[622] | 1 | #!/bin/ksh |
---|
| 2 | |
---|
| 3 | #************************************************************** |
---|
| 4 | # Author: Sebastien Denvil, Martial Mancip, Arnaud Caubel |
---|
| 5 | # Contact: Arnaud.Caubel__at__lsce.ipsl.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 |
---|
[623] | 16 | # if you use the prefix #D- for comments to be extracted. |
---|
[622] | 17 | # Extract with command: cat lib* | grep "^#D-" | cut -c "4-" |
---|
| 18 | #========================================================= |
---|
| 19 | |
---|
| 20 | #D-#================================================== |
---|
[623] | 21 | #D-LibIGCM_sys for Curie |
---|
[622] | 22 | #D-#================================================== |
---|
| 23 | #D- |
---|
[623] | 24 | #D- This ksh library if a layer under some usefull |
---|
[622] | 25 | #D-environment variables and shell commands. |
---|
| 26 | #D-All those definitions depend on host particularities. |
---|
| 27 | #D-It manages a stack mechanism and test validity of operations. |
---|
| 28 | #D-All function described bellow must be prefixed by IGCM_sys. |
---|
| 29 | |
---|
| 30 | #==================================================== |
---|
| 31 | # libIGCM_sys PARAMETERS |
---|
| 32 | #==================================================== |
---|
| 33 | |
---|
| 34 | #==================================================== |
---|
| 35 | # set DEBUG_sys to true to output calls of function |
---|
| 36 | typeset -r DEBUG_sys=${DEBUG_sys:=true} |
---|
| 37 | |
---|
| 38 | #==================================================== |
---|
| 39 | # Turn in dry run mode ? (sys_Put_Rest, sys_Put_Out, sys_Get) |
---|
| 40 | typeset -r DRYRUN=${DRYRUN:=0} |
---|
| 41 | |
---|
| 42 | # YOU MUST COMPILE YOUR EXE FILES FOR DRYRUN MODE ! |
---|
| 43 | # ------------------------------------------------------------------------------------- |
---|
| 44 | # | DRYRUN= | Date computations, | sys_Get | Exe | sys_Put_Out; sys_Put_Rest | |
---|
[809] | 45 | # | | Cp/Exe/param/files | | | | |
---|
| 46 | # | | Chmod Qsub | | | | |
---|
[622] | 47 | # ------------------------------------------------------------------------------------- |
---|
| 48 | # | 0 | yes | yes | yes | yes | |
---|
| 49 | # ------------------------------------------------------------------------------------- |
---|
| 50 | # | 1 | yes | yes | yes | no | |
---|
| 51 | # ------------------------------------------------------------------------------------- |
---|
| 52 | # | 2 | yes | yes | no | no | |
---|
| 53 | # ------------------------------------------------------------------------------------- |
---|
| 54 | # | 3 | yes | no | no | no | |
---|
| 55 | # ------------------------------------------------------------------------------------- |
---|
| 56 | |
---|
| 57 | #===================================================== |
---|
| 58 | # Global Variables : |
---|
| 59 | #===================================================== |
---|
| 60 | # Language : "fr" or "en" |
---|
| 61 | typeset -r MYLANG="fr" |
---|
| 62 | |
---|
| 63 | #===================================================== |
---|
[1201] | 64 | # Host user names project maxCpuTime |
---|
[622] | 65 | # $hostname ou hostname |
---|
[1158] | 66 | typeset HOST=${HOST:=$( hostname )} |
---|
[622] | 67 | # $username ou whoami |
---|
[1158] | 68 | typeset LOGIN=${LOGIN:=$( whoami )} |
---|
[622] | 69 | # $hostname of the MASTER job |
---|
| 70 | typeset MASTER=curie |
---|
[880] | 71 | # add default project on curie |
---|
[1201] | 72 | typeset PROJECT=$(echo ${BRIDGE_MSUB_PROJECT:=gen0826} | cut -d@ -f1 ) |
---|
| 73 | # jobWarningDelay in seconds |
---|
| 74 | typeset jobWarningDelay=${BRIDGE_MSUB_MAXTIME} |
---|
[622] | 75 | |
---|
| 76 | #D- |
---|
| 77 | #D-#================================================== |
---|
| 78 | #D-Program used in libIGCM |
---|
| 79 | #D-#================================================== |
---|
| 80 | |
---|
[742] | 81 | # Submit command |
---|
[747] | 82 | typeset SUBMIT=${SUBMIT:=ccc_msub} |
---|
[622] | 83 | # rsync with path |
---|
| 84 | typeset -r RSYNC=/usr/bin/rsync |
---|
| 85 | # RSYNC_opt args to rsync |
---|
| 86 | typeset -r RSYNC_opt="-va" |
---|
| 87 | # ie storage filesystem |
---|
[623] | 88 | typeset -r STOREHOST=${MASTER} |
---|
[1180] | 89 | typeset -r REMOTE_RSYNC=/usr/bin/rsync |
---|
[622] | 90 | |
---|
| 91 | #==================================================== |
---|
[1180] | 92 | # Set environment tools (ferret, nco, cdo, rebuild, ...) |
---|
[622] | 93 | #==================================================== |
---|
[996] | 94 | if [ X${TaskType} = Xcomputing ] ; then |
---|
[1409] | 95 | IGCM_debug_Print 1 "Module will be loaded once we know the compiler version" |
---|
[996] | 96 | else |
---|
[1413] | 97 | module purge > /dev/null 2>&1 |
---|
[1409] | 98 | . /ccc/cont003/home/dsm/p86ipsl/.env_intel17.0.2_curie > /dev/null 2>&1 |
---|
| 99 | . /ccc/cont003/home/dsm/p86ipsl/.env_netcdf4.3.3.1_curie > /dev/null 2>&1 |
---|
| 100 | . /ccc/cont003/home/dsm/p86ipsl/.env_atlas_curie > /dev/null 2>&1 |
---|
[1358] | 101 | export PCMDI_MP=/ccc/work/cont003/igcmg/igcmg/PCMDI-MP |
---|
[1364] | 102 | export UVCDAT_ANONYMOUS_LOG=FALSE |
---|
[996] | 103 | fi |
---|
[1315] | 104 | |
---|
| 105 | # Use CMIP6 storage space when using CMIP6 cpu ressources |
---|
[1444] | 106 | [[ X"$( echo ${PROJECT} | grep cmip6 )" != "X" && ! $(module list --terse 2>&1 | grep dfldatadir/gencmip6) ]] && module switch dfldatadir dfldatadir/gencmip6 |
---|
[1315] | 107 | |
---|
| 108 | # FYI |
---|
[1210] | 109 | [ ! X${TaskType} = Xchecking ] && IGCM_debug_Print 1 "List of loaded modules:" |
---|
| 110 | [ ! X${TaskType} = Xchecking ] && module list |
---|
[1208] | 111 | |
---|
[1179] | 112 | export PATH=${PATH}:/ccc/cont003/home/dsm/p86ipsl/AddNoise/src_X64_CURIE/bin |
---|
| 113 | export PATH=${PATH}:/ccc/cont003/home/dsm/p86ipsl/AddPerturbation/src_X64_CURIE/bin |
---|
| 114 | export PATH=${PATH}:$( ccc_home -u p86ipsl )/rebuild/src_X64_CURIE/modipsl_v2_2_2_netcdf4.2/bin/ |
---|
[622] | 115 | |
---|
| 116 | #==================================================== |
---|
| 117 | # Specific for ocean additionnal diagnostic |
---|
| 118 | export FER_GO="$FER_GO /home/cont003/p86denv/IGCM_POST_UTIL/JNL /home/cont003/p86denv/GRAF /home/cont003/p86denv/GRAF/GO" |
---|
| 119 | export FER_PALETTE="$FER_PALETTE /home/cont003/p86denv/GRAF/PALET" |
---|
| 120 | |
---|
| 121 | #==================================================== |
---|
| 122 | # Host specific DIRECTORIES |
---|
| 123 | #==================================================== |
---|
| 124 | |
---|
[623] | 125 | # ============ CESIUM START ============ # |
---|
| 126 | |
---|
[622] | 127 | #==================================================== |
---|
[623] | 128 | #- Mirror libIGCM from titane to cesium if needed |
---|
| 129 | #ROOTSYS=$( echo ${libIGCM} | gawk -F"/" '{print $3}' ) |
---|
| 130 | #if [ ! ${ROOTSYS} = "home" ] ; then |
---|
| 131 | # typeset -r MirrorlibIGCM=${MirrorlibIGCM:=true} |
---|
| 132 | #else |
---|
| 133 | # typeset -r MirrorlibIGCM=${MirrorlibIGCM:=false} |
---|
| 134 | #fi |
---|
| 135 | |
---|
| 136 | #==================================================== |
---|
| 137 | #- libIGCM_POST |
---|
| 138 | #if ( ${MirrorlibIGCM} ) ; then |
---|
| 139 | # PATHlibIGCM=$( echo ${libIGCM} | gawk -F"${LOGIN}/" '{print $2}' | sed -e "s&/libIGCM&&" ) |
---|
| 140 | # typeset -r libIGCM_POST=${HOME}/MIRROR/${PATHlibIGCM}/libIGCM |
---|
| 141 | #else |
---|
| 142 | # typeset -r libIGCM_POST=${libIGCM} |
---|
| 143 | #fi |
---|
| 144 | |
---|
| 145 | # ============ CESIUM END ============ # |
---|
| 146 | |
---|
| 147 | #==================================================== |
---|
| 148 | #- MirrorlibIGCM for frontend |
---|
| 149 | typeset -r MirrorlibIGCM=${MirrorlibIGCM:=false} |
---|
| 150 | |
---|
| 151 | #==================================================== |
---|
| 152 | #- libIGCM_POST for frontend |
---|
| 153 | typeset -r libIGCM_POST=${libIGCM} |
---|
| 154 | |
---|
| 155 | #==================================================== |
---|
[622] | 156 | #- R_EXE (==> BIN_DIR = ${MODIPSL}/bin ) |
---|
| 157 | typeset -r R_EXE="${MODIPSL}/bin" |
---|
| 158 | |
---|
| 159 | #==================================================== |
---|
| 160 | #- SUBMIT_DIR : submission dir |
---|
[1010] | 161 | typeset -x SUBMIT_DIR=${SUBMIT_DIR:=${BRIDGE_MSUB_PWD}} |
---|
[622] | 162 | |
---|
| 163 | #==================================================== |
---|
| 164 | #- IN |
---|
[1276] | 165 | typeset -r R_IN=${R_IN:=/ccc/work/cont003/igcmg/igcmg/IGCM} |
---|
[622] | 166 | typeset -r R_IN_ECMWF=${R_IN_ECMWF:=/ccc/work/cont003/dsm/p24data} |
---|
| 167 | |
---|
| 168 | #==================================================== |
---|
| 169 | #- RUN_DIR_PATH : Temporary working directory (=> TMP) |
---|
[850] | 170 | typeset -r RUN_DIR_PATH=${RUN_DIR_PATH:=${SCRATCHDIR}/RUN_DIR/${BRIDGE_MSUB_JOBID}_${$}} |
---|
[622] | 171 | |
---|
| 172 | #==================================================== |
---|
[1179] | 173 | #- OUTCOMMAND_PATH : tmp place to store command lines standard error and outputs |
---|
[1186] | 174 | typeset -r OUTCOMMAND_PATH=/tmp |
---|
[1179] | 175 | |
---|
| 176 | #==================================================== |
---|
[622] | 177 | #- HOST_MPIRUN_COMMAND |
---|
[1382] | 178 | typeset -r HOST_MPIRUN_COMMAND=${HOST_MPIRUN_COMMAND:="/usr/bin/time ccc_mprun -E-K1 "} |
---|
[622] | 179 | |
---|
| 180 | #==================================================== |
---|
| 181 | #- Max number of arguments passed to nco operator or demigration command |
---|
| 182 | UNIX_MAX_LIMIT=360 |
---|
| 183 | |
---|
[623] | 184 | #==================================================== |
---|
[1180] | 185 | #- set PackDefault to true on curie |
---|
[623] | 186 | PackDefault=true |
---|
| 187 | |
---|
[664] | 188 | #==================================================== |
---|
[623] | 189 | #- Default number of MPI task for IPSL coupled model |
---|
| 190 | #- required for backward compatibility |
---|
| 191 | #- |
---|
| 192 | DEFAULT_NUM_PROC_OCE=5 |
---|
| 193 | DEFAULT_NUM_PROC_CPL=1 |
---|
[637] | 194 | (( DEFAULT_NUM_PROC_ATM = BATCH_NUM_PROC_TOT - DEFAULT_NUM_PROC_OCE - DEFAULT_NUM_PROC_CPL )) |
---|
[623] | 195 | |
---|
[1426] | 196 | #==================================================== |
---|
| 197 | #- Number of computing cores per node |
---|
| 198 | #- |
---|
| 199 | NB_CORE_PER_NODE=16 |
---|
| 200 | |
---|
[725] | 201 | #D-#================================================== |
---|
[1057] | 202 | #D-function IGCM_sys_defineArchives |
---|
| 203 | #D-* Purpose: |
---|
| 204 | #D-* Define ARCHIVE : Dedicated to large files |
---|
| 205 | #D-* Define STORAGE : Dedicated to small/medium files |
---|
| 206 | #D-* Define R_OUT : Output tree located on ARCHIVE |
---|
| 207 | #D-* Define R_FIG : Output tree located on STORAGE hosting figures (monitoring and atlas, and/or small files) |
