[1841] | 1 | #!/bin/bash |
---|
[1046] | 2 | |
---|
[1059] | 3 | # mrun - script for running PALM jobs |
---|
| 4 | |
---|
[1046] | 5 | #--------------------------------------------------------------------------------# |
---|
| 6 | # This file is part of PALM. |
---|
| 7 | # |
---|
| 8 | # PALM is free software: you can redistribute it and/or modify it under the terms |
---|
| 9 | # of the GNU General Public License as published by the Free Software Foundation, |
---|
| 10 | # either version 3 of the License, or (at your option) any later version. |
---|
| 11 | # |
---|
| 12 | # PALM is distributed in the hope that it will be useful, but WITHOUT ANY |
---|
| 13 | # WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR |
---|
| 14 | # A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
---|
| 15 | # |
---|
| 16 | # You should have received a copy of the GNU General Public License along with |
---|
| 17 | # PALM. If not, see <http://www.gnu.org/licenses/>. |
---|
| 18 | # |
---|
[1753] | 19 | # Copyright 1997-2015 Leibniz Universitaet Hannover |
---|
[1046] | 20 | #--------------------------------------------------------------------------------# |
---|
| 21 | # |
---|
| 22 | # Current revisions: |
---|
[1090] | 23 | # ------------------ |
---|
[1805] | 24 | # |
---|
[1867] | 25 | # |
---|
[1805] | 26 | # Former revisions: |
---|
| 27 | # ----------------- |
---|
| 28 | # $Id: mrun 1917 2016-05-27 14:28:12Z hellstea $ |
---|
[1867] | 29 | # |
---|
| 30 | # 1866 2016-04-15 06:50:59Z raasch |
---|
| 31 | # adjusted for lcocean |
---|
[1805] | 32 | # |
---|
[1842] | 33 | # 1841 2016-04-07 19:14:06Z raasch |
---|
| 34 | # script now running under bash |
---|
| 35 | # |
---|
[1805] | 36 | # 1804 2016-04-05 16:30:18Z maronga |
---|
[1779] | 37 | # test: implementing an execute mechanism where the execute command is given in the |
---|
| 38 | # configuration file |
---|
| 39 | # ROPTS removed from execution commands |
---|
| 40 | # Warnings about missing optional files or user code changed to informative messages |
---|
[1804] | 41 | # Removed parameter file check |
---|
[1305] | 42 | # |
---|
[1756] | 43 | # 1755 2016-02-22 13:53:39Z maronga |
---|
| 44 | # Minor bugfix to last commit |
---|
| 45 | # |
---|
[1754] | 46 | # 1753 2016-02-22 13:49:49Z maronga |
---|
| 47 | # Bugfix: use random job number when PBS job nummer retrieval fails on lccray* |
---|
| 48 | # |
---|
[1623] | 49 | # 1622 2015-07-20 06:21:36Z heinze |
---|
| 50 | # bugfix for Mistral (lcbullhh) |
---|
| 51 | # |
---|
| 52 | # 1621 2015-07-17 11:39:33Z heinze |
---|
[1621] | 53 | # adjustments for Mistral at DKRZ Hamburg (lcbullhh) |
---|
[1200] | 54 | # |
---|
[1610] | 55 | # 1609 2015-07-03 15:37:58Z maronga |
---|
| 56 | # Modified the random number of the temporary working directory to match the |
---|
| 57 | # queuing system number on HLRN-III (lccray*). |
---|
| 58 | # |
---|
[1605] | 59 | # 1604 2015-06-23 15:48:03Z suehring |
---|
| 60 | # Enable coupled runs for lcmuk |
---|
| 61 | # |
---|
[1599] | 62 | # 1598 2015-05-29 06:45:40Z raasch |
---|
| 63 | # bugfix for creating hostfile and total number of MPI tasks if run in openmp-mode on one node |
---|
| 64 | # |
---|
[1550] | 65 | # 1549 2015-01-30 14:26:16Z suehring |
---|
| 66 | # Enable interactive debug sessions with allinea debugger |
---|
| 67 | # |
---|
[1548] | 68 | # 1547 2015-01-29 15:09:12Z witha |
---|
| 69 | # adjustments for ForWind computing cluster (lcflow) |
---|
| 70 | # |
---|
[1492] | 71 | # 1491 2014-11-12 08:12:57Z heinze |
---|
| 72 | # correct setting of batch_job allowed for ibm-machines |
---|
| 73 | # |
---|
[1469] | 74 | # 1468 2014-09-24 14:06:57Z maronga |
---|
| 75 | # Typo removed (addres->address) |
---|
| 76 | # Bugfix: returning files to IMUK via ssh did not work for lccrayh and lcycrayb |
---|
| 77 | # Added support for restart runs (remote) for lcxe6 |
---|
| 78 | # |
---|
[1443] | 79 | # 1442 2014-07-28 07:09:10Z raasch |
---|
| 80 | # default queues for HLRN-III (lccrayb/lccrayh) defined |
---|
| 81 | # |
---|
[1423] | 82 | # 1422 2014-07-02 10:01:25Z kanani |
---|
| 83 | # automatic restarts enabled for lccrayb, |
---|
| 84 | # adjustment of path to compilervars.sh at IMUK (see automatic restarts) |
---|
| 85 | # |
---|
[1421] | 86 | # 1420 2014-06-10 06:24:14Z raasch |
---|
| 87 | # -j1 option added to aprung command for lccrayf (CSC) |
---|
| 88 | # |
---|
[1403] | 89 | # 1402 2014-05-09 14:25:13Z raasch |
---|
| 90 | # batch_job added to envpar-NAMELIST |
---|
| 91 | # |
---|
[1358] | 92 | # 1357 2014-04-11 15:02:03Z witha |
---|
| 93 | # adjustments for lccrayh (automatic restarts on lccrayh outside of IMUK are now |
---|
| 94 | # possible) |
---|
| 95 | # |
---|
[1328] | 96 | # 1320 2014-03-20 08:40:49Z raasch |
---|
| 97 | # check namelist file set false by default |
---|
| 98 | # |
---|
[1305] | 99 | # 1304 2014-03-12 10:29:42Z raasch |
---|
| 100 | # ulimit option changed from -Ss to -s |
---|
| 101 | # |
---|
| 102 | # bugfix: missing "fi" in r1289 |
---|
| 103 | # |
---|
[1290] | 104 | # 1289 2014-03-04 07:12:34Z raasch |
---|
| 105 | # comments translated to English |
---|
| 106 | # necriam-, fimm-, ibmy-, and sgi-specific code removed |
---|
| 107 | # export of variables for palm and interpret_config removed |
---|
| 108 | # |
---|
[1282] | 109 | # 1281 2014-02-01 07:55:49Z raasch |
---|
| 110 | # rsync-copy restricted to Cray machines, since command is unavailable on some |
---|
| 111 | # other systems |
---|
| 112 | # |
---|
[1280] | 113 | # 1279 2014-01-28 12:10:14Z raasch |
---|
| 114 | # tasks_per_node must not be an integral divisor of numprocs any more. This was done |
---|
| 115 | # in order to remove annoying restrictions concerning the number of processors which |
---|
| 116 | # appear on machines with larger nodes (e.g. containing 24 cores). Now without this |
---|
| 117 | # restriction, one of the nodes will be filled with less than the given number of |
---|
| 118 | # tasks per node. A respective warning is given. |
---|
| 119 | # |
---|
[1275] | 120 | # 1274 2014-01-09 13:14:54Z heinze |
---|
| 121 | # adjustments for lccrayh |
---|
| 122 | # |
---|
[1273] | 123 | # 1272 2014-01-08 10:19:32Z witha |
---|
| 124 | # small adjustment for lcflow |
---|
| 125 | # |
---|
[1271] | 126 | # 1270 2013-12-16 11:05:01Z fricke |
---|
| 127 | # call of combine_plot_fields adjusted for lccrayb/lccrayh |
---|
| 128 | # |
---|
[1256] | 129 | # 1255 2013-11-07 14:43:35Z raasch |
---|
| 130 | # further adjustments for lccrayb remote access |
---|
| 131 | # |
---|
[1242] | 132 | # 1241 2013-10-30 11:36:58Z heinze |
---|
| 133 | # Enable use of nudging input and input of large scale forcing from |
---|
| 134 | # external files |
---|
| 135 | # |
---|
[1230] | 136 | # 1229 2013-09-20 06:55:19Z raasch |
---|
| 137 | # further adjustments for lccrayb |
---|
| 138 | # |
---|
[1225] | 139 | # 1224 2013-09-16 07:27:23Z raasch |
---|
| 140 | # first adjustments for lccrayb |
---|
| 141 | # |
---|
[1211] | 142 | # 1210 2013-08-14 10:58:20Z raasch |
---|
| 143 | # fftw support added |
---|
| 144 | # |
---|
[1203] | 145 | # 1201 2013-07-10 16:17:59Z witha |
---|
| 146 | # adjustments for Forwind cluster (lcflow) |
---|
| 147 | # |
---|
[1200] | 148 | # 1199 2013-07-05 14:52:22Z raasch |
---|
[1199] | 149 | # adjustments for CSC Helsinki (lccrayf), |
---|
| 150 | # executables for batch jobs can be created in advance, in order to avoid calling |
---|
| 151 | # the compiler within the batch job (only works if batch jobs are submitted on |
---|
| 152 | # local host) |
---|
[1104] | 153 | # |
---|
[1191] | 154 | # 1190 2013-06-25 09:39:21Z heinze |
---|
| 155 | # enable use of automatic restarts for ibmh |
---|
[1185] | 156 | # use of cluster/express queue enabled (ibmh) |
---|
[1104] | 157 | # |
---|
[1125] | 158 | # 1124 2013-04-09 15:46:52Z raasch |
---|
| 159 | # variable "memory" is exported via typeset option -x, because otherwise an unknown |
---|
| 160 | # side effect may lead to data loss while getopts is reading the script-option arguments |
---|
| 161 | # |
---|
[1123] | 162 | # 1122 2013-04-09 08:37:16Z heinze |
---|
| 163 | # Bugfix: change type of variable last_char |
---|
| 164 | # |
---|
[1121] | 165 | # 1119 2013-04-05 15:11:19Z raasch |
---|
| 166 | # Bugfix for setting -T option for subjob |
---|
| 167 | # |
---|
[1109] | 168 | # 1108 2013-03-05 07:03:32Z raasch |
---|
| 169 | # bugfix for coupled runs on lckyut/lckyuh |
---|
| 170 | # |
---|
[1107] | 171 | # 1106 2013-03-04 05:31:38Z raasch |
---|
| 172 | # --stdin argument for mpiexec on lckyuh |
---|
| 173 | # -y and -Y settings output to header |
---|
| 174 | # |
---|
[1104] | 175 | # 1103 2013-02-20 02:15:53Z raasch |
---|
[1103] | 176 | # default script runs again under ksh, because of unsolved problems with read |
---|
| 177 | # from stdin: when bash script is called from a ksh, message "read error: 0: |
---|
| 178 | # Resource temporarily unavailable" appears and script does not stop, |
---|
| 179 | # further bash compatibility adjustments, |
---|
| 180 | # shebang line replaced by /bin/bash when running jobs on lckyuh; no restarts |
---|
| 181 | # on lckyuh, but mrun does not terminate and issues a warning instead |
---|
[1046] | 182 | # |
---|
[1102] | 183 | # 1101 2013-02-17 10:20:21Z raasch |
---|
| 184 | # script now running under bash instead of ksh, which required small adjustments |
---|
| 185 | # (output formatting with printf instead "typeset -L/-R", print replaced by echo, |
---|
| 186 | # read from stdin), |
---|
| 187 | # cross compilername on lckyuh compute nodes replaced by real compiler name |
---|
| 188 | # |
---|
[1100] | 189 | # 1099 2013-02-10 01:47:43Z raasch |
---|
| 190 | # adjustments for Kyushu-University computing center (lckyuh - hayaka) |
---|
| 191 | # and for Forwind cluster (lcflow) |
---|
| 192 | # small further adjustments for lckyut |
---|
| 193 | # |
---|
[1096] | 194 | # 1094 2013-02-03 01:52:12Z raasch |
---|
| 195 | # explicit ssh/scp port can be set in config file with environment variable |
---|
| 196 | # scp_port. This port is handled to all ssh/scp/batch_scp calls. |
---|
| 197 | # decalpha parts (yonsei) removed |
---|
| 198 | # |
---|
[1091] | 199 | # 2013-02-02 07:06:13Z raasch |
---|
[1099] | 200 | # adjustments for Kyushu-University computing center (lckyut - tatara) |
---|
[1091] | 201 | # |
---|
[1084] | 202 | # 1083 2013-01-04 10:22:09Z maronga |
---|
| 203 | # bugfix in parameter file check (read %cpp_options was missing) |
---|
| 204 | # |
---|
[1070] | 205 | # 1069 2012-11-28 16:18:43Z maronga |
---|
| 206 | # bugfix: coupling mode was always set to mpi2, typos removed |
---|
| 207 | # |
---|
[1059] | 208 | # 1058 2012-11-21 07:00:35Z raasch |
---|
| 209 | # Intel inspector (inspxe) is given the number of PEs instead of the number of |
---|
| 210 | # nodes |
---|
| 211 | # |
---|
[1047] | 212 | # 1046 2012-11-09 14:38:45Z maronga |
---|
| 213 | # code put under GPL (PALM 3.9) |
---|
| 214 | # |
---|
[1090] | 215 | # 21/03/94 - Siggi - first version finished |
---|
| 216 | # 03/03/94 - Siggi - script development started |
---|
| 217 | # |
---|
[1059] | 218 | #--------------------------------------------------------------------------------# |
---|
[1090] | 219 | # mrun - script for running PALM jobs |
---|
| 220 | #--------------------------------------------------------------------------------# |
---|
[1] | 221 | |
---|
[1047] | 222 | |
---|
[1] | 223 | |
---|
[1289] | 224 | # DECLARATION OF VARIABLES AND THEIR DEFUALT VALUES |
---|
[1] | 225 | |
---|
[1289] | 226 | set +o allexport # SUPPRESS EXPORT OF ALL VARIABLES, SINCE IN THE PAST THIS |
---|
| 227 | # LES TO PROBLEMS IN ROUTINES CALLED BY MRUN |
---|
| 228 | # (TOO MANY ARGUMENTS - PROBLEM) |
---|
| 229 | set +o noclobber # EXISTING FILES ARE ALLOWED TO BE OVERWRITTEN |
---|
[1] | 230 | |
---|
| 231 | AddFilenames="" |
---|
| 232 | additional_conditions="" |
---|
| 233 | add_source_path="" |
---|
| 234 | afname="" |
---|
| 235 | archive_save=true |
---|
| 236 | archive_system=none |
---|
[1320] | 237 | check_namelist_files=false |
---|
[920] | 238 | combine_plot_fields=true |
---|
[1] | 239 | compiler_name="" |
---|
| 240 | cond1="" |
---|
| 241 | cond2="" |
---|
| 242 | config_file=.mrun.config |
---|
[206] | 243 | coupled_dist="" |
---|
| 244 | coupled_mode="mpi1" |
---|
[1] | 245 | cpp_opts="" |
---|
[82] | 246 | cpp_options="" |
---|
[1] | 247 | cpumax=0 |
---|
| 248 | cpurest=0 |
---|
[1199] | 249 | create_executable_for_batch=false |
---|
[1] | 250 | delete_temporary_catalog=true |
---|
| 251 | do_batch=false |
---|
| 252 | do_compile=true |
---|
| 253 | do_remote=false |
---|
| 254 | do_stagein=true |
---|
| 255 | do_stageout=true |
---|
| 256 | do_trace=false |
---|
[352] | 257 | email_notification="none" |
---|
[1] | 258 | exclude="" |
---|
| 259 | executable="" |
---|
[1779] | 260 | execute_command="none" |
---|
[1] | 261 | execution_error=false |
---|
[1210] | 262 | fftw_inc="" |
---|
| 263 | fftw_lib="" |
---|
| 264 | fftw_support=false |
---|
[1289] | 265 | fname=test |
---|
[1] | 266 | fromhost="" |
---|
[78] | 267 | global_revision="" |
---|
[122] | 268 | group_number=none |
---|
[1] | 269 | host="" |
---|
| 270 | host_file="" |
---|
| 271 | hp="" |
---|
| 272 | ignore_archive_error=false |
---|
| 273 | input_list="" |
---|
| 274 | interpreted_config_file="" |
---|
[164] | 275 | job_catalog="~/job_queue" |
---|
[1] | 276 | job_on_file="" |
---|
| 277 | keep_data_from_previous_run=false |
---|
[183] | 278 | link_local_input=false |
---|
[149] | 279 | link_local_output=false |
---|
[1] | 280 | localhost_realname=$(hostname) |
---|
[260] | 281 | local_dvrserver_running=.FALSE. |
---|
[1] | 282 | locat=normal |
---|
| 283 | mainprog="" |
---|
| 284 | makefile="" |
---|
[759] | 285 | max_par_io_str="" |
---|
[1] | 286 | mc=$0 |
---|
| 287 | while [[ $(echo $mc | grep -c "/") != 0 ]] |
---|
| 288 | do |
---|
| 289 | mc=`echo $mc | cut -f2- -d"/"` |
---|
| 290 | done |
---|
[493] | 291 | module_calls="" |
---|
[1] | 292 | mrun_script_name=$mc |
---|
[82] | 293 | netcdf_inc="" |
---|
| 294 | netcdf_lib="" |
---|
[1] | 295 | netcdf_support=false |
---|
| 296 | node_usage=default |
---|
| 297 | numprocs="" |
---|
[206] | 298 | numprocs_atmos=0 |
---|
| 299 | numprocs_ocean=0 |
---|
[1] | 300 | OOPT="" |
---|
| 301 | openmp=false |
---|
| 302 | output_list="" |
---|
| 303 | package_list="" |
---|
| 304 | queue=none |
---|
| 305 | read_from_config="" |
---|
[5] | 306 | restart_run=false |
---|
[622] | 307 | if [[ `hostname` = rte10 ]] |
---|
[1] | 308 | then |
---|
[1468] | 309 | return_address=133.5.185.60 |
---|
| 310 | echo "+++ WARNING: fixed return_address = $return_address is used !!!!!" |
---|
[693] | 311 | elif [[ `hostname` = climate0 ]] |
---|
| 312 | then |
---|
[1468] | 313 | return_address=165.132.26.68 |
---|
| 314 | echo "+++ WARNING: fixed return_address = $return_address is used !!!!!" |
---|
[1866] | 315 | elif [[ `hostname` = "schultzl-Latitude-E6540" ]] |
