/${StackFileName} |
---|
| 381 | fi |
---|
[926] | 382 | |
---|
[913] | 383 | # Save input list in an indexed array |
---|
[823] | 384 | INPUTS=( $@ ) |
---|
[913] | 385 | |
---|
| 386 | # Get timing information |
---|
| 387 | startTime_ms=$( IGCM_debug_getDate_ms ) |
---|
| 388 | |
---|
[544] | 389 | # We add function call name on beginning of the stack |
---|
| 390 | set +A IGCM_debug_Stack -- ${1} ${IGCM_debug_Stack[*]} |
---|
[2] | 391 | |
---|
[913] | 392 | # Save timing in milliseconds in an indexed array |
---|
| 393 | set +A IGCM_debug_StackTiming -- ${startTime_ms} ${IGCM_debug_StackTiming[*]} |
---|
| 394 | |
---|
[544] | 395 | # We include the "null" Args in the beginning of the StackArgs |
---|
[913] | 396 | set +A IGCM_debug_StackArgs ${NULL_STR} ${IGCM_debug_StackArgs[*]} |
---|
| 397 | |
---|
[544] | 398 | # Then, we shift StackArgs tabular |
---|
[1065] | 399 | # Replacing blank separated list by comma separated list of quoted elements (except the first and last element) |
---|
[913] | 400 | if [ $# -gt 1 ]; then |
---|
[1065] | 401 | IGCM_debug_StackArgs[0]=$(echo ${INPUTS[*]:1} | sed -e "s/\ /\",\"/g" ) |
---|
[544] | 402 | fi |
---|
[855] | 403 | |
---|
| 404 | # Increment LenStack |
---|
[544] | 405 | (( IGCM_debug_LenStack = IGCM_debug_LenStack + 1 )) |
---|
[2] | 406 | |
---|
[869] | 407 | #IGCM_debug_CallStack |
---|
[544] | 408 | fi |
---|
[2] | 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 { |
---|
[544] | 416 | if ( $DEBUG_debug ) ; then |
---|
[926] | 417 | typeset i decal command arguments startTime_ms endTime_ms |
---|
[941] | 418 | typeset instrumentation dest prefix |
---|
[926] | 419 | # they are not typeset because they are send "by adress" to son functions |
---|
| 420 | # we unset them to avoid "memory effect" |
---|
| 421 | unset fileList source |
---|
[913] | 422 | |
---|
| 423 | # INTRODUCE SIMPLE ERROR GENERATOR TO TEST SUPERVISOR |
---|
| 424 | # PROBABILITY ERROR IS 0.0001 PER COMMAND OR FUNCTION CALL |
---|
| 425 | # THERE ARE ~500 COMMAND OR FUNCTION CALL PER PERIOD |
---|
[1277] | 426 | # ONLY WHEN TaskType is "computing". |
---|
| 427 | if [ X${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 |
---|
[1115] | 437 | fi |
---|
[913] | 438 | fi |
---|
| 439 | fi |
---|
| 440 | |
---|
[544] | 441 | if [ "${IGCM_debug_Stack[0]}" = "${1}" ]; then |
---|
[913] | 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 | |
---|
[1084] | 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 |
---|
[1083] | 452 | set -A arguments -- $( echo ${IGCM_debug_StackArgs[0]} | sed -e "s/\",\"/\ /g" ) |
---|
[913] | 453 | |
---|
| 454 | # Save Stack information before poping the stack |
---|
| 455 | startTime_ms=${IGCM_debug_StackTiming[0]} |
---|
| 456 | |
---|
| 457 | # Pop the stack |
---|
[823] | 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} |
---|
[913] | 461 | set -A IGCM_debug_StackTiming -- ${IGCM_debug_StackTiming[*]:1} |
---|
[544] | 462 | else |
---|
| 463 | echo 'IGCM_debug_Exit : stack is corrupted ! LenStack =' ${IGCM_debug_LenStack} |
---|
| 464 | IGCM_debug_Exit $@ |
---|
| 465 | fi |
---|
[913] | 466 | |
---|
[914] | 467 | # Special actions depending on command to prepare IGCM_debug_PrintInfosActions call |
---|
[913] | 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 | |
---|
[915] | 475 | instrumentation=false |