---|
| 208 | #D-* Define R_BUF : Output tree located on SCRATCHDIR hosting files waiting for rebuild or pack processes |
---|
| 209 | #D-* if SpaceName=TEST everything is stored on SCRATCHDIR |
---|
[725] | 210 | #D-* Examples: |
---|
| 211 | #D- |
---|
[1057] | 212 | function IGCM_sys_defineArchives { |
---|
| 213 | IGCM_debug_PushStack "IGCM_sys_defineArchives" |
---|
[676] | 214 | |
---|
[1060] | 215 | if [ ! X${config_UserChoices_ARCHIVE} = X ]; then |
---|
[1057] | 216 | #==================================================== |
---|
| 217 | #- ARCHIVE (dedicated to large files) |
---|
[1063] | 218 | ARCHIVE=${config_UserChoices_ARCHIVE} |
---|
[1057] | 219 | else |
---|
| 220 | #==================================================== |
---|
| 221 | #- ARCHIVE (dedicated to large files) |
---|
[1063] | 222 | ARCHIVE=${CCCSTOREDIR} |
---|
[1057] | 223 | fi |
---|
[725] | 224 | |
---|
[1060] | 225 | if [ ! X${config_UserChoices_STORAGE} = X ]; then |
---|
[1057] | 226 | #==================================================== |
---|
| 227 | #- STORAGE (dedicated to small/medium files) |
---|
[1063] | 228 | STORAGE=${config_UserChoices_STORAGE} |
---|
[1057] | 229 | else |
---|
| 230 | #==================================================== |
---|
| 231 | #- STORAGE (dedicated to small/medium files) |
---|
[1063] | 232 | STORAGE=${CCCWORKDIR} |
---|
[1057] | 233 | fi |
---|
[725] | 234 | |
---|
[1057] | 235 | if [ X${config_UserChoices_SpaceName} = XTEST ]; then |
---|
| 236 | #==================================================== |
---|
| 237 | #- R_OUT |
---|
[1063] | 238 | R_OUT=${SCRATCHDIR}/IGCM_OUT |
---|
[1057] | 239 | |
---|
| 240 | #==================================================== |
---|
| 241 | #- R_FIG (hosting figures : monitoring and atlas, and/or small files) |
---|
[1063] | 242 | R_FIG=${SCRATCHDIR}/IGCM_OUT |
---|
[1057] | 243 | |
---|
| 244 | IGCM_debug_Print 1 "SpaceName=TEST ==> OVERRULE destination path directories" |
---|
| 245 | |
---|
| 246 | else |
---|
| 247 | #==================================================== |
---|
| 248 | #- R_OUT |
---|
[1063] | 249 | R_OUT=${ARCHIVE}/IGCM_OUT |
---|
[1057] | 250 | |
---|
| 251 | #==================================================== |
---|
| 252 | #- R_FIG (hosting figures : monitoring and atlas, and/or small files) |
---|
[1063] | 253 | R_FIG=${STORAGE}/IGCM_OUT |
---|
[1057] | 254 | fi |
---|
| 255 | |
---|
| 256 | #==================================================== |
---|
[1431] | 257 | #- CMIP6 (hosting CMIP6 files produced by XIOS2 and configured by dr2xml) |
---|
| 258 | CMIP6_BUF=${STORAGE}/IGCM_OUT |
---|
| 259 | |
---|
| 260 | #==================================================== |
---|
[1057] | 261 | #- R_BUF (ONLY FOR double copy an scratch) |
---|
[1063] | 262 | R_BUF=${SCRATCHDIR}/IGCM_OUT |
---|
[1057] | 263 | |
---|
| 264 | IGCM_debug_Print 1 "R_OUT has been defined = ${R_OUT}" |
---|
| 265 | IGCM_debug_Print 1 "R_BUF has been defined = ${R_BUF}" |
---|
| 266 | IGCM_debug_Print 1 "R_FIG has been defined = ${R_FIG}" |
---|
| 267 | |
---|
| 268 | IGCM_debug_PopStack "IGCM_sys_defineArchives" |
---|
[725] | 269 | } |
---|
| 270 | |
---|
[622] | 271 | #D-#================================================== |
---|
| 272 | #D-function IGCM_sys_RshArchive |
---|
| 273 | #D-* Purpose: Archive rsh command |
---|
| 274 | #D-* Examples: |
---|
| 275 | #D- |
---|
| 276 | function IGCM_sys_RshArchive { |
---|
[623] | 277 | IGCM_debug_PushStack "IGCM_sys_RshArchive" $@ |
---|
| 278 | /bin/ksh <<-EOF |
---|
[622] | 279 | ${@} |
---|
| 280 | EOF |
---|
[811] | 281 | status=$? |
---|
| 282 | if [ ${status} -gt 0 ] ; then |
---|
| 283 | IGCM_debug_Print 2 "IGCM_sys_RshArchive : command failed error code ${status}" |
---|
[623] | 284 | IGCM_debug_Exit "IGCM_sys_RshArchive" |
---|
| 285 | fi |
---|
| 286 | IGCM_debug_PopStack "IGCM_sys_RshArchive" |
---|
[622] | 287 | } |
---|
| 288 | |
---|
| 289 | #D-#================================================== |
---|
[845] | 290 | #D-function IGCM_sys_RshArchive_NoError |
---|
| 291 | #D-* Purpose: Archive rsh command, without error |
---|
| 292 | #D-* used only in monitoring.job |
---|
| 293 | #D-* Examples: |
---|
| 294 | #D- |
---|
| 295 | function IGCM_sys_RshArchive_NoError { |
---|
| 296 | IGCM_debug_PushStack "IGCM_sys_RshArchive_NoError" $@ |
---|
| 297 | /bin/ksh <<-EOF |
---|
[1181] | 298 | ${@} 2> /dev/null |
---|
[845] | 299 | EOF |
---|
| 300 | IGCM_debug_PopStack "IGCM_sys_RshArchive_NoError" |
---|
| 301 | } |
---|
| 302 | |
---|
| 303 | #D-#================================================== |
---|
[622] | 304 | #D-function IGCM_sys_MkdirArchive |
---|
| 305 | #D-* Purpose: Mkdir on Archive |
---|
| 306 | #D-* Examples: |
---|
| 307 | #D- |
---|
| 308 | function IGCM_sys_MkdirArchive { |
---|
[623] | 309 | IGCM_debug_PushStack "IGCM_sys_MkdirArchive" $@ |
---|
| 310 | if ( $DEBUG_sys ) ; then |
---|
| 311 | echo "IGCM_sys_MkdirArchive :" $@ |
---|
| 312 | fi |
---|
| 313 | #- creation de repertoire sur le serveur fichier |
---|
| 314 | if [ ! -d ${1} ]; then |
---|
| 315 | \mkdir -p $1 |
---|
[811] | 316 | status=$? |
---|
| 317 | |
---|
| 318 | if [ ${status} -gt 0 ] ; then |
---|
| 319 | IGCM_debug_Print 2 "IGCM_sys_MkdirArchive : mkdir failed error code ${status}" |
---|
[623] | 320 | IGCM_debug_Exit "IGCM_sys_MkdirArchive" |
---|
[622] | 321 | fi |
---|
[623] | 322 | fi |
---|
| 323 | IGCM_debug_PopStack "IGCM_sys_MkdirArchive" |
---|
[622] | 324 | } |
---|
| 325 | |
---|
| 326 | #D-#================================================== |
---|
| 327 | #D-function IGCM_sys_TestDirArchive |
---|
| 328 | #D-* Purpose: Test Directory that must exists on Archive |
---|
| 329 | #D-* Examples: |
---|
| 330 | #D- |
---|
| 331 | function IGCM_sys_TestDirArchive { |
---|
[623] | 332 | IGCM_debug_PushStack "IGCM_sys_TestDirArchive" $@ |
---|
| 333 | if ( $DEBUG_sys ) ; then |
---|
| 334 | echo "IGCM_sys_TestDirArchive :" $@ |
---|
| 335 | fi |
---|
| 336 | typeset ExistFlag |
---|
| 337 | ExistFlag=$( [ -d $1 ] && echo 0 || echo 1 ) |
---|
| 338 | IGCM_debug_PopStack "IGCM_sys_TestDirArchive" |
---|
| 339 | return ${ExistFlag} |
---|
[622] | 340 | } |
---|
| 341 | |
---|
| 342 | #D-#================================================== |
---|
[623] | 343 | #D-function IGCM_sys_IsFileArchived |
---|
[1179] | 344 | #D-* Purpose: Test file that must NOT EXISTS on Archive based on filename only |
---|
[623] | 345 | #D-* Examples: |
---|
| 346 | #D- |
---|
| 347 | function IGCM_sys_IsFileArchived { |
---|
| 348 | IGCM_debug_PushStack "IGCM_sys_IsFileArchived" $@ |
---|
| 349 | if ( $DEBUG_sys ) ; then |
---|
| 350 | echo "IGCM_sys_IsFileArchived :" $@ |
---|
| 351 | fi |
---|
| 352 | typeset IsArchivedFlag |
---|
[818] | 353 | IsArchivedFlag=$( [ X$( echo $@ | grep ^\/ccc\/store ) != X ] && echo 0 || echo 1 ) |
---|
[623] | 354 | IGCM_debug_PopStack "IGCM_sys_IsFileArchived" |
---|
| 355 | |
---|
| 356 | return ${IsArchivedFlag} |
---|
| 357 | } |
---|
| 358 | |
---|
| 359 | #D-#================================================== |
---|
[622] | 360 | #D-function IGCM_sys_TestFileArchive |
---|
| 361 | #D-* Purpose: Test file that must NOT EXISTS on Archive |
---|
| 362 | #D-* Examples: |
---|
| 363 | #D- |
---|
| 364 | function IGCM_sys_TestFileArchive { |
---|
[623] | 365 | IGCM_debug_PushStack "IGCM_sys_TestFileArchive" $@ |
---|
| 366 | typeset ExistFlag |
---|
| 367 | ExistFlag=$( [ -f $1 ] && echo 0 || echo 1 ) |
---|
| 368 | IGCM_debug_PopStack "IGCM_sys_TestFileArchive" |
---|
[622] | 369 | |
---|
[623] | 370 | return ${ExistFlag} |
---|
[622] | 371 | } |
---|
| 372 | |
---|
| 373 | #D-#================================================== |
---|
| 374 | #D-function IGCM_sys_CountFileArchive |
---|
| 375 | #D-* Purpose: Count files on Archive filesystem |
---|
| 376 | #D-* Examples: |
---|
| 377 | #D- |
---|
| 378 | function IGCM_sys_CountFileArchive { |
---|
[623] | 379 | IGCM_debug_PushStack "IGCM_sys_CountFileArchive" $@ |
---|
| 380 | ls ${@} 2>/dev/null | wc -l |
---|
| 381 | if [ $? -gt 0 ] ; then |
---|
| 382 | echo "IGCM_sys_CountFileArchive : erreur." |
---|
| 383 | fi |
---|
| 384 | IGCM_debug_PopStack "IGCM_sys_CountFileArchive" |
---|
[622] | 385 | } |
---|
| 386 | |
---|
| 387 | #D-#================================================== |
---|
| 388 | #D-function IGCM_sys_Tree |
---|
| 389 | #D-* Purpose: Tree directories with files on ${ARCHIVE} |
---|
| 390 | #D-* Examples: IGCM_sys_Tree ${R_IN} ${R_OUT} |
---|
| 391 | #D- |
---|
| 392 | function IGCM_sys_Tree { |
---|
[623] | 393 | IGCM_debug_PushStack "IGCM_sys_Tree" $@ |
---|
| 394 | if ( $DEBUG_sys ) ; then |
---|
| 395 | echo "IGCM_sys_Tree :" $@ |
---|
| 396 | fi |
---|
[622] | 397 | |
---|
[623] | 398 | \ls -lR ${@} |
---|
[622] | 399 | |
---|
[623] | 400 | IGCM_debug_PopStack "IGCM_sys_Tree" |
---|
[622] | 401 | } |
---|
| 402 | |
---|
| 403 | #D-#================================================== |
---|
| 404 | #D-function IGCM_sys_Qsub |
---|
| 405 | #D-* Purpose: Qsub new job |
---|
| 406 | #D-* Examples: |
---|
| 407 | #D- |
---|
| 408 | function IGCM_sys_Qsub { |
---|
[623] | 409 | IGCM_debug_PushStack "IGCM_sys_Qsub" $@ |
---|
| 410 | if ( $DEBUG_sys ) ; then |
---|
| 411 | echo "IGCM_sys_Qsub :" $@ |
---|
| 412 | fi |
---|
[811] | 413 | typeset options status |
---|
[923] | 414 | options="-o ${SUBMIT_DIR}/${Script_Output} -e ${SUBMIT_DIR}/${Script_Output}" |
---|
[622] | 415 | |
---|
[1179] | 416 | /usr/bin/ccc_msub ${options} $1 > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[811] | 417 | status=$? |
---|
[948] | 418 | |
---|
[1179] | 419 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[811] | 420 | if [ ${status} -gt 0 ] ; then |
---|
| 421 | IGCM_debug_Print 2 "IGCM_sys_Qsub ${options} $1 : error code ${status}" |
---|
[623] | 422 | IGCM_debug_Exit "IGCM_sys_Qsub" |
---|
[948] | 423 | else |
---|
[1179] | 424 | JobID=$( gawk {'print $4'} ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ ) |
---|
[623] | 425 | fi |
---|
| 426 | IGCM_debug_PopStack "IGCM_sys_Qsub" |
---|
[622] | 427 | } |
---|
| 428 | |
---|
| 429 | #D-#================================================== |
---|
| 430 | #D-function IGCM_sys_QsubPost |
---|
| 431 | #D-* Purpose: Qsub new job on scalaire |
---|
| 432 | #D-* Examples: |
---|
| 433 | #D- |
---|
| 434 | function IGCM_sys_QsubPost { |
---|
[623] | 435 | IGCM_debug_PushStack "IGCM_sys_QsubPost" $@ |
---|
| 436 | if ( $DEBUG_sys ) ; then |
---|
| 437 | echo "IGCM_sys_QsubPost :" $@ |
---|
| 438 | fi |
---|
[923] | 439 | typeset options status |
---|
| 440 | |
---|
[1062] | 441 | # EASIER TO DO THIS INSTEAD OF DUPLICATING libIGCM_sys_curie.ksh |
---|
| 442 | case $( hostname -s ) in |
---|
| 443 | curie*) |
---|
[1201] | 444 | options="-Q normal -A ${PROJECT} -o ${POST_DIR}/${Script_Post_Output}.out -e ${POST_DIR}/${Script_Post_Output}.out";; |
---|
[1062] | 445 | airain*) |
---|