---|
| 316 | then |
---|
| 317 | return_address="schultzl-Latitude-E6540" |
---|
| 318 | echo "+++ WARNING: fixed return_address = $return_address is used !!!!!" |
---|
[1021] | 319 | elif [[ `hostname` = urban00 ]] |
---|
| 320 | then |
---|
[1468] | 321 | return_address=147.46.30.151 |
---|
| 322 | echo "+++ WARNING: fixed return_address = $return_address is used !!!!!" |
---|
[622] | 323 | else |
---|
[1468] | 324 | return_address=$(nslookup `hostname` 2>&1 | grep "Address:" | tail -1 | awk '{print $2}') |
---|
[1] | 325 | fi |
---|
| 326 | return_password="" |
---|
| 327 | return_username=$LOGNAME |
---|
| 328 | remotecall=false |
---|
| 329 | remote_username="" |
---|
[102] | 330 | run_coupled_model=false |
---|
[1] | 331 | run_mode="" |
---|
[397] | 332 | store_on_archive_system=false |
---|
[1289] | 333 | dashes=" ----------------------------------------------------------------------------" |
---|
[1] | 334 | silent=false |
---|
| 335 | source_list="" |
---|
| 336 | source_path=SOURCE |
---|
| 337 | tasks_per_node="" |
---|
| 338 | threads_per_task=1 |
---|
| 339 | tmpcreate=false |
---|
| 340 | tmp_data_catalog="" |
---|
| 341 | transfer_problems=false |
---|
| 342 | usern=$LOGNAME |
---|
[377] | 343 | use_openmp=false |
---|
[1289] | 344 | version="MRUN 2.1 Rev$Rev: 1917 $" |
---|
[1] | 345 | working_directory=`pwd` |
---|
| 346 | TOPT="" |
---|
| 347 | XOPT="" |
---|
| 348 | zeit=$( date | cut -c 12-19 ) |
---|
| 349 | |
---|
[1124] | 350 | typeset -i iec=0 iic=0 iin=0 ioc=0 iout=0 stagein_anz=0 stageout_anz=0 |
---|
[1289] | 351 | typeset -x -i memory=0 # HAS TO BE EXPORTED HERE, OTHERWISE AN UNKNOWN SIDE |
---|
| 352 | # SIDE EFFECT MAY CAUSE DATA LOSS WHEN GETOPTS IS READING THE |
---|
| 353 | # SCRIPT-OPTION ARGUMENTS |
---|
[1279] | 354 | typeset -i cputime i ii iia iii iio icycle inode ival jobges jobsek last_char_int maxcycle minuten nodes pes remaining_pes sekunden tp1 |
---|
[1] | 355 | |
---|
[1289] | 356 | # ERROR HANDLING IN CASE OF EXIT |
---|
[83] | 357 | trap 'rm -rf $working_directory/tmp_mrun |
---|
[1289] | 358 | if [[ $locat != normal && $locat != control_c ]] |
---|
[1] | 359 | then |
---|
| 360 | |
---|
[1289] | 361 | # CARRY OUT ERROR-COMMANDS GIVEN IN THE CONFIGURATION FILE (EC:) |
---|
[1] | 362 | (( i = 0 )) |
---|
| 363 | while (( i < iec )) |
---|
| 364 | do |
---|
| 365 | (( i = i + 1 )) |
---|
| 366 | printf "\n *** Execution of ERROR-command:\n" |
---|
| 367 | printf " >>> ${err_command[$i]}\n" |
---|
| 368 | eval ${err_command[$i]} |
---|
| 369 | done |
---|
[1101] | 370 | if [[ -n $interpreted_config_file ]] |
---|
| 371 | then |
---|
[1] | 372 | rm -rf $interpreted_config_file |
---|
| 373 | fi |
---|
[1101] | 374 | if [[ -n .mrun_environment ]] |
---|
| 375 | then |
---|
[71] | 376 | rm -rf .mrun_environment |
---|
| 377 | fi |
---|
[1] | 378 | if [[ $tmpcreate = true ]] |
---|
| 379 | then |
---|
| 380 | printf "\n *** Contents of \"$TEMPDIR\":\n" |
---|
| 381 | ls -al; cd |
---|
| 382 | [[ $delete_temporary_catalog = true ]] && rm -rf $TEMPDIR |
---|
| 383 | fi |
---|
[260] | 384 | if [[ "$dvrserver_id" != "" ]] |
---|
| 385 | then |
---|
| 386 | echo "+++ killing dvrserver_id=$dvrserver_id" |
---|
| 387 | kill $dvrserver_id |
---|
| 388 | fi |
---|
[1] | 389 | if [[ -f ~/job_queue/JOBINFO.$QSUB_REQID ]] |
---|
| 390 | then |
---|
| 391 | rm -rf ~/job_queue/JOBINFO.$QSUB_REQID |
---|
| 392 | fi |
---|
| 393 | printf "\n\n+++ MRUN killed \n\n" |
---|
| 394 | elif [[ $locat != control_c ]] |
---|
| 395 | then |
---|
| 396 | printf "\n\n --> all actions finished\n\n" |
---|
| 397 | printf " Bye, bye $usern !!\n\n" |
---|
| 398 | fi' exit |
---|
| 399 | |
---|
| 400 | |
---|
[1289] | 401 | # ACTIONS IN CASE OF TERMINAL-BREAK (CONTROL-C): |
---|
[83] | 402 | trap 'rm -rf $working_directory/tmp_mrun |
---|
[807] | 403 | rm -rf $working_directory/tmp_check_namelist_files |
---|
[69] | 404 | [[ $tmpcreate = true ]] && (cd; rm -rf $TEMPDIR) |
---|
[1] | 405 | if [[ -f ~/job_queue/JOBINFO.$QSUB_REQID ]] |
---|
| 406 | then |
---|
| 407 | rm -rf ~/job_queue/JOBINFO.$QSUB_REQID |
---|
| 408 | fi |
---|
[260] | 409 | if [[ "$dvrserver_id" != "" ]] |
---|
| 410 | then |
---|
| 411 | echo "+++ killing dvrserver_id=$dvrserver_id" |
---|
| 412 | kill $dvrserver_id |
---|
| 413 | fi |
---|
[1] | 414 | printf "\n+++ MRUN killed by \"^C\" \n\n" |
---|
| 415 | locat=control_c |
---|
| 416 | exit |
---|
| 417 | ' 2 |
---|
| 418 | |
---|
| 419 | |
---|
[22] | 420 | # CHECK IF THE PATH FOR THE PALM BINARIES (SCRIPTS+UTILITY-PROGRAMS) HAS |
---|
| 421 | # BEEN SET |
---|
| 422 | if [[ "$PALM_BIN" = "" ]] |
---|
| 423 | then |
---|
| 424 | printf "\n +++ environment variable PALM_BIN has not been set" |
---|
| 425 | printf "\n please set it to the directory where the PALM scripts are located" |
---|
| 426 | locat=palm_bin; exit |
---|
| 427 | fi |
---|
[180] | 428 | export PATH=$PALM_BIN:$PATH |
---|
[22] | 429 | |
---|
| 430 | |
---|
[1] | 431 | |
---|
[1289] | 432 | # READ SHELLSCRIPT-OPTIONS AND REBUILD THE MRUN-COMMAND STRING (MC), |
---|
| 433 | # WHICH WILL BE USED TO START RESTART-JOBS |
---|
[920] | 434 | while getopts :a:AbBc:Cd:D:Fg:G:h:H:i:IkK:m:M:n:o:O:p:P:q:r:R:s:St:T:u:U:vw:xX:yY:zZ option |
---|
[1] | 435 | do |
---|
| 436 | case $option in |
---|
| 437 | (a) afname=$OPTARG;; |
---|
[397] | 438 | (A) store_on_archive_system=true; mc="$mc -A";; |
---|
[1] | 439 | (b) do_batch=true; mc="$mc -b";; |
---|
| 440 | (B) delete_temporary_catalog=false; mc="$mc -B";; |
---|
| 441 | (c) config_file=$OPTARG; mc="$mc -c$OPTARG";; |
---|
[5] | 442 | (C) restart_run=true; mc="$mc -C";; |
---|
[1] | 443 | (d) fname=$OPTARG; mc="$mc -d$OPTARG";; |
---|
[210] | 444 | (D) cpp_opts="$cpp_opts $OPTARG"; mc="$mc -D'$OPTARG'";; |
---|
[79] | 445 | (F) job_on_file="-D"; mc="$mc -F";; |
---|
[122] | 446 | (g) group_number=$OPTARG; mc="$mc -g$OPTARG";; |
---|
[78] | 447 | (G) global_revision=$OPTARG; mc="$mc -G'$OPTARG'";; |
---|
[1] | 448 | (h) host=$OPTARG; mc="$mc -h$OPTARG";; |
---|
| 449 | (H) fromhost=$OPTARG; mc="$mc -H$OPTARG";; |
---|
| 450 | (i) input_list=$OPTARG; mc="$mc -i'$OPTARG'";; |
---|
| 451 | (I) ignore_archive_error=true; mc="$mc -I";; |
---|
| 452 | (k) keep_data_from_previous_run=true; mc="$mc -k";; |
---|
[210] | 453 | (K) additional_conditions="$OPTARG"; mc="$mc -K'$OPTARG'";; |
---|
[1] | 454 | (m) memory=$OPTARG; mc="$mc -m$OPTARG";; |
---|
| 455 | (M) makefile=$OPTARG; mc="$mc -M$OPTARG";; |
---|
| 456 | (n) node_usage=$OPTARG; mc="$mc -n$OPTARG";; |
---|
| 457 | (o) output_list=$OPTARG; mc="$mc -o'$OPTARG'";; |
---|
[622] | 458 | (O) use_openmp=true; threads_per_task=$OPTARG; mc="$mc -O$OPTARG";; |
---|
[1] | 459 | (p) package_list=$OPTARG; mc="$mc -p'$OPTARG'";; |
---|
| 460 | (P) return_password=$OPTARG; mc="$mc -P$OPTARG";; |
---|
| 461 | (q) queue=$OPTARG; mc="$mc -q$OPTARG";; |
---|
| 462 | (r) run_mode=$OPTARG; mc="$mc -r'$OPTARG'";; |
---|
[1468] | 463 | (R) remotecall=true;return_address=$OPTARG; mc="$mc -R$OPTARG";; |
---|
[1] | 464 | (s) source_list=$OPTARG; mc="$mc -s'$OPTARG'";; |
---|
| 465 | (S) read_from_config=false; mc="$mc -S";; |
---|
| 466 | (t) cpumax=$OPTARG; mc="$mc -t$OPTARG";; |
---|
[759] | 467 | (T) mrun_tasks_per_node=$OPTARG; mc="$mc -T$OPTARG";; |
---|
[1] | 468 | (u) remote_username=$OPTARG; mc="$mc -u$OPTARG";; |
---|
| 469 | (U) return_username=$OPTARG; mc="$mc -U$OPTARG";; |
---|
| 470 | (v) silent=true; mc="$mc -v";; |
---|
[759] | 471 | (w) max_par_io_str=$OPTARG; mc="$mc -w$OPTARG";; |
---|
[1] | 472 | (x) do_trace=true;set -x; mc="$mc -x";; |
---|
| 473 | (X) numprocs=$OPTARG; mc="$mc -X$OPTARG";; |
---|
[291] | 474 | (y) ocean_file_appendix=true; mc="$mc -y";; |
---|
[206] | 475 | (Y) run_coupled_model=true; coupled_dist=$OPTARG; mc="$mc -Y'$OPTARG'";; |
---|
[807] | 476 | (z) check_namelist_files=false; mc="$mc -z";; |
---|
[920] | 477 | (Z) combine_plot_fields=false; mc="$mc -Z";; |
---|
[1] | 478 | (\?) printf "\n +++ unknown option $OPTARG \n" |
---|
| 479 | printf "\n --> type \"$0 ?\" for available options \n" |
---|
| 480 | locat=parameter;exit;; |
---|
| 481 | esac |
---|
| 482 | done |
---|
| 483 | |
---|
| 484 | |
---|
[1289] | 485 | # SKIP GIVEN OPTIONS TO READ POSITIONAL PARAMETER, IF GIVEN |
---|
| 486 | # CURRENTLY ONLY PARAMETER ? (TO OUTPUT A SHORT COMMAND INFO) IS ALLOWED |
---|
[1101] | 487 | (( to_shift = $OPTIND - 1 )) |
---|
| 488 | shift $to_shift |
---|
[1] | 489 | |
---|
[1289] | 490 | # PRINT SHORT DESCRIPTION OF MRUN OPTIONS |
---|
[1] | 491 | if [[ "$1" = "?" ]] |
---|
| 492 | then |
---|
| 493 | (printf "\n *** mrun can be called as follows:\n" |
---|
[1804] | 494 | printf "\n $mrun_script_name -b -c.. -d.. -D.. -f.. -F -h.. -i.. -I -K.. -m.. -o.. -p.. -r.. -R -s.. -t.. -T.. -v -x -X.. -y -Y.. -Z <modus> \n" |
---|
[1] | 495 | printf "\n Description of available options:\n" |
---|
| 496 | printf "\n Option Description Default-Value" |
---|
| 497 | printf "\n -a base name of input files equiv. -d" |
---|
[400] | 498 | printf "\n -A archiving when using file-attribute fl" |
---|
[1] | 499 | printf "\n -b batch-job on local machine ---" |
---|
| 500 | printf "\n -B do not delete temporary directory at end ---" |
---|
| 501 | printf "\n -c configuration file .mrun.config" |
---|
| 502 | printf "\n -d base name of files attached to program test" |
---|
| 503 | printf "\n -D preprocessor(cpp)-directives \"\" " |
---|
| 504 | printf "\n -F create remote job file only ---" |
---|
| 505 | printf "\n -h execution host $localhost_realname" |
---|
| 506 | printf "\n -i INPUT control list \"\" " |
---|
| 507 | printf "\n -I archiving errors of previous batch-jobs" |
---|
| 508 | printf "\n will be ignored" |
---|
| 509 | printf "\n -k keep data from previous run" |
---|
| 510 | printf "\n -K additional conditions for controling" |
---|
| 511 | printf "\n usage of conditional code and" |
---|
| 512 | printf "\n env-variables in configuration file \"\" " |
---|
| 513 | printf "\n -m memory demand in MB (batch-jobs) 0 MB" |
---|
| 514 | printf "\n -M Makefile name Makefile" |
---|
| 515 | printf "\n -n node usage (shared/not_shared) depending on -h" |
---|
| 516 | printf "\n -o OUTPUT control list \"\" " |
---|
[1304] | 517 | printf "\n -O threads per openMP task ---" |
---|
[1] | 518 | printf "\n -p software package list \"\" " |
---|
| 519 | printf "\n -q queue \"$queue\" " |
---|
| 520 | printf "\n -r run control list (combines -i -o) \"\" " |
---|
| 521 | printf "\n -s filenames of routines to be compiled \"\" " |
---|
| 522 | printf "\n must end with .f, .f90, .F, or .c !" |
---|
| 523 | printf "\n use \"..\" for more than one file and wildcards" |
---|
[837] | 524 | printf "\n -s LM compiles all locally modified files" |
---|
[1] | 525 | printf "\n -S config file interpreted by shellscript ---" |
---|
| 526 | printf "\n -t allowed cpu-time in seconds (batch) 0" |
---|
| 527 | printf "\n -T tasks per node depending on -h" |
---|
| 528 | printf "\n -u username on remote machine \"\" " |
---|
| 529 | printf "\n -v no prompt for confirmation ---" |
---|
[759] | 530 | printf "\n -w maximum parallel io streams as given by -X" |
---|
[1] | 531 | printf "\n -x tracing of mrun for debug purposes ---" |
---|
| 532 | printf "\n -X # of processors (on parallel machines) 1" |
---|
[291] | 533 | printf "\n -y add appendix \"_O\" to all local output" |
---|
| 534 | printf "\n files (ocean precursor runs followed by" |
---|
| 535 | printf "\n coupled atmosphere-ocean runs) ---" |
---|
[206] | 536 | printf "\n -Y run coupled model, \"#1 #2\" with" |
---|
| 537 | printf "\n #1 atmosphere and #2 ocean processors \"#/2 #/2\" depending on -X" |
---|
[920] | 538 | printf "\n -Z skip combine_plot_fields at the end of " |
---|
| 539 | printf "\n the simulation ---" |
---|
[1] | 540 | printf "\n " |
---|
| 541 | printf "\n Possible values of positional parameter <modus>:" |
---|
| 542 | printf "\n \"?\" - this outline \n\n") | more |
---|
| 543 | exit |
---|
| 544 | elif [[ "$1" != "" ]] |
---|
| 545 | then |
---|
| 546 | printf "\n +++ positional parameter $1 unknown \n" |
---|
| 547 | locat=parameter; exit |
---|
| 548 | fi |
---|
| 549 | |
---|
| 550 | |
---|
| 551 | |
---|
[1289] | 552 | # SHORT STARTING MESSAGE |
---|
[1] | 553 | printf "\n*** $version " |
---|
| 554 | printf "\n will be executed. Please wait ..." |
---|
| 555 | |
---|
| 556 | |
---|
[82] | 557 | |
---|
[1289] | 558 | # CHECK, IF CONFIGURATION FILE EXISTS |
---|
[82] | 559 | if [[ ! -f $config_file ]] |
---|
| 560 | then |
---|
| 561 | printf "\n\n +++ configuration file: " |
---|
| 562 | printf "\n $config_file" |
---|
| 563 | printf "\n does not exist" |
---|
| 564 | locat=connect; exit |
---|
| 565 | fi |
---|
| 566 | |
---|
| 567 | |
---|
[1289] | 568 | # DETERMINE THE HOST-IDENTIFIER (localhost) FROM THE CONFIGURATION FILE |
---|
[82] | 569 | line="" |
---|
| 570 | grep "%host_identifier" $config_file > tmp_mrun |
---|
| 571 | while read line |
---|
| 572 | do |
---|
[206] | 573 | if [[ "$line" != "" && $(echo $line | cut -c1) != "#" ]] |
---|
[82] | 574 | then |
---|
| 575 | HOSTNAME=`echo $line | cut -d" " -s -f2` |
---|
| 576 | host_identifier=`echo $line | cut -d" " -s -f3` |
---|
[83] | 577 | if [[ $localhost_realname = $HOSTNAME ]] |
---|
[82] | 578 | then |
---|
| 579 | localhost=$host_identifier |
---|
| 580 | break |
---|
| 581 | fi |
---|
| 582 | fi |
---|
| 583 | done < tmp_mrun |
---|
| 584 | |
---|
| 585 | if [[ "$localhost" = "" ]] |
---|
| 586 | then |
---|
| 587 | printf "\n\n +++ no host identifier found in configuration file \"$config_file\"" |
---|
| 588 | printf "\n for local host \"$localhost_realname\"." |
---|
| 589 | printf "\n Please add line" |
---|
| 590 | printf "\n \"\%host_identifier $localhost_realname <identifier>\"" |
---|
| 591 | printf "\n to the configuration file." |
---|
| 592 | locat=localhost; exit |
---|
| 593 | fi |
---|
| 594 | |
---|
| 595 | |
---|
[1289] | 596 | # SET HOST-SPECIFIC VARIABLES |
---|
[82] | 597 | case $localhost_realname in |
---|
[397] | 598 | (r1*|r2*|h01*|b01*) archive_system=tivoli;; |
---|
[82] | 599 | (cs*) archive_system=ut;; |
---|
| 600 | esac |
---|
| 601 | |
---|
| 602 | |
---|
[1289] | 603 | # ??? THIS SHOULD BE DECRIBED IN THE MRUN DOCUMENTATION ??? |
---|
| 604 | # SET BASENAME OF THE INPUT-FILES TO THE GENERAL BASENAME (GIVEN BY OPTION -d), |
---|
| 605 | # IF IT HAS NOT BEEN SET BY THE USER DIFFERENTLY (USING OPTION -a) |
---|
[1] | 606 | [[ "$afname" = "" ]] && afname=$fname |
---|
| 607 | |
---|
| 608 | |
---|
[1289] | 609 | # ADD RUN-FILE ACTIVATION STRINGS (GIVEN BY OPTION -r) TO THE |
---|
| 610 | # RESPECTIVE LISTS FOR ACTIVATING THE INPUT/OUTPUT FILE CONNECTION |
---|
| 611 | # STATEMENTS IN THE CONFIGURATION FILE |
---|
[1] | 612 | if [[ "$run_mode" != "" ]] |
---|
| 613 | then |
---|
| 614 | input_list="$input_list $run_mode" |
---|
| 615 | output_list="$output_list $run_mode" |
---|
| 616 | fi |
---|
| 617 | |
---|
| 618 | |
---|