---|
| 476 | |
---|
[913] | 477 | case ${command} in |
---|
[925] | 478 | # Classical copy (only files are given to IGCM_sys_Cp as options) |
---|
| 479 | IGCM_sys_Cp) |
---|
| 480 | instrumentation=true |
---|
| 481 | # All but the latest |
---|
| 482 | fileList=${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 | |
---|
[913] | 491 | # Copy from archive machine or from buffer |
---|
| 492 | IGCM_sys_Get|IGCM_sys_GetBuffer) |
---|
[915] | 493 | instrumentation=true |
---|
[913] | 494 | if [ ${#arguments[*]} -eq 2 ] ; then |
---|
| 495 | source=${arguments[0]} |
---|
| 496 | dest=${arguments[1]} |
---|
| 497 | # Size of file whose name are stored in a variable |
---|
[917] | 498 | entitySize=$( IGCM_debug_sizeOfTabContent source ${dest} ) |
---|
[913] | 499 | elif ( [ ${#arguments[*]} -eq 3 ] && [ ${arguments[0]} = '/l' ] ) ; then |
---|
[936] | 500 | # IGCM_sys_Get /l liste_file[*] /ccc/scratch/cont003/dsm/p86denv/RUN_DIR/985998_14754/ |
---|
[913] | 501 | # Keep the array name hosting the all list |
---|
[936] | 502 | eval set +A fileList \${${arguments[1]}} |
---|
[913] | 503 | # just need the first file to get the directory |
---|
[936] | 504 | source=${fileList[0]} |
---|
[931] | 505 | dest=${arguments[2]} |
---|
[934] | 506 | # Size of file whose name are stored in a list |
---|
[936] | 507 | entitySize=$( IGCM_debug_sizeOfTabContent fileList[*] ${dest} ) |
---|
[913] | 508 | elif [ [ ${#arguments[*]} -ge 3 ] ; then |
---|
| 509 | # All but the latest |
---|
[916] | 510 | fileList=${arguments[*]:0:${#arguments[*]}-1} |
---|
[913] | 511 | # just need the first file to get the directory |
---|
| 512 | source=${arguments[0]} |
---|
| 513 | # Nothing but the latest |
---|
| 514 | dest=${arguments[${#arguments[*]}-1]} |
---|
| 515 | # Size of file whose name are stored in a list |
---|
[917] | 516 | entitySize=$( IGCM_debug_sizeOfTabContent fileList ${dest} ) |
---|
[913] | 517 | fi |
---|
| 518 | ;; |
---|
| 519 | |
---|
[925] | 520 | # Copy from compute node or copy to archive/buffer |
---|
| 521 | IGCM_sys_Get_Master|IGCM_sys_Get_Dir|IGCM_sys_Put_Out|IGCM_sys_PutBuffer_Out) |
---|
[924] | 522 | instrumentation=true |
---|
[916] | 523 | source=${arguments[0]} |
---|
[924] | 524 | dest=${arguments[1]} |
---|
| 525 | # Size of file whose name are stored in a variable |
---|
| 526 | entitySize=$( IGCM_debug_sizeOfTabContent source ${dest} ) |
---|
[913] | 527 | ;; |
---|
| 528 | |
---|
| 529 | # Rebuild command |
---|
| 530 | IGCM_sys_rebuild|IGCM_sys_rebuild_station) |
---|
[915] | 531 | instrumentation=true |
---|
[913] | 532 | # All but the first |
---|
| 533 | fileList=${arguments[*]:1:${#arguments[*]}-1} |
---|
| 534 | # just need a file to get the directory |
---|
| 535 | source=${arguments[1]} |
---|
| 536 | # Nothing but the first |
---|
| 537 | dest=${arguments[0]} |
---|
| 538 | # Size of file whose name are stored in a list |
---|
[917] | 539 | entitySize=$( IGCM_debug_sizeOfTabContent fileList ${dest} ) |
---|
[913] | 540 | ;; |
---|
[941] | 541 | |
---|
[926] | 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 |
---|
[941] | 550 | keepGoing=true |
---|
| 551 | prefix=. |
---|
[926] | 552 | i=0 |
---|
| 553 | while ( ${keepGoing} ) ; do |
---|
[941] | 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 |
---|
[926] | 559 | ((i = i + 1)) |