| 446 | options="-q ivybridge -A dsm -o ${POST_DIR}/${Script_Post_Output}.out -e ${POST_DIR}/${Script_Post_Output}.out" |
---|
| 447 | esac |
---|
| 448 | |
---|
[1179] | 449 | /usr/bin/ccc_msub ${options} ${libIGCM_POST}/$1.job > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[811] | 450 | status=$? |
---|
[948] | 451 | |
---|
[1179] | 452 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[811] | 453 | if [ ${status} -gt 0 ] ; then |
---|
[923] | 454 | IGCM_debug_Print 2 "IGCM_sys_QsubPost ${options} ${libIGCM_POST}/$1.job : error code ${status}" |
---|
[623] | 455 | IGCM_debug_Exit "IGCM_sys_QsubPost" |
---|
[948] | 456 | else |
---|
[1179] | 457 | JobID=$( gawk {'print $4'} ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ ) |
---|
[623] | 458 | fi |
---|
| 459 | IGCM_debug_PopStack "IGCM_sys_QsubPost" |
---|
[622] | 460 | } |
---|
| 461 | |
---|
| 462 | #D-************************* |
---|
[623] | 463 | #D- File transfer functions |
---|
[622] | 464 | #D-************************* |
---|
| 465 | #D- |
---|
| 466 | |
---|
| 467 | #D-#================================================== |
---|
| 468 | #D-function IGCM_sys_RmRunDir |
---|
| 469 | #D-* Purpose: rm tmpdir (dummy function most of the time batch |
---|
| 470 | #D- scheduler will do the job) |
---|
| 471 | #D-* Examples: |
---|
| 472 | #D- |
---|
| 473 | function IGCM_sys_RmRunDir { |
---|
[1057] | 474 | IGCM_debug_PushStack "IGCM_sys_RmRunDir" $@ |
---|
[623] | 475 | if ( $DEBUG_sys ) ; then |
---|
| 476 | echo "IGCM_sys_RmRunDir :" $@ |
---|
| 477 | fi |
---|
[622] | 478 | |
---|
[715] | 479 | typeset status |
---|
[622] | 480 | |
---|
[1179] | 481 | echo rm $@ > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 482 | \rm $@ >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[715] | 483 | status=$? |
---|
[622] | 484 | |
---|
[715] | 485 | if [ ${status} -gt 0 ] ; then |
---|
[1183] | 486 | IGCM_debug_Print 1 "IGCM_sys_RmRunDir : rm error code is ${status}." |
---|
[1179] | 487 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[623] | 488 | IGCM_debug_Exit "IGCM_sys_RmRunDir" |
---|
| 489 | else |
---|
[1179] | 490 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[623] | 491 | fi |
---|
| 492 | IGCM_debug_PopStack "IGCM_sys_RmRunDir" |
---|
[622] | 493 | } |
---|
| 494 | |
---|
| 495 | #D-#================================================== |
---|
| 496 | #D-function IGCM_sys_Put_Dir |
---|
| 497 | #D-* Purpose: Copy a complete directory on $(ARCHIVE) |
---|
| 498 | #D-* Examples: |
---|
| 499 | #D- |
---|
| 500 | function IGCM_sys_Put_Dir { |
---|
[623] | 501 | IGCM_debug_PushStack "IGCM_sys_Put_Dir" $@ |
---|
| 502 | if ( $DEBUG_sys ) ; then |
---|
| 503 | echo "IGCM_sys_Put_Dir :" $@ |
---|
| 504 | fi |
---|
| 505 | if [ $DRYRUN = 0 ]; then |
---|
| 506 | if [ ! -d ${1} ] ; then |
---|
| 507 | echo "WARNING : IGCM_sys_Put_Dir ${1} DOES NOT EXIST ." |
---|
| 508 | IGCM_debug_PopStack "IGCM_sys_Put_Dir" |
---|
| 509 | return |
---|
[622] | 510 | fi |
---|
| 511 | |
---|
[715] | 512 | typeset status |
---|
[622] | 513 | |
---|
[623] | 514 | # Only if we use rsync |
---|
| 515 | #IGCM_sys_TestDirArchive $( dirname $2 ) |
---|
| 516 | # |
---|
| 517 | #USUAL WAY |
---|
[1179] | 518 | \cp -r $1 $2 > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[715] | 519 | status=$? |
---|
[622] | 520 | |
---|
[715] | 521 | if [ ${status} -gt 0 ] ; then |
---|
[811] | 522 | IGCM_debug_Print 2 "IGCM_sys_Put_Dir : cp failed error code ${status}" |
---|
[1179] | 523 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[623] | 524 | IGCM_debug_Exit "IGCM_sys_Put_Dir" |
---|
[622] | 525 | else |
---|
[1179] | 526 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[622] | 527 | fi |
---|
[623] | 528 | fi |
---|
| 529 | IGCM_debug_PopStack "IGCM_sys_Put_Dir" |
---|
[622] | 530 | } |
---|
| 531 | |
---|
| 532 | #D-#================================================== |
---|
| 533 | #D-function IGCM_sys_Get_Dir |
---|
[623] | 534 | #D-* Purpose: Copy a complete directory from ${ARCHIVE} |
---|
[622] | 535 | #D-* Examples: |
---|
| 536 | #D- |
---|
| 537 | function IGCM_sys_Get_Dir { |
---|
[623] | 538 | IGCM_debug_PushStack "IGCM_sys_Get_Dir" $@ |
---|
| 539 | if ( $DEBUG_sys ) ; then |
---|
| 540 | echo "IGCM_sys_Get_Dir :" $@ |
---|
| 541 | fi |
---|
| 542 | if [ $DRYRUN = 0 ]; then |
---|
[708] | 543 | typeset NB_ESSAI DELAI status i |
---|
| 544 | # number of tentative |
---|
| 545 | NB_ESSAI=3 |
---|
| 546 | # time delay between tentative |
---|
| 547 | DELAI=2 |
---|
[622] | 548 | |
---|
[676] | 549 | # |
---|
[708] | 550 | # USUAL WAY |
---|
| 551 | # add 'ccc_hsm get' (to demigrate all offline files) to reduce time of this command : |
---|
[713] | 552 | ccc_hsm get -r $1 |
---|
[622] | 553 | |
---|
[709] | 554 | i=0 |
---|
[708] | 555 | while [ $i -lt $NB_ESSAI ] ; do |
---|
[1179] | 556 | \cp -ur $1 $2 >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[708] | 557 | status=$? |
---|
[811] | 558 | if [ ${status} -gt 0 ] ; then |
---|
[713] | 559 | IGCM_debug_Print 2 "IGCM_sys_Get_Dir : cp failed error code ${status} ${i}/${NB_ESSAI}" |
---|
[708] | 560 | IGCM_debug_Print 2 "IGCM_sys_Get_Dir : sleep ${DELAI} seconds and try again." |
---|
| 561 | sleep $DELAI |
---|
| 562 | else |
---|
| 563 | break |
---|
| 564 | fi |
---|
| 565 | (( i = i + 1 )) |
---|
| 566 | done |
---|
| 567 | |
---|
| 568 | if [ ${status} -gt 0 ] ; then |
---|
[1178] | 569 | IGCM_debug_Print 2 "IGCM_sys_Get_Dir : cp failed error code ${status}" |
---|
[1179] | 570 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[623] | 571 | IGCM_debug_Exit "IGCM_sys_Get_Dir" |
---|
[622] | 572 | else |
---|
[1179] | 573 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[622] | 574 | fi |
---|
[623] | 575 | fi |
---|
| 576 | IGCM_debug_PopStack "IGCM_sys_Get_Dir" |
---|
[622] | 577 | } |
---|
| 578 | |
---|
| 579 | #D-#================================================== |
---|
| 580 | #D-function IGCM_sys_Put_Rest |
---|
[623] | 581 | #D-* Purpose: Put computied restarts on ${ARCHIVE}. |
---|
[622] | 582 | #D- File and target directory must exist. |
---|
| 583 | #D-* Examples: |
---|
| 584 | #D- |
---|
| 585 | function IGCM_sys_Put_Rest { |
---|
[623] | 586 | IGCM_debug_PushStack "IGCM_sys_Put_Rest" $@ |
---|
| 587 | if ( $DEBUG_sys ) ; then |
---|
| 588 | echo "IGCM_sys_Put_Rest :" $@ |
---|
| 589 | fi |
---|
| 590 | if [ $DRYRUN = 0 ]; then |
---|
| 591 | if [ ! -f ${1} ] ; then |
---|
| 592 | echo "ERROR : IGCM_sys_Put_Rest ${1} DOES NOT EXIST ." |
---|
| 593 | IGCM_debug_Exit "IGCM_sys_Put_Rest" |
---|
[622] | 594 | fi |
---|
| 595 | |
---|
[715] | 596 | typeset status |
---|
[623] | 597 | # |
---|
| 598 | # USUAL WAY |
---|
[1179] | 599 | \cp $1 $2 > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[715] | 600 | status=$? |
---|
[622] | 601 | |
---|
[623] | 602 | # #RSYNC WITH NETWORK SSH CALL |
---|
[1179] | 603 | # echo ${RSYNC} ${RSYNC_opt} -e ssh ${RUN_DIR}/$1 ${STOREHOST}:${2} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 604 | # ${RSYNC} ${RSYNC_opt} -e ssh ${RUN_DIR}/$1 ${STOREHOST}:${2} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[622] | 605 | |
---|
[623] | 606 | # #RSYNC WITH NFS USE |
---|
[1179] | 607 | # echo ${RSYNC} ${RSYNC_opt} ${RUN_DIR}/$1 ${2} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 608 | # ${RSYNC} ${RSYNC_opt} ${RUN_DIR}/$1 ${2} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[622] | 609 | |
---|
[715] | 610 | # status=$? |
---|
| 611 | # IGCM_sys_Rsync_out $status |
---|
[622] | 612 | |
---|
[1179] | 613 | # ${libIGCM}/libIGCM_sys/IGCM_analyse_rsync_out.awk ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[715] | 614 | # (( status=status+$? )) |
---|
[623] | 615 | |
---|
[715] | 616 | if [ ${status} -gt 0 ] ; then |
---|
[1179] | 617 | IGCM_debug_Print 2 "IGCM_sys_Put_Rest : cp failed error code ${status}" |
---|
[765] | 618 | [ -f ${1} ] && ls -l ${1} |
---|
[759] | 619 | [ -f ${2} ] && ls -l ${2} |
---|
| 620 | [ -f ${2}/${1} ] && ls -l ${2}/${1} |
---|
[1179] | 621 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[623] | 622 | IGCM_debug_Exit "IGCM_sys_Put_Rest" |
---|
[622] | 623 | else |
---|
[759] | 624 | |
---|
| 625 | if [ X${JobType} = XRUN ] ; then |
---|
| 626 | [ -f ${2} ] && IGCM_sys_Chmod 444 ${2} |
---|
| 627 | [ -f ${2}/${1} ] && IGCM_sys_Chmod 444 ${2}/${1} |
---|
| 628 | fi |
---|
| 629 | |
---|
[1179] | 630 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[622] | 631 | fi |
---|
[623] | 632 | fi |
---|
| 633 | IGCM_debug_PopStack "IGCM_sys_Put_Rest" |
---|
[622] | 634 | } |
---|
| 635 | |
---|
| 636 | #D-#================================================== |
---|
| 637 | #D-function IGCM_sys_Put_Out |
---|
[623] | 638 | #D-* Purpose: Copy a file on ${ARCHIVE} after having chmod it in readonly |
---|
[622] | 639 | #D-* Examples: |
---|
| 640 | #D- |
---|
| 641 | function IGCM_sys_Put_Out { |
---|
[623] | 642 | IGCM_debug_PushStack "IGCM_sys_Put_Out" $@ |
---|
| 643 | if ( $DEBUG_sys ) ; then |
---|
| 644 | echo "IGCM_sys_Put_Out :" $@ |
---|
| 645 | fi |
---|
[667] | 646 | |
---|
[711] | 647 | typeset NB_ESSAI DELAI status i exist skip |
---|
[1234] | 648 | typeset fileDeviceNumberInHex directoryDeviceNumberInHex |
---|
[667] | 649 | |
---|
| 650 | # number of tentative |
---|
[668] | 651 | NB_ESSAI=3 |
---|
[667] | 652 | # time delay between tentative |
---|
[668] | 653 | DELAI=2 |
---|
[667] | 654 | |
---|
[623] | 655 | if [ $DRYRUN = 0 ]; then |
---|
| 656 | if [ ! -f ${1} ] ; then |
---|
| 657 | echo "WARNING : IGCM_sys_Put_Out ${1} DOES NOT EXIST ." |
---|
| 658 | IGCM_debug_PopStack "IGCM_sys_Put_Out" |
---|
| 659 | return 1 |
---|
[622] | 660 | fi |
---|
[623] | 661 | # |
---|
| 662 | IGCM_sys_MkdirArchive $( dirname $2 ) |
---|
| 663 | # |
---|
[711] | 664 | exist=false |
---|
| 665 | skip=false |
---|
| 666 | if [ -f $2 ] ; then |
---|
| 667 | IGCM_debug_Print 1 "$2 already exist" |
---|
| 668 | ccc_hsm get $2 |
---|
| 669 | exist=true |
---|
| 670 | if [ "X$( diff $1 $2 )" = X ] ; then |
---|
| 671 | IGCM_debug_Print 2 "$1 and $2 are the same file, we skip the copy" |
---|
| 672 | skip=true |
---|
[667] | 673 | else |
---|
[711] | 674 | IGCM_debug_Print 2 "$1 and $2 are not the same file, we force the copy" |
---|
| 675 | skip=false |
---|
[667] | 676 | fi |
---|
[711] | 677 | fi |
---|
| 678 | # |
---|
| 679 | if ( [ X${exist} = Xtrue ] && [ X${skip} = Xfalse ] ) ; then |
---|
| 680 | IGCM_sys_Chmod u+w $2 |
---|
| 681 | fi |
---|
[622] | 682 | |
---|
[711] | 683 | if [ X${skip} = Xfalse ] ; then |
---|
[719] | 684 | i=0 |
---|
[711] | 685 | while [ $i -lt $NB_ESSAI ] ; do |
---|
[1234] | 686 | # Identify file system |
---|
| 687 | fileDeviceNumberInHex=$( stat -c %d $1 ) |
---|
| 688 | status=$? |
---|
| 689 | if [ ${status} -gt 0 ] ; then |
---|
| 690 | IGCM_debug_Exit "IGCM_sys_Put_Out" |
---|
| 691 | fi |
---|
| 692 | # Identify file system |
---|
| 693 | directoryDeviceNumberInHex=$( stat -c %d $( dirname $2 ) ) |