[1289] | 619 | # ??? is this (and the respective option -H) really required ??? |
---|
| 620 | # STORE HOSTNAME, FROM WHICH THE JOB IS STARTED, |
---|
| 621 | # BUT ONLY IF IT HASN'T BEEN ALREADY SET BY OPTION -H |
---|
| 622 | # (MRUN IS AUTOMATICALLY SETTING THIS OPTION IN THE MRUN-CALLS WITHIN |
---|
| 623 | # THOSE BATCH-JOBS, WHICH ARE CREATED BY MRUN ITSELF) |
---|
[1] | 624 | if [[ "$fromhost" = "" ]] |
---|
| 625 | then |
---|
| 626 | fromhost=$localhost |
---|
| 627 | fi |
---|
| 628 | |
---|
| 629 | |
---|
[1289] | 630 | # CHECK, IF EXECUTION SHALL TO BE DONE ON THE REMOTE-MACHINE. |
---|
| 631 | # VALUE OF do_remote IS IMPORTANT FOR THE FILE CONNECTIONS. |
---|
| 632 | # IN CASE OF EXECUTION ON A REMOTE-MACHINE, A BATCH-JOB HAS |
---|
[1609] | 633 | # TO BE SUBMITTED (I.E. do_batch=true) |
---|
[1] | 634 | if [[ -n $host && "$host" != $localhost ]] |
---|
| 635 | then |
---|
| 636 | do_batch=true |
---|
| 637 | do_remote=true |
---|
| 638 | case $host in |
---|
[1866] | 639 | (ibm|ibmh|ibmkisti|ibmku|ibms|nech|lcbullhh|lccrayb|lccrayh|lccrayf|lcflow|lckyoto|lcocean|unics|lcxe6|lcxt5m|lck|lckiaps|lckordi|lckyuh|lckyut|lcsb) true;; |
---|
[1] | 640 | (*) printf "\n" |
---|
[83] | 641 | printf "\n +++ sorry: execution of batch jobs on remote host \"$host\"" |
---|
| 642 | printf "\n is not available" |
---|
[1] | 643 | locat=nqs; (( iec = 0 )); exit;; |
---|
| 644 | esac |
---|
| 645 | else |
---|
| 646 | host=$localhost |
---|
| 647 | fi |
---|
| 648 | |
---|
| 649 | |
---|
[1289] | 650 | # EVALUATE ADDITIONAL CONDITIONS GIVEN BY OPTION -K |
---|
[1] | 651 | if [[ -n $additional_conditions ]] |
---|
| 652 | then |
---|
[69] | 653 | cond1=`echo $additional_conditions | cut -d" " -f1` |
---|
| 654 | cond2=`echo $additional_conditions | cut -d" " -s -f2` |
---|
| 655 | dummy=`echo $additional_conditions | cut -d" " -s -f3` |
---|
[1] | 656 | if [[ -n $dummy ]] |
---|
| 657 | then |
---|
| 658 | printf "\n +++ more than 2 additional conditions given for Option \"-K\"" |
---|
| 659 | locat=options; exit |
---|
| 660 | fi |
---|
[503] | 661 | block=_$cond1 |
---|
| 662 | [[ -n $cond2 ]] && block=${block}_$cond2 |
---|
[1] | 663 | fi |
---|
| 664 | |
---|
| 665 | |
---|
[1289] | 666 | # EVALUATE MODEL COUPLING FEATURES (OPTION -Y) AND DETERMINE coupled_mode |
---|
[206] | 667 | if [[ $run_coupled_model = true ]] |
---|
| 668 | then |
---|
| 669 | |
---|
[1289] | 670 | numprocs_atmos=`echo $coupled_dist | cut -d" " -s -f1` |
---|
| 671 | numprocs_ocean=`echo $coupled_dist | cut -d" " -s -f2` |
---|
| 672 | |
---|
| 673 | if (( $numprocs_ocean + $numprocs_atmos != $numprocs )) |
---|
[206] | 674 | then |
---|
| 675 | |
---|
[1289] | 676 | printf "\n +++ number of processors does not fit to specification by \"-Y\"." |
---|
| 677 | printf "\n PEs (total) : $numprocs" |
---|
| 678 | printf "\n PEs (atmosphere): $numprocs_atmos" |
---|
| 679 | printf "\n PEs (ocean) : $numprocs_ocean" |
---|
| 680 | locat=coupling; exit |
---|
[206] | 681 | |
---|
[1289] | 682 | fi |
---|
[206] | 683 | |
---|
| 684 | |
---|
| 685 | # GET coupled_mode FROM THE CONFIG FILE |
---|
| 686 | line="" |
---|
| 687 | grep "%cpp_options.*-D__mpi2.*$host" $config_file > tmp_mrun |
---|
| 688 | while read line |
---|
| 689 | do |
---|
[1779] | 690 | echo line=\"$line\" |
---|
| 691 | if [[ "$line" != "" && $(echo $line | cut -c1) != "#" && $(echo $line | cut -d" " -s -f4) = $cond1 && $(echo $line | cut -d" " -s -f5) = $cond2 ]] |
---|
[206] | 692 | then |
---|
| 693 | coupled_mode="mpi2" |
---|
| 694 | fi |
---|
| 695 | done < tmp_mrun |
---|
| 696 | |
---|
| 697 | fi |
---|
| 698 | |
---|
[1289] | 699 | # CHECK, IF FILE-ARCHIVING HAS FAILED IN PREVIOUS JOB (OF A JOB-CHAIN) |
---|
[1] | 700 | if [[ -f ~/job_queue/ARCHIVE_ERROR_$fname ]] |
---|
| 701 | then |
---|
| 702 | if [[ $ignore_archive_error = false ]] |
---|
| 703 | then |
---|
| 704 | printf "\n +++ data archiving of previous run failed" |
---|
| 705 | printf "\n see directory \~/job_queue on remote machine" |
---|
| 706 | locat=archive; exit |
---|
| 707 | else |
---|
| 708 | printf "\n +++ warning: data archiving in a previous run failed" |
---|
| 709 | printf "\n MRUN continues, trying to get backup copy" |
---|
| 710 | fi |
---|
| 711 | fi |
---|
| 712 | |
---|
| 713 | |
---|
[1289] | 714 | # SAVE VALUES OF MRUN-OPTIONS SICHERN IN ORDER TO OVERWRITE |
---|
| 715 | # THOSE VALUES GIVEN IN THE CONFIGURATION-FILE |
---|
[1099] | 716 | mrun_memory=$memory |
---|
| 717 | mrun_group_number=$group_number |
---|
| 718 | mrun_cpumax=$cpumax |
---|
| 719 | mrun_numprocs=$numprocs |
---|
[1] | 720 | |
---|
[1099] | 721 | |
---|
[1289] | 722 | # READ AND EVALUATE THE CONFIGURATION-FILE FROM WITHIN THIS SHELLSCRIPT |
---|
[1468] | 723 | # (OPTION -S). THE DEFAULT IS USING THE ROUTINE <<<< |
---|
[1] | 724 | if [[ "$read_from_config" = false ]] |
---|
| 725 | then |
---|
| 726 | [[ $silent = false ]] && printf "\n Reading the configuration file... " |
---|
[1289] | 727 | |
---|
| 728 | # READ THE CONFIGURATION FILE LINE BY LINE |
---|
[1] | 729 | while read zeile |
---|
| 730 | do |
---|
| 731 | [[ $silent = false ]] && printf "." |
---|
| 732 | |
---|
| 733 | |
---|
[1289] | 734 | # FIRST REPLACE ENVIRONMENT-VARIABLES BY THEIR RESPECTIVE VALUES |
---|
[1] | 735 | eval zeile=\"$zeile\" |
---|
| 736 | |
---|
| 737 | |
---|
[1289] | 738 | # INTERPRET THE LINE |
---|
[21] | 739 | if [[ "$(echo $zeile)" = "" ]] |
---|
[1] | 740 | then |
---|
[1289] | 741 | # EMPTY LINE, NO ACTION |
---|
[21] | 742 | continue |
---|
[1] | 743 | |
---|
[21] | 744 | elif [[ "$(echo $zeile | cut -c1)" = "#" ]] |
---|
| 745 | then |
---|
| 746 | |
---|
[1289] | 747 | # LINE IS A COMMENT LINE |
---|
[1] | 748 | true |
---|
| 749 | |
---|
| 750 | elif [[ "$(echo $zeile | cut -c1)" = "%" ]] |
---|
| 751 | then |
---|
| 752 | |
---|
[1289] | 753 | # LINE DEFINES AN ENVIRONMENT-VARIABLE |
---|
[1] | 754 | zeile=$(echo $zeile | cut -c2-) |
---|
[69] | 755 | var=`echo $zeile | cut -d" " -f1` |
---|
| 756 | value=`echo $zeile | cut -d" " -s -f2` |
---|
| 757 | for_host=`echo $zeile | cut -d" " -s -f3` |
---|
| 758 | for_cond1=`echo $zeile | cut -d" " -s -f4` |
---|
| 759 | for_cond2=`echo $zeile | cut -d" " -s -f5` |
---|
[1] | 760 | |
---|
| 761 | if [[ "$for_host" = "" || ( "$for_host" = $host && "$for_cond1" = "$cond1" && "$for_cond2" = "$cond2" ) || $(echo "$input_list$output_list"|grep -c "$for_host") != 0 ]] |
---|
| 762 | then |
---|
| 763 | |
---|
[1289] | 764 | # REPLACE ":" BY " " IN COMPILER- CPP- OR LINKER-OPTIONS, |
---|
| 765 | # "::" IS REPLACED BY ":". |
---|
[1] | 766 | value=`echo $value | sed 's/::/%DUM%/g' | sed 's/:/ /g' | sed 's/%DUM%/:/g'` |
---|
| 767 | |
---|
| 768 | |
---|
[1289] | 769 | # VALUE FROM THE CONFIGURATION-FILE IS ASSIGNED TO THE |
---|
| 770 | # ENVIRONMENT-VARIABLE, BUT ONLY IF NO VALUE HAS BEEN ALREADY |
---|
| 771 | # ASSIGNED WITHIN THIS SCRIPT (E.G. BY SCRIPT-OPTIONS). |
---|
| 772 | # NON-ASSIGNED VARIABLES HAVE VALUE "" OR 0 (IN CASE OF INTEGER). |
---|
| 773 | # HENCE THE GENERAL RULE IS: SCRIPT-OPTION OVERWRITES THE |
---|
| 774 | # CONFIGURATION-FILE. |
---|
[1] | 775 | if [[ "$(eval echo \$$var)" = "" || "$(eval echo \$$var)" = "0" ]] |
---|
| 776 | then |
---|
| 777 | eval $var=\$value |
---|
| 778 | |
---|
[1289] | 779 | # TERMINAL OUTPUT OF ENVIRONMENT-VARIABLES, IF TRACEBACK IS SWITCHED on |
---|
[1] | 780 | if [[ $do_trace = true ]] |
---|
| 781 | then |
---|
| 782 | printf "\n*** ENVIRONMENT-VARIABLE $var = $value" |
---|
| 783 | fi |
---|
| 784 | fi |
---|
| 785 | |
---|
[1289] | 786 | # IF AN ENVIRONMENT-VARIABLE DETERMINES THE HOST, THEN EVALUATE IT IMMEDIATELY: |
---|
| 787 | # VALUE OF do-remote IS REQUIRED FOR THE FILE CONNECTIONS (COPY OF I/O FILES). |
---|
| 788 | # IF EXECUTION IS SCHEDULED FOR A REMOTE-MASCHINE, A BATCH-JOB |
---|
| 789 | # MUST HAVE TO BE STARTED |
---|
[1] | 790 | if [[ $var = host ]] |
---|
| 791 | then |
---|
| 792 | if [[ -n $host && "$host" != $localhost ]] |
---|
| 793 | then |
---|
| 794 | do_batch=true |
---|
| 795 | do_remote=true |
---|
| 796 | case $host in |
---|
[1866] | 797 | (ibm|ibmh|ibmkisti|ibmku|ibms|lcbullhh|lccrayb|lccrayh|lccrayf|lcflow|lckyoto|lcocean|nech|unics|lcxe6|lcxt5m|lck|lckiaps|lckordi|lckyuh|lckyut|lcsb) true;; |
---|
[83] | 798 | (*) printf "\n +++ sorry: execution of batch jobs on remote host \"$host\"" |
---|
| 799 | printf "\n is not available" |
---|
[1] | 800 | locat=nqs; exit;; |
---|
| 801 | esac |
---|
| 802 | else |
---|
| 803 | host=$localhost |
---|
| 804 | fi |
---|
| 805 | fi |
---|
| 806 | |
---|
[1289] | 807 | # USER-DEFINED ENVIRONMENT VARIABLES MUST BE EXPORTED, |
---|
| 808 | # BECAUSE THEY MAY BE REQUIRED IN OTHER SCRIPTS CALLED |
---|
| 809 | # BY MRUN (E.G. subjob) |
---|
[1] | 810 | export $var |
---|
| 811 | fi |
---|
| 812 | |
---|
| 813 | |
---|
| 814 | |
---|
| 815 | |
---|
| 816 | elif [[ "$(echo $zeile | cut -c1-3)" = "EC:" ]] |
---|
| 817 | then |
---|
| 818 | |
---|
[1289] | 819 | # LINE DEFINES ERROR-COMMAND |
---|
[1] | 820 | (( iec = iec + 1 )) |
---|
| 821 | zeile=$(echo $zeile | cut -c4-) |
---|
| 822 | err_command[$iec]="$zeile" |
---|
| 823 | |
---|
| 824 | elif [[ "$(echo $zeile | cut -c1-3)" = "IC:" ]] |
---|
| 825 | then |
---|
| 826 | |
---|
[1289] | 827 | # LINE DEFINES INPUT-COMMAND |
---|
[1] | 828 | (( iic = iic + 1 )) |
---|
| 829 | zeile=$(echo $zeile | cut -c4-) |
---|
| 830 | in_command[$iic]="$zeile" |
---|
| 831 | |
---|
| 832 | elif [[ "$(echo $zeile | cut -c1-3)" = "OC:" ]] |
---|
| 833 | then |
---|
| 834 | |
---|
[1289] | 835 | # LINE DEFINES OUTPUT-COMMAND |
---|
[1] | 836 | (( ioc = ioc + 1 )) |
---|
| 837 | zeile=$(echo $zeile | cut -c4-) |
---|
| 838 | out_command[$ioc]="$zeile" |
---|
| 839 | |
---|
| 840 | else |
---|
| 841 | |
---|
[1289] | 842 | # LINE DEFINES FILE CONNECTION. READ THE FILE ATTRIBUTES. |
---|
| 843 | # s2a: in/out - field |
---|
| 844 | # s2b: loc - field (optional) |
---|
| 845 | # s2c: tr/ar - field (optional) |
---|
[69] | 846 | s1=`echo $zeile | cut -d" " -f1` |
---|
| 847 | s2=`echo $zeile | cut -d" " -s -f2` |
---|
[1] | 848 | s2a=$(echo $s2 | cut -d":" -f1) |
---|
| 849 | if [[ $(echo $s2 | grep -c ":") = 0 ]] |
---|
| 850 | then |
---|
| 851 | s2b="" |
---|
| 852 | s2c="" |
---|
| 853 | else |
---|
[69] | 854 | s2b=`echo $s2 | cut -d":" -f2 | sed 's/:/ /g'` |
---|
| 855 | s2c=`echo $s2 | cut -d":" -s -f3 | sed 's/:/ /g'` |
---|
[1] | 856 | fi |
---|
[69] | 857 | s3=`echo $zeile | cut -d" " -f3` |
---|
| 858 | s4=`echo $zeile | cut -d" " -s -f4` |
---|
| 859 | s5=`echo $zeile | cut -d" " -s -f5` |
---|
| 860 | s6=`echo $zeile | cut -d" " -s -f6` |
---|
[1] | 861 | |
---|
| 862 | |
---|
[1289] | 863 | # STORE FILE CONNECTION, IF ACTIVATED BY ACTIVATION-STRING FROM |
---|
| 864 | # INPUT- OR OUTPUT-LIST. |
---|
| 865 | # VARIABLE S3 MAY CONTAIN A LIST OF ACTIVATION STRINGS (FIELD-SEPERATOR ":"). |
---|
| 866 | # IF EXECUTION IS SCHEDULED FOR A REMOTE-MACHINE AND THE FILE IS ONLY |
---|
| 867 | # LOCALLY REQUIRED ON THAT MACHINE (I.E. s2b = loc), THE FILE CONNECTION |
---|
| 868 | # IS NOT CHECKED AND STORED. |
---|
[1] | 869 | IFSALT="$IFS"; IFS="$IFS:" |
---|
| 870 | if [[ "$s2a" = in && ! ( $do_remote = true && ( "$s2b" = loc || "$s2b" = locopt ) ) ]] |
---|
| 871 | then |
---|
| 872 | found=false |
---|
| 873 | for actual in $input_list |
---|
| 874 | do |
---|
| 875 | for formal in $s3 |
---|
| 876 | do |
---|
| 877 | [[ $actual = $formal || "$formal" = "-" ]] && found=true |
---|
| 878 | done |
---|
| 879 | done |
---|
| 880 | if [[ $found = true ]] |
---|
| 881 | then |
---|
| 882 | (( iin = iin + 1 )) |
---|
| 883 | localin[$iin]=$s1; transin[$iin]=$s2b; actionin[$iin]=$s2c; |
---|
| 884 | typein[$iin]=$s3; pathin[$iin]=$s4; endin[$iin]=$s5; |
---|
| 885 | extin[$iin]=$s6 |
---|
| 886 | fi |
---|
| 887 | elif [[ "$s2a" = out && ! ( $do_remote = true && "$s2b" = loc ) ]] |
---|
| 888 | then |
---|
| 889 | found=false |
---|
| 890 | for actual in $output_list |
---|
| 891 | do |
---|
| 892 | for formal in $s3 |
---|
| 893 | do |
---|
| 894 | [[ $actual = $formal || "$formal" = "-" ]] && found=true |
---|
| 895 | done |
---|
| 896 | done |
---|
| 897 | if [[ $found = true ]] |
---|
| 898 | then |
---|
| 899 | (( iout = iout + 1 )) |
---|
| 900 | localout[$iout]=$s1; actionout[$iout]=$s2c; typeout[$iout]=$s3; |
---|
| 901 | pathout[$iout]=$s4; endout[$iout]=$s5; extout[$iout]=$s6 |
---|
| 902 | fi |
---|
| 903 | elif [[ "$s2a" != in && "$s2a" != out ]] |
---|
| 904 | then |
---|
| 905 | printf "\n +++ I/O-attribute in configuration file $config_file has the invalid" |
---|
| 906 | printf "\n value \"$s2\". Only \"in\" and \"out\" are allowed!" |
---|
| 907 | locat=connect; exit |
---|
| 908 | fi |
---|
| 909 | IFS="$IFSALT" |
---|
| 910 | fi |
---|
| 911 | done < $config_file |
---|
| 912 | |
---|
| 913 | else |
---|
| 914 | |
---|
[1289] | 915 | # EVALUATE THE CONFIGURATION-FILE BY FORTRAN-PROGRAM |
---|
[1] | 916 | [[ $silent = false ]] && printf "..." |
---|
[1289] | 917 | interpreted_config_file=.icf.$RANDOM |
---|
[1] | 918 | |
---|
| 919 | |
---|
[1917] | 920 | |
---|
[1289] | 921 | # PROVIDE VALUES OF ENVIRONMENT-VARIABLE FOR interpret_config VIA NAMELIST-FILE |
---|
[69] | 922 | cat > .mrun_environment << %%END%% |
---|
| 923 | &mrun_environment cond1 = '$cond1', cond2 = '$cond2', |
---|
| 924 | config_file = '$config_file', do_remote = '$do_remote', |
---|
| 925 | do_trace = '$do_trace', host = '$host', |
---|
| 926 | input_list = '$input_list', icf = '$interpreted_config_file', |
---|
| 927 | localhost = '$localhost', output_list = '$output_list' / |
---|
| 928 | |
---|
| 929 | %%END%% |
---|
| 930 | |
---|
[1289] | 931 | if [[ "$host" != $localhost ]] |
---|
[66] | 932 | then |
---|
[503] | 933 | |
---|
[1289] | 934 | # REMOTE JOB FROM LOCAL HOST: JUST TAKE THE FIRST EXECUTABLE FOUND |
---|
| 935 | interpret_config_executable=`ls -1 ${PALM_BIN}/interpret_config*.x 2>/dev/null` |
---|
| 936 | if [[ $? != 0 ]] |
---|
[503] | 937 | then |
---|
[1289] | 938 | printf "\n\n +++ no interpret_config found" |
---|
| 939 | printf "\n run \"mbuild -u -h ...\" to generate utilities for this host" |
---|
| 940 | locat=interpret_config; exit |
---|
| 941 | fi |
---|
| 942 | interpret_config_executable=`echo $interpret_config_executable | cut -d" " -f1` |
---|
| 943 | $interpret_config_executable |
---|
[503] | 944 | |
---|
[1289] | 945 | else |
---|
[503] | 946 | |
---|
[1289] | 947 | # CHECK, IF THERE IS AN EXECUTABLE FOR THE BLOCK |
---|
| 948 | if [[ ! -f ${PALM_BIN}/interpret_config${block}.x ]] |
---|
| 949 | then |
---|
| 950 | printf "\n\n +++ no interpret_config found for given block \"$cond1 $cond2\"" |
---|
| 951 | printf "\n run \"mbuild -u -h ...\" to generate utilities for this block" |
---|
| 952 | locat=interpret_config; exit |
---|
[503] | 953 | else |
---|
[1289] | 954 | interpret_config${block}.x |
---|
| 955 | fi |
---|
[503] | 956 | |
---|
[66] | 957 | fi |
---|
[71] | 958 | rm .mrun_environment |