---|
[941] | 560 | prefix=${arguments[${i}]} |
---|
| 561 | ((i = i + 1)) |
---|
| 562 | elif [ ${i} -eq ${#arguments[*]}-1 ] ; then |
---|
[926] | 563 | keepGoing=false |
---|
[941] | 564 | # looking for files |
---|
| 565 | elif [ -f ${prefix}/${arguments[${i}]} ] ; then |
---|
| 566 | fileList="${fileList} ${prefix}/${arguments[${i}]}" |
---|
| 567 | ((i = i + 1)) |
---|
| 568 | # other options are not interesting |
---|
[926] | 569 | else |
---|
| 570 | ((i = i + 1)) |
---|
| 571 | fi |
---|
| 572 | done |
---|
[941] | 573 | |
---|
[926] | 574 | # i value is at least 1 |
---|
| 575 | # just need one file to get the directory |
---|
[941] | 576 | source=$( echo ${fileList} | gawk '{print $1}' ) |
---|
[926] | 577 | # Nothing but the latest |
---|
| 578 | dest=${arguments[${#arguments[*]}-1]} |
---|
| 579 | # Size of file whose name are stored in a list |
---|
| 580 | entitySize=$( IGCM_debug_sizeOfTabContent fileList ${dest} ) |
---|
| 581 | ;; |
---|
[913] | 582 | esac |
---|
| 583 | |
---|
| 584 | # Print information related to instrumentation |
---|
[915] | 585 | ( ${instrumentation} ) && IGCM_debug_PrintInfosActions ${command} ${entitySize} ${startTime_ms} ${endTime_ms} ${dest} ${source} |
---|
[913] | 586 | |
---|
| 587 | # Only cosmetics : stack file |
---|
[1115] | 588 | if [ X${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 |
---|
[2] | 595 | |
---|
[855] | 596 | if ( ${ExitFlag} ) ; then |
---|
| 597 | # Inform the stack file |
---|
[1115] | 598 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
| 599 | echo '!!! ExitFlag has been activated !!!' >> ${StackFileLocation}/${StackFileName} |
---|
| 600 | fi |
---|
[874] | 601 | |
---|
[1216] | 602 | # Unplugged message 4900 handling for now. To ease downstream treatment. |
---|
[1244] | 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 |
---|
[855] | 620 | else |
---|
| 621 | # Inform the stack file |
---|
[1115] | 622 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
| 623 | echo "< ${IGCM_debug_LenStack} : ${@}" >> ${StackFileLocation}/${StackFileName} |
---|
| 624 | fi |
---|
[855] | 625 | fi |
---|
| 626 | |
---|
[913] | 627 | # Reset array if necessary |
---|
[544] | 628 | if [ ${IGCM_debug_LenStack} = 0 ]; then |
---|
| 629 | #echo |
---|
| 630 | #IGCM_debug_Print 3 "Clean stack array" |
---|
| 631 | #echo |
---|
| 632 | unset IGCM_debug_Stack |
---|
| 633 | unset IGCM_debug_StackArgs |
---|
[913] | 634 | unset IGCM_debug_StackTiming |
---|
[544] | 635 | IGCM_debug_Stack[0]=${NULL_STR} |
---|
| 636 | IGCM_debug_StackArgs[0]=${NULL_STR} |
---|
[913] | 637 | IGCM_debug_StackTiming[0]=${NULL_STR} |
---|
[2] | 638 | fi |
---|
[544] | 639 | fi |
---|
[869] | 640 | #IGCM_debug_CallStack |
---|
[2] | 641 | } |
---|
| 642 | |
---|
| 643 | #D-#================================================================== |
---|
[1189] | 644 | #D-function IGCM_debug_BigBro_Initialize |
---|
[855] | 645 | #D-* Purpose: switch rabbitMQ on |
---|
| 646 | #D- |
---|
[1189] | 647 | function IGCM_debug_BigBro_Initialize { |
---|
| 648 | IGCM_debug_PushStack "IGCM_debug_BigBro_Initialize" |
---|
[855] | 649 | |
---|
[1347] | 650 | typeset postProcessingIDLength postProcessingName postProcessingDate postProcessingDimn postProcessingComp postProcessingFile accountingFile |
---|
[1229] | 651 | |
---|