---|
| 694 | status=$? |
---|
| 695 | if [ ${status} -gt 0 ] ; then |
---|
| 696 | IGCM_debug_Exit "IGCM_sys_Put_Out" |
---|
| 697 | fi |
---|
| 698 | |
---|
[1412] | 699 | if ( [ ${fileDeviceNumberInHex} -ne ${directoryDeviceNumberInHex} ] || [ X$3 = XNOMOVE ] ) ; then |
---|
[1234] | 700 | # They are not on the same device. USUAL WAY |
---|
[1179] | 701 | \cp $1 $2 > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[1037] | 702 | status=$? |
---|
| 703 | else |
---|
[1234] | 704 | # They are on the same device. NOT SO USUAL WAY |
---|
[1179] | 705 | \mv $1 $2 > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[1037] | 706 | status=$? |
---|
| 707 | fi |
---|
[711] | 708 | if [ ${status} -gt 0 ]; then |
---|
[713] | 709 | IGCM_debug_Print 2 "IGCM_sys_Put_Out : cp failed error code ${status} ${i}/${NB_ESSAI}" |
---|
[711] | 710 | IGCM_debug_Print 2 "IGCM_sys_Put_Out : sleep ${DELAI} seconds and try again." |
---|
[765] | 711 | [ -f ${1} ] && ls -l ${1} |
---|
[759] | 712 | [ -f ${2} ] && ls -l ${2} |
---|
| 713 | [ -f ${2}/${1} ] && ls -l ${2}/${1} |
---|
[711] | 714 | sleep $DELAI |
---|
| 715 | else |
---|
| 716 | break |
---|
| 717 | fi |
---|
| 718 | (( i = i + 1 )) |
---|
| 719 | done |
---|
| 720 | fi |
---|
| 721 | |
---|
[623] | 722 | # #RSYNC WITH NETWORK SSH CALL |
---|
[1179] | 723 | # echo ${RSYNC} ${RSYNC_opt} -e ssh ${RUN_DIR}/$1 ${STOREHOST}:${2} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 724 | # ${RSYNC} ${RSYNC_opt} -e ssh ${RUN_DIR}/$1 ${STOREHOST}:${2} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[622] | 725 | |
---|
[623] | 726 | # #RSYNC WITH NFS USE |
---|
[1179] | 727 | # echo ${RSYNC} ${RSYNC_opt} ${RUN_DIR}/$1 ${2} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 728 | # ${RSYNC} ${RSYNC_opt} ${RUN_DIR}/$1 ${2} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[622] | 729 | |
---|
[711] | 730 | # status=$? |
---|
| 731 | # IGCM_sys_Rsync_out $status |
---|
[622] | 732 | |
---|
[1179] | 733 | # ${libIGCM}/libIGCM_sys/IGCM_analyse_rsync_out.awk ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[711] | 734 | # (( status=status+$? )) |
---|
[622] | 735 | |
---|
[711] | 736 | if [ ${status} -gt 0 ] ; then |
---|
[1179] | 737 | IGCM_debug_Print 2 "IGCM_sys_Put_Out : cp failed error code ${status}" |
---|
[765] | 738 | [ -f ${1} ] && ls -l ${1} |
---|
[759] | 739 | [ -f ${2} ] && ls -l ${2} |
---|
| 740 | [ -f ${2}/${1} ] && ls -l ${2}/${1} |
---|
[1179] | 741 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[623] | 742 | IGCM_debug_Exit "IGCM_sys_Put_Out" |
---|
[622] | 743 | else |
---|
[759] | 744 | |
---|
| 745 | if [ X${JobType} = XRUN ] ; then |
---|
| 746 | if [ X${3} = X ] ; then |
---|
| 747 | [ -f ${2} ] && IGCM_sys_Chmod 444 ${2} |
---|
| 748 | [ -f ${2}/${1} ] && IGCM_sys_Chmod 444 ${2}/${1} |
---|
| 749 | fi |
---|
| 750 | fi |
---|
| 751 | |
---|
[1179] | 752 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[622] | 753 | fi |
---|
[623] | 754 | fi |
---|
| 755 | IGCM_debug_PopStack "IGCM_sys_Put_Out" |
---|
| 756 | return 0 |
---|
[622] | 757 | } |
---|
| 758 | |
---|
| 759 | #D-#================================================== |
---|
| 760 | #D-function IGCM_sys_Get |
---|
| 761 | #D-* Purpose: Get a file from ${ARCHIVE} |
---|
| 762 | #D-* Examples: IGCM_sys_Get myfile /destpath/myfile_with_PREFIX |
---|
| 763 | #D- IGCM_sys_Get /l Array_contain_myfiles /destpath/ |
---|
| 764 | function IGCM_sys_Get { |
---|
[623] | 765 | IGCM_debug_PushStack "IGCM_sys_Get" $@ |
---|
[622] | 766 | |
---|
[773] | 767 | typeset DEST dm_liste target file_work |
---|
[667] | 768 | typeset NB_ESSAI DELAI status i |
---|
[622] | 769 | |
---|
[623] | 770 | if ( $DEBUG_sys ) ; then |
---|
| 771 | echo "IGCM_sys_Get :" $@ |
---|
| 772 | fi |
---|
[667] | 773 | |
---|
| 774 | # number of tentative |
---|
[668] | 775 | NB_ESSAI=3 |
---|
[667] | 776 | # time delay between tentative |
---|
[668] | 777 | DELAI=2 |
---|
[667] | 778 | |
---|
[623] | 779 | if [ $DRYRUN -le 2 ]; then |
---|
| 780 | if [ X${1} = X'/l' ] ; then |
---|
| 781 | eval set +A dm_liste \${${2}} |
---|
| 782 | else |
---|
| 783 | eval set +A dm_liste ${1} |
---|
[622] | 784 | fi |
---|
[623] | 785 | eval DEST=\${${#}} |
---|
[1179] | 786 | ccc_hsm get ${dm_liste[*]} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[715] | 787 | status=$? |
---|
| 788 | if [ ${status} -gt 0 ] ; then |
---|
| 789 | echo "WARNING IGCM_sys_Get : error code ${status}" |
---|
[1179] | 790 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[623] | 791 | echo "WARNING IGCM_sys_Get : will stop later if the cp fails." |
---|
| 792 | fi |
---|
[622] | 793 | |
---|
[715] | 794 | #if [ ${status} -gt 0 ] ; then |
---|
[1179] | 795 | # if [ ! "X$( grep "Lost dmusrcmd connection" ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ )" = "X" ] ; then |
---|
| 796 | # cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[785] | 797 | # echo "WARNING IGCM_sys_Get : Lost dmusrcmd connection : " |
---|
| 798 | # sleep 30 |
---|
| 799 | # echo "We try another time" |
---|
[1179] | 800 | ## dmget ${dm_liste[*]} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 801 | # ccc_hsm get ${dm_liste[*]} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[785] | 802 | # status=$? |
---|
| 803 | # if [ ${status} -gt 0 ] ; then |
---|
| 804 | # echo "ERROR IGCM_sys_Get : again demigration error :" |
---|
[1179] | 805 | # cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[785] | 806 | # IGCM_debug_Exit "IGCM_sys_Get" |
---|
[623] | 807 | # fi |
---|
[785] | 808 | # else |
---|
| 809 | # echo "ERROR IGCM_sys_Get : demigration error :" |
---|
[1179] | 810 | # cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[785] | 811 | # IGCM_debug_Exit "IGCM_sys_Get" |
---|
| 812 | # fi |
---|
[623] | 813 | #fi |
---|
[622] | 814 | |
---|
[719] | 815 | # #RSYNC WITH NETWORK SSH CALL |
---|
[1179] | 816 | # echo ${RSYNC} ${RSYNC_opt} -e ssh ${STOREHOST}:"${dm_liste}" ${STOREHOST}:${RUN_DIR}/${DEST} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 817 | # ${RSYNC} ${RSYNC_opt} -e ssh ${STOREHOST}:"${dm_liste}" ${STOREHOST}:${RUN_DIR}/${DEST} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[719] | 818 | |
---|
| 819 | # #RSYNC WITH NFS USE |
---|
[1179] | 820 | # echo ${RSYNC} ${RSYNC_opt} ${dm_liste} ${RUN_DIR}/${DEST} > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 821 | # ${RSYNC} ${RSYNC_opt} ${dm_liste} ${RUN_DIR}/${DEST} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[719] | 822 | |
---|
| 823 | # status=$? |
---|
| 824 | # IGCM_sys_Rsync_out $status |
---|
| 825 | |
---|
[1179] | 826 | # ${libIGCM}/libIGCM_sys/IGCM_analyse_rsync_out.awk ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[719] | 827 | # (( status=status+$? )) |
---|
| 828 | |
---|
[623] | 829 | #USUAL WAY |
---|
| 830 | if [ X${1} = X'/l' ] ; then |
---|
| 831 | for target in ${dm_liste[*]} ; do |
---|
| 832 | local_file=$( basename ${target} ) |
---|
[668] | 833 | # test if the target file is present before the loop |
---|
| 834 | IGCM_sys_TestFileArchive ${target} |
---|
[715] | 835 | status=$? |
---|
| 836 | if [ ${status} -gt 0 ] ; then |
---|
[701] | 837 | echo "IGCM_sys_Get, ERROR : regular file ${target} DOES NOT EXIST ." |
---|
[668] | 838 | IGCM_debug_Exit "IGCM_sys_Get" |
---|
| 839 | else |
---|
[719] | 840 | i=0 |
---|
[668] | 841 | while [ $i -lt $NB_ESSAI ] ; do |
---|
[1037] | 842 | #if [ X${DoLink} = Xtrue ] ; then |
---|
[1179] | 843 | # \ln -s ${target} ${DEST}/${local_file} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[1037] | 844 | # status=$? |
---|
| 845 | # else |
---|
[1179] | 846 | # \cp ${target} ${DEST}/${local_file} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[1037] | 847 | # status=$? |
---|
| 848 | #fi |
---|
[1179] | 849 | \ln -s ${target} ${DEST}/${local_file} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[668] | 850 | status=$? |
---|
| 851 | if [ ${status} -gt 0 ]; then |
---|
[713] | 852 | IGCM_debug_Print 2 "IGCM_sys_Get : cp failed error code ${status} ${i}/${NB_ESSAI}" |
---|
[676] | 853 | IGCM_debug_Print 2 "IGCM_sys_Get : sleep ${DELAI} seconds and try again." |
---|
| 854 | sleep $DELAI |
---|
[668] | 855 | else |
---|
[676] | 856 | break |
---|
[668] | 857 | fi |
---|
| 858 | (( i = i + 1 )) |
---|
| 859 | done |
---|
[719] | 860 | if [ ${status} -gt 0 ] ; then |
---|
| 861 | echo "IGCM_sys_Get : error" |
---|
[1179] | 862 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
| 863 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[719] | 864 | IGCM_debug_Exit "IGCM_sys_Get" |
---|
| 865 | else |
---|
[1179] | 866 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[719] | 867 | fi |
---|
[668] | 868 | fi |
---|
[623] | 869 | done |
---|
| 870 | else |
---|
[719] | 871 | i=0 |
---|
| 872 | while [ $i -lt $NB_ESSAI ] ; do |
---|
[1179] | 873 | \cp ${dm_liste} ${DEST} >> ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
[719] | 874 | status=$? |
---|
| 875 | if [ ${status} -gt 0 ]; then |
---|
| 876 | IGCM_debug_Print 2 "IGCM_sys_Get : cp failed error code ${status} ${i}/${NB_ESSAI}" |
---|
| 877 | IGCM_debug_Print 2 "IGCM_sys_Get : sleep ${DELAI} seconds and try again." |
---|
| 878 | sleep $DELAI |
---|
| 879 | else |
---|
| 880 | break |
---|
| 881 | fi |
---|
| 882 | (( i = i + 1 )) |
---|
| 883 | done |
---|
| 884 | if [ ${status} -gt 0 ] ; then |
---|
| 885 | echo "IGCM_sys_Get : error" |
---|
[1179] | 886 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
| 887 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[719] | 888 | IGCM_debug_Exit "IGCM_sys_Get" |
---|
| 889 | else |
---|
[1179] | 890 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
[719] | 891 | fi |
---|
[623] | 892 | fi |
---|
| 893 | fi |
---|
| 894 | IGCM_debug_PopStack "IGCM_sys_Get" |
---|
| 895 | } |
---|
[622] | 896 | |
---|
[623] | 897 | #D-#================================================== |
---|
[835] | 898 | #D-function IGCM_sys_GetDate_Monitoring |
---|
| 899 | #D-* Purpose: get the last year for which the monitoring has been computed |
---|
| 900 | #D-* Examples: |
---|
| 901 | #D- |
---|
| 902 | function IGCM_sys_GetDate_Monitoring { |
---|
| 903 | IGCM_debug_PushStack "IGCM_sys_GetDate_Monitoring" $@ |
---|
| 904 | if ( $DEBUG_sys ) ; then |
---|
| 905 | echo "IGCM_sys_GetDate_Monitoring :" $@ |
---|
| 906 | fi |
---|
| 907 | |
---|
| 908 | eval ${2}=$( cdo showyear ${1} 2> /dev/null | gawk '{ print $NF }' ) |
---|
| 909 | |
---|
| 910 | IGCM_debug_PopStack "IGCM_sys_GetDate_Monitoring" |
---|
| 911 | } |
---|
| 912 | |
---|
| 913 | #D-#================================================== |
---|
[623] | 914 | #D-function IGCM_sys_Dods_Rm |
---|
| 915 | #D-* Purpose: DO NOTHING ! Put ${ARCHIVE} files on DODS internet protocole. |
---|
| 916 | #D-* Examples: |
---|
| 917 | #D- |
---|
| 918 | function IGCM_sys_Dods_Rm { |
---|
| 919 | if ( $DEBUG_sys ) ; then |
---|
| 920 | echo "IGCM_sys_Dods_Rm :" $@ |
---|