---|
[1] | 959 | |
---|
[1917] | 960 | |
---|
[1289] | 961 | # EXECUTE SHELL-COMMANDS GENERATED BY interpret_config WITHIN THIS SHELL |
---|
[1] | 962 | chmod u+x $interpreted_config_file |
---|
| 963 | export PATH=$PATH:. |
---|
| 964 | . $interpreted_config_file |
---|
[71] | 965 | rm $interpreted_config_file |
---|
[1] | 966 | |
---|
[1099] | 967 | fi |
---|
[71] | 968 | |
---|
[1917] | 969 | |
---|
[1289] | 970 | # VALUES OF MRUN-OPTIONS OVERWRITE THOSE FROM THE CONFIGURATION-FILE |
---|
[1099] | 971 | [[ $mrun_memory != 0 ]] && memory=$mrun_memory |
---|
| 972 | [[ "$mrun_group_number" != "none" ]] && group_number=$mrun_group_number |
---|
| 973 | [[ $mrun_cpumax != 0 ]] && cpumax=$mrun_cpumax |
---|
| 974 | [[ "$mrun_numprocs" != "" ]] && numprocs=$mrun_numprocs |
---|
| 975 | [[ "$max_par_io_str" != "" ]] && maximum_parallel_io_streams=$max_par_io_str |
---|
| 976 | [[ "$mrun_tasks_per_node" != "" ]] && tasks_per_node=$mrun_tasks_per_node |
---|
[1] | 977 | |
---|
| 978 | |
---|
[1289] | 979 | # ON THE LOCAL MACHINE, DETERMINE (FROM THE CONFIGURATION-FILE) THE PATH |
---|
| 980 | # FOR SOURCE-FILES TO BE COMPILED. |
---|
| 981 | # IN A BATCH-JOB, SOURCES (ROUTINES TO BE COMPILED) ARE COMPLETE ALREADY. |
---|
| 982 | # BEFORE MRUN IS CALLED IN THE JOB, SOURCES_COMPLETE = true IS ASSIGNED (SEE FURTHER BELOW). |
---|
[1] | 983 | if [[ "$SOURCES_COMPLETE" = "" ]] |
---|
| 984 | then |
---|
| 985 | |
---|
[1289] | 986 | # FIRST CHECK, IF A GLOBAL SOURCE PATH (TO BE USED FOR ALL HOSTS) |
---|
| 987 | # HAS BEEN DEFINED |
---|
[1] | 988 | source_path="" |
---|
| 989 | line="" |
---|
| 990 | grep "%source_path" $config_file > tmp_mrun |
---|
| 991 | while read line |
---|
| 992 | do |
---|
| 993 | if [[ "$line" != "" && $(echo $line | cut -c1) != "#" ]] |
---|
| 994 | then |
---|
| 995 | if [[ "$(echo $line | cut -d" " -f3)" = "" ]] |
---|
| 996 | then |
---|
| 997 | global_source_path=`echo $line | cut -d" " -f2` |
---|
| 998 | fi |
---|
| 999 | fi |
---|
| 1000 | done < tmp_mrun |
---|
| 1001 | |
---|
| 1002 | line="" |
---|
| 1003 | found=false |
---|
| 1004 | grep " $localhost" $config_file | grep "%source_path" > tmp_mrun |
---|
| 1005 | while read line |
---|
| 1006 | do |
---|
| 1007 | if [[ "$line" != "" && $(echo $line | cut -c1) != "#" ]] |
---|
| 1008 | then |
---|
| 1009 | if [[ $found = true ]] |
---|
| 1010 | then |
---|
| 1011 | printf "\n\n +++ more than one source path found in configuration file" |
---|
| 1012 | printf "\n for local host \"$localhost\" " |
---|
| 1013 | locat=source_path; exit |
---|
| 1014 | fi |
---|
| 1015 | source_path=`echo $line | cut -d" " -f2` |
---|
| 1016 | found=true |
---|
| 1017 | fi |
---|
| 1018 | done < tmp_mrun |
---|
| 1019 | rm tmp_mrun |
---|
| 1020 | |
---|
| 1021 | if [[ "$source_path" = "" ]] |
---|
| 1022 | then |
---|
| 1023 | if [[ "$global_source_path" != "" ]] |
---|
| 1024 | then |
---|
| 1025 | source_path=$global_source_path |
---|
| 1026 | else |
---|
| 1027 | printf "\n\n +++ no source path found in configuration file" |
---|
| 1028 | printf "\n for local host \"$localhost\" " |
---|
| 1029 | locat=source_path; exit |
---|
| 1030 | fi |
---|
| 1031 | fi |
---|
| 1032 | eval source_path=$source_path |
---|
| 1033 | |
---|
| 1034 | if [[ ! -d $source_path ]] |
---|
| 1035 | then |
---|
| 1036 | printf "\n\n +++ source path \"$source_path\" on local host" |
---|
| 1037 | printf "\n \"$localhost\" does not exist" |
---|
| 1038 | locat=source_path; exit |
---|
| 1039 | fi |
---|
| 1040 | |
---|
| 1041 | fi |
---|
| 1042 | |
---|
| 1043 | |
---|
[1289] | 1044 | # GET THE GLOBAL REVISION-NUMBER OF THE SVN-REPOSITORY |
---|
| 1045 | # (HANDED OVER TO RESTART-RUNS USING OPTION -G) |
---|
| 1046 | if [[ "$global_revision" = "" && $host != "ibmkisti" ]] |
---|
[78] | 1047 | then |
---|
| 1048 | global_revision=`svnversion $source_path 2>/dev/null` |
---|
| 1049 | global_revision="Rev: $global_revision" |
---|
| 1050 | fi |
---|
[1] | 1051 | |
---|
[78] | 1052 | |
---|
[1289] | 1053 | # ??? das darf doch eigentlich garnicht passieren, weil optionen config-datei uebersteuern ??? |
---|
| 1054 | # CHECK AGAIN, IF EXECUTION SHALL BE DONE ON A REMOTE-HOST (BATCH-JOB). |
---|
| 1055 | # (HOST MAY HAVE CHANGED DUE TO SETTINGS IN THE CONFIGURATION-FILE) |
---|
[1] | 1056 | if [[ -n $host && "$host" != $localhost ]] |
---|
| 1057 | then |
---|
| 1058 | do_batch=true |
---|
| 1059 | do_remote=true |
---|
| 1060 | case $host in |
---|
[1866] | 1061 | (ibm|ibmh|ibmkisti|ibmku|ibms|lcbullhh|lccrayb|lccrayh|lccrayf|lcflow|lckyoto|lcocean|nech|unics|lcxe6|lcxt5m|lck|lckiaps|lckordi|lckyuh|lckyut|lcsb) true;; |
---|
[1] | 1062 | (*) printf "\n" |
---|
[83] | 1063 | printf "\n +++ sorry: execution of batch jobs on remote host \"$host\"" |
---|
| 1064 | printf "\n is not available" |
---|
[1] | 1065 | locat=nqs; (( iec = 0 )); exit;; |
---|
| 1066 | esac |
---|
| 1067 | else |
---|
| 1068 | host=$localhost |
---|
| 1069 | fi |
---|
| 1070 | |
---|
| 1071 | |
---|
[1620] | 1072 | |
---|
[1289] | 1073 | # IN CASE OF PARALLEL EXECUTION, CHECK SOME SPECIFICATIONS CONCERNING PROCESSOR NUMBERS |
---|
[1] | 1074 | if [[ "$cond1" = parallel || "$cond2" = parallel ]] |
---|
| 1075 | then |
---|
| 1076 | |
---|
[1289] | 1077 | # CHECK, IF NUMBER OF CORES TO BE USED HAS BEEN GIVEN |
---|
[1] | 1078 | if [[ ! -n $numprocs ]] |
---|
| 1079 | then |
---|
| 1080 | printf "\n" |
---|
| 1081 | printf "\n +++ option \"-K parallel\" requires additional specification" |
---|
| 1082 | printf "\n of the number of processors to be used by" |
---|
| 1083 | printf "\n mrun-option \"-X\" or by environment-variable" |
---|
| 1084 | printf "\n \"numprocs\" in the configuration file" |
---|
| 1085 | locat=numprocs; (( iec = 0 )); exit |
---|
| 1086 | fi |
---|
| 1087 | |
---|
[1289] | 1088 | # CHECK, IF THE NUMBER OF CORES PER NODE HAS BEEN GIVEN UND IF IT IS AN |
---|
| 1089 | # INTEGRAL DIVISOR OF THE TOTAL NUMBER OF CORES GIVEN BY OPTION -X |
---|
[1039] | 1090 | if [[ "$tasks_per_node" = "" && $host != lcflow && $host != lcxt5m ]] |
---|
[1] | 1091 | then |
---|
[759] | 1092 | printf "\n" |
---|
| 1093 | printf "\n +++ option \"-T\" (tasks per node) is missing" |
---|
| 1094 | printf "\n set -T option or define tasks_per_node in the config file" |
---|
| 1095 | locat=tasks_per_node; (( iec = 0 )); exit |
---|
| 1096 | fi |
---|
[1279] | 1097 | |
---|
| 1098 | if (( numprocs < tasks_per_node )) |
---|
| 1099 | then |
---|
| 1100 | printf "\n" |
---|
| 1101 | printf "\n +++ tasks per node (-T) cannot exceed total number of processors (-X)" |
---|
| 1102 | printf "\n given values: -T $tasks_per_node -X $numprocs" |
---|
| 1103 | locat=tasks_per_node; (( iec = 0 )); exit |
---|
| 1104 | fi |
---|
| 1105 | |
---|
[1039] | 1106 | if [[ $host != lcflow && $host != lcxt5m ]] |
---|
[759] | 1107 | then |
---|
[1279] | 1108 | (( nodes = numprocs / ( tasks_per_node * threads_per_task ) )) |
---|
[848] | 1109 | (( ival = $tasks_per_node )) |
---|
| 1110 | (( pes = numprocs )) |
---|
| 1111 | (( ii = pes / ival )) |
---|
[1279] | 1112 | (( remaining_pes = pes - ii * ival )) |
---|
| 1113 | if (( remaining_pes > 0 )) |
---|
[848] | 1114 | then |
---|
| 1115 | printf "\n" |
---|
[1279] | 1116 | printf "\n +++ WARNING: tasks per node (option \"-T\") is not an integral" |
---|
[848] | 1117 | printf "\n divisor of the total number of processors (option \"-X\")" |
---|
| 1118 | printf "\n values of this mrun-call: \"-T $tasks_per_node\" \"-X $numprocs\"" |
---|
[1279] | 1119 | printf "\n One of the nodes is filled with $remaining_pes instead of $tasks_per_node tasks" |
---|
| 1120 | (( nodes = nodes + 1 )) |
---|
[848] | 1121 | fi |
---|
[1] | 1122 | fi |
---|
| 1123 | |
---|
[759] | 1124 | |
---|
[377] | 1125 | # SETTINGS FOR SUBJOB-COMMAND |
---|
[1] | 1126 | OOPT="-O $threads_per_task" |
---|
| 1127 | |
---|
[1289] | 1128 | # SET THE TOTAL NUMBER OF NODES, REQUIRED FOR THE SUBJOB-COMMAND (SEE FURTHER BELOW) |
---|
[1] | 1129 | if [[ "$tasks_per_node" != "" ]] |
---|
| 1130 | then |
---|
[1119] | 1131 | TOPT="-T $tasks_per_node" |
---|
[1] | 1132 | fi |
---|
| 1133 | |
---|
[1289] | 1134 | # CHECK node_usage FOR ALLOWED VALUES AND SET DEFAULT VALUE, IF NECESSARY |
---|
[1] | 1135 | if [[ $node_usage = default ]] |
---|
| 1136 | then |
---|
| 1137 | if [[ $host = ibms ]] |
---|
| 1138 | then |
---|
| 1139 | node_usage=shared |
---|
| 1140 | else |
---|
| 1141 | node_usage=not_shared |
---|
| 1142 | fi |
---|
| 1143 | fi |
---|
[181] | 1144 | |
---|
[635] | 1145 | if [[ $node_usage != shared && $node_usage != not_shared && $node_usage != singlejob && "$(echo $node_usage | cut -c1-3)" != "sla" ]] |
---|
[1] | 1146 | then |
---|
| 1147 | printf "\n" |
---|
| 1148 | printf "\n +++ node usage (option \"-n\") is only allowed to be set" |
---|
| 1149 | printf "\n \"shared\" or \"not_shared\"" |
---|
| 1150 | locat=tasks_per_node; (( iec = 0 )); exit |
---|
| 1151 | fi |
---|
| 1152 | |
---|
| 1153 | fi |
---|
| 1154 | |
---|
[1289] | 1155 | # CHECK IF HOSTFILE EXISTS |
---|
[1] | 1156 | if [[ -n $hostfile ]] |
---|
| 1157 | then |
---|
| 1158 | if [[ ! -f $hostfile ]] |
---|
| 1159 | then |
---|
| 1160 | printf "\n" |
---|
| 1161 | printf "\n +++ hostfile \"$hostfile\" does not exist" |
---|
| 1162 | locat=hostfile; exit |
---|
| 1163 | fi |
---|
| 1164 | fi |
---|
| 1165 | |
---|
| 1166 | |
---|
[1289] | 1167 | # SET DEFAULT VALUE FOR THE MAXIMUM NUMBER OF PARALLEL IO STREAMS |
---|
[759] | 1168 | if [[ "$maximum_parallel_io_streams" = "" ]] |
---|
| 1169 | then |
---|
| 1170 | maximum_parallel_io_streams=$numprocs |
---|
| 1171 | fi |
---|
[1] | 1172 | |
---|
[759] | 1173 | |
---|
[1289] | 1174 | # SET PORT NUMBER OPTION FOR CALLS OF SSH/SCP, subjob AND batch_scp SCRIPTS |
---|
[1094] | 1175 | if [[ "$scp_port" != "" ]] |
---|
| 1176 | then |
---|
| 1177 | PORTOPT="-P $scp_port" |
---|
| 1178 | SSH_PORTOPT="-p $scp_port" |
---|
| 1179 | fi |
---|
| 1180 | |
---|
| 1181 | |
---|
[1289] | 1182 | # SET DEFAULT-QUEUE, IF NOT GIVEN |
---|
[1] | 1183 | if [[ $queue = none ]] |
---|
| 1184 | then |
---|
| 1185 | case $host in |
---|
[1184] | 1186 | (ibmh) queue=cluster;; |
---|
[693] | 1187 | (ibmkisti) queue=class.32plus;; |
---|
[1620] | 1188 | (lcbullhh) queue=compute;; |
---|
[1442] | 1189 | (lccrayb) queue=mpp1q;; |
---|
| 1190 | (lccrayh) queue=mpp1q;; |
---|
[1040] | 1191 | (lckiaps) queue=normal;; |
---|
[440] | 1192 | (lckyoto) queue=ph;; |
---|
[1099] | 1193 | (lckyuh) queue=fx-single;; |
---|
| 1194 | (lckyut) queue=cx-single;; |
---|
[635] | 1195 | (lctit) queue=S;; |
---|
[1] | 1196 | (nech) queue=none;; |
---|
| 1197 | (unics) queue=unics;; |
---|
| 1198 | esac |
---|
| 1199 | fi |
---|
| 1200 | |
---|
| 1201 | |
---|
[1289] | 1202 | # GENERATE FULL FILENAMES OF INPUT-FILES, INCLUDING THEIR PATH |
---|
| 1203 | # CHECK, IF INPUT-FILES EXIST, AND DETERMINE HIGHEST CYCLE NUMBER (IF CYCLES EXIST) |
---|
[1] | 1204 | (( i = 0 )) |
---|
| 1205 | while (( i < iin )) |
---|
| 1206 | do |
---|
| 1207 | (( i = i + 1 )) |
---|
| 1208 | (( maxcycle = 0 )) |
---|
| 1209 | |
---|
[1289] | 1210 | # GENERATE PATH AND FULL FILE NAME (then-BRANCH: FIXED FULL NAME IS GIVEN, I.E. THE |
---|
| 1211 | # FILE IDENTIFIER IS NOT PART OF THE FILENAME)) |
---|
[1] | 1212 | if [[ "${actionin[$i]}" = di ]] |
---|
| 1213 | then |
---|
[1289] | 1214 | remotepathin[$i]=${pathin[$i]}/${endin[$i]} # EVALUATE REMOTE-PATH ON THE REMOTE |
---|
| 1215 | # HOST ONLY |
---|
[1] | 1216 | eval filename=${pathin[$i]}/${endin[$i]} |
---|
| 1217 | else |
---|
[1289] | 1218 | remotepathin[$i]=${pathin[$i]}/${afname}${endin[$i]} # EVALUATE REMOTE-PATH ON THE REMOTE |
---|
| 1219 | # HOST ONLY |
---|
[1] | 1220 | eval filename=${pathin[$i]}/${afname}${endin[$i]} |
---|
| 1221 | fi |
---|
[892] | 1222 | eval pathname=${pathin[$i]} |
---|
[1] | 1223 | |
---|
[1289] | 1224 | # CHECK IF FILE EXISTS |
---|
[1] | 1225 | if [[ $(ls $filename* 2>&1 | grep -c "not found") = 1 || \ |
---|
[746] | 1226 | $(ls $filename* 2>&1 | grep -c "nicht gefunden") = 1 || \ |
---|
[1] | 1227 | $(ls $filename* 2>&1 | grep -c "No such file") = 1 || \ |
---|
| 1228 | $(ls $filename* 2>&1 | grep -c "does not exist") = 1 ]] |
---|
| 1229 | then |
---|
[1289] | 1230 | |
---|
| 1231 | # FILES WITH ATTRIBUTE locopt ARE OPTIONAL. NO ABORT, IF THEY DO NOT EXIST. |
---|
[1] | 1232 | if [[ "${transin[$i]}" != "locopt" ]] |
---|
| 1233 | then |
---|
| 1234 | printf "\n\n +++ INPUT-file: " |
---|
| 1235 | if [[ "${extin[$i]}" = "" ]] |
---|
| 1236 | then |
---|
| 1237 | printf "\n $filename" |
---|
| 1238 | else |
---|
| 1239 | printf "\n $filename.${extin[$i]}" |
---|
| 1240 | fi |
---|
| 1241 | printf "\n does not exist\n" |
---|
| 1242 | locat=input; exit |
---|
| 1243 | else |
---|
| 1244 | transin[$i]="unavailable" |
---|
| 1245 | fi |
---|
| 1246 | else |
---|
| 1247 | |
---|
[1289] | 1248 | # DETERMINE THE FILE'S CYCLE NUMBER |
---|
[1] | 1249 | ls -1 -d $filename > filelist 2>/dev/null |
---|
| 1250 | ls -1 -d $filename.* >> filelist 2>/dev/null |
---|
| 1251 | while read zeile |
---|
| 1252 | do |
---|
[1101] | 1253 | cycle=$(echo $zeile | cut -f2 -d".") |
---|
[1] | 1254 | if [[ "$cycle" = "$zeile" ]] |
---|
| 1255 | then |
---|
| 1256 | (( icycle = 0 )) |
---|
| 1257 | elif [[ "$cycle" = "${extin[$i]}" ]] |
---|
| 1258 | then |
---|
| 1259 | (( icycle = 0 )) |
---|
| 1260 | else |
---|
| 1261 | (( icycle = $cycle )) |
---|
| 1262 | fi |
---|
| 1263 | if (( icycle > maxcycle )) |
---|
| 1264 | then |
---|
| 1265 | (( maxcycle = icycle )) |
---|
| 1266 | file_to_be_used=$zeile |
---|
| 1267 | fi |
---|
| 1268 | done <filelist |
---|
| 1269 | rm filelist |
---|
| 1270 | |
---|
[1289] | 1271 | # APPEND CYCLE NUMBER TO FILENAME |
---|
[1] | 1272 | if (( maxcycle > 0 )) |
---|
| 1273 | then |
---|
| 1274 | if [[ "${extin[$i]}" != " " && "${extin[$i]}" != "" ]] |
---|
| 1275 | then |
---|
| 1276 | filename=${filename}.$maxcycle.${extin[$i]} |
---|
| 1277 | else |
---|
| 1278 | filename=${filename}.$maxcycle |
---|
| 1279 | fi |
---|
| 1280 | else |
---|
| 1281 | if [[ "${extin[$i]}" != " " && "${extin[$i]}" != "" ]] |
---|
| 1282 | then |
---|
| 1283 | filename=${filename}.${extin[$i]} |
---|
| 1284 | fi |
---|
| 1285 | fi |
---|
| 1286 | |
---|
[1289] | 1287 | # STORE FILENAME WITHOUT PATH BUT WITH CYCLE NUMBER, |
---|
| 1288 | # BECAUSE IT MIGHT BE REQUIRED LATER TO RESTORE THE FILE FROM AN ARCHIVE-SYSTEM |
---|
[1] | 1289 | absnamein[$i]=$filename |
---|
| 1290 | if (( maxcycle > 0 )) |
---|
| 1291 | then |
---|
| 1292 | if [[ "${actionin[$i]}" = di ]] |
---|
| 1293 | then |
---|
| 1294 | frelin[$i]=${endin[$i]}.$maxcycle |
---|
| 1295 | else |
---|
| 1296 | frelin[$i]=${afname}${endin[$i]}.$maxcycle |
---|
| 1297 | fi |
---|
| 1298 | else |
---|
| 1299 | if [[ "${actionin[$i]}" = di ]] |
---|
| 1300 | then |
---|
| 1301 | frelin[$i]=${endin[$i]} |
---|
| 1302 | else |
---|
| 1303 | frelin[$i]=${afname}${endin[$i]} |
---|
| 1304 | fi |
---|
| 1305 | fi |