[1051] | 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 | |
---|
[868] | 658 | if [ X${BigBrother} = Xtrue ] ; then |
---|
[1051] | 659 | # create a unique ID for this specific job |
---|
| 660 | jobuid=$(uuidgen) |
---|
[913] | 661 | |
---|
[1244] | 662 | # get the assigned id by the scheduler for that job |
---|
| 663 | IGCM_sys_getJobSchedulerID jobSchedulerID |
---|
| 664 | |
---|
[1189] | 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 ${simuid} |
---|
| 672 | # Standard fields for the first message |
---|
[1244] | 673 | genericSimulationID=$( 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}\"" ) |
---|
[1189] | 674 | # RabbitMQ message body with specific fields associated message codes treated here |
---|
[1353] | 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" )\"}" ) |
---|
[1189] | 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 |
---|
[1244] | 685 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"simuid\":\"${simuid}\",\"jobuid\":\"${jobuid}\"" ) |
---|
[1189] | 686 | # RabbitMQ message body with specific fields associated message codes treated here |
---|
[1353] | 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" )\"}" ) |
---|
[1189] | 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 |
---|
[1244] | 695 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"simuid\":\"${simuid}\",\"jobuid\":\"${jobuid}\"" ) |
---|
[1189] | 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 |
---|
[1076] | 700 | # retrieve this simulation's unique id |
---|
[1051] | 701 | IGCM_card_DefineVariableFromOption ${SUBMIT_DIR}/run.card Configuration simuid |
---|
[1086] | 702 | simuid=${run_Configuration_simuid} |
---|
[1076] | 703 | # Using standard fields for message others than the first one. Still subject to change |
---|
[1244] | 704 | genericSimulationID=$( echo "\"msgApplication\":\"monitoring\",\"msgProducer\":\"libigcm\",\"msgProducerVersion\":\"${libIGCMVersion}\",\"simuid\":\"${simuid}\",\"jobuid\":\"${jobuid}\"" ) |
---|
[1229] | 705 | |
---|
| 706 | # Specify the post-processing task we are dealing with |
---|
[1244] | 707 | postProcessingIDLength=$( echo "${Script_Post_Output}" | tr -d -c "\." | wc -c ) |
---|
[1229] | 708 | postProcessingName=$( echo "${Script_Post_Output}" | gawk -F. '{print $1}' ) |
---|
| 709 | postProcessingDate=$( echo "${Script_Post_Output}" | gawk -F. '{print $2}' ) |
---|
[1231] | 710 | postProcessingDimn="null" |
---|
| 711 | postProcessingComp="null" |
---|
| 712 | postProcessingFile="null" |
---|
[1254] | 713 | if [ ${postProcessingIDLength} -eq 2 ] ; then |
---|
[1231] | 714 | postProcessingDimn=$( echo "${Script_Post_Output}" | gawk -F. '{print $3}' ) |
---|
[1254] | 715 | elif [ ${postProcessingIDLength} -eq 4 ] ; then |
---|
[1229] | 716 | postProcessingComp=$( echo "${Script_Post_Output}" | gawk -F. '{print $4}' ) |
---|
| 717 | postProcessingFile=$( echo "${Script_Post_Output}" | gawk -F. '{print $5}' ) |
---|
| 718 | fi |
---|
| 719 | |
---|
[1087] | 720 | # RabbitMQ message body with specific fields associated message codes treated here |
---|
[1244] | 721 | Body=$( 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\":\"${postProcessingFile}\"}" ) |
---|
[1087] | 722 | # Fill the rabbitMQ queue |
---|
| 723 | IGCM_debug_sendAMQP |
---|
[1355] | 724 | |
---|