| 921 | fi |
---|
[715] | 922 | typeset status |
---|
[623] | 923 | if [ $DRYRUN = 0 ]; then |
---|
| 924 | |
---|
| 925 | # if [ ! -d /dmnfs/cont003/dods/public/${LOGIN}/${R_DODS}/${1} ] ; then |
---|
| 926 | # echo "WARNING : IGCM_sys_Dods_Rm /dmnfs/cont003/dods/public/${LOGIN}/${R_DODS}/${1} DOES NOT EXIST ." |
---|
| 927 | # echo "Nothing has been done." |
---|
| 928 | # return |
---|
| 929 | # fi |
---|
| 930 | |
---|
[1392] | 931 | if [ "$#" -eq 1 ]; then |
---|
| 932 | /ccc/cont003/home/dsm/p86ipsl/bin/dods_rm public/${LOGIN}/${R_DODS}/${1} # > out_dods_rm 2>&1 |
---|
| 933 | status=$? |
---|
| 934 | else |
---|
| 935 | /ccc/cont003/home/dsm/p86ipsl/bin/dods_rm public/${LOGIN}/${R_DODS} # > out_dods_rm 2>&1 |
---|
| 936 | status=$? |
---|
| 937 | fi |
---|
[623] | 938 | |
---|
[785] | 939 | # if [ ${status} -gt 0 ] ; then |
---|
| 940 | # echo "IGCM_sys_Dods_Rm : error." |
---|
| 941 | # cat out_dods_rm |
---|
| 942 | # IGCM_debug_Exit "IGCM_sys_Dods_Rm" |
---|
| 943 | # else |
---|
| 944 | # rm out_dods_rm |
---|
| 945 | # fi |
---|
[623] | 946 | |
---|
| 947 | fi |
---|
[715] | 948 | return $status |
---|
[623] | 949 | } |
---|
| 950 | |
---|
| 951 | #D-#================================================== |
---|
| 952 | #D-function IGCM_sys_Dods_Cp |
---|
| 953 | #D-* Purpose: Copy $(ARCHIVE) files on DODS internet protocole. |
---|
| 954 | #D-* Examples: |
---|
| 955 | #D- |
---|
| 956 | function IGCM_sys_Dods_Cp { |
---|
| 957 | if ( $DEBUG_sys ) ; then |
---|
| 958 | echo "IGCM_sys_Dods_Cp :" $@ |
---|
| 959 | fi |
---|
[715] | 960 | typeset status |
---|
[623] | 961 | if [ $DRYRUN = 0 ]; then |
---|
| 962 | |
---|
| 963 | # if [ ! -d ${R_SAVE}/${1} ] ; then |
---|
| 964 | # echo "WARNING : IGCM_sys_Dods_Cp ${R_SAVE}/${1} DOES NOT EXIST ." |
---|
| 965 | # echo "Nothing has been done." |
---|
| 966 | # return |
---|
| 967 | # fi |
---|
| 968 | |
---|
[649] | 969 | /ccc/cont003/home/dsm/p86ipsl/bin/dods_cp ${1} public/${LOGIN}/${R_DODS} # > out_dods_cp 2>&1 |
---|
[715] | 970 | status=$? |
---|
[623] | 971 | |
---|
[715] | 972 | # if [ ${status} -gt 0 ] ; then |
---|
[623] | 973 | # echo "IGCM_sys_Dods_Cp : error." |
---|
| 974 | # cat out_dods_cp |
---|
| 975 | # IGCM_debug_Exit "IGCM_sys_Dods_Cp" |
---|
| 976 | # else |
---|
| 977 | # rm out_dods_cp |
---|
| 978 | # fi |
---|
| 979 | |
---|
| 980 | fi |
---|
[715] | 981 | return $status |
---|
[623] | 982 | } |
---|
| 983 | |
---|
| 984 | #D-#================================================== |
---|
| 985 | #D-function IGCM_sys_Put_Dods |
---|
[1178] | 986 | #D-* Purpose: Put ${ARCHIVE} files on DODS internet protocole. |
---|
[623] | 987 | #D-* Examples: |
---|
| 988 | #D- |
---|
| 989 | function IGCM_sys_Put_Dods { |
---|
| 990 | IGCM_debug_PushStack "IGCM_sys_Put_Dods" $@ |
---|
| 991 | if ( $DEBUG_sys ) ; then |
---|
| 992 | echo "IGCM_sys_Put_Dods :" $@ |
---|
| 993 | fi |
---|
[715] | 994 | typeset status |
---|
[623] | 995 | if [ $DRYRUN = 0 ]; then |
---|
| 996 | if ( [ ! -d ${R_SAVE}/${1} ] && [ ! -d ${R_FIGR}/${1} ] ) ; then |
---|
| 997 | echo "WARNING IGCM_sys_Put_Dods : None of the following directories exist. Exactly one should." |
---|
| 998 | echo "WARNING IGCM_sys_Put_Dods : ${R_SAVE}/${1} DOES NOT EXIST." |
---|
| 999 | echo "WARNING IGCM_sys_Put_Dods : ${R_FIGR}/${1} DOES NOT EXIST." |
---|
| 1000 | IGCM_debug_PopStack "IGCM_sys_Put_Dods" |
---|
| 1001 | return |
---|
[622] | 1002 | fi |
---|
[623] | 1003 | |
---|
| 1004 | if ( [ -d ${R_SAVE}/${1} ] && [ -d ${R_FIGR}/${1} ] ) ; then |
---|
| 1005 | echo "WARNING IGCM_sys_Put_Dods : Both of the following directories exist. Exactly one should." |
---|
| 1006 | echo "WARNING IGCM_sys_Put_Dods : ${R_SAVE}/${1} EXISTS." |
---|
| 1007 | echo "WARNING IGCM_sys_Put_Dods : ${R_FIGR}/${1} EXISTS." |
---|
| 1008 | IGCM_debug_PopStack "IGCM_sys_Put_Dods" |
---|
| 1009 | return |
---|
| 1010 | fi |
---|
| 1011 | # |
---|
| 1012 | if [ -d ${R_SAVE}/${1} ] ; then |
---|
| 1013 | cd ${R_SAVE} |
---|
| 1014 | elif [ -d ${R_FIGR}/${1} ] ; then |
---|
| 1015 | cd ${R_FIGR} |
---|
| 1016 | fi |
---|
| 1017 | |
---|
| 1018 | IGCM_sys_Dods_Rm ${1} |
---|
| 1019 | IGCM_sys_Dods_Cp ${1} |
---|
[715] | 1020 | status=0 |
---|
[623] | 1021 | |
---|
[715] | 1022 | if [ ${status} -gt 0 ] ; then |
---|
[623] | 1023 | echo "IGCM_sys_Put_Dods : error." |
---|
| 1024 | IGCM_debug_Exit "IGCM_sys_Put_Dods" |
---|
| 1025 | fi |
---|
| 1026 | fi |
---|
| 1027 | IGCM_debug_PopStack "IGCM_sys_Put_Dods" |
---|
[622] | 1028 | } |
---|
| 1029 | |
---|
| 1030 | ############################################################## |
---|
| 1031 | # REBUILD OPERATOR |
---|
| 1032 | |
---|
[811] | 1033 | #D-#================================================== |
---|
[1163] | 1034 | #D-function IGCM_sys_sync |
---|
| 1035 | #D-* Purpose: flush buffer on disk |
---|
| 1036 | #D-* Examples: |
---|
| 1037 | #D- |
---|
| 1038 | function IGCM_sys_sync { |
---|
| 1039 | IGCM_debug_PushStack "IGCM_sys_sync" $@ |
---|
| 1040 | if ( $DEBUG_sys ) ; then |
---|
| 1041 | echo "IGCM_sys_sync :" $@ |
---|
| 1042 | fi |
---|
| 1043 | |
---|
| 1044 | /bin/sync |
---|
| 1045 | |
---|
| 1046 | IGCM_debug_PopStack "IGCM_sys_sync" |
---|
| 1047 | } |
---|
| 1048 | |
---|
[622] | 1049 | ############################################################ |
---|
| 1050 | # Activate Running Environnment Variables |
---|
| 1051 | |
---|
[811] | 1052 | #D-#================================================== |
---|
[1239] | 1053 | #D-function IGCM_sys_activ_variables |
---|
[811] | 1054 | #D-* Purpose: set environement variables prior to execution |
---|
| 1055 | #D-* Examples: |
---|
| 1056 | #D- |
---|
[622] | 1057 | function IGCM_sys_activ_variables { |
---|
[623] | 1058 | IGCM_debug_PushStack "IGCM_sys_activ_variables" |
---|
| 1059 | if ( $DEBUG_sys ) ; then |
---|
| 1060 | echo "IGCM_sys_activ_variables" |
---|
| 1061 | fi |
---|
[622] | 1062 | |
---|
[811] | 1063 | # -------------------------------------------------------------------- |
---|
[1409] | 1064 | #D- MODULE specifications |
---|
| 1065 | # -------------------------------------------------------------------- |
---|
[1413] | 1066 | module purge > /dev/null 2>&1 |
---|
[1410] | 1067 | if [ X${compilerVersion} = XV17 ] ; then |
---|
[1409] | 1068 | . /ccc/cont003/home/dsm/p86ipsl/.env_intel17.0.2_curie > /dev/null 2>&1 |
---|
[1410] | 1069 | elif [ X${compilerVersion} = XV14 ] ; then |
---|
[1409] | 1070 | . /ccc/cont003/home/dsm/p86ipsl/.env_intel14.0.3_curie > /dev/null 2>&1 |
---|
| 1071 | else |
---|
| 1072 | IGCM_debug_Exit "This compiler version is not yet supported : ${compilerVersion}" |
---|
| 1073 | fi |
---|
| 1074 | . /ccc/cont003/home/dsm/p86ipsl/.env_netcdf4.3.3.1_curie > /dev/null 2>&1 |
---|
| 1075 | |
---|
| 1076 | # Keep track of this |
---|
| 1077 | [ ! X${TaskType} = Xchecking ] && IGCM_debug_Print 1 "List of loaded modules:" |
---|
| 1078 | [ ! X${TaskType} = Xchecking ] && module list |
---|
| 1079 | |
---|
| 1080 | # -------------------------------------------------------------------- |
---|
[811] | 1081 | #D- MPI specifications |
---|
| 1082 | # -------------------------------------------------------------------- |
---|
[1191] | 1083 | |
---|
[811] | 1084 | # -------------------------------------------------------------------- |
---|
| 1085 | #D- Other specifications |
---|
| 1086 | # -------------------------------------------------------------------- |
---|
| 1087 | |
---|
[623] | 1088 | ulimit -s unlimited |
---|
[622] | 1089 | |
---|
[623] | 1090 | IGCM_debug_PopStack "IGCM_sys_activ_variables" |
---|
[622] | 1091 | } |
---|
| 1092 | |
---|
| 1093 | ############################################################ |
---|
| 1094 | # Desactivate Running Environnment Variables |
---|
| 1095 | |
---|
[811] | 1096 | #D-#================================================== |
---|
| 1097 | #D-function IGCM_sys_desactiv_variables |
---|
| 1098 | #D-* Purpose: unset environement variables after execution |
---|
| 1099 | #D-* Examples: |
---|
| 1100 | #D- |
---|
[622] | 1101 | function IGCM_sys_desactiv_variables { |
---|
[623] | 1102 | IGCM_debug_PushStack "IGCM_sys_desactiv_variables" |
---|
| 1103 | if ( $DEBUG_sys ) ; then |
---|
| 1104 | echo "IGCM_sys_desactiv_variables" |
---|
| 1105 | fi |
---|
[811] | 1106 | # -------------------------------------------------------------------- |
---|
| 1107 | #D- MPI specifications |
---|
| 1108 | # -------------------------------------------------------------------- |
---|
| 1109 | |
---|
| 1110 | # -------------------------------------------------------------------- |
---|
| 1111 | #D- Other specifications |
---|
| 1112 | # -------------------------------------------------------------------- |
---|
| 1113 | |
---|
[623] | 1114 | IGCM_debug_PopStack "IGCM_sys_desactiv_variables" |
---|
[622] | 1115 | } |
---|
| 1116 | |
---|
[623] | 1117 | ############################################################ |
---|
[1238] | 1118 | # Update job headers to be used by the scheduler |
---|
| 1119 | |
---|
| 1120 | #D-#================================================== |
---|
| 1121 | #D-function IGCM_sys_updateHeaders |
---|
| 1122 | #D-* Purpose: Update job headers to be used by the scheduler |
---|
| 1123 | #D-* Examples: IGCM_sys_updateHeaders /path/to/Job_MYEXP |
---|
| 1124 | #D- |
---|
| 1125 | function IGCM_sys_updateHeaders { |
---|
| 1126 | IGCM_debug_PushStack "IGCM_sys_updateHeaders" |
---|
| 1127 | if ( $DEBUG_sys ) ; then |
---|
| 1128 | echo "IGCM_sys_updateHeaders" |
---|
| 1129 | fi |
---|
| 1130 | typeset file |
---|
| 1131 | file=$1 |
---|
| 1132 | |
---|
| 1133 | if [ ${executionType} -eq 1 ] ; then |
---|
| 1134 | # MPMD + MPI |
---|
| 1135 | sed -e "/::openMPthreads::/d" \ |
---|
| 1136 | -e "s/::JobNumProcTot::/${coreNumber}/" \ |
---|
[1256] | 1137 | -e "/#MSUB -x/d" \ |
---|
| 1138 | -e "/--cpu_bind=none/d" \ |
---|
[1238] | 1139 | ${file} > ${file}.tmp |
---|
| 1140 | |
---|
| 1141 | elif [ ${executionType} -eq 2 ] ; then |
---|
[1390] | 1142 | # MPMD + MPI + OMP : mpirun/ccc_mprun/error |
---|
| 1143 | if ( [ "X${config_UserChoices_ExecutionModeOnCurie}" = "X" ] || [ "X${config_UserChoices_ExecutionModeOnCurie}" = "Xmpirun" ] ) ; then |
---|
| 1144 | sed -e "/::openMPthreads::/d" \ |
---|
| 1145 | -e "s/::JobNumProcTot::/${coreNumber}/" \ |
---|
| 1146 | ${file} > ${file}.tmp |
---|
| 1147 | elif ( [ "X${config_UserChoices_ExecutionModeOnCurie}" = "Xccc_mprun" ] ) ; then |
---|
| 1148 | sed -e "/::openMPthreads::/d" \ |
---|
| 1149 | -e "s/::JobNumProcTot::/${coreNumber}/" \ |
---|
| 1150 | -e "/--cpu_bind=none/d" \ |
---|
| 1151 | ${file} > ${file}.tmp |
---|
| 1152 | else |
---|
| 1153 | IGCM_debug_Print 1 "You have to set ExecutionModeOnCurie=ccc_mprun or mpirun in config.card" |
---|
| 1154 | IGCM_debug_Exit "IGCM_sys_updateHeaders" |
---|
| 1155 | fi |
---|
[1238] | 1156 | elif [ ${executionType} -eq 3 ] ; then |
---|
| 1157 | # SPMD + MPI/OMP |
---|