---|
| 1306 | |
---|
| 1307 | fi |
---|
| 1308 | done |
---|
[920] | 1309 | |
---|
[1289] | 1310 | # GENERATE FULL FILENAMES OF OUTPUT-FILES (WITHOUT $ OR ~), |
---|
| 1311 | # CHECK, IF OUTPUT-FILES EXIST, AND DETERMINE HIGHEST CYCLE NUMBER (IF CYCLES EXIST), |
---|
| 1312 | # OR, IN CASE THAT FILE DOES NOT EXIST, CHECK, IF IT CAN BE CREATED |
---|
| 1313 | # THESE ACTIONS ARE NOT CARRIED OUT, IF FILES SHALL BE TRASFERRED FROM THE REMOTE TO |
---|
| 1314 | # THE LOCAL HOST (BECAUSE THEIR IS NO DIRECT ACCESS TO THE LOCAL DIRECTORIES FROM THE |
---|
| 1315 | # REMOTE HOST) |
---|
[1] | 1316 | (( i = 0 )) |
---|
| 1317 | while (( i < iout )) |
---|
| 1318 | do |
---|
| 1319 | (( i = i + 1 )) |
---|
| 1320 | if [[ ! ( $fromhost != $localhost && ( "${actionout[$i]}" = tr || "${actionout[$i]}" = tra || "${actionout[$i]}" = trpe ) ) ]] |
---|
| 1321 | then |
---|
[260] | 1322 | if [[ "${actionout[$i]}" = tr ]] |
---|
[1] | 1323 | then |
---|
| 1324 | actionout[$i]="" |
---|
[260] | 1325 | elif [[ "${actionout[$i]}" = trpe ]] |
---|
| 1326 | then |
---|
| 1327 | actionout[$i]=pe |
---|
[1] | 1328 | elif [[ "${actionout[$i]}" = tra ]] |
---|
| 1329 | then |
---|
| 1330 | actionout[$i]=a |
---|
| 1331 | fi |
---|
| 1332 | (( maxcycle = 0 )) |
---|
| 1333 | eval filename=${pathout[$i]}/${fname}${endout[$i]} |
---|
| 1334 | eval catalogname=${pathout[$i]} |
---|
| 1335 | if [[ $(ls $filename* 2>&1 | grep -c "not found") = 1 || \ |
---|
[746] | 1336 | $(ls $filename* 2>&1 | grep -c "nicht gefunden") = 1 || \ |
---|
[1] | 1337 | $(ls $filename* 2>&1 | grep -c "No such file") = 1 || \ |
---|
| 1338 | $(ls $filename* 2>&1 | grep -c "does not exist") = 1 ]] |
---|
| 1339 | then |
---|
| 1340 | |
---|
[1289] | 1341 | # IF OUTPUT-FILE DOES NOT EXIST CHECK, IF IT CAN BE CREATED |
---|
[1] | 1342 | if cat /dev/null > $filename |
---|
| 1343 | then |
---|
| 1344 | rm $filename |
---|
| 1345 | else |
---|
| 1346 | |
---|
[1289] | 1347 | # CHECK, IF THE DIRECTORY WHERE FILE SHALL BE COPIED TO EXISTS |
---|
| 1348 | # IF IT DOES NOT EXIST, TRY TO CREATE IT |
---|
[1] | 1349 | if [[ ! -d $catalogname ]] |
---|
| 1350 | then |
---|
| 1351 | if mkdir -p $catalogname |
---|
| 1352 | then |
---|
| 1353 | printf "\n\n *** directory:" |
---|
| 1354 | printf "\n $catalogname" |
---|
| 1355 | printf "\n was created\n" |
---|
| 1356 | else |
---|
| 1357 | printf "\n\n +++ OUTPUT-file:" |
---|
| 1358 | printf "\n $filename" |
---|
| 1359 | printf "\n cannot be created, because directory does not exist" |
---|
| 1360 | printf "\n and cannot be created either" |
---|
| 1361 | printf "\n" |
---|
| 1362 | locat=output ; exit |
---|
| 1363 | fi 2>/dev/null |
---|
| 1364 | else |
---|
| 1365 | printf "\n\n +++ OUTPUT-file:" |
---|
| 1366 | printf "\n $filename" |
---|
| 1367 | printf "\n cannot be created, although directory exists" |
---|
| 1368 | printf "\n" |
---|
| 1369 | locat=output ; exit |
---|
| 1370 | fi |
---|
| 1371 | fi 2>/dev/null |
---|
| 1372 | else |
---|
| 1373 | |
---|
[1289] | 1374 | # DETERMINE THE CYCLE NUMBER |
---|
[1] | 1375 | ls -1 -d $filename > filelist 2>/dev/null |
---|
| 1376 | ls -1 -d $filename.* >> filelist 2>/dev/null |
---|
| 1377 | while read zeile |
---|
| 1378 | do |
---|
[1101] | 1379 | cycle=$(echo $zeile | cut -f2 -d".") |
---|
[1] | 1380 | if [[ "$cycle" = "$zeile" || "$cycle" = ${extout[$i]} ]] |
---|
| 1381 | then |
---|
| 1382 | (( icycle = 1 )) |
---|
| 1383 | else |
---|
| 1384 | (( icycle = $cycle + 1 )) |
---|
| 1385 | fi |
---|
| 1386 | if (( icycle > maxcycle )) |
---|
| 1387 | then |
---|
| 1388 | (( maxcycle = icycle )) |
---|
| 1389 | fi |
---|
| 1390 | done <filelist |
---|
| 1391 | rm filelist |
---|
| 1392 | fi |
---|
| 1393 | |
---|
[1289] | 1394 | # APPEND CYCLE NUMBER TO FILENAME AND CHECK, IF FILE CAN BE CREATED |
---|
| 1395 | # IN CASE OF FILE-APPEND, FILE MUST BE APPENDED TO THE ONE WITH HIGHEST CYCLE NUMBER |
---|
[1] | 1396 | if [[ "${actionout[$i]}" != a ]] |
---|
| 1397 | then |
---|
| 1398 | if (( maxcycle > 0 )) |
---|
| 1399 | then |
---|
| 1400 | filename=${filename}.$maxcycle |
---|
| 1401 | if cat /dev/null > $filename |
---|
| 1402 | then |
---|
| 1403 | rm $filename |
---|
| 1404 | else |
---|
| 1405 | printf "\n +++ OUTPUT-file:" |
---|
| 1406 | printf "\n $filename" |
---|
| 1407 | printf "\n cannot be created" |
---|
| 1408 | locat=output ; exit |
---|
| 1409 | fi |
---|
| 1410 | fi |
---|
| 1411 | else |
---|
| 1412 | (( maxcycle = maxcycle - 1 )) |
---|
| 1413 | if (( maxcycle > 0 )) |
---|
| 1414 | then |
---|
| 1415 | filename=${filename}.$maxcycle |
---|
| 1416 | fi |
---|
| 1417 | fi |
---|
| 1418 | |
---|
[1289] | 1419 | # STORE FILENAME WITHOUT PATH BUT WITH CYCLE NUMBER, |
---|
| 1420 | # BECAUSE IT MIGHT BE REQUIRED LATER TO STORE THE FILE ON AN ARCHIVE-SYSTEM |
---|
| 1421 | # OR TO PUT THIS FILENAME ON FILE OUTPUT_FILE_CONNECTIONS |
---|
[1] | 1422 | pathout[$i]=$filename |
---|
| 1423 | if (( maxcycle > 0 )) |
---|
| 1424 | then |
---|
| 1425 | frelout[$i]=${fname}${endout[$i]}.$maxcycle |
---|
| 1426 | else |
---|
| 1427 | frelout[$i]=${fname}${endout[$i]} |
---|
| 1428 | fi |
---|
| 1429 | |
---|
| 1430 | fi |
---|
| 1431 | done |
---|
| 1432 | |
---|
| 1433 | |
---|
[1289] | 1434 | # THE DVR-PACKAGE REQUIRES ITS OWN LIBRARY |
---|
[83] | 1435 | if [[ $(echo $package_list | grep -c dvrp_graphics) != 0 ]] |
---|
| 1436 | then |
---|
[260] | 1437 | if [[ "$dvr_inc" = "" ]] |
---|
[83] | 1438 | then |
---|
[260] | 1439 | printf "\n\n +++ no value for \"dvr_inc\" given in configuration file" |
---|
[83] | 1440 | printf "\n This is required for the dvrp_graphics package.\n" |
---|
[260] | 1441 | locat=dvr; exit |
---|
[83] | 1442 | fi |
---|
[260] | 1443 | if [[ "$dvr_lib" = "" ]] |
---|
[83] | 1444 | then |
---|
[260] | 1445 | printf "\n\n +++ no value for \"dvr_lib\" given in configuration file" |
---|
[83] | 1446 | printf "\n This is required for the dvrp_graphics package.\n" |
---|
[260] | 1447 | locat=dvr; exit |
---|
[83] | 1448 | fi |
---|
| 1449 | fi |
---|
| 1450 | |
---|
| 1451 | |
---|
[1289] | 1452 | # CHECK, WETHER A MAIN PROGRAM OR AN EXECUTABLE HAVE BEEN DECLARED. |
---|
| 1453 | # EXECUTABLES DO NOT NEED TO BE COMPILED. |
---|
[1] | 1454 | if [[ "$mainprog" = "" && "$executable" = "" ]] |
---|
| 1455 | then |
---|
| 1456 | printf "\n +++ neither main program nor executable defined" |
---|
| 1457 | locat=source; exit |
---|
| 1458 | elif [[ "$mainprog" != "" && "$executable" != "" ]] |
---|
| 1459 | then |
---|
| 1460 | printf "\n +++ main program as well as executable defined" |
---|
| 1461 | locat=source; exit |
---|
| 1462 | elif [[ "$mainprog" = "" && "$executable" != "" ]] |
---|
| 1463 | then |
---|
| 1464 | do_compile=false |
---|
| 1465 | fi |
---|
| 1466 | |
---|
| 1467 | |
---|
[1289] | 1468 | # CREATE SOURCE-DIRECTORY TO COLLECT ROUTINES TO BE COMPILED. |
---|
| 1469 | # LATER THE MRUN-SCRIPT AND CONFIGURATION-FILE WILL ALSO BE COPIED TO THIS DIRECTORY. |
---|
[193] | 1470 | if [[ $restart_run != true && "$SOURCES_COMPLETE" = "" ]] |
---|
[191] | 1471 | then |
---|
| 1472 | rm -rf SOURCES_FOR_RUN_$fname |
---|
| 1473 | mkdir SOURCES_FOR_RUN_$fname |
---|
| 1474 | fi |
---|
| 1475 | |
---|
| 1476 | |
---|
[1289] | 1477 | # COLLECT ALL ROUTINES TO BE COMPILED |
---|
| 1478 | # THIS IS NOT REQUIRED WITHIN BATCH-JOBS, BECAUSE ROUTINES HAVE ALREADY BEEN COLLECTED |
---|
| 1479 | # BY THE MRUN-CALL WHICH CREATED THE BATCH-JOB. |
---|
[1] | 1480 | if [[ $do_compile = true && "$SOURCES_COMPLETE" = "" ]] |
---|
| 1481 | then |
---|
| 1482 | |
---|
[5] | 1483 | [[ "$source_list" = LM ]] && source_list=LOCALLY_MODIFIED |
---|
[1] | 1484 | |
---|
[838] | 1485 | if [[ "$source_list" = LOCALLY_MODIFIED ]] |
---|
[1] | 1486 | then |
---|
| 1487 | |
---|
[1289] | 1488 | # DETERMINE MODIFIED FILES OF THE SVN WORKING COPY |
---|
[5] | 1489 | source_list="" |
---|
| 1490 | cd $source_path |
---|
| 1491 | |
---|
| 1492 | |
---|
[1289] | 1493 | # CHECK, IF DIRECTORY IS UNDER SVN CONTROL |
---|
[5] | 1494 | if [[ ! -d .svn ]] |
---|
| 1495 | then |
---|
| 1496 | printf "\n\n +++ source directory" |
---|
| 1497 | printf "\n \"$source_path\" " |
---|
| 1498 | printf "\n is not under control of \"subversion\"." |
---|
[306] | 1499 | printf "\n Please do not use mrun-option \"-s LOCALLY_MODIFIED\"\n" |
---|
[5] | 1500 | fi |
---|
| 1501 | |
---|
| 1502 | |
---|
[1289] | 1503 | # LIST ALL MODIFIED SOURCE CODE FILES |
---|
[5] | 1504 | Filenames="" |
---|
[69] | 1505 | svn status > tmp_mrun |
---|
| 1506 | while read line |
---|
[5] | 1507 | do |
---|
| 1508 | firstc=`echo $line | cut -c1` |
---|
| 1509 | if [[ $firstc = M || $firstc = "?" ]] |
---|
| 1510 | then |
---|
| 1511 | Name=`echo "$line" | cut -c8-` |
---|
| 1512 | extension=`echo $Name | cut -d. -f2` |
---|
| 1513 | if [[ "$extension" = f90 || "$extension" = F90 || "$extension" = f || "$extension" = F || "$extension" = c ]] |
---|
| 1514 | then |
---|
| 1515 | Filenames="$Filenames "$Name |
---|
| 1516 | fi |
---|
| 1517 | fi |
---|
[69] | 1518 | done < tmp_mrun |
---|
[5] | 1519 | |
---|
| 1520 | |
---|
[1289] | 1521 | # COPY FILES TO SOURCES_FOR_RUN_... |
---|
[5] | 1522 | for dateiname in $Filenames |
---|
| 1523 | do |
---|
| 1524 | cp $dateiname $working_directory/SOURCES_FOR_RUN_$fname |
---|
| 1525 | source_list=$source_list"$dateiname " |
---|
| 1526 | done |
---|
| 1527 | |
---|
| 1528 | cd - > /dev/null |
---|
| 1529 | |
---|
| 1530 | |
---|
[1289] | 1531 | # COPY FILES GIVEN BY OPTION -s TO DIRECTORY SOURCES_FOR_RUN_... |
---|
| 1532 | # AUTOMATIC RESTART RUNS JUST ACCESS THE DIRECTORY CREATED BY THE INITIAL RUN |
---|
[5] | 1533 | elif [[ "$source_list" != "" && $restart_run != true ]] |
---|
[1] | 1534 | then |
---|
| 1535 | |
---|
| 1536 | cd $source_path |
---|
| 1537 | |
---|
| 1538 | for filename in $source_list |
---|
| 1539 | do |
---|
| 1540 | |
---|
[1289] | 1541 | # SOURCE CODE FILE IS NOT ALLOWED TO INCLUDE PATH |
---|
[1103] | 1542 | if [[ $(echo $filename | grep -c "/") != 0 ]] |
---|
[1] | 1543 | then |
---|
| 1544 | printf "\n +++ source code file: $filename" |
---|
| 1545 | printf "\n must not contain (\"/\") " |
---|
| 1546 | locat=source; exit |
---|
| 1547 | fi |
---|
| 1548 | |
---|
| 1549 | if [[ ! -f $filename ]] |
---|
| 1550 | then |
---|
| 1551 | printf "\n +++ source code file: $filename" |
---|
| 1552 | printf "\n does not exist" |
---|
| 1553 | locat=source; exit |
---|
| 1554 | else |
---|
| 1555 | cp $filename $working_directory/SOURCES_FOR_RUN_$fname |
---|
| 1556 | fi |
---|
| 1557 | |
---|
| 1558 | done |
---|
| 1559 | |
---|
| 1560 | cd - > /dev/null |
---|
| 1561 | |
---|
| 1562 | fi |
---|
| 1563 | |
---|
| 1564 | |
---|
[1289] | 1565 | # CHECK, IF A MAIN PROGRAM EXISTS AND IF IT IS PART OF THE LIST OF FILES |
---|
| 1566 | # TO BE COMPILED. IF NOT, ADD IT TO THE LIST. |
---|
[5] | 1567 | if [[ $restart_run != true ]] |
---|
[1] | 1568 | then |
---|
| 1569 | |
---|
| 1570 | if [[ ! -f "$source_path/$mainprog" ]] |
---|
| 1571 | then |
---|
| 1572 | printf "\n\n +++ main program: $mainprog" |
---|
| 1573 | printf "\n does not exist in source directory" |
---|
| 1574 | printf "\n \"$source_path\"\n" |
---|
| 1575 | locat=source; exit |
---|
| 1576 | else |
---|
| 1577 | if [[ $(echo $source_list | grep -c $mainprog) = 0 ]] |
---|
| 1578 | then |
---|
| 1579 | cp $source_path/$mainprog SOURCES_FOR_RUN_$fname |
---|
| 1580 | source_list=${mainprog}" $source_list" |
---|
| 1581 | fi |
---|
| 1582 | fi |
---|
| 1583 | fi |
---|
| 1584 | |
---|
| 1585 | |
---|
[1289] | 1586 | # CHECK, IF MAKEFILE EXISTS AND COPY IT TO THE SOURCES_FOR_RUN... DIRECTORY |
---|
| 1587 | # NOT REQUIRED FOR RESTART RUNS, SOURCES_FOR_RUN... HAS BEEN CREATED BY THE INITIAL RUN |
---|
[211] | 1588 | if [[ "$restart_run" != true ]] |
---|
| 1589 | then |
---|
| 1590 | [[ "$makefile" = "" ]] && makefile=$source_path/Makefile |
---|
| 1591 | if [[ ! -f $makefile ]] |
---|
| 1592 | then |
---|
| 1593 | printf "\n +++ file \"$makefile\" does not exist" |
---|
| 1594 | locat=make; exit |
---|
| 1595 | else |
---|
| 1596 | cp $makefile SOURCES_FOR_RUN_$fname/Makefile |
---|
| 1597 | fi |
---|
| 1598 | fi |
---|
| 1599 | |
---|
| 1600 | |
---|
[1289] | 1601 | # COPY FILES FROM OPTIONAL SOURCE PATH GIVEN IN THE CONFIGURATION FILE |
---|
[5] | 1602 | if [[ $restart_run != true && "$add_source_path" != "" ]] |
---|
[1] | 1603 | then |
---|
| 1604 | |
---|
[1289] | 1605 | # DOES THE DIRECTORY EXIST? |
---|
[1] | 1606 | if [[ ! -d $add_source_path ]] |
---|
| 1607 | then |
---|
[1779] | 1608 | printf "\n\n *** INFORMATIVE: additional source code directory" |
---|
[1] | 1609 | printf "\n \"$add_source_path\" " |
---|
| 1610 | printf "\n does not exist or is not a directory." |
---|
| 1611 | printf "\n No source code will be used from this directory!\n" |
---|
| 1612 | add_source_path="" |
---|
[931] | 1613 | if [[ $silent == false ]] |
---|
| 1614 | then |
---|
| 1615 | sleep 3 |
---|
| 1616 | fi |
---|
[1] | 1617 | else |
---|
| 1618 | |
---|
| 1619 | cd $add_source_path |
---|
[211] | 1620 | found=false |
---|
[1] | 1621 | |
---|
| 1622 | Names=$(ls -1 *.f90 2>&1) |
---|
| 1623 | [[ $(echo $Names | grep -c '*.f90') = 0 ]] && AddFilenames="$Names" |
---|
| 1624 | Names=$(ls -1 *.F90 2>&1) |
---|
| 1625 | [[ $(echo $Names | grep -c '*.F90') = 0 ]] && AddFilenames="$AddFilenames $Names" |
---|
| 1626 | Names=$(ls -1 *.F 2>&1) |
---|
| 1627 | [[ $(echo $Names | grep -c '*.F') = 0 ]] && AddFilenames="$AddFilenames $Names" |
---|
| 1628 | Names=$(ls -1 *.f 2>&1) |
---|
| 1629 | [[ $(echo $Names | grep -c '*.f') = 0 ]] && AddFilenames="$AddFilenames $Names" |
---|
| 1630 | Names=$(ls -1 *.c 2>&1) |
---|
| 1631 | [[ $(echo $Names | grep -c '*.c') = 0 ]] && AddFilenames="$AddFilenames $Names" |
---|
| 1632 | |
---|
| 1633 | cd - > /dev/null |
---|
| 1634 | cd SOURCES_FOR_RUN_$fname |
---|
| 1635 | |
---|
[211] | 1636 | # COPY MAKEFILE IF EXISTING |
---|
| 1637 | if [[ -f $add_source_path/Makefile ]] |
---|
| 1638 | then |
---|
| 1639 | printf "\n\n *** user Makefile from directory" |
---|
| 1640 | printf "\n \"$add_source_path\" is used \n" |
---|
[931] | 1641 | if [[ $silent == false ]] |
---|
| 1642 | then |
---|
| 1643 | sleep 1 |
---|
| 1644 | fi |
---|
[211] | 1645 | cp $add_source_path/Makefile . |
---|
| 1646 | fi |
---|
| 1647 | |
---|
[1] | 1648 | for dateiname in $AddFilenames |
---|
| 1649 | do |
---|
| 1650 | if [[ -f $dateiname ]] |
---|
| 1651 | then |
---|
| 1652 | printf "\n +++ source code file \"$dateiname\" found in additional" |
---|
| 1653 | printf "\n source code directory \"$add_source_path\" " |
---|
[211] | 1654 | printf "\n but was also given with option \"-s\" which means that it should be taken" |
---|
| 1655 | printf "\n from directory \"$source_path\"." |
---|
[1] | 1656 | locat=source; exit |
---|
| 1657 | fi |
---|
| 1658 | |
---|
| 1659 | cp $add_source_path/$dateiname . |
---|
| 1660 | source_list="$source_list $dateiname" |
---|
| 1661 | |
---|