| 725 | elif [ X${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\":\"${simuid}\",\"jobuid\":\"N\/A\"" ) |
---|
[855] | 731 | fi |
---|
[1285] | 732 | |
---|
[1295] | 733 | # Turn the flag on |
---|
| 734 | ActivateBigBro=true |
---|
| 735 | |
---|
[1354] | 736 | if [ X${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} |
---|
[1285] | 740 | |
---|
[1354] | 741 | # And send it |
---|
| 742 | IGCM_debug_sendAMQP_projectAccounting ${accountingFile} |
---|
| 743 | fi |
---|
[855] | 744 | fi |
---|
[1189] | 745 | IGCM_debug_PopStack "IGCM_debug_BigBro_Initialize" |
---|
[855] | 746 | } |
---|
| 747 | |
---|
| 748 | #D-#================================================================== |
---|
[1332] | 749 | #D-function IGCM_debug_BigBro_Update |
---|
| 750 | #D-* Purpose: Update rabbitMQ messages exchanges during the run |
---|
| 751 | #D- |
---|
| 752 | function IGCM_debug_BigBro_Update { |
---|
| 753 | IGCM_debug_PushStack "IGCM_debug_BigBro_Update" |
---|
| 754 | |
---|
| 755 | # Message type standard fields: |
---|
| 756 | # https://github.com/Prodiguer/prodiguer-docs/wiki/MQ-Standard-Message-Fields |
---|
| 757 | |
---|
| 758 | # Message type dictionnary and custom fields: |
---|
| 759 | # https://github.com/Prodiguer/prodiguer-docs/wiki/Monitoring-Message-Dictionary |
---|
| 760 | |
---|
| 761 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
[1354] | 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 |
---|
[1332] | 769 | fi |
---|
| 770 | IGCM_debug_PopStack "IGCM_debug_BigBro_Update" |
---|
| 771 | } |
---|
| 772 | |
---|
| 773 | #D-#================================================================== |
---|
[1189] | 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 |
---|
[1207] | 793 | elif ( ${ExitFlag} ) ; then |
---|
| 794 | # RabbitMQ message code "EXIT THE JOBS BECAUSE ERROR(S) HAS BEEN TRIGGERED" |
---|
[1244] | 795 | code=1999 |
---|
[1207] | 796 | FlushAMQP=true |
---|
[1189] | 797 | else |
---|
| 798 | # RabbitMQ message code "COMPUTING JOB ENDS" |
---|
| 799 | code=1100 |
---|
| 800 | fi |
---|
| 801 | elif [ X${TaskType} = Xpost-processing ]; then |
---|
[1207] | 802 | if ( ${ExitFlag} ) ; then |
---|
| 803 | # RabbitMQ message code "POST-PROCESSING JOB FAILS" |
---|
[1244] | 804 | code=2999 |
---|
[1207] | 805 | FlushAMQP=true |
---|
[1244] | 806 | else |
---|
[1207] | 807 | # RabbitMQ message code "POST-PROCESSING JOB ENDS" |
---|
| 808 | code=2100 |
---|
| 809 | FlushAMQP=true |
---|
| 810 | fi |
---|
[1244] | 811 | elif [ X${TaskType} = Xchecking ]; then |
---|
| 812 | if ( ${ExitFlag} ) ; then |
---|
| 813 | # RabbitMQ message code "POST-PROCESSING JOB FAILS" |
---|
| 814 | code=3999 |
---|
| 815 | FlushAMQP=true |
---|
| 816 | else |
---|
| 817 | # RabbitMQ message code "POST-PROCESSING JOB ENDS" |
---|
| 818 | code=3100 |
---|
| 819 | FlushAMQP=true |
---|
| 820 | fi |
---|
[1189] | 821 | fi |
---|
| 822 | # RabbitMQ message body |
---|
| 823 | Body=$( 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-#================================================================== |
---|
[2] | 833 | #D-function IGCM_debug_Exit |
---|
| 834 | #D-* Purpose: Print Call Stack and set ExitFlag to true |
---|
| 835 | #D- |
---|
| 836 | function IGCM_debug_Exit { |
---|
[544] | 837 | IGCM_debug_PushStack "IGCM_debug_Exit" |
---|
| 838 | echo "IGCM_debug_Exit : " "${@}" |
---|
[913] | 839 | echo |