| 1158 | sed -e "s/::openMPthreads::/${openMPthreads}/" \ |
---|
| 1159 | -e "s/::JobNumProcTot::/${mpiTasks}/" \ |
---|
[1240] | 1160 | -e "/#MSUB -x/d" \ |
---|
[1238] | 1161 | -e "/--cpu_bind=none/d" \ |
---|
| 1162 | ${file} > ${file}.tmp |
---|
| 1163 | |
---|
| 1164 | elif [ ${executionType} -eq 4 ] ; then |
---|
| 1165 | # SPMD + MPI only |
---|
| 1166 | sed -e "s/::JobNumProcTot::/${mpiTasks}/" \ |
---|
| 1167 | -e "/::openMPthreads::/d" \ |
---|
[1240] | 1168 | -e "/#MSUB -x/d" \ |
---|
[1238] | 1169 | -e "/--cpu_bind=none/d" \ |
---|
| 1170 | ${file} > ${file}.tmp |
---|
| 1171 | |
---|
| 1172 | elif [ ${executionType} -eq 5 ] ; then |
---|
| 1173 | # SPMD + OMP only |
---|
| 1174 | sed -e "s/::openMPthreads::/${openMPthreads}/" \ |
---|
| 1175 | -e "/::JobNumProcTot::/d" \ |
---|
[1240] | 1176 | -e "/#MSUB -x/d" \ |
---|
[1238] | 1177 | -e "/--cpu_bind=none/d" \ |
---|
| 1178 | ${file} > ${file}.tmp |
---|
| 1179 | |
---|
| 1180 | elif [ ${executionType} -eq 6 ] ; then |
---|
| 1181 | # SEQUENTIAL THEN |
---|
| 1182 | sed -e "s/::JobNumProcTot::/1/" \ |
---|
| 1183 | -e "/::openMPthreads::/d" \ |
---|
[1240] | 1184 | -e "/#MSUB -x/d" \ |
---|
[1238] | 1185 | -e "/--cpu_bind=none/d" \ |
---|
| 1186 | ${file} > ${file}.tmp |
---|
| 1187 | |
---|
| 1188 | fi |
---|
| 1189 | |
---|
| 1190 | IGCM_sys_Mv ${file}.tmp ${file} |
---|
| 1191 | |
---|
| 1192 | IGCM_debug_PopStack "IGCM_sys_updateHeaders" |
---|
| 1193 | } |
---|
| 1194 | |
---|
| 1195 | ############################################################ |
---|
[623] | 1196 | # Build MPI/OMP scripts run file (dummy function) |
---|
[622] | 1197 | |
---|
[811] | 1198 | #D-#================================================== |
---|
| 1199 | #D-function IGCM_sys_build_run_file |
---|
| 1200 | #D-* Purpose: build run file (deprecated) |
---|
| 1201 | #D-* Examples: |
---|
| 1202 | #D- |
---|
[622] | 1203 | function IGCM_sys_build_run_file { |
---|
[811] | 1204 | IGCM_debug_Print 3 " dummy function : IGCM_sys_build_run_file " |
---|
[622] | 1205 | } |
---|
| 1206 | |
---|
| 1207 | ############################################################ |
---|
| 1208 | # Build MPI/OMP scripts |
---|
[811] | 1209 | |
---|
| 1210 | #D-#================================================== |
---|
| 1211 | #D-function IGCM_sys_build_execution_scripts |
---|
| 1212 | #D-* Purpose: build execution scripts to be launch by ${HOST_MPIRUN_COMMAND} |
---|
| 1213 | #D-* Examples: |
---|
| 1214 | #D- |
---|
[623] | 1215 | function IGCM_sys_build_execution_scripts |
---|
[622] | 1216 | { |
---|
[623] | 1217 | IGCM_debug_PushStack "IGCM_sys_build_execution_scripts" $@ |
---|
| 1218 | if ( $DEBUG_sys ) ; then |
---|
| 1219 | echo "IGCM_sys_build_execution_scripts " $@ |
---|
| 1220 | fi |
---|
[622] | 1221 | |
---|
[623] | 1222 | EXECUTION=${HOST_MPIRUN_COMMAND} |
---|
[622] | 1223 | |
---|
[623] | 1224 | if ( ${OK_PARA_MPMD} ) ; then |
---|
[622] | 1225 | |
---|
[1390] | 1226 | # MPMD mode |
---|
| 1227 | # 1 MPI only |
---|
| 1228 | # 2 MPI/OpenMP mpirun method |
---|
| 1229 | # 3 MPI/OpenMP ccc_mprun method |
---|
| 1230 | |
---|
[1381] | 1231 | if [ -f run_file ] ; then |
---|
| 1232 | IGCM_sys_Rm -f run_file |
---|
| 1233 | fi |
---|
| 1234 | touch run_file |
---|
| 1235 | |
---|
[1390] | 1236 | # case 1 : Only MPI (MPMD) |
---|
[1078] | 1237 | if ( ! ${OK_PARA_OMP} ) ; then |
---|
[622] | 1238 | |
---|
[1078] | 1239 | # Build run_file |
---|
[622] | 1240 | |
---|
[1078] | 1241 | # First loop on the components for the coupler ie oasis (only if oasis3) |
---|
| 1242 | # the coupler ie oasis3 must be the first one |
---|
| 1243 | for comp in ${config_ListOfComponents[*]} ; do |
---|
[622] | 1244 | |
---|
[1078] | 1245 | eval ExeNameIn=\${config_Executable_${comp}[0]} |
---|
| 1246 | eval ExeNameOut=\${config_Executable_${comp}[1]} |
---|
[622] | 1247 | |
---|
[1078] | 1248 | # for CPL component only |
---|
| 1249 | if [ "X${comp}" = "XCPL" ] && [ "X${ExeNameOut}" != X\"\" ] ; then |
---|
[1000] | 1250 | |
---|
[1078] | 1251 | eval comp_proc_mpi_loc=\${${comp}_PROC_MPI} |
---|
| 1252 | eval comp_proc_omp_loc=\${${comp}_PROC_OMP} |
---|
| 1253 | echo "${comp_proc_mpi_loc} ./${ExeNameOut}" >> run_file |
---|
| 1254 | fi |
---|
| 1255 | done |
---|
[622] | 1256 | |
---|
[1078] | 1257 | # Then second loop on the components |
---|
| 1258 | for comp in ${config_ListOfComponents[*]} ; do |
---|
[622] | 1259 | |
---|
[1078] | 1260 | eval ExeNameIn=\${config_Executable_${comp}[0]} |
---|
| 1261 | eval ExeNameOut=\${config_Executable_${comp}[1]} |
---|
[622] | 1262 | |
---|
[1078] | 1263 | # Only if we really have an executable for the component and not the coupler ie oasis: |
---|
| 1264 | if ( [ "X${ExeNameOut}" != X\"\" ] && [ "X${comp}" != "XCPL" ] ) ; then |
---|
[622] | 1265 | |
---|
[1078] | 1266 | eval comp_proc_mpi_loc=\${${comp}_PROC_MPI} |
---|
| 1267 | eval comp_proc_omp_loc=\${${comp}_PROC_OMP} |
---|
| 1268 | echo "${comp_proc_mpi_loc} ./${ExeNameOut}" >> run_file |
---|
| 1269 | fi |
---|
| 1270 | done |
---|
[622] | 1271 | |
---|
[1390] | 1272 | EXECUTION="${HOST_MPIRUN_COMMAND} -f ./run_file" |
---|
| 1273 | |
---|
| 1274 | IGCM_sys_Chmod u+x run_file |
---|
| 1275 | if ( $DEBUG_sys ) ; then |
---|
| 1276 | echo "run_file contains : " |
---|
| 1277 | cat run_file |
---|
| 1278 | fi |
---|
| 1279 | |
---|
| 1280 | elif ( [ "X${config_UserChoices_ExecutionModeOnCurie}" = "X" ] || [ "X${config_UserChoices_ExecutionModeOnCurie}" = "Xmpirun" ] ) ; then |
---|
| 1281 | |
---|
| 1282 | # 2 MPI/OpenMP mpirun method |
---|
| 1283 | |
---|
| 1284 | # Use of mpirun instead of ccc_mprun |
---|
| 1285 | EXECUTION="time mpirun" |
---|
| 1286 | |
---|
| 1287 | # Hosts treatment |
---|
| 1288 | ${EXECUTION} hostname | sort | uniq > hosts.tmp |
---|
| 1289 | |
---|
| 1290 | i=0 |
---|
| 1291 | rm -f hosts rankfile |
---|
| 1292 | IGCM_debug_Print 1 "sys Curie, Hosts available :" |
---|
| 1293 | for nodes in `cat hosts.tmp` |
---|
| 1294 | do |
---|
| 1295 | host[$i]=$nodes |
---|
| 1296 | echo "${host[$i]}" >> hosts |
---|
| 1297 | IGCM_debug_Print 1 ${host[$i]} |
---|
| 1298 | i=$((i+1)) |
---|
| 1299 | done |
---|
| 1300 | rm -f hosts.tmp |
---|
| 1301 | |
---|
| 1302 | listnodes=${host[*]} |
---|
| 1303 | |
---|
| 1304 | EXECUTION="${EXECUTION} -hostfile hosts -rankfile rankfile" |
---|
| 1305 | |
---|
| 1306 | # Initialisation |
---|
| 1307 | rank=0 |
---|
| 1308 | current_core=0 |
---|
| 1309 | init_exec=n |
---|
| 1310 | |
---|
| 1311 | # Loop on the components |
---|
| 1312 | for comp in ${config_ListOfComponents[*]} ; do |
---|
| 1313 | |
---|
| 1314 | eval ExeNameIn=\${config_Executable_${comp}[0]} |
---|
| 1315 | eval ExeNameOut=\${config_Executable_${comp}[1]} |
---|
| 1316 | |
---|
| 1317 | # Not possible if oasis has an executable (i.e old version of oasis3) |
---|
| 1318 | if ( [ "X${ExeNameOut}" != X\"\" ] && [ "X${comp}" = "XCPL" ] ) ; then |
---|
| 1319 | IGCM_debug_Exit "ERROR MPMD with hybrid MPI-OpenMP is not available with oasis3 version" |
---|
| 1320 | IGCM_debug_Print 2 "Only available with oasis3-MCT version coupler" |
---|
| 1321 | IGCM_debug_Verif_Exit |
---|
| 1322 | fi |
---|
| 1323 | |
---|
| 1324 | # Only if we really have an executable for the component : |
---|
| 1325 | if [ "X${ExeNameOut}" != X\"\" ] ; then |
---|
| 1326 | |
---|
| 1327 | eval comp_proc_mpi_loc=\${${comp}_PROC_MPI} |
---|
| 1328 | eval comp_proc_omp_loc=\${${comp}_PROC_OMP} |
---|
[1426] | 1329 | eval comp_proc_nod_loc=\${${comp}_PROC_NOD} |
---|
[1390] | 1330 | |
---|
| 1331 | echo "#!/bin/ksh" > script_${ExeNameOut}.ksh |
---|
| 1332 | echo "" >> script_${ExeNameOut}.ksh |
---|
| 1333 | if [ ${comp_proc_omp_loc} -gt 1 ] ; then |
---|
| 1334 | |
---|
| 1335 | # Check if the number of threads is correct |
---|
| 1336 | case ${comp_proc_omp_loc} in |
---|
| 1337 | 2|4|8|16) |
---|
| 1338 | IGCM_debug_Print 1 "You run ${ExeNameOut} on ${comp_proc_omp_loc} OMP threads" |
---|
| 1339 | ;; |
---|
| 1340 | *) |
---|
| 1341 | IGCM_debug_Exit "ERROR with OMP parameters !" |
---|
| 1342 | IGCM_debug_Print 2 "${comp_proc_omp_loc} is not possible as number of OMP threads" |
---|
| 1343 | IGCM_debug_Print 2 "Only 2,4,8,16 as number of OMP threads are possible " |
---|
| 1344 | IGCM_debug_Verif_Exit |
---|
| 1345 | ;; |
---|
| 1346 | esac |
---|
| 1347 | echo "export KMP_STACKSIZE=3g" >> script_${ExeNameOut}.ksh |
---|
| 1348 | echo "export KMP_LIBRARY=turnaround" >> script_${ExeNameOut}.ksh |
---|
| 1349 | echo "export MKL_SERIAL=YES" >> script_${ExeNameOut}.ksh |
---|
| 1350 | echo "OMP_NUM_THREADS=${comp_proc_omp_loc}" >> script_${ExeNameOut}.ksh |
---|
| 1351 | fi |
---|
| 1352 | |
---|
| 1353 | #echo "./${ExeNameOut}" >> script_${ExeNameOut}.ksh |
---|
| 1354 | echo "(( MYMPIRANK = OMPI_COMM_WORLD_RANK )) " >> script_${ExeNameOut}.ksh |
---|
| 1355 | echo "MYMPIRANK=\$(printf '%3.3d\n' \${MYMPIRANK})" >> script_${ExeNameOut}.ksh |
---|
| 1356 | echo "./${ExeNameOut} > out_${ExeNameOut}.out.\${MYMPIRANK} 2>out_${ExeNameOut}.err.\${MYMPIRANK}" >> script_${ExeNameOut}.ksh |
---|
| 1357 | IGCM_sys_Chmod u+x script_${ExeNameOut}.ksh |
---|
| 1358 | |
---|
| 1359 | if [ ${init_exec} = y ] ; then |
---|
| 1360 | EXECUTION="${EXECUTION} : -np ${comp_proc_mpi_loc} ./script_${ExeNameOut}.ksh" |
---|
| 1361 | else |
---|
| 1362 | EXECUTION="${EXECUTION} -np ${comp_proc_mpi_loc} ./script_${ExeNameOut}.ksh" |
---|
| 1363 | init_exec=y |
---|
| 1364 | fi |
---|
| 1365 | |
---|
| 1366 | # Build rankfile : method used to assign cores and nodes for the MPI process |
---|
| 1367 | # Ex : |
---|
| 1368 | #rank 0=curie5296 slot=0,1,2,3 |
---|
| 1369 | #rank 1=curie5296 slot=4,5,6,7 |
---|
| 1370 | # Example of final command : |
---|
| 1371 | # mpirun -hostfile hosts -rankfile rankfile -np 27 ./script_lmdz.x.ksh : -np 5 ./script_opa.xx.ksh |
---|
| 1372 | # with script_lmdz.x.ksh : |
---|
| 1373 | # #!/bin/ksh |
---|
| 1374 | #export KMP_STACKSIZE=3g |
---|
| 1375 | #export KMP_LIBRARY=turnaround |
---|
| 1376 | #export MKL_SERIAL=YES |
---|
| 1377 | #OMP_NUM_THREADS=4 |
---|
| 1378 | #./lmdz.x |
---|
[1426] | 1379 | if [ ${comp_proc_nod_loc} -gt 1 ] ; then |
---|
[1430] | 1380 | (( offset_comp_proc_loc = NB_CORE_PER_NODE / (comp_proc_mpi_loc / comp_proc_nod_loc) )) |
---|
[1426] | 1381 | else |
---|
| 1382 | (( offset_comp_proc_loc = comp_proc_omp_loc )) |
---|
| 1383 | fi |
---|
[1390] | 1384 | for nb_proc_mpi in `seq 0 $(($comp_proc_mpi_loc-1))`; do |
---|
[1426] | 1385 | (( index_host = current_core / NB_CORE_PER_NODE )) |
---|
| 1386 | host_value=${host[${index_host}]} |
---|
| 1387 | (( slot = current_core % NB_CORE_PER_NODE )) |
---|
| 1388 | virg="," |
---|
| 1389 | string_final="" |
---|
| 1390 | for index in `seq $slot $(($slot+$comp_proc_omp_loc-1))`; do |
---|
| 1391 | string=$index$virg |
---|
| 1392 | string_final=$string_final$string |
---|
| 1393 | done |
---|
| 1394 | string_final=$( echo $string_final | sed "s/.$//" ) |
---|