[211] | 1662 | # CHECK IF FILE IS CONTAINED IN MAKEFILE |
---|
| 1663 | if [[ $(grep -c $dateiname Makefile) = 0 ]] |
---|
| 1664 | then |
---|
| 1665 | printf "\n\n +++ user file \"$dateiname\" " |
---|
| 1666 | printf "\n is not listed in Makefile \n" |
---|
| 1667 | locat=source; exit |
---|
| 1668 | else |
---|
| 1669 | |
---|
| 1670 | if [[ $found = false ]] |
---|
[1] | 1671 | then |
---|
[211] | 1672 | found=true |
---|
| 1673 | printf "\n\n *** following user file(s) added to the" |
---|
| 1674 | printf " files to be translated:\n " |
---|
[1] | 1675 | fi |
---|
[211] | 1676 | printf "$dateiname " |
---|
[931] | 1677 | if [[ $silent == false ]] |
---|
| 1678 | then |
---|
| 1679 | sleep 0.5 |
---|
| 1680 | fi |
---|
| 1681 | |
---|
[1] | 1682 | fi |
---|
| 1683 | done |
---|
[211] | 1684 | [[ $found = true ]] && printf "\n" |
---|
[1] | 1685 | cd - > /dev/null |
---|
| 1686 | fi |
---|
| 1687 | fi |
---|
| 1688 | |
---|
| 1689 | |
---|
[1289] | 1690 | # ADD ALL ROUTINES BELONGING TO SOFTWARE PACKAGES (GIVEN BY OPTION -p) |
---|
| 1691 | # TO THE LIST OF FILES TO BE COMPILED |
---|
[5] | 1692 | if [[ $restart_run != true && -n $package_list ]] |
---|
[1] | 1693 | then |
---|
| 1694 | |
---|
| 1695 | cd $source_path |
---|
| 1696 | |
---|
| 1697 | for package in $package_list |
---|
| 1698 | do |
---|
| 1699 | |
---|
| 1700 | [[ $package = "dvrp_graphics+1PE" ]] && package=dvrp_graphics |
---|
| 1701 | |
---|
[1289] | 1702 | # DETERMINE FILES BELONGING TO THE PACKAGE |
---|
| 1703 | # ERROR MESSAGE ARE REDIRECTED TO /dev/null, BECAUSE WILDCARD (*) ALSO GIVES |
---|
| 1704 | # THE NAME OF THE DIRECTORY |
---|
[1] | 1705 | package_source_list=`grep "defined( __$package " * 2>/dev/null | cut -f1 -d:` |
---|
| 1706 | |
---|
| 1707 | |
---|
[1289] | 1708 | # ADD THESE FILES TO THE LIST OF FILES TO BE COMPILED, |
---|
| 1709 | # IF THEY ARE NOT ALREADY PART OF THE LIST |
---|
[1] | 1710 | for source_list_name in $package_source_list |
---|
| 1711 | do |
---|
| 1712 | if [[ $(echo $source_list | grep -c $source_list_name) = 0 ]] |
---|
| 1713 | then |
---|
| 1714 | |
---|
[1289] | 1715 | # ONLY TAKE FILES WITH VALID SUFFIX |
---|
[1] | 1716 | ending=`echo $source_list_name | cut -f2 -d.` |
---|
| 1717 | if [[ "$ending" = f90 || "$ending" = F90 || "$ending" = f || "$ending" = F || "$ending" = c ]] |
---|
| 1718 | then |
---|
| 1719 | cp $source_list_name $working_directory/SOURCES_FOR_RUN_$fname |
---|
| 1720 | source_list="$source_list $source_list_name" |
---|
| 1721 | fi |
---|
| 1722 | fi |
---|
| 1723 | done |
---|
| 1724 | done |
---|
| 1725 | |
---|
| 1726 | cd - > /dev/null |
---|
| 1727 | fi |
---|
| 1728 | |
---|
| 1729 | fi # do_compile=true |
---|
| 1730 | |
---|
| 1731 | |
---|
[1289] | 1732 | # IF SOURCE CODE IS TO BE COMPILED, DO SOME MORE CHECKS |
---|
| 1733 | # AND SET PRE-PROCESSOR DIRECTIVES |
---|
[1199] | 1734 | if [[ $do_compile = true || $create_executable_for_batch = true ]] |
---|
[1] | 1735 | then |
---|
| 1736 | |
---|
[1289] | 1737 | # SET PREPROCESSOR-DIRECTIVES TO SELECT OPERATING SYSTEM SPECIFIC CODE |
---|
[1] | 1738 | if [[ $(echo $localhost | cut -c1-3) = ibm ]] |
---|
| 1739 | then |
---|
[82] | 1740 | cpp_options="${cpp_options},-D__ibm=__ibm" |
---|
[1] | 1741 | elif [[ $(echo $localhost | cut -c1-3) = nec ]] |
---|
| 1742 | then |
---|
[82] | 1743 | cpp_options="$cpp_options -D__nec" |
---|
| 1744 | elif [[ $(echo $localhost | cut -c1-2) = lc ]] |
---|
| 1745 | then |
---|
| 1746 | cpp_options="$cpp_options -D__lc" |
---|
[1] | 1747 | else |
---|
[82] | 1748 | cpp_options="$cpp_options -D__$localhost" |
---|
[1] | 1749 | fi |
---|
[83] | 1750 | |
---|
[1289] | 1751 | # SET DIRECTIVES GIVEN BY OPTION -K (E.G. parallel) |
---|
[1] | 1752 | if [[ $(echo $localhost | cut -c1-3) = ibm ]] |
---|
| 1753 | then |
---|
[82] | 1754 | [[ -n $cond1 ]] && cpp_options="${cpp_options},-D__$cond1=__$cond1" |
---|
| 1755 | [[ -n $cond2 ]] && cpp_options="${cpp_options},-D__$cond2=__$cond2" |
---|
[1] | 1756 | else |
---|
[82] | 1757 | [[ -n $cond1 ]] && cpp_options="$cpp_options -D__$cond1" |
---|
| 1758 | [[ -n $cond2 ]] && cpp_options="$cpp_options -D__$cond2" |
---|
[1] | 1759 | fi |
---|
[83] | 1760 | |
---|
[1289] | 1761 | # SET DIRECTIVES FOR ACTIVATING SOFTWARE-PACKAGES (OPTION -p) |
---|
[1] | 1762 | if [[ -n $package_list ]] |
---|
| 1763 | then |
---|
| 1764 | for package in $package_list |
---|
| 1765 | do |
---|
| 1766 | if [[ $(echo $localhost | cut -c1-3) = ibm ]] |
---|
| 1767 | then |
---|
| 1768 | if [[ $package != "dvrp_graphics+1PE" ]] |
---|
| 1769 | then |
---|
[82] | 1770 | cpp_options="${cpp_options},-D__$package=__$package" |
---|
[1] | 1771 | else |
---|
[82] | 1772 | cpp_options="${cpp_options},-D__dvrp_graphics=__dvrp_graphics" |
---|
[1] | 1773 | export use_seperate_pe_for_dvrp_output=true |
---|
| 1774 | fi |
---|
| 1775 | else |
---|
| 1776 | if [[ $package != "dvrp_graphics+1PE" ]] |
---|
| 1777 | then |
---|
[82] | 1778 | cpp_options="$cpp_options -D__$package" |
---|
[1] | 1779 | else |
---|
[82] | 1780 | cpp_options="$cpp_options -D__dvrp_graphics" |
---|
[1] | 1781 | export use_seperate_pe_for_dvrp_output=true |
---|
| 1782 | fi |
---|
| 1783 | fi |
---|
| 1784 | done |
---|
| 1785 | fi |
---|
[83] | 1786 | |
---|
[1289] | 1787 | # SET DIRECTIVES GIVEN BY OPTION -D |
---|
[1] | 1788 | if [[ -n $cpp_opts ]] |
---|
| 1789 | then |
---|
| 1790 | for popts in $cpp_opts |
---|
| 1791 | do |
---|
[82] | 1792 | if [[ $(echo $localhost | cut -c1-3) = ibm ]] |
---|
| 1793 | then |
---|
| 1794 | cpp_options="${cpp_options},-D__$popts=__$popts" |
---|
| 1795 | else |
---|
| 1796 | cpp_options="$cpp_options -D__$popts" |
---|
| 1797 | fi |
---|
[1] | 1798 | done |
---|
| 1799 | fi |
---|
| 1800 | |
---|
| 1801 | else |
---|
| 1802 | |
---|
| 1803 | |
---|
[1289] | 1804 | # FOR LOCAL RUNS CHECK AGAIN, IF EXECUTABLE EXISTS |
---|
[1] | 1805 | if [[ $do_remote = false ]] |
---|
| 1806 | then |
---|
| 1807 | if [[ ! -f $executable ]] |
---|
| 1808 | then |
---|
| 1809 | printf "\n +++ executable file: $executable" |
---|
| 1810 | printf "\n does not exist" |
---|
| 1811 | locat=executable; exit |
---|
| 1812 | fi |
---|
| 1813 | fi |
---|
| 1814 | fi |
---|
| 1815 | |
---|
| 1816 | |
---|
[1289] | 1817 | # DETERMINE THE JOB MODE |
---|
[1491] | 1818 | if [[ $(echo $localhost | cut -c1-3) = ibm ]] |
---|
[1] | 1819 | then |
---|
[1491] | 1820 | if [[ "$LOADLBATCH" = yes ]] |
---|
| 1821 | then |
---|
| 1822 | batch_job=.TRUE. |
---|
| 1823 | jobmo=BATCH |
---|
| 1824 | else |
---|
| 1825 | batch_job=.FALSE. |
---|
| 1826 | jobmo=INTERACTIVE |
---|
| 1827 | fi |
---|
[1] | 1828 | else |
---|
[1491] | 1829 | if [[ "$ENVIRONMENT" = BATCH ]] |
---|
| 1830 | then |
---|
| 1831 | batch_job=.TRUE. |
---|
| 1832 | jobmo=BATCH |
---|
| 1833 | else |
---|
| 1834 | batch_job=.FALSE. |
---|
| 1835 | jobmo=INTERACTIVE |
---|
| 1836 | fi |
---|
[1] | 1837 | fi |
---|
| 1838 | |
---|
[1289] | 1839 | # NO INTERACTIVE RUNS ALLOWED ON LCTIT |
---|
[635] | 1840 | if [[ $host = lctit && $jobmo = INTERACTIVE && $do_batch = false ]] |
---|
| 1841 | then |
---|
| 1842 | printf "\n +++ no interactive runs allowed on host \"$host\" " |
---|
| 1843 | printf "\n please submit batch job using mrun option \"-b\" \n" |
---|
| 1844 | locat=normal; exit |
---|
| 1845 | fi |
---|
| 1846 | |
---|
| 1847 | |
---|
[1289] | 1848 | # CHECK, IF USER DEFINED A COMPILER |
---|
[1] | 1849 | if [[ "$compiler_name" = "" ]] |
---|
| 1850 | then |
---|
[82] | 1851 | |
---|
[220] | 1852 | printf "\n +++ no compiler specified for \"$host $cond1 $cond2\"" |
---|
[82] | 1853 | locat=compiler_name; exit |
---|
| 1854 | |
---|
[1] | 1855 | fi |
---|
| 1856 | |
---|
| 1857 | |
---|
[1289] | 1858 | # DETERMINE THE NAME OF MRUN'S TEMPORARY WORKING DIRECTORY |
---|
[1609] | 1859 | |
---|
| 1860 | # ON HLRN-III, USE THE QUEUING NAME. OTHERWISE USE USERNAME AND RANDOM NUMBER |
---|
| 1861 | if [[ $do_batch = false && $(echo $host | cut -c1-6) = lccray ]] |
---|
| 1862 | then |
---|
| 1863 | kennung=$(checkjob $PBS_JOBID | grep Reservation | cut -d" " -s -f2 | cut -d"." -s -f2 | sed "s/['\"]//g") |
---|
[1755] | 1864 | if [[ "$kennung" = "" ]] |
---|
[1753] | 1865 | then |
---|
| 1866 | kennung=$RANDOM |
---|
| 1867 | fi |
---|
[1609] | 1868 | else |
---|
| 1869 | kennung=$RANDOM |
---|
| 1870 | fi |
---|
| 1871 | |
---|
[1] | 1872 | if [[ "$tmp_user_catalog" = "" ]] |
---|
| 1873 | then |
---|
[312] | 1874 | if [[ $localhost = ibmh ]] |
---|
[1] | 1875 | then |
---|
[312] | 1876 | tmp_user_catalog=$SCRATCH |
---|
[1] | 1877 | elif [[ $localhost = nech ]] |
---|
| 1878 | then |
---|
| 1879 | tmp_user_catalog=$WRKSHR |
---|
| 1880 | else |
---|
| 1881 | tmp_user_catalog=/tmp |
---|
| 1882 | fi |
---|
| 1883 | fi |
---|
| 1884 | TEMPDIR=$tmp_user_catalog/${usern}.$kennung |
---|
| 1885 | |
---|
| 1886 | |
---|
[1289] | 1887 | # DETERMINE THE NAME OF THE DIRECTORY WHICH IS USED TO TEMPORARILY STORE DATA FOR RESTART RUNS |
---|
[1] | 1888 | if [[ "$tmp_data_catalog" = "" ]] |
---|
| 1889 | then |
---|
[622] | 1890 | if [[ $localhost = nech ]] |
---|
[1] | 1891 | then |
---|
| 1892 | tmp_data_catalog=$WRKSHR/mrun_restart_data |
---|
| 1893 | else |
---|
| 1894 | tmp_data_catalog=/tmp/mrun_restart_data |
---|
| 1895 | fi |
---|
| 1896 | fi |
---|
| 1897 | |
---|
| 1898 | |
---|
[1289] | 1899 | # IN CASE OF LOCAL RUNS REPLACE ENVIRONMENT VARIABLES BY THEIR VALUES |
---|
[1199] | 1900 | if [[ $do_remote = false && $do_compile = true || $create_executable_for_batch = true ]] |
---|
[1] | 1901 | then |
---|
| 1902 | eval fopts=\"$fopts\" |
---|
| 1903 | eval lopts=\"$lopts\" |
---|
| 1904 | fi |
---|
| 1905 | |
---|
| 1906 | |
---|
[1289] | 1907 | # DETERMINE COMPILE- AND LINK-OPTIONS |
---|
[1210] | 1908 | fopts="$fopts $netcdf_inc $fftw_inc $dvr_inc" |
---|
| 1909 | lopts="$lopts $netcdf_lib $fftw_lib $dvr_lib" |
---|
[1779] | 1910 | XOPT="-X $numprocs" |
---|
[1] | 1911 | |
---|
| 1912 | |
---|
[1289] | 1913 | # CHECK THE CPU-LIMIT. IT MUST BE GIVEN FOR BATCH-JOBS AND IS COMMUNICATED TO THE |
---|
| 1914 | # EXECUTABLE VIA NAMELIST-PARAMETER cputime |
---|
[1] | 1915 | done=false |
---|
| 1916 | while [[ $done = false ]] |
---|
| 1917 | do |
---|
| 1918 | cputime=$cpumax |
---|
| 1919 | if (( $cputime == 0 )) |
---|
| 1920 | then |
---|
[246] | 1921 | if [[ $do_batch = true ]] |
---|
| 1922 | then |
---|
| 1923 | printf "\n +++ cpu-time is undefined" |
---|
| 1924 | printf "\n >>> Please type CPU-time in seconds as INTEGER:" |
---|
| 1925 | printf "\n >>> " |
---|
| 1926 | read cputime 1>/dev/null 2>&1 |
---|
| 1927 | else |
---|
| 1928 | cputime=10000000 # NO CPU LIMIT FOR INTERACTIVE RUNS |
---|
| 1929 | fi |
---|
[1] | 1930 | else |
---|
| 1931 | done=true |
---|
| 1932 | fi |
---|
| 1933 | cpumax=$cputime |
---|
| 1934 | done |
---|
| 1935 | |
---|
| 1936 | (( minuten = cputime / 60 )) |
---|
| 1937 | (( sekunden = cputime - minuten * 60 )) |
---|
| 1938 | |
---|
| 1939 | |
---|
[1289] | 1940 | # CHECK THE MEMORY DEMAND |
---|
[1] | 1941 | if [[ $do_batch = true ]] |
---|
| 1942 | then |
---|
| 1943 | done=false |
---|
| 1944 | while [[ $done = false ]] |
---|
| 1945 | do |
---|
| 1946 | if (( memory == 0 )) |
---|
| 1947 | then |
---|
| 1948 | printf "\n +++ memory demand is undefined" |
---|
| 1949 | printf "\n >>> Please type memory in MByte per process as INTEGER:" |
---|
| 1950 | printf "\n >>> " |
---|
| 1951 | read memory 1>/dev/null 2>&1 |
---|
| 1952 | else |
---|
| 1953 | done=true |
---|
| 1954 | fi |
---|
| 1955 | done |
---|
| 1956 | fi |
---|
| 1957 | |
---|
| 1958 | |
---|
[1289] | 1959 | # IN CASE OF REMOTE-JOBS CHECK, IF A USERNAME FOR THE REMOTE HOST IS GIVEN |
---|
[1] | 1960 | if [[ $do_remote = true && -z $remote_username ]] |
---|
| 1961 | then |
---|
| 1962 | while [[ -z $remote_username ]] |
---|
| 1963 | do |
---|
| 1964 | printf "\n +++ username on remote host \"$host\" is undefined" |
---|
| 1965 | printf "\n >>> Please type username:" |
---|
| 1966 | printf "\n >>> " |
---|
| 1967 | read remote_username |
---|
| 1968 | done |
---|
| 1969 | mc="$mc -u$remote_username" |
---|
| 1970 | fi |
---|
| 1971 | |
---|
[892] | 1972 | # CHECK FOR INITIAL COMMANDS AFTER LOGIN |
---|
| 1973 | if [[ "$login_init_cmd" != "" ]] |
---|
| 1974 | then |
---|
| 1975 | export init_cmds="${login_init_cmd};" |
---|
| 1976 | fi |
---|
[221] | 1977 | |
---|
[892] | 1978 | |
---|
[1289] | 1979 | # SET THE MODULE-LOAD COMMAD AND EXPORT IT FOR subjob |
---|
[493] | 1980 | if [[ "$modules" != "" ]] |
---|
| 1981 | then |
---|
[678] | 1982 | if [[ $host = lctit ]] |
---|
| 1983 | then |
---|
| 1984 | export module_calls=". $modules" |
---|
| 1985 | else |
---|
| 1986 | export module_calls="module load ${modules};" |
---|
| 1987 | fi |
---|
[493] | 1988 | fi |
---|
| 1989 | |
---|
[1289] | 1990 | # OUTPUT OF THE MRUN-HEADER |
---|
[1] | 1991 | calltime=$(date) |
---|
| 1992 | printf "\n" |
---|
[1101] | 1993 | printf "#------------------------------------------------------------------------# \n" |
---|
| 1994 | printf "| %-35s%35s | \n" "$version" "$calltime" |
---|
| 1995 | printf "| | \n" |
---|
[1] | 1996 | spalte1="called on:"; spalte2=$localhost_realname |
---|
[1101] | 1997 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1289] | 1998 | if [[ $do_remote = true ]] |
---|
[1] | 1999 | then |
---|
[1289] | 2000 | spalte1="execution on:"; spalte2="$host (username: $remote_username)" |
---|
[1] | 2001 | else |
---|
[1289] | 2002 | spalte1="execution on:"; spalte2="$host ($localhost_realname)" |
---|
[1] | 2003 | fi |
---|
[1101] | 2004 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2005 | if [[ -n $numprocs ]] |
---|
| 2006 | then |
---|
[1106] | 2007 | if [[ $run_coupled_model = false ]] |
---|
| 2008 | then |
---|
| 2009 | spalte1="number of PEs:"; spalte2=$numprocs |
---|
| 2010 | else |
---|
| 2011 | spalte1="number of PEs:"; spalte2="$numprocs (atmosphere: $numprocs_atmos, ocean: $numprocs_ocean)" |
---|
| 2012 | fi |
---|
[1101] | 2013 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2014 | fi |
---|
| 2015 | if [[ -n $tasks_per_node ]] |
---|
| 2016 | then |
---|
| 2017 | spalte1="tasks per node:"; spalte2="$tasks_per_node (number of nodes: $nodes)" |
---|
[1101] | 2018 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1279] | 2019 | if (( remaining_pes > 0 )) |
---|
| 2020 | then |
---|
| 2021 | spalte1=" "; spalte2="one of the nodes only filled with $remaining_pes tasks" |
---|
| 2022 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2023 | fi |
---|
[1] | 2024 | fi |
---|
[759] | 2025 | if [[ $maximum_parallel_io_streams != $numprocs ]] |
---|
| 2026 | then |
---|
| 2027 | spalte1="max par io streams:"; spalte2="$maximum_parallel_io_streams" |
---|
[1101] | 2028 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[759] | 2029 | fi |
---|
[622] | 2030 | if [[ $use_openmp = true ]] |
---|
[1] | 2031 | then |
---|
| 2032 | spalte1="threads per task:"; spalte2="$threads_per_task" |
---|