---|
[894] | 840 | echo "!!!!!!!!!!!!!!!!!!!!!!!!!!" |
---|
[913] | 841 | echo "!! ERROR TRIGGERED !!" |
---|
| 842 | echo "!! EXIT FLAG SET !!" |
---|
| 843 | echo "!------------------------!" |
---|
| 844 | echo |
---|
[894] | 845 | IGCM_debug_CallStack |
---|
[544] | 846 | ExitFlag=true |
---|
| 847 | IGCM_debug_PopStack "IGCM_debug_Exit" |
---|
[2] | 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 { |
---|
[544] | 855 | if ( ${ExitFlag} ) ; then |
---|
[1206] | 856 | echo "IGCM_debug_Verif_Exit : Something wrong happened previously." |
---|
[1207] | 857 | echo "IGCM_debug_Verif_Exit : ERROR and EXIT keyword will help find out where." |
---|
[1206] | 858 | # Only computing TaskType stops the job for now. |
---|
| 859 | if [ X${TaskType} = Xcomputing ] ; then |
---|
[775] | 860 | IGCM_card_WriteOption ${SUBMIT_DIR}/run.card Configuration PeriodState "Fatal" |
---|
| 861 | echo " EXIT THE JOB." |
---|
| 862 | echo |
---|
[869] | 863 | IGCM_debug_CallStack |
---|
[874] | 864 | |
---|
[1206] | 865 | # Mail notification |
---|
| 866 | IGCM_sys_SendMail |
---|
[913] | 867 | |
---|
[1207] | 868 | # Inform the rabbitMQ queue |
---|
| 869 | IGCM_debug_BigBro_Finalize |
---|
| 870 | |
---|
[544] | 871 | # And Good Bye |
---|
| 872 | date |
---|
| 873 | exit 1 |
---|
[1206] | 874 | |
---|
| 875 | elif [ X${TaskType} = Xpost-processing ] ; then |
---|
[1302] | 876 | StopAll=false |
---|
[1478] | 877 | typeset postProcessingName |
---|
| 878 | postProcessingName=$( echo "${Script_Post_Output}" | gawk -F. '{print $1}' ) |
---|
[1282] | 879 | # Test if we need to stop the computing job |
---|
| 880 | case ${postProcessingName} in |
---|
| 881 | atlas*|monitoring*|metrics*) |
---|
[1603] | 882 | [ ${postProcessingStopLevel} -gt 2 ] && StopAll=true |
---|
| 883 | echo "The job is not a pack job " |
---|
| 884 | echo "SO WE DO NOT EXIT THE JOB." |
---|
| 885 | echo |
---|
| 886 | date |
---|
| 887 | ;; |
---|
[1282] | 888 | create_*) |
---|
[1603] | 889 | [ ${postProcessingStopLevel} -gt 1 ] && StopAll=true |
---|
| 890 | echo "The job is not a pack job " |
---|
| 891 | echo "SO WE DO NOT EXIT THE JOB." |
---|
| 892 | echo |
---|
| 893 | date |
---|
| 894 | ;; |
---|
[1282] | 895 | rebuild*|pack_*) |
---|
[1603] | 896 | if [ ${postProcessingStopLevel} -gt 0 ] ; then |
---|
[1282] | 897 | # Notify the computing job that something wrong happened and stop it. |
---|
[1603] | 898 | IGCM_card_WriteOption ${SUBMIT_DIR}/run.card Configuration PeriodState "Fatal.${Script_Post_Output}" |
---|
| 899 | fi |
---|
[1220] | 900 | echo " EXIT THE POST-PROCESSING JOB." |
---|
[1206] | 901 | echo |
---|
[1207] | 902 | IGCM_debug_CallStack |
---|
| 903 | |
---|
| 904 | # Inform the rabbitMQ queue |
---|
| 905 | IGCM_debug_BigBro_Finalize |
---|
| 906 | |
---|
[1206] | 907 | # And Good Bye |
---|
| 908 | date |
---|
| 909 | exit 1 |
---|
[1603] | 910 | ;; |
---|
| 911 | esac |
---|
| 912 | # Notify the computing job that something wrong happened and stop it. |
---|
| 913 | ( ${StopAll} ) && IGCM_card_WriteOption ${SUBMIT_DIR}/run.card Configuration PeriodState "Fatal.${Script_Post_Output}" |
---|
| 914 | |
---|
[1206] | 915 | elif [ X${TaskType} = Xchecking ] ; then |
---|
[1207] | 916 | echo "Nothing will happen for now" |
---|
[2] | 917 | fi |
---|
[544] | 918 | fi |
---|
[2] | 919 | } |
---|