| 1395 | echo "rank $rank=$host_value slot=$string_final" >> rankfile |
---|
| 1396 | (( rank = rank + 1 )) |
---|
| 1397 | (( current_core = current_core + offset_comp_proc_loc )) |
---|
[1390] | 1398 | done |
---|
| 1399 | fi |
---|
| 1400 | done |
---|
| 1401 | |
---|
| 1402 | elif ( [ "X${config_UserChoices_ExecutionModeOnCurie}" = "Xccc_mprun" ] ) ; then |
---|
| 1403 | |
---|
| 1404 | # 3 MPI/OpenMP ccc_mprun method |
---|
| 1405 | |
---|
[1381] | 1406 | # MPI-OpenMP (MPMD) |
---|
| 1407 | # example of run_file |
---|
| 1408 | # 71-8 env OMP_NUM_THREADS=8 ./script_lmdz.x.ksh |
---|
| 1409 | # 480-1 env OMP_NUM_THREADS=1 ./script_opa.xx.ksh |
---|
| 1410 | # 1-1 env OMP_NUM_THREADS=1 ./script_xios.x.ksh |
---|
[622] | 1411 | |
---|
[1078] | 1412 | # Loop on the components |
---|
| 1413 | for comp in ${config_ListOfComponents[*]} ; do |
---|
[1000] | 1414 | |
---|
[1078] | 1415 | eval ExeNameIn=\${config_Executable_${comp}[0]} |
---|
| 1416 | eval ExeNameOut=\${config_Executable_${comp}[1]} |
---|
[1000] | 1417 | |
---|
[1078] | 1418 | # Not possible if oasis has an executable (i.e old version of oasis3) |
---|
| 1419 | if ( [ "X${ExeNameOut}" != X\"\" ] && [ "X${comp}" = "XCPL" ] ) ; then |
---|
| 1420 | IGCM_debug_Exit "ERROR MPMD with hybrid MPI-OpenMP is not available with oasis3 version" |
---|
| 1421 | IGCM_debug_Print 2 "Only available with oasis3-MCT version coupler" |
---|
| 1422 | IGCM_debug_Verif_Exit |
---|
| 1423 | fi |
---|
[1000] | 1424 | |
---|
[1078] | 1425 | # Only if we really have an executable for the component : |
---|
| 1426 | if [ "X${ExeNameOut}" != X\"\" ] ; then |
---|
[1000] | 1427 | |
---|
[1078] | 1428 | eval comp_proc_mpi_loc=\${${comp}_PROC_MPI} |
---|
| 1429 | eval comp_proc_omp_loc=\${${comp}_PROC_OMP} |
---|
[1000] | 1430 | |
---|
[1381] | 1431 | # Build script files |
---|
| 1432 | |
---|
[1078] | 1433 | echo "#!/bin/ksh" > script_${ExeNameOut}.ksh |
---|
| 1434 | echo "" >> script_${ExeNameOut}.ksh |
---|
| 1435 | if [ ${comp_proc_omp_loc} -gt 1 ] ; then |
---|
| 1436 | |
---|
| 1437 | # Check if the number of threads is correct |
---|
| 1438 | case ${comp_proc_omp_loc} in |
---|
| 1439 | 2|4|8|16) |
---|
| 1440 | IGCM_debug_Print 1 "You run ${ExeNameOut} on ${comp_proc_omp_loc} OMP threads" |
---|
| 1441 | ;; |
---|
| 1442 | *) |
---|
| 1443 | IGCM_debug_Exit "ERROR with OMP parameters !" |
---|
| 1444 | IGCM_debug_Print 2 "${comp_proc_omp_loc} is not possible as number of OMP threads" |
---|
| 1445 | IGCM_debug_Print 2 "Only 2,4,8,16 as number of OMP threads are possible " |
---|
| 1446 | IGCM_debug_Verif_Exit |
---|
| 1447 | ;; |
---|
| 1448 | esac |
---|
[1381] | 1449 | ### only for LMDZ? |
---|
[1078] | 1450 | echo "export KMP_STACKSIZE=3g" >> script_${ExeNameOut}.ksh |
---|
| 1451 | echo "export KMP_LIBRARY=turnaround" >> script_${ExeNameOut}.ksh |
---|
| 1452 | echo "export MKL_SERIAL=YES" >> script_${ExeNameOut}.ksh |
---|
| 1453 | echo "OMP_NUM_THREADS=${comp_proc_omp_loc}" >> script_${ExeNameOut}.ksh |
---|
[1381] | 1454 | |
---|
[1078] | 1455 | fi |
---|
[1318] | 1456 | |
---|
[1381] | 1457 | # to have out/err per process on different files |
---|
| 1458 | echo "./${ExeNameOut} > out_${ExeNameOut}.out.\${SLURM_PROCID} 2>out_${ExeNameOut}.err.\${SLURM_PROCID}" >> script_${ExeNameOut}.ksh |
---|
[1317] | 1459 | #echo "./${ExeNameOut}" >> script_${ExeNameOut}.ksh |
---|
[1078] | 1460 | |
---|
[1381] | 1461 | IGCM_sys_Chmod u+x script_${ExeNameOut}.ksh |
---|
[1078] | 1462 | |
---|
[1381] | 1463 | # Complete run_file |
---|
[1000] | 1464 | |
---|
[1381] | 1465 | echo "${comp_proc_mpi_loc}-${comp_proc_omp_loc} env OMP_NUM_THREADS=${comp_proc_omp_loc} ./script_${ExeNameOut}.ksh " >>run_file |
---|
| 1466 | |
---|
[1078] | 1467 | fi |
---|
| 1468 | done |
---|
[1381] | 1469 | |
---|
[1385] | 1470 | ## variable added to stop after 60s instead of 600s by default. |
---|
| 1471 | ## This is used when no error comes from executables and when something stopped an executable without notice. |
---|
| 1472 | export SLURM_WAIT=60 |
---|
[1381] | 1473 | |
---|
[1385] | 1474 | ## module advised by TGCC (instead of 2 variables) |
---|
| 1475 | module load feature/bridge/heterogenous_mpmd |
---|
[1381] | 1476 | |
---|
[1390] | 1477 | # set EXECUTION for ccc_mprun case (similar to #1 : MPI only MPMD method) |
---|
[1000] | 1478 | |
---|
[1390] | 1479 | EXECUTION="${HOST_MPIRUN_COMMAND} -f ./run_file" |
---|
[1381] | 1480 | |
---|
[1390] | 1481 | IGCM_sys_Chmod u+x run_file |
---|
| 1482 | if ( $DEBUG_sys ) ; then |
---|
| 1483 | echo "run_file contains : " |
---|
| 1484 | cat run_file |
---|
| 1485 | fi |
---|
[1381] | 1486 | |
---|
[1390] | 1487 | else |
---|
[1381] | 1488 | |
---|
[1390] | 1489 | IGCM_debug_Print 1 "sys Curie : choose mpirun or ccc_mprun in config.card for ExecutionModeOnCurie" |
---|
| 1490 | |
---|
| 1491 | fi # 1 2 or 3 if ${OK_PARA_MPMD} |
---|
| 1492 | |
---|
| 1493 | else |
---|
[1078] | 1494 | # Only one executable (SPMD mode). |
---|
[1000] | 1495 | |
---|
[622] | 1496 | for comp in ${config_ListOfComponents[*]} ; do |
---|
| 1497 | |
---|
[685] | 1498 | # Only if we really have an executable for the component : |
---|
[623] | 1499 | eval ExeNameOut=\${config_Executable_${comp}[1]} |
---|
[685] | 1500 | if ( [ "X${ExeNameOut}" != X\"\" ] && [ "X${ExeNameOut}" != "Xinca.dat" ] ) ; then |
---|
[622] | 1501 | |
---|
[1381] | 1502 | # Build script files |
---|
| 1503 | |
---|
[685] | 1504 | echo "#!/bin/ksh" > script_${ExeNameOut}.ksh |
---|
| 1505 | echo "" >> script_${ExeNameOut}.ksh |
---|
| 1506 | IGCM_sys_Chmod u+x script_${ExeNameOut}.ksh |
---|
| 1507 | |
---|
| 1508 | if ( ${OK_PARA_OMP} ) ; then |
---|
| 1509 | eval comp_proc_omp_loc=\${${comp}_PROC_OMP} |
---|
| 1510 | echo "" >> script_${ExeNameOut}.ksh |
---|
| 1511 | echo "export KMP_STACKSIZE=3g" >> script_${ExeNameOut}.ksh |
---|
| 1512 | echo "export KMP_LIBRARY=turnaround" >> script_${ExeNameOut}.ksh |
---|
| 1513 | echo "export MKL_SERIAL=YES" >> script_${ExeNameOut}.ksh |
---|
| 1514 | echo "OMP_NUM_THREADS=${comp_proc_omp_loc}" >> script_${ExeNameOut}.ksh |
---|
| 1515 | fi |
---|
| 1516 | |
---|
| 1517 | if ( ${OK_PARA_MPI} ) ; then |
---|
[623] | 1518 | eval comp_proc_mpi_loc=\${${comp}_PROC_MPI} |
---|
[685] | 1519 | # Default : ccc_mprun used if nb_proc gt 1 |
---|
| 1520 | # to have out/err per process on different files |
---|
[1317] | 1521 | echo "./${ExeNameOut} > out_${ExeNameOut}.out.\${SLURM_PROCID} 2>out_${ExeNameOut}.err.\${SLURM_PROCID}" >> script_${ExeNameOut}.ksh |
---|
| 1522 | #echo "./${ExeNameOut}" >> script_${ExeNameOut}.ksh |
---|
[685] | 1523 | EXECUTION="${HOST_MPIRUN_COMMAND} -n ${comp_proc_mpi_loc} ./script_${ExeNameOut}.ksh" |
---|
[623] | 1524 | else |
---|
[685] | 1525 | # Default : ccc_mprun is NOT used if nb_proc eq 1 |
---|
| 1526 | # to have out/err per process on different files |
---|
[1317] | 1527 | echo "./${ExeNameOut} > out_${ExeNameOut}.out 2>out_${ExeNameOut}.err" >> script_${ExeNameOut}.ksh |
---|
| 1528 | #echo "./${ExeNameOut}" >> script_${ExeNameOut}.ksh |
---|
[685] | 1529 | EXECUTION="/usr/bin/time ./script_${ExeNameOut}.ksh" |
---|
[622] | 1530 | fi |
---|
[685] | 1531 | |
---|
| 1532 | IGCM_debug_Print 1 "sys Curie : script_${ExeNameOut}.ksh contains" |
---|
| 1533 | cat script_${ExeNameOut}.ksh |
---|
| 1534 | |
---|
[623] | 1535 | fi |
---|
[622] | 1536 | |
---|
| 1537 | done |
---|
| 1538 | |
---|
[1390] | 1539 | fi # ${OK_PARA_MPMD} |
---|
[622] | 1540 | |
---|
[623] | 1541 | IGCM_debug_Print 1 "sys Curie : execution command is " |
---|
| 1542 | IGCM_debug_Print 1 "$EXECUTION" |
---|
[622] | 1543 | |
---|
[623] | 1544 | IGCM_debug_PopStack "IGCM_sys_build_execution_scripts" |
---|
[622] | 1545 | } |
---|
| 1546 | |
---|
[933] | 1547 | #D-#================================================== |
---|
| 1548 | #D-function IGCM_sys_check_path |
---|
| 1549 | #D-* Purpose: check that RUN_DIR_PATH that will be removed on some machine |
---|
| 1550 | #D-* do not point to an important use directory. Stop immediately in that case. |
---|
| 1551 | #D-* Examples: |
---|
| 1552 | #D- |
---|
| 1553 | function IGCM_sys_check_path { |
---|
| 1554 | IGCM_debug_PushStack "IGCM_sys_check_path" |
---|
| 1555 | if ( $DEBUG_sys ) ; then |
---|
| 1556 | echo "IGCM_sys_check_path" |
---|
| 1557 | fi |
---|
[811] | 1558 | |
---|
[933] | 1559 | if ( [ X${RUN_DIR_PATH} = X${HOME} ] || [ X${RUN_DIR_PATH} = X${WORKDIR} ] || [ X${RUN_DIR_PATH} = X${SCRATCHDIR} ] || [ X${RUN_DIR_PATH} = X${CCCWORKDIR} ] || [ X${RUN_DIR_PATH} = X${CCCSTOREDIR} ] ) ; then |
---|
| 1560 | IGCM_debug_Print 1 "Variable RUN_DIR_PATH is pointing to an important directory : ${RUN_DIR_PATH}" |
---|
| 1561 | IGCM_debug_Print 1 "Please check the RUN_DIR_PATH definition in your Job : Job_${config_UserChoices_JobName}" |
---|
[936] | 1562 | IGCM_debug_Exit "This will stop the job" |
---|
[933] | 1563 | fi |
---|
| 1564 | IGCM_debug_PopStack "IGCM_sys_check_path" |
---|
| 1565 | } |
---|
| 1566 | |
---|
[811] | 1567 | #D-#================================================== |
---|
| 1568 | #D-function IGCM_sys_check_quota |
---|
| 1569 | #D-* Purpose: check user quota. Stop the simulation if quota above 90% |
---|
| 1570 | #D-* Examples: |
---|
| 1571 | #D- |
---|
[714] | 1572 | function IGCM_sys_check_quota { |
---|
[776] | 1573 | IGCM_debug_PushStack "IGCM_sys_check_quota" |
---|
| 1574 | if ( $DEBUG_sys ) ; then |
---|
| 1575 | echo "IGCM_sys_check_quota" |
---|
| 1576 | fi |
---|
| 1577 | # Limit of quota (in %) |
---|
| 1578 | limit_quota=90 |
---|
| 1579 | |
---|
| 1580 | # Check of the volume |
---|
[912] | 1581 | volume_quota=$(ccc_quota | grep ' scratch' | gawk '{print $2}') |
---|
| 1582 | volume_avail=$(ccc_quota | grep ' scratch' | gawk '{print $3}') |
---|
[776] | 1583 | |
---|
| 1584 | if ( [ ! X${volume_quota} = X ] && [ ! ${volume_quota} = "-" ] ) ; then |
---|
| 1585 | |
---|
| 1586 | unit_avail=${volume_avail: -1} |
---|
| 1587 | unit_quota=${volume_quota: -1} |
---|
| 1588 | |
---|
| 1589 | if [ "${unit_quota}" = "*" ] ; then |
---|
[811] | 1590 | IGCM_debug_Print 1 "Please, check your quota of volume on scratch" |
---|
| 1591 | IGCM_debug_Print 1 "More than 100% of your quota is used" |
---|
| 1592 | IGCM_debug_Print 1 "Use the ccc_quota command to check" |
---|
| 1593 | IGCM_debug_Print 1 "You must have more than 10% available to run" |
---|
| 1594 | IGCM_debug_Exit "Not enough space to run ! STOP HERE" |
---|
| 1595 | IGCM_debug_Verif_Exit |
---|
[714] | 1596 | fi |
---|
[776] | 1597 | |
---|
| 1598 | temp_avail=${volume_avail%%${volume_avail: -1}*} |
---|
| 1599 | temp_quota=${volume_quota%%${volume_quota: -1}*} |
---|
| 1600 | |
---|
| 1601 | if [ ! ${unit_avail} = ${unit_quota} ] ; then |
---|
| 1602 | |
---|
| 1603 | # Convertion |
---|
| 1604 | if [ ${volume_avail: -1} = "T" ] ; then |
---|
| 1605 | (( temp_avail = temp_avail * 1000000000000 )) |
---|
| 1606 | elif [ ${volume_avail: -1} = "G" ] ; then |
---|