[1101] | 2033 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2034 | fi |
---|
[1101] | 2035 | printf "| | \n" |
---|
[1] | 2036 | if [[ $do_compile = true ]] |
---|
| 2037 | then |
---|
[475] | 2038 | if [[ "$mopts" != "" ]] |
---|
| 2039 | then |
---|
[1101] | 2040 | spalte1="make options:"; spalte2=$(echo "$mopts" | cut -c-45) |
---|
| 2041 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2042 | zeile=$(echo "$mopts" | cut -c46-) |
---|
[475] | 2043 | while [[ "$zeile" != "" ]] |
---|
| 2044 | do |
---|
| 2045 | spalte1="" |
---|
[1101] | 2046 | spalte2=$(echo "$zeile" | cut -c-45) |
---|
| 2047 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2048 | zeile=$(echo "$zeile" | cut -c46-) |
---|
[475] | 2049 | done |
---|
| 2050 | fi |
---|
| 2051 | |
---|
[1101] | 2052 | spalte1="cpp directives:"; spalte2=$(echo "$cpp_options" | cut -c-45) |
---|
| 2053 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2054 | zeile=$(echo "$cpp_options" | cut -c46-) |
---|
[1] | 2055 | while [[ "$zeile" != "" ]] |
---|
| 2056 | do |
---|
| 2057 | spalte1="" |
---|
[1101] | 2058 | spalte2=$(echo "$zeile" | cut -c-45) |
---|
| 2059 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2060 | zeile=$(echo "$zeile" | cut -c46-) |
---|
[1] | 2061 | done |
---|
| 2062 | |
---|
[1101] | 2063 | spalte1="compiler options:"; spalte2=$(echo "$fopts" | cut -c-45) |
---|
| 2064 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2065 | zeile=$(echo "$fopts" | cut -c46-) |
---|
[1] | 2066 | while [[ "$zeile" != "" ]] |
---|
| 2067 | do |
---|
| 2068 | spalte1="" |
---|
[1101] | 2069 | spalte2=$(echo "$zeile" | cut -c-45) |
---|
| 2070 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2071 | zeile=$(echo "$zeile" | cut -c46-) |
---|
[1] | 2072 | done |
---|
| 2073 | |
---|
[1101] | 2074 | spalte1="linker options:"; spalte2=$(echo "$lopts" | cut -c-45) |
---|
| 2075 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2076 | zeile=$(echo "$lopts" | cut -c46-) |
---|
[1] | 2077 | while [[ "$zeile" != "" ]] |
---|
| 2078 | do |
---|
| 2079 | spalte1="" |
---|
[1101] | 2080 | spalte2=$(echo "$zeile" | cut -c-45) |
---|
| 2081 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2082 | zeile=$(echo "$zeile" | cut -c46-) |
---|
[1] | 2083 | done |
---|
| 2084 | |
---|
[1101] | 2085 | spalte1="modules to be load:"; spalte2=$(echo "$modules" | cut -c-45) |
---|
| 2086 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2087 | zeile=$(echo "$modules" | cut -c46-) |
---|
[493] | 2088 | while [[ "$zeile" != "" ]] |
---|
| 2089 | do |
---|
| 2090 | spalte1="" |
---|
[1101] | 2091 | spalte2=$(echo "$zeile" | cut -c-45) |
---|
| 2092 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2093 | zeile=$(echo "$zeile" | cut -c46-) |
---|
[493] | 2094 | done |
---|
| 2095 | |
---|
[1101] | 2096 | spalte1="main program:"; spalte2=$mainprog |
---|
| 2097 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2098 | else |
---|
| 2099 | spalte1=executable:; spalte2=$executable |
---|
[1101] | 2100 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2101 | fi |
---|
[1101] | 2102 | printf "| | \n" |
---|
| 2103 | spalte1="base name of files:"; spalte2=$fname |
---|
| 2104 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2105 | if [[ $fname != $afname ]] |
---|
| 2106 | then |
---|
[1101] | 2107 | spalte1="base name of input files:"; spalte2=$afname |
---|
| 2108 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2109 | fi |
---|
[1101] | 2110 | spalte1="INPUT control list:"; spalte2=$(echo $input_list) |
---|
| 2111 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
| 2112 | spalte1="OUTPUT control list:"; spalte2=$(echo $output_list) |
---|
| 2113 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2114 | |
---|
[1106] | 2115 | if [[ "$ocean_file_appendix" = true ]] |
---|
| 2116 | then |
---|
| 2117 | printf "| %-35s%-35s | \n" "suffix \"_O\" is added to local files" " " |
---|
| 2118 | fi |
---|
| 2119 | |
---|
[1] | 2120 | if [[ $do_batch = true || "$LOADLBATCH" = yes ]] |
---|
| 2121 | then |
---|
| 2122 | spalte1="memory demand / PE":; spalte2="$memory MB" |
---|
[1101] | 2123 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2124 | spalte1=CPU-time:; spalte2="$minuten:$sekunden" |
---|
[1101] | 2125 | printf "| %-25s%-45s | \n" "$spalte1" "$spalte2" |
---|
[1] | 2126 | fi |
---|
| 2127 | |
---|
| 2128 | if [[ $do_compile = true ]] |
---|
| 2129 | then |
---|
[1101] | 2130 | printf "| | \n" |
---|
| 2131 | printf "| Files to be compiled: | \n" |
---|
[1] | 2132 | zeile=$source_list |
---|
| 2133 | while [[ "$zeile" != "" ]] |
---|
| 2134 | do |
---|
[1101] | 2135 | linestart=$(echo $zeile | cut -c-70) |
---|
| 2136 | printf "| %-70s | \n" "$linestart" |
---|
| 2137 | zeile=$(echo "$zeile" | cut -c71-) |
---|
[1] | 2138 | done |
---|
| 2139 | fi |
---|
[1101] | 2140 | printf "#------------------------------------------------------------------------#" |
---|
[1] | 2141 | |
---|
| 2142 | |
---|
| 2143 | |
---|
[1289] | 2144 | # OUTPUT OF FILE CONNECTIONS IN CASE OF TRACEBACK |
---|
[1] | 2145 | if [[ $do_trace = true ]] |
---|
| 2146 | then |
---|
| 2147 | (( i = 0 )) |
---|
| 2148 | while (( i < iin )) |
---|
| 2149 | do |
---|
| 2150 | (( i = i + 1 )) |
---|
| 2151 | if (( i == 1 )) |
---|
| 2152 | then |
---|
| 2153 | printf "\n\n >>> INPUT-file assignments:\n" |
---|
| 2154 | fi |
---|
| 2155 | printf "\n ${localin[$i]} : ${absnamein[$i]}" |
---|
| 2156 | done |
---|
| 2157 | (( i = 0 )) |
---|
| 2158 | while (( i < iout )) |
---|
| 2159 | do |
---|
| 2160 | (( i = i + 1 )) |
---|
| 2161 | if (( i == 1 )) |
---|
| 2162 | then |
---|
| 2163 | printf "\n\n >>> OUTPUT-file assignments:\n" |
---|
| 2164 | fi |
---|
| 2165 | printf "\n ${localout[$i]} : ${pathout[$i]}" |
---|
| 2166 | done |
---|
| 2167 | (( i = 0 )) |
---|
| 2168 | while (( i < iic )) |
---|
| 2169 | do |
---|
| 2170 | (( i = i + 1 )) |
---|
| 2171 | if (( i == 1 )) |
---|
| 2172 | then |
---|
| 2173 | printf "\n\n >>> INPUT-commands:\n" |
---|
| 2174 | fi |
---|
| 2175 | printf "\n ${in_command[$i]}" |
---|
| 2176 | done |
---|
| 2177 | (( i = 0 )) |
---|
| 2178 | while (( i < ioc )) |
---|
| 2179 | do |
---|
| 2180 | (( i = i + 1 )) |
---|
| 2181 | if (( i == 1 )) |
---|
| 2182 | then |
---|
| 2183 | printf "\n\n >>> OUTPUT-commands:\n" |
---|
| 2184 | fi |
---|
| 2185 | printf "\n ${out_command[$i]}" |
---|
| 2186 | done |
---|
| 2187 | fi |
---|
| 2188 | |
---|
[1289] | 2189 | # QUERY FOR CONTINUE (ON LOCAL MACHINES ONLY) |
---|
[1] | 2190 | if [[ $remotecall = false && $silent = false && $jobmo != BATCH ]] |
---|
| 2191 | then |
---|
| 2192 | antwort=dummy |
---|
| 2193 | printf "\n\n" |
---|
[1101] | 2194 | printf " >>> everything o.k. (y/n) ? " |
---|
| 2195 | while read antwort |
---|
[1] | 2196 | do |
---|
[1101] | 2197 | if [[ "$antwort" != y && "$antwort" != Y && "$antwort" != n && "$antwort" != N ]] |
---|
| 2198 | then |
---|
| 2199 | printf " >>> everything o.k. (y/n) ? " |
---|
| 2200 | else |
---|
| 2201 | break |
---|
| 2202 | fi |
---|
[1] | 2203 | done |
---|
| 2204 | if [[ $antwort = n || $antwort = N ]] |
---|
| 2205 | then |
---|
| 2206 | locat=user_abort; (( iec = 0 )); exit |
---|
| 2207 | fi |
---|
| 2208 | if [[ $do_batch = true ]] |
---|
| 2209 | then |
---|
| 2210 | printf " >>> batch-job will be created and submitted" |
---|
| 2211 | else |
---|
[1289] | 2212 | printf " >>> MRUN will now continue to execute on this machine" |
---|
[1] | 2213 | fi |
---|
| 2214 | fi |
---|
| 2215 | |
---|
[1804] | 2216 | |
---|
[1199] | 2217 | # DETERMINE PATH FOR MAKE DEPOSITORY |
---|
| 2218 | if [[ $do_batch = false || $create_executable_for_batch = true ]] |
---|
| 2219 | then |
---|
| 2220 | |
---|
| 2221 | line="" |
---|
| 2222 | grep "%depository_path" $config_file > tmp_mrun |
---|
| 2223 | while read line |
---|
| 2224 | do |
---|
| 2225 | if [[ "$line" != "" && $(echo $line | cut -c1) != "#" ]] |
---|
| 2226 | then |
---|
| 2227 | if [[ "$(echo $line | cut -d" " -s -f3)" = "" ]] |
---|
| 2228 | then |
---|
| 2229 | global_depository_path=`echo $line | cut -d" " -s -f2` |
---|
| 2230 | fi |
---|
| 2231 | fi |
---|
| 2232 | done < tmp_mrun |
---|
| 2233 | |
---|
| 2234 | line="" |
---|
| 2235 | grep " $localhost" $config_file | grep "%depository_path" > tmp_mrun |
---|
| 2236 | while read line |
---|
| 2237 | do |
---|
| 2238 | if [[ "$line" != "" && $(echo $line | cut -c1) != "#" ]] |
---|
| 2239 | then |
---|
| 2240 | if [[ "$(echo $line | cut -d" " -s -f4)" = "$cond1" && "$(echo $line | cut -d" " -s -f5)" = "$cond2" ]] |
---|
| 2241 | then |
---|
| 2242 | local_depository_path=`echo $line | cut -d" " -s -f2` |
---|
| 2243 | fi |
---|
| 2244 | fi |
---|
| 2245 | done < tmp_mrun |
---|
| 2246 | |
---|
| 2247 | if [[ "$local_depository_path" = "" ]] |
---|
| 2248 | then |
---|
| 2249 | if [[ "$global_depository_path" != "" ]] |
---|
| 2250 | then |
---|
| 2251 | local_depository_path=$global_depository_path |
---|
| 2252 | else |
---|
| 2253 | printf "\n\n +++ no depository path found in configuration file" |
---|
| 2254 | printf "\n for local host \"$localhost\" " |
---|
| 2255 | printf "\n please set \"\%depository_path\" in configuration file\n" |
---|
| 2256 | locat=config_file; exit |
---|
| 2257 | fi |
---|
| 2258 | fi |
---|
| 2259 | eval local_depository_path=$local_depository_path |
---|
| 2260 | [[ "$cond1" != "" ]] && local_depository_path=${local_depository_path}_$cond1 |
---|
| 2261 | [[ "$cond2" != "" ]] && local_depository_path=${local_depository_path}_$cond2 |
---|
| 2262 | |
---|
| 2263 | |
---|
| 2264 | basename=`echo $mainprog | cut -f1 -d"."` |
---|
| 2265 | eval make_depository=${local_depository_path}/${basename}_current_version.tar |
---|
| 2266 | if [[ ! -f $make_depository ]] |
---|
| 2267 | then |
---|
| 2268 | printf "\n" |
---|
| 2269 | printf "\n *** WARNING: make depository \"$make_depository\" not found" |
---|
| 2270 | printf "\n \"make\" will fail, if the Makefile or other source files are missing\n" |
---|
| 2271 | fi |
---|
| 2272 | |
---|
| 2273 | fi |
---|
| 2274 | |
---|
| 2275 | |
---|
[1289] | 2276 | # NOW PERFORM THOSE ACTIONS REQUIRED TO EXECUTE THE PROGRAM (PALM) ON THIS MACHINE |
---|
| 2277 | # (COMPILING/LINKING, EXECUTING, COPYING I/O FILES) |
---|
[1] | 2278 | if [[ $do_batch = false ]] |
---|
| 2279 | then |
---|
| 2280 | |
---|
| 2281 | |
---|
[1289] | 2282 | # CREATE THE TEMPORARY WORKING DIRECTORY |
---|
[1] | 2283 | mkdir -p $TEMPDIR |
---|
| 2284 | chmod go+rx $TEMPDIR |
---|
| 2285 | tmpcreate=true |
---|
| 2286 | |
---|
| 2287 | |
---|
[1289] | 2288 | # COPY EITHER THE COMPLETE SOURCE CODE FILES TO BE COMPILED OR THE EXECUTABLE |
---|
| 2289 | # INTO THE TEMPORARY WORKING DIRECTORY |
---|
[1] | 2290 | if [[ $do_compile = true ]] |
---|
| 2291 | then |
---|
| 2292 | |
---|
| 2293 | # ON NEC, COMPILATION IS DONE ON HOST CROSS VIA CROSS COMPILING |
---|
| 2294 | # CREATE A TEMPORARY DIRECTORY ON THAT MACHINE (HOME MOUNTED VIA NFS) |
---|
| 2295 | if [[ $localhost = nech ]] |
---|
| 2296 | then |
---|
| 2297 | TEMPDIR_COMPILE=$HOME/work/${usern}.$kennung |
---|
| 2298 | if mkdir -p $TEMPDIR_COMPILE |
---|
| 2299 | then |
---|
| 2300 | printf "\n *** \"$TEMPDIR_COMPILE\" " |
---|
| 2301 | printf "\n is generated as temporary directory for cross compiling\n" |
---|
| 2302 | else |
---|
| 2303 | printf "\n +++ creating directory \"$TEMPDIR_COMPILE\" " |
---|
| 2304 | printf "\n needed for cross compilation failed" |
---|
| 2305 | locat=compile |
---|
| 2306 | exit |
---|
| 2307 | fi |
---|
| 2308 | else |
---|
| 2309 | TEMPDIR_COMPILE=$TEMPDIR |
---|
| 2310 | fi |
---|
| 2311 | |
---|
[1199] | 2312 | cp $make_depository $TEMPDIR_COMPILE |
---|
| 2313 | cd $TEMPDIR_COMPILE |
---|
| 2314 | tar -xf $make_depository > /dev/null 2>&1 |
---|
| 2315 | cd - > /dev/null |
---|
[1] | 2316 | |
---|
[1255] | 2317 | # DUE TO UNKNOWN REASONS, COPY WITH cp COMMAND CREATES CORRUPT |
---|
| 2318 | # FILES ON CRAY XC30 SYSTEMS (HLRN III), rsync IS USED INSTEAD |
---|
[1281] | 2319 | if [[ $(echo $host | cut -c1-6) = lccray ]] |
---|
| 2320 | then |
---|
| 2321 | rsync -av -t SOURCES_FOR_RUN_$fname/* $TEMPDIR_COMPILE > /dev/null |
---|
| 2322 | else |
---|
| 2323 | cp SOURCES_FOR_RUN_$fname/* $TEMPDIR_COMPILE > /dev/null |
---|
| 2324 | fi |
---|
[82] | 2325 | |
---|
[1] | 2326 | else |
---|
[475] | 2327 | |
---|
[1] | 2328 | cp $executable ${TEMPDIR}/a.out |
---|
[475] | 2329 | |
---|
[1] | 2330 | fi |
---|
| 2331 | |
---|
| 2332 | |
---|
[1289] | 2333 | # CHANGE TO THE TEMPORARY WORKING DIRECTORY |
---|
[475] | 2334 | cd $TEMPDIR |
---|
[1] | 2335 | printf "\n *** changed to temporary directory: $TEMPDIR" |
---|
| 2336 | |
---|
| 2337 | |
---|
[1289] | 2338 | # THE FOLLOWING IS REQUIRED FPR AVS-OUTPUT WITH PALM ??? REMOVE ??? |
---|
| 2339 | # WRITE INFORMATIONS ABOUT THE OUTPUT-FILE CONNECTIONS TO A TEMPORARY FILE |
---|
| 2340 | # THIS FILE CAN LATER BE READ FROM USER-DEFINED CODE TO DETERMINE THE |
---|
| 2341 | # RELATION BETWEEN THE LOCAL TEMPORARY AND PERMANENT FILE NAMES |
---|
[1] | 2342 | (( i = 0 )) |
---|
| 2343 | while (( i < iout )) |
---|
| 2344 | do |
---|
| 2345 | (( i = i + 1 )) |
---|
| 2346 | if [[ "${actionout[$i]}" = tr || "${actionout[$i]}" = tra || "${actionout[$i]}" = trpe ]] |
---|
| 2347 | then |
---|
| 2348 | printf "${localout[$i]} ${actionout[$i]}\n${pathout[$i]}\n${localhost}_${fname}${endout[$i]}\n" >> OUTPUT_FILE_CONNECTIONS |
---|
| 2349 | else |
---|
| 2350 | printf "${localout[$i]} ${actionout[$i]}\n${pathout[$i]}\n${frelout[$i]}\n" >> OUTPUT_FILE_CONNECTIONS |
---|
| 2351 | fi |
---|
| 2352 | done |
---|
| 2353 | |
---|
| 2354 | |
---|
[1289] | 2355 | # IF REQUIRED, START WITH COMPILING |
---|
[1] | 2356 | if [[ $do_compile = true ]] |
---|
| 2357 | then |
---|
| 2358 | |
---|
| 2359 | |
---|
[1199] | 2360 | if [[ -f a.out ]] |
---|
[475] | 2361 | then |
---|
[82] | 2362 | |
---|
[1199] | 2363 | # EXECUTABLE WAS CREATED DURING INTERACTIVE CALL OF MRUN |
---|
| 2364 | printf "\n\n\n *** executable a.out found" |
---|
| 2365 | printf "\n no compilation required \n" |
---|
| 2366 | |
---|
[1] | 2367 | else |
---|
[807] | 2368 | |
---|
[1199] | 2369 | # COMPILING WITH MAKE (ON NEC COMPILER IS CALLED ON HOST CROSS) |
---|
[1289] | 2370 | printf "\n\n\n *** compilation starts \n$dashes\n" |
---|
[1199] | 2371 | printf " *** compilation with make using following options:\n" |
---|
| 2372 | printf " make depository: $make_depository" |
---|
| 2373 | if [[ "$mopts" != "" ]] |
---|
| 2374 | then |
---|
| 2375 | printf " make options: $mopts\n" |
---|
| 2376 | fi |
---|
| 2377 | printf " compilername: $compiler_name\n" |
---|
| 2378 | printf " compiler options: $fopts\n" |
---|
| 2379 | printf " preprocessor directives: $cpp_options \n" |
---|
| 2380 | printf " linker options: $lopts \n" |
---|
| 2381 | if [[ "$modules" != "" ]] |
---|
| 2382 | then |
---|
| 2383 | printf " modules to be load: $modules \n" |
---|
| 2384 | fi |
---|
| 2385 | printf " source code files: $source_list \n" |