| 920 | |
---|
| 921 | #D-#================================================================== |
---|
| 922 | #D-function IGCM_debug_Print |
---|
| 923 | #D-* Purpose: Print arguments according to a level of verbosity. |
---|
| 924 | #D- |
---|
| 925 | function IGCM_debug_Print |
---|
| 926 | { |
---|
[544] | 927 | typeset level=$1 |
---|
| 928 | shift |
---|
| 929 | |
---|
| 930 | if [ X"${1}" = X"-e" ]; then |
---|
| 931 | typeset cmd_echo="echo -e" |
---|
[2] | 932 | shift |
---|
[544] | 933 | else |
---|
| 934 | typeset cmd_echo="echo" |
---|
| 935 | fi |
---|
[2] | 936 | |
---|
[544] | 937 | if [ ${level} -le ${Verbosity} ] ; then |
---|
| 938 | typeset i |
---|
| 939 | case "${level}" in |
---|
| 940 | 1) for i in "$@" ; do |
---|
[734] | 941 | ${cmd_echo} $(date +"%Y-%m-%d %T") "--Debug1-->" ${i} |
---|
[913] | 942 | done ;; |
---|
[544] | 943 | 2) for i in "$@" ; do |
---|
[734] | 944 | ${cmd_echo} $(date +"%Y-%m-%d %T") "--------Debug2-->" ${i} |
---|
[913] | 945 | done ;; |
---|
[544] | 946 | 3) for i in "$@" ; do |
---|
[734] | 947 | ${cmd_echo} $(date +"%Y-%m-%d %T") "--------------Debug3-->" ${i} |
---|
[913] | 948 | done ;; |
---|
[544] | 949 | esac |
---|
| 950 | fi |
---|
[2] | 951 | } |
---|
| 952 | |
---|
| 953 | #D-#================================================================== |
---|
| 954 | #D-function IGCM_debug_PrintVariables |
---|
| 955 | #D-* Purpose: Print arguments when match a pattern |
---|
| 956 | #D- according to a level of verbosity. |
---|
| 957 | function IGCM_debug_PrintVariables |
---|
| 958 | { |
---|
[544] | 959 | typeset level=$1 |
---|
| 960 | shift |
---|
[2] | 961 | |
---|
[830] | 962 | list=$( set | grep ^$1 | sed -e "s/'//g" ) |
---|
[54] | 963 | |
---|
[544] | 964 | if [ "X${list}" != X ] ; then |
---|
| 965 | IGCM_debug_Print ${level} ${list} |
---|
| 966 | fi |
---|
[2] | 967 | } |
---|
| 968 | |
---|
| 969 | #D-#================================================================== |
---|
[914] | 970 | #D-function IGCM_debug_PrintInfosActions |
---|
[913] | 971 | #D-* Purpose: Print information related to instrumentation |
---|
| 972 | function IGCM_debug_PrintInfosActions |
---|
| 973 | { |
---|
| 974 | typeset actionType=$1 |
---|
| 975 | typeset entitySize=$2 |
---|
| 976 | typeset start_ms=$3 |
---|
| 977 | typeset end_ms=$4 |
---|
| 978 | |
---|
| 979 | typeset dest=$5 |
---|
| 980 | typeset source=$6 |
---|
| 981 | |
---|
| 982 | typeset diff_ms entitySizeKo entitySizeMo flux_Ko_ms flux_Ko_s flux_Mo_s |
---|
[1090] | 983 | typeset dirFrom dirTo |
---|
[913] | 984 | |
---|
| 985 | diff_ms=$(( $end_ms - $start_ms )) |
---|
| 986 | # echo "diff_ms=$diff_ms" |
---|
| 987 | |
---|
| 988 | entitySizeKo=$( echo ${entitySize} | gawk -F"|" '{print $1}' ) |
---|
| 989 | # echo "entitySizeKo=$entitySizeKo" |
---|
| 990 | entitySizeMo=$( echo ${entitySize} | gawk -F"|" '{print $2}' ) |
---|
| 991 | |
---|
| 992 | # flux en Ko / ms |
---|
| 993 | flux_Ko_ms=$( echo "scale=6;${entitySizeKo}/${diff_ms}" | bc ) |
---|
| 994 | # echo "flux_Ko_ms=$flux_Ko_ms" |
---|
| 995 | |
---|
| 996 | # flux en Ko / s |
---|
| 997 | flux_Ko_s=$(( $flux_Ko_ms * 1000 )) |
---|
| 998 | # echo "flux_Ko_s=$flux_Ko_s" |
---|
| 999 | |
---|
| 1000 | # flux en Mo / s |
---|
| 1001 | flux_Mo_s=$( echo "scale=6;${flux_Ko_s}/1024" | bc ) |
---|