| 1607 | (( temp_avail = temp_avail * 1000000000 )) |
---|
| 1608 | elif [ ${volume_avail: -1} = "M" ] ; then |
---|
| 1609 | (( temp_avail = temp_avail * 1000000 )) |
---|
| 1610 | elif [ ${volume_avail: -1} = "k" ] ; then |
---|
| 1611 | (( temp_avail = temp_avail * 1000 )) |
---|
| 1612 | else |
---|
| 1613 | (( temp_avail = volume_avail )) |
---|
| 1614 | fi |
---|
| 1615 | if [ ${volume_quota: -1} = "T" ] ; then |
---|
| 1616 | (( temp_quota = temp_quota * 1000000000000 )) |
---|
| 1617 | elif [ ${volume_quota: -1} = "G" ] ; then |
---|
| 1618 | (( temp_quota = temp_quota * 1000000000 )) |
---|
| 1619 | elif [ ${volume_quota: -1} = "M" ] ; then |
---|
| 1620 | (( temp_quota = temp_quota * 1000000 )) |
---|
| 1621 | elif [ ${volume_quota: -1} = "k" ] ; then |
---|
| 1622 | (( temp_quota = temp_quota * 1000 )) |
---|
| 1623 | else |
---|
| 1624 | (( temp_quota = volume_quota )) |
---|
| 1625 | fi |
---|
| 1626 | fi |
---|
| 1627 | |
---|
| 1628 | quota_volume=$(echo "scale=2 ; $temp_quota/$temp_avail*100" | bc) |
---|
| 1629 | # echo "volume ratio is " $quota_volume |
---|
| 1630 | |
---|
| 1631 | if [ ${quota_volume} -ge ${limit_quota} ] ; then |
---|
| 1632 | IGCM_debug_Print 1 "Please, check your quota of volume on scratch" |
---|
| 1633 | IGCM_debug_Print 1 "${quota_volume}% of your quota is used" |
---|
| 1634 | IGCM_debug_Print 1 "Use the ccc_quota command to check" |
---|
| 1635 | IGCM_debug_Print 1 "You must have more than 10% available to run" |
---|
| 1636 | IGCM_debug_Exit "Not enough space to run ! STOP HERE" |
---|
| 1637 | IGCM_debug_Verif_Exit |
---|
| 1638 | fi |
---|
| 1639 | |
---|
| 1640 | fi |
---|
| 1641 | |
---|
| 1642 | # Check of the number of inodes |
---|
| 1643 | |
---|
[912] | 1644 | inode_quota=$(ccc_quota | grep ' scratch' | gawk '{print $6}') |
---|
| 1645 | inode_avail=$(ccc_quota | grep ' scratch' | gawk '{print $7}') |
---|
[776] | 1646 | |
---|
| 1647 | if ( [ ! X${inode_quota} = X ] && [ ! ${inode_quota} = "-" ] ) ; then |
---|
| 1648 | |
---|
| 1649 | unit_avail=${inode_avail: -1} |
---|
| 1650 | unit_quota=${inode_quota: -1} |
---|
| 1651 | |
---|
| 1652 | if [ "${unit_quota}" = "*" ] ; then |
---|
[811] | 1653 | IGCM_debug_Print 1 "Please, check your quota of inode on scratch" |
---|
| 1654 | IGCM_debug_Print 1 "More than 100% of your quota is used" |
---|
| 1655 | IGCM_debug_Print 1 "Use the ccc_quota command to check" |
---|
| 1656 | IGCM_debug_Print 1 "You must have more than 10% available to run" |
---|
| 1657 | IGCM_debug_Exit "Not enough space to run ! STOP HERE" |
---|
| 1658 | IGCM_debug_Verif_Exit |
---|
[776] | 1659 | fi |
---|
| 1660 | |
---|
| 1661 | temp_avail=${inode_avail%%${inode_avail: -1}*} |
---|
| 1662 | temp_quota=${inode_quota%%${inode_quota: -1}*} |
---|
| 1663 | |
---|
| 1664 | if [ ! ${unit_avail} = ${unit_quota} ] ; then |
---|
| 1665 | |
---|
| 1666 | # Convertion |
---|
| 1667 | if [ ${inode_avail: -1} = "T" ] ; then |
---|
| 1668 | (( temp_avail = temp_avail * 1000000000000 )) |
---|
| 1669 | elif [ ${inode_avail: -1} = "G" ] ; then |
---|
| 1670 | (( temp_avail = temp_avail * 1000000000 )) |
---|
| 1671 | elif [ ${inode_avail: -1} = "M" ] ; then |
---|
| 1672 | (( temp_avail = temp_avail * 1000000 )) |
---|
| 1673 | elif [ ${inode_avail: -1} = "k" ] ; then |
---|
| 1674 | (( temp_avail = temp_avail * 1000 )) |
---|
| 1675 | else |
---|
| 1676 | (( temp_avail = inode_avail )) |
---|
| 1677 | fi |
---|
| 1678 | |
---|
| 1679 | if [ ${inode_quota: -1} = "T" ] ; then |
---|
| 1680 | (( temp_quota = temp_quota * 1000000000000 )) |
---|
| 1681 | elif [ ${inode_quota: -1} = "G" ] ; then |
---|
| 1682 | (( temp_quota = temp_quota * 1000000000 )) |
---|
| 1683 | elif [ ${inode_quota: -1} = "M" ] ; then |
---|
| 1684 | (( temp_quota = temp_quota * 1000000 )) |
---|
| 1685 | elif [ ${inode_quota: -1} = "k" ] ; then |
---|
| 1686 | (( temp_quota = temp_quota * 1000 )) |
---|
| 1687 | else |
---|
| 1688 | (( temp_quota = inode_quota )) |
---|
| 1689 | fi |
---|
| 1690 | fi |
---|
| 1691 | quota_inode=$(echo "scale=2 ; $temp_quota/$temp_avail*100" | bc) |
---|
| 1692 | # echo "inode ratio is " $quota_inode |
---|
| 1693 | |
---|
| 1694 | if [ ${quota_inode} -ge ${limit_quota} ] ; then |
---|
| 1695 | IGCM_debug_Print 1 "Please, check your quota of inode on scratch" |
---|
| 1696 | IGCM_debug_Print 1 "${quota_inode}% of your quota is used" |
---|
| 1697 | IGCM_debug_Print 1 "Use the ccc_quota command to check" |
---|
| 1698 | IGCM_debug_Print 1 "You must have more than 10% available to run" |
---|
| 1699 | IGCM_debug_Exit "Not enough space to run ! STOP HERE" |
---|
| 1700 | IGCM_debug_Verif_Exit |
---|
| 1701 | fi |
---|
| 1702 | fi |
---|
| 1703 | IGCM_debug_PopStack "IGCM_sys_check_quota" |
---|
[714] | 1704 | } |
---|
| 1705 | |
---|
[773] | 1706 | #D-#================================================== |
---|
[1286] | 1707 | #D-function IGCM_sys_projectAccounting |
---|
| 1708 | #D-* Purpose: store project accounting information in a file |
---|
| 1709 | #D-* Examples: |
---|
| 1710 | #D- |
---|
| 1711 | function IGCM_sys_projectAccounting { |
---|
| 1712 | IGCM_debug_PushStack "IGCM_sys_projectAccounting" |
---|
| 1713 | if ( $DEBUG_sys ) ; then |
---|
| 1714 | echo "IGCM_sys_check_quota" |
---|
| 1715 | fi |
---|
| 1716 | |
---|
[1288] | 1717 | ssh curie91 /usr/bin/ccc_myproject > $1 |
---|
[1286] | 1718 | |
---|
| 1719 | IGCM_debug_PopStack "IGCM_sys_projectAccounting" |
---|
| 1720 | } |
---|
| 1721 | |
---|
| 1722 | #D-#================================================== |
---|
[1244] | 1723 | #D-function IGCM_sys_getJobSchedulerID |
---|
| 1724 | #D-* Purpose: Get the job ID during execution |
---|
| 1725 | #D-* Examples: IGCM_sys_getJobSchedulerID jobSchedulerID |
---|
| 1726 | #D- |
---|
| 1727 | function IGCM_sys_getJobSchedulerID { |
---|
| 1728 | IGCM_debug_PushStack "IGCM_sys_getJobSchedulerID" |
---|
| 1729 | if ( $DEBUG_sys ) ; then |
---|
| 1730 | echo "IGCM_sys_getJobSchedulerID" |
---|
| 1731 | fi |
---|
| 1732 | |
---|
| 1733 | eval ${1}=${BRIDGE_MSUB_JOBID} |
---|
| 1734 | |
---|
| 1735 | IGCM_debug_PopStack "IGCM_sys_getJobSchedulerID" |
---|
| 1736 | } |
---|
| 1737 | |
---|
| 1738 | #D-#================================================== |
---|
[958] | 1739 | #D-function IGCM_sys_GetJobID |
---|
[1244] | 1740 | #D-* Purpose: Get the job ID from the JobName |
---|
[958] | 1741 | #D-* Examples: IGCM_sys_GetJobID ${JobName} ${TargetUsr} JobID |
---|
| 1742 | #D- |
---|
| 1743 | function IGCM_sys_GetJobID { |
---|
| 1744 | IGCM_debug_PushStack "IGCM_sys_GetJobID" |
---|
| 1745 | if ( $DEBUG_sys ) ; then |
---|
| 1746 | echo "IGCM_sys_GetJobID" |
---|
| 1747 | fi |
---|
| 1748 | |
---|
| 1749 | # With -f option, the full job name is given in the last column |
---|
| 1750 | ID=$( ccc_mstat -f -u $2 | \ |
---|
| 1751 | gawk -v JobName=$1 '( $NF ~ JobName ) { print $1 }' ) |
---|
| 1752 | |
---|
| 1753 | eval ${3}=${ID} |
---|
| 1754 | IGCM_debug_PopStack "IGCM_sys_GetJobID" |
---|
| 1755 | } |
---|
| 1756 | |
---|
| 1757 | #D-#================================================== |
---|
[773] | 1758 | #D-function IGCM_sys_CountJobInQueue |
---|
[1244] | 1759 | #D-* Purpose: Count number of users job |
---|
[773] | 1760 | #D-* Examples: IGCM_sys_CountJobInQueue ${JobName} NbRun |
---|
| 1761 | #D- |
---|
| 1762 | function IGCM_sys_CountJobInQueue { |
---|
| 1763 | IGCM_debug_PushStack "IGCM_sys_CountJobInQueue" |
---|
| 1764 | if ( $DEBUG_sys ) ; then |
---|
| 1765 | echo "IGCM_sys_CountJobInQueue" |
---|
| 1766 | fi |
---|
| 1767 | |
---|
| 1768 | # With -f option, the full job name is given in the last column |
---|
[834] | 1769 | NbRun=$( ccc_mstat -f | gawk -v JobName=$1 'BEGIN { x=0 } ( $NF ~ JobName ) { x=x+1 } END { print x }' ) |
---|
[773] | 1770 | |
---|
| 1771 | eval ${2}=${NbRun} |
---|
| 1772 | |
---|
| 1773 | IGCM_debug_PopStack "IGCM_sys_CountJobInQueue" |
---|
| 1774 | } |
---|
| 1775 | |
---|
[955] | 1776 | #D-#================================================== |
---|
| 1777 | #D-function IGCM_sys_ListJobInQueue |
---|
[1244] | 1778 | #D-* Purpose: Produce a list of users computing jobs (excluding post-processing) |
---|
[955] | 1779 | #D-* Examples: IGCM_sys_ListJobInQueue ${User} JobNameList |
---|
| 1780 | #D- |
---|
| 1781 | function IGCM_sys_ListJobInQueue { |
---|
| 1782 | IGCM_debug_PushStack "IGCM_sys_ListJobInQueue" |
---|
| 1783 | if ( $DEBUG_sys ) ; then |
---|
| 1784 | echo "IGCM_sys_ListJobInQueue" |
---|
| 1785 | fi |
---|
| 1786 | |
---|
| 1787 | # With -f option, the full job name is given in the last column |
---|
| 1788 | set -A JobList $( ccc_mstat -f | gawk -v User=$1 \ |
---|
| 1789 | '( $2 == User && \ |
---|
| 1790 | $NF != /TS/ && \ |
---|
| 1791 | $NF !~ /PACK/ && \ |
---|
| 1792 | $NF !~ /REBUILD/ && \ |
---|
| 1793 | $NF !~ /pack/ ) \ |
---|
| 1794 | { print $NF }' | sed -e "s/\(.*\)\.[0-9]*/\1/" ) |
---|
| 1795 | |
---|
| 1796 | eval set -A ${2} ${JobList[*]} |
---|
| 1797 | |
---|
| 1798 | IGCM_debug_PopStack "IGCM_sys_ListJobInQueue" |
---|
| 1799 | } |
---|
[1272] | 1800 | |
---|
| 1801 | #D-#================================================== |
---|
| 1802 | #D-function IGCM_sys_atlas |
---|
| 1803 | #D-* Purpose: encapsulate atlas call so as to manage error code and curie specificity |
---|
| 1804 | #D-* Examples: |
---|
| 1805 | #D- |
---|
| 1806 | function IGCM_sys_atlas { |
---|
| 1807 | IGCM_debug_PushStack "IGCM_sys_atlas" $@ |
---|
| 1808 | if ( $DEBUG_sys ) ; then |
---|
| 1809 | echo "IGCM_sys_atlas :" $@ |
---|
| 1810 | fi |
---|
| 1811 | |
---|
| 1812 | typeset status |
---|
| 1813 | |
---|
| 1814 | \ccc_mprun atlas $@ > ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ 2>&1 |
---|
| 1815 | status=$? |
---|
| 1816 | if [ ${status} -gt 0 ] ; then |
---|
| 1817 | echo "IGCM_sys_atlas : error code ${status}" |
---|
| 1818 | cat ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
| 1819 | \rm ${OUTCOMMAND_PATH}/out_command_${LOGIN}.$$ |
---|
| 1820 | IGCM_debug_PopStack "IGCM_sys_atlas" |
---|
| 1821 | return 1 |
---|
| 1822 | else |
---|
| 1823 | IGCM_debug_PopStack "IGCM_sys_atlas" |
---|
| 1824 | return 0 |
---|
| 1825 | fi |
---|
| 1826 | |
---|
| 1827 | IGCM_debug_PopStack "IGCM_sys_atlas" |
---|
| 1828 | } |
---|
| 1829 | |
---|
[1380] | 1830 | #D-#================================================== |
---|
| 1831 | #D-function IGCM_sys_rebuild_nemo |
---|
| 1832 | #D-* Purpose: rebuild nemo parallel files with nemo specific rebuild on curie |
---|
| 1833 | #D-* Examples: |
---|
| 1834 | #D- |
---|
| 1835 | |
---|
| 1836 | function IGCM_sys_rebuild_nemo { |
---|
| 1837 | IGCM_debug_PushStack "IGCM_sys_rebuild_nemo" $@ |
---|
| 1838 | if ( $DEBUG_sys ) ; then |
---|
| 1839 | echo "IGCM_sys_rebuild_nemo :" $@ |
---|
| 1840 | fi |
---|
| 1841 | |
---|
| 1842 | $( ccc_home -u p86ipsl )/REBUILD_NEMO/rebuild_nemo ${1} ${2} |
---|
| 1843 | |
---|
| 1844 | IGCM_debug_PopStack "IGCM_sys_rebuild_nemo" |
---|
| 1845 | } |
---|
| 1846 | |
---|