| 1002 | # echo "flux_Mo_s=$flux_Mo_s" |
---|
| 1003 | |
---|
| 1004 | if [ -d $dest ] ; then |
---|
[1090] | 1005 | dirTo=$( readlink -f ${dest} ) |
---|
[913] | 1006 | else |
---|
[1090] | 1007 | dirTo=$( readlink -f $( dirname ${dest} ) ) |
---|
[913] | 1008 | fi |
---|
| 1009 | |
---|
| 1010 | if [ -d $source ] ; then |
---|
[1090] | 1011 | dirFrom=$( readlink -f ${source} ) |
---|
[913] | 1012 | else |
---|
[1090] | 1013 | dirFrom=$( readlink -f $( dirname ${source} ) ) |
---|
[913] | 1014 | fi |
---|
| 1015 | |
---|
[1094] | 1016 | instrumentationContent=$( echo "\"actionName\":\"${actionType}\",\"size_Mo\":\"${entitySizeMo}\",\"duration_ms\":\"${diff_ms}\",\"throughput_Mo_s\":\"${flux_Mo_s}\",\"dirFrom\":\"${dirFrom}\",\"dirTo\":\"${dirTo}\"" ) |
---|
[1050] | 1017 | |
---|
[1115] | 1018 | if [ X${ActivateStackFilling} = Xtrue ] ; then |
---|
| 1019 | echo "{${instrumentationContent}}" >> ${StackFileLocation}/${StackFileName} |
---|
| 1020 | fi |
---|
[1050] | 1021 | |
---|
| 1022 | # Inform the rabbitMQ queue |
---|
| 1023 | if [ X${ActivateBigBro} = Xtrue ] ; then |
---|
| 1024 | # RabbitMQ message body |
---|
[1065] | 1025 | Body=$( echo "{${genericSimulationID},\"msgCode\":\"7000\",\"msgUID\":\"$(uuidgen)\",${instrumentationContent},\"msgTimestamp\":\"$( date +"%Y-%m-%dT%H:%M:%S.%N%z" )\"}" ) |
---|
[1050] | 1026 | # Fill the rabbitMQ queue |
---|
| 1027 | IGCM_debug_sendAMQP |
---|
| 1028 | fi |
---|
[913] | 1029 | } |
---|
| 1030 | |
---|
| 1031 | #D-#================================================================== |
---|
[2] | 1032 | #D-function IGCM_debug_Check |
---|
| 1033 | #D- * Purpose: Check the present file by comparison with a reference file |
---|
| 1034 | function IGCM_debug_Check |
---|
| 1035 | { |
---|
[544] | 1036 | #--------------------- |
---|
| 1037 | if [ ! -n "${libIGCM}" ] ; then |
---|
| 1038 | echo "Check libIGCM_debug ..........................................[ FAILED ]" |
---|
| 1039 | echo "--Error--> libIGCM variable is not defined" |
---|
| 1040 | exit 2 |
---|
| 1041 | fi |
---|
[2] | 1042 | |
---|
[544] | 1043 | #--------------------- |
---|
| 1044 | if [ ! -n "${Verbosity}" ] ; then |
---|
| 1045 | echo "Check libIGCM_debug ..........................................[ FAILED ]" |
---|
| 1046 | echo "--Error--> Verbosity variable is not defined" |
---|
| 1047 | exit 3 |
---|
| 1048 | fi |
---|
[2] | 1049 | |
---|
[544] | 1050 | #--------------------- |
---|
[1118] | 1051 | # Need to remove timestamps here |
---|
| 1052 | 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 |
---|
[1106] | 1053 | status=$? |
---|
[2] | 1054 | |
---|
[1118] | 1055 | if [ ${status} -eq 0 ] ; then |
---|
[544] | 1056 | echo "Check libIGCM_debug ..............................................[ OK ]" |
---|
| 1057 | else |
---|
| 1058 | echo "Check libIGCM_debug ..........................................[ FAILED ]" |
---|
| 1059 | echo "--Error--> Execution of ${libIGCM}/libIGCM_debug/IGCM_debug_Test.ksh" |
---|
[1578] | 1060 | echo " has produced an error in the checking phase" |
---|
| 1061 | echo " To solve it: verify you didn't made any modifications in libIGCM_debug" |
---|
[1118] | 1062 | 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") |
---|
[544] | 1063 | exit 4 |
---|
| 1064 | fi |
---|
| 1065 | #--------------------- |
---|
[2] | 1066 | } |
---|