---|
[1] | 2386 | |
---|
[1199] | 2387 | if [[ $localhost = nech ]] |
---|
| 2388 | then |
---|
| 2389 | ssh $SSH_PORTOPT 136.172.44.192 -l $usern "$init_cmds $module_calls cd \$HOME/work/${usern}.$kennung; sxmake $mopts -f Makefile PROG=a.out F90=$compiler_name COPT=\"$cpp_options\" F90FLAGS=\"$fopts\" LDFLAGS=\"$lopts\" " |
---|
| 2390 | cp $TEMPDIR_COMPILE/a.out . |
---|
| 2391 | [[ $? != 0 ]] && compile_error=true |
---|
| 2392 | rm -rf $TEMPDIR_COMPILE |
---|
| 2393 | elif [[ $localhost = ibmh ]] |
---|
| 2394 | then |
---|
| 2395 | printf " compiler is called via ssh on \"plogin1\" \n" |
---|
| 2396 | ssh $SSH_PORTOPT plogin1 -l $usern "$init_cmds export PATH=/sw/ibm/xlf/13.1.0.8/usr/bin:$PATH; $module_calls cd $TEMPDIR; make $mopts -f Makefile PROG=a.out F90=$compiler_name COPT=\"$cpp_options\" F90FLAGS=\"$fopts\" LDFLAGS=\"$lopts\" " |
---|
| 2397 | [[ ! -f a.out ]] && compile_error=true |
---|
[1289] | 2398 | continue # STATUS=1, IF a.out EXISTS |
---|
[1866] | 2399 | elif [[ $localhost = lcocean ]] |
---|
| 2400 | then |
---|
| 2401 | printf " compiler is called via ssh on \"ocean\" \n" |
---|
| 2402 | echo $PWD |
---|
| 2403 | ssh $SSH_PORTOPT ocean -l $usern "$init_cmds $module_calls cd $TEMPDIR; make $mopts -f Makefile PROG=a.out F90=$compiler_name COPT=\"$cpp_options\" F90FLAGS=\"$fopts\" LDFLAGS=\"$lopts\" " |
---|
| 2404 | [[ ! -f a.out ]] && compile_error=true |
---|
| 2405 | continue # STATUS=1, IF a.out EXISTS |
---|
[1199] | 2406 | elif [[ $localhost = lcflow ]] |
---|
| 2407 | then |
---|
| 2408 | printf " compiler is called via ssh on \"flow\" \n" |
---|
[1201] | 2409 | /usr/bin/ssh $SSH_PORTOPT flow02.hpc.uni-oldenburg.de -l $usern "$init_cmds $module_calls cd $TEMPDIR; make $mopts -f Makefile PROG=a.out F90=$compiler_name COPT=\"$cpp_options\" F90FLAGS=\"$fopts\" LDFLAGS=\"$lopts\" " |
---|
[1272] | 2410 | check_for_file=`/usr/bin/ssh $SSH_PORTOPT flow02.hpc.uni-oldenburg.de -l $usern "ls $TEMPDIR/a.out 2> /dev/null"` |
---|
| 2411 | [[ "$check_for_file" = "" ]] && compile_error=true |
---|
[1289] | 2412 | continue # STATUS=1, IF a.out EXISTS |
---|
[1620] | 2413 | elif [[ $localhost = lcbullhh || $localhost = lccrayb || $localhost = lccrayf || $localhost = lccrayh ]] |
---|
[1199] | 2414 | then |
---|
| 2415 | make $mopts -f Makefile PROG=a.out F90=$compiler_name COPT="$cpp_options" F90FLAGS="$fopts" LDFLAGS="$lopts" |
---|
| 2416 | else |
---|
| 2417 | [[ "$init_cmds" != "" ]] && eval $init_cmds |
---|
| 2418 | [[ "$module_calls" != "" ]] && eval $module_calls |
---|
| 2419 | make $mopts -f Makefile PROG=a.out F90=$compiler_name COPT="$cpp_options" F90FLAGS="$fopts" LDFLAGS="$lopts" |
---|
| 2420 | fi |
---|
[181] | 2421 | |
---|
[1199] | 2422 | if [[ $? != 0 || "$compile_error" = true || "$module_compile_error" = true ]] |
---|
| 2423 | then |
---|
| 2424 | printf "\n +++ error occured while compiling or linking" |
---|
| 2425 | locat=compile |
---|
| 2426 | |
---|
[181] | 2427 | exit |
---|
[1199] | 2428 | else |
---|
[1289] | 2429 | printf "$dashes\n *** compilation finished \n" |
---|
[1199] | 2430 | fi |
---|
| 2431 | |
---|
[1] | 2432 | fi |
---|
[1199] | 2433 | |
---|
[1] | 2434 | fi |
---|
| 2435 | |
---|
| 2436 | |
---|
[1289] | 2437 | # PROVIDE THE INPUT FILES |
---|
| 2438 | # LOOP OVER ALL ACTIVATED FILES (LISTED IN THE CONFIGURATION FILE) |
---|
[1] | 2439 | (( i = 0 )) |
---|
| 2440 | while (( i < iin )) |
---|
| 2441 | do |
---|
| 2442 | (( i = i + 1 )) |
---|
| 2443 | if (( i == 1 )) |
---|
| 2444 | then |
---|
[1289] | 2445 | printf "\n\n *** providing INPUT-files:\n$dashes" |
---|
[1] | 2446 | fi |
---|
| 2447 | |
---|
| 2448 | |
---|
[1289] | 2449 | # SKIP OPTIONAL FILES, IF THEY DO NOT EXIST |
---|
[1] | 2450 | if [[ "${transin[$i]}" = unavailable ]] |
---|
| 2451 | then |
---|
| 2452 | if [[ "${extin[$i]}" = "" || "${extin[$i]}" = " " ]] |
---|
| 2453 | then |
---|
[1779] | 2454 | printf "\n *** INFORMATIVE: input file \"${pathin[$i]}/${afname}${endin[$i]}\" " |
---|
| 2455 | printf "\n is not available!" |
---|
[1] | 2456 | else |
---|
[1779] | 2457 | printf "\n *** INFORMATIVE: input file \"${pathin[$i]}/${afname}${endin[$i]}.${extin[$i]}\" " |
---|
| 2458 | printf "\n is not available!" |
---|
[1] | 2459 | fi |
---|
| 2460 | continue |
---|
| 2461 | fi |
---|
| 2462 | |
---|
[1289] | 2463 | # CHECK FOR SINGLE FILE (SERIAL RUN) OR DIRECTORY (ONE FILE PER CORE FOR PARELLEL EXECUTION) |
---|
[1] | 2464 | files_for_pes=false; datentyp=file |
---|
| 2465 | if [[ "${actionin[$i]}" = pe && -n $numprocs ]] |
---|
| 2466 | then |
---|
| 2467 | files_for_pes=true; datentyp=directory |
---|
| 2468 | actionin[$i]="" |
---|
| 2469 | elif [[ "${actionin[$i]}" = pe && ! -n $numprocs ]] |
---|
| 2470 | then |
---|
| 2471 | actionin[$i]="" |
---|
| 2472 | elif [[ "${actionin[$i]}" = arpe && -n $numprocs ]] |
---|
| 2473 | then |
---|
| 2474 | files_for_pes=true; datentyp=directory |
---|
| 2475 | actionin[$i]="ar" |
---|
| 2476 | elif [[ "${actionin[$i]}" = arpe && ! -n $numprocs ]] |
---|
| 2477 | then |
---|
| 2478 | actionin[$i]="ar" |
---|
| 2479 | elif [[ "${actionin[$i]}" = flpe && -n $numprocs ]] |
---|
| 2480 | then |
---|
| 2481 | files_for_pes=true; datentyp=directory |
---|
| 2482 | actionin[$i]="fl" |
---|
| 2483 | elif [[ "${actionin[$i]}" = flpe && ! -n $numprocs ]] |
---|
| 2484 | then |
---|
| 2485 | actionin[$i]="fl" |
---|
| 2486 | fi |
---|
| 2487 | |
---|
| 2488 | if [[ $files_for_pes = true ]] |
---|
| 2489 | then |
---|
| 2490 | printf "\n >>> INPUT: ${absnamein[$i]}/.... to ${localin[$i]}" |
---|
| 2491 | else |
---|
| 2492 | printf "\n >>> INPUT: ${absnamein[$i]} to ${localin[$i]}" |
---|
| 2493 | fi |
---|
| 2494 | |
---|
[1289] | 2495 | # INPUT-FILE FOR A RESTART RUN. |
---|
| 2496 | # CHECK, IF THIS FILE STILL EXISTS ON THE TEMPORARY DATA DIRECTORY. |
---|
| 2497 | # IF NOT, TRY TO GET IT FROM THE ARCHIVE SYSTEM |
---|
[1] | 2498 | if [[ "${actionin[$i]}" = fl ]] |
---|
| 2499 | then |
---|
| 2500 | printf "\n $datentyp will be fetched from temporary directory \"${tmp_data_catalog}\" !" |
---|
| 2501 | if [[ $files_for_pes = false ]] |
---|
| 2502 | then |
---|
| 2503 | if [[ -f "$tmp_data_catalog/${frelin[$i]}" ]] |
---|
| 2504 | then |
---|
| 2505 | ln $tmp_data_catalog/${frelin[$i]} ${localin[$i]} |
---|
| 2506 | got_tmp[$i]=true |
---|
| 2507 | elif [[ -f "$WORK/${frelin[$i]}" && $ignore_archive_error = true ]] |
---|
| 2508 | then |
---|
| 2509 | printf "\n +++ $datentyp not found in \"$tmp_data_catalog\" !" |
---|
| 2510 | printf "\n *** trying to use backup copy in \"$WORK\" " |
---|
| 2511 | cp $WORK/${frelin[$i]} ${localin[$i]} |
---|
| 2512 | else |
---|
| 2513 | printf "\n +++ $datentyp not found in \"$tmp_data_catalog\" " |
---|
| 2514 | printf "\n or \"$tmp_data_catalog\" does not exist!" |
---|
| 2515 | printf "\n *** trying to get copy from archive" |
---|
| 2516 | actionin[$i]=ar |
---|
| 2517 | fi |
---|
| 2518 | else |
---|
| 2519 | if [[ -d "$tmp_data_catalog/${frelin[$i]}" ]] |
---|
| 2520 | then |
---|
| 2521 | mkdir ${localin[$i]} |
---|
[149] | 2522 | cd $tmp_data_catalog/${frelin[$i]} |
---|
| 2523 | for file in $(ls *) |
---|
| 2524 | do |
---|
| 2525 | ln $file $TEMPDIR/${localin[$i]} |
---|
| 2526 | done |
---|
| 2527 | cd $TEMPDIR |
---|
[1] | 2528 | got_tmp[$i]=true |
---|
| 2529 | elif [[ -d "$WORK/${frelin[$i]}" && $ignore_archive_error = true ]] |
---|
| 2530 | then |
---|
| 2531 | printf "\n +++ $datentyp not found in \"$tmp_data_catalog\" !" |
---|
| 2532 | printf "\n *** trying to use backup copy in \"$WORK\" " |
---|
| 2533 | cp -r $WORK/${frelin[$i]} ${localin[$i]} |
---|
| 2534 | else |
---|
| 2535 | printf "\n +++ $datentyp not found in \"$tmp_data_catalog\" " |
---|
| 2536 | printf "\n or \"$tmp_data_catalog\" does not exist!" |
---|
| 2537 | printf "\n *** trying to get copy from archive" |
---|
| 2538 | actionin[$i]=ar |
---|
| 2539 | fi |
---|
| 2540 | fi |
---|
| 2541 | fi |
---|
| 2542 | |
---|
| 2543 | |
---|
[1289] | 2544 | # FILE IS STORED ON THE ARCHIVE SYSTEM |
---|
[1] | 2545 | if [[ "${actionin[$i]}" = ar ]] |
---|
| 2546 | then |
---|
| 2547 | |
---|
| 2548 | if [[ $files_for_pes = false ]] |
---|
| 2549 | then |
---|
| 2550 | printf "\n file will be restored from archive-system ($archive_system)!" |
---|
| 2551 | else |
---|
| 2552 | printf "\n directory will be restored from archive-system ($archive_system)!" |
---|
| 2553 | fi |
---|
| 2554 | |
---|
| 2555 | file_restored=false |
---|
| 2556 | |
---|
| 2557 | if [[ $archive_system = asterix ]] |
---|
| 2558 | then |
---|
| 2559 | do_stagein=true |
---|
| 2560 | (( stagein_anz = 0 )) |
---|
| 2561 | while [[ $do_stagein = true ]] |
---|
| 2562 | do |
---|
| 2563 | if [[ $files_for_pes = false ]] |
---|
| 2564 | then |
---|
| 2565 | stagein -O ${frelin[$i]} > STAGEIN_OUTPUT |
---|
| 2566 | else |
---|
| 2567 | stagein -t -O ${frelin[$i]} > STAGEIN_OUTPUT |
---|
| 2568 | fi |
---|
| 2569 | cat STAGEIN_OUTPUT |
---|
| 2570 | if [[ $(grep -c "st.msg:i24" STAGEIN_OUTPUT) != 0 ]] |
---|
| 2571 | then |
---|
| 2572 | file_restored=true |
---|
| 2573 | do_stagein=false |
---|
| 2574 | else |
---|
| 2575 | (( stagein_anz = stagein_anz + 1 )) |
---|
| 2576 | if (( stagein_anz == 10 )) |
---|
| 2577 | then |
---|
| 2578 | printf "\n +++ stagein stoped after 10 tries" |
---|
| 2579 | locat=stage |
---|
| 2580 | exit |
---|
| 2581 | fi |
---|
| 2582 | printf "\n +++ restoring from archive failed, trying again:" |
---|
| 2583 | sleep 900 |
---|
| 2584 | fi |
---|
| 2585 | done |
---|
| 2586 | elif [[ $archive_system = DMF ]] |
---|
| 2587 | then |
---|
| 2588 | if [[ $files_for_pes = false ]] |
---|
| 2589 | then |
---|
| 2590 | printf "\n +++ restoring of single files impossible with $archive_system !\n" |
---|
| 2591 | locat=DMF |
---|
| 2592 | exit |
---|
| 2593 | else |
---|
| 2594 | find $ARCHIVE/${frelin[$i]} -type m -print | dmget |
---|
| 2595 | cp -r $ARCHIVE/${frelin[$i]} $PWD |
---|
| 2596 | file_restored=true |
---|
| 2597 | fi |
---|
| 2598 | elif [[ $archive_system = tivoli ]] |
---|
| 2599 | then |
---|
| 2600 | if [[ $files_for_pes = false ]] |
---|
| 2601 | then |
---|
[1289] | 2602 | ssh $SSH_PORTOPT $usern@bicedata.hlrn.de "cp $PERM/${frelin[$i]} $PWD" |
---|
[1] | 2603 | else |
---|
| 2604 | (( inode = 0 )) |
---|
| 2605 | while (( inode < nodes )) |
---|
| 2606 | do |
---|
[1289] | 2607 | ssh $SSH_PORTOPT $usern@bicedata.hlrn.de "cd $PWD; tar xf $PERM/${frelin[$i]}/${frelin[$i]}.node_$inode.tar" |
---|
[1] | 2608 | (( inode = inode + 1 )) |
---|
| 2609 | done |
---|
| 2610 | fi |
---|
| 2611 | file_restored=true |
---|
| 2612 | elif [[ $archive_system = ut ]] |
---|
| 2613 | then |
---|
| 2614 | if [[ $files_for_pes = false ]] |
---|
| 2615 | then |
---|
| 2616 | cp $UT/${frelin[$i]} . |
---|
| 2617 | else |
---|
| 2618 | (( inode = 0 )) |
---|
| 2619 | while (( inode < nodes )) |
---|
| 2620 | do |
---|
| 2621 | tar xf $UT/${frelin[$i]}/${frelin[$i]}.node_$inode.tar |
---|
| 2622 | (( inode = inode + 1 )) |
---|
| 2623 | done |
---|
| 2624 | fi |
---|
| 2625 | file_restored=true |
---|
| 2626 | else |
---|
| 2627 | printf "\n +++ archive_system=\"$archive_system\" restore impossible!" |
---|
| 2628 | locat=rearchive |
---|
| 2629 | exit |
---|
| 2630 | fi |
---|
| 2631 | |
---|
| 2632 | if [[ $file_restored = true ]] |
---|
| 2633 | then |
---|
| 2634 | |
---|
[1289] | 2635 | # PUT FILE ON THE TEMPORARY DATA DIRECTORY TOO. |
---|
| 2636 | # THIS WILL MAKE FILE ACCESS EASIER FOR LATER RUNS |
---|
[1] | 2637 | [[ ! -d $tmp_data_catalog ]] && mkdir -p $tmp_data_catalog; chmod g+rx $tmp_data_catalog |
---|
| 2638 | if [[ $files_for_pes = false ]] |
---|
| 2639 | then |
---|
| 2640 | ln -f ${frelin[$i]} $tmp_data_catalog/${frelin[$i]} |
---|
| 2641 | else |
---|
| 2642 | mkdir $tmp_data_catalog/${frelin[$i]} |
---|
| 2643 | ln -f ${frelin[$i]}/* $tmp_data_catalog/${frelin[$i]} |
---|
| 2644 | fi |
---|
| 2645 | got_tmp[$i]=true |
---|
| 2646 | |
---|
[1289] | 2647 | # PROVIDE THE FILE UNDER ITS LOCAL NAME |
---|
[1] | 2648 | mv ${frelin[$i]} ${localin[$i]} |
---|
| 2649 | |
---|
| 2650 | fi |
---|
| 2651 | fi |
---|
| 2652 | |
---|
| 2653 | |
---|
[1289] | 2654 | # FILE IS STORED IN THE RESPECTIVE DIRECTORY GIVEN IN THE CONFIGURATION FILE |
---|
[1] | 2655 | if [[ "${actionin[$i]}" = "" || "${actionin[$i]}" = "di" || "${actionin[$i]}" = "npe" ]] |
---|
| 2656 | then |
---|
| 2657 | |
---|
| 2658 | if [[ "${actionin[$i]}" = "npe" && -n $numprocs ]] |
---|
| 2659 | then |
---|
| 2660 | |
---|
[1289] | 2661 | # FILE COPIES ARE PROVIDED FOR ALL CORES |
---|
| 2662 | # EACH FILE GETS A UNIQUE FILENAME WITH A FOUR DIGIT NUMBER |
---|
[1] | 2663 | printf "\n file will be provided for $numprocs processors" |
---|
| 2664 | mkdir ${localin[$i]} |
---|
| 2665 | ival=$numprocs |
---|
| 2666 | (( ii = 0 )) |
---|
| 2667 | while (( ii <= ival-1 )) |
---|
| 2668 | do |
---|
| 2669 | if (( ii < 10 )) |
---|
| 2670 | then |
---|
| 2671 | cp ${absnamein[$i]} ${localin[$i]}/_000$ii |
---|
| 2672 | elif (( ii < 100 )) |
---|
| 2673 | then |
---|
| 2674 | cp ${absnamein[$i]} ${localin[$i]}/_00$ii |
---|
| 2675 | elif (( ii < 1000 )) |
---|
| 2676 | then |
---|
| 2677 | cp ${absnamein[$i]} ${localin[$i]}/_0$ii |
---|
| 2678 | else |
---|
| 2679 | cp ${absnamein[$i]} ${localin[$i]}/_$ii |
---|
| 2680 | fi |
---|
| 2681 | (( ii = ii + 1 )) |
---|
| 2682 | done |
---|
| 2683 | |
---|
| 2684 | else |
---|
| 2685 | |
---|
| 2686 | if [[ $files_for_pes = true ]] |
---|
| 2687 | then |
---|
| 2688 | |
---|
[1289] | 2689 | # PROVIDE FILES FOR EACH CORE |
---|
| 2690 | # FIRST CREATE THE LOCAL DIRECTORY, THEN COPY FILES |
---|
| 2691 | # FROM THE PERMANENT DIRECTORY BY LINKING THEM TO THE LOCAL ONE |
---|
[1] | 2692 | printf "\n providing $numprocs files for the respective processors" |
---|
| 2693 | mkdir ${localin[$i]} |
---|
[183] | 2694 | if [[ $link_local_input = true ]] |
---|
| 2695 | then |
---|
| 2696 | printf " using ln -f\n" |
---|
| 2697 | cd ${absnamein[$i]} |
---|
| 2698 | for file in $(ls *) |
---|
| 2699 | do |
---|
| 2700 | ln -f $file ${localin[$i]} |
---|
| 2701 | done |
---|
| 2702 | cd $TEMPDIR |
---|
| 2703 | fi |
---|
[1289] | 2704 | |
---|
| 2705 | # IF "ln -f" FAILED OR IF "$link_local_input = false" DO A NORMAL "cp -r" |
---|
[183] | 2706 | if [[ ! -f "${localin[$i]}/_0000" ]] |
---|
| 2707 | then |
---|
| 2708 | if [[ $link_local_input = true ]] |
---|
| 2709 | then |
---|
| 2710 | printf " ln failed for .../_0000, using cp...\n" |
---|
| 2711 | fi |
---|
| 2712 | cp -r ${absnamein[$i]}/* ${localin[$i]} |
---|
| 2713 | fi |
---|
[1] | 2714 | |
---|
| 2715 | else |
---|
[1289] | 2716 | |
---|
| 2717 | # PROVIDE FILE FOR RUNS ON A SINGLE CORE |
---|
[183] | 2718 | if [[ $link_local_input = true ]] |
---|
| 2719 | then |
---|
| 2720 | printf " using ln -f\n" |
---|
| 2721 | ln -f ${absnamein[$i]} ${localin[$i]} |
---|
| 2722 | fi |
---|
| 2723 | # If "ln -f" fails of if "$link_local_input = false" do a normal "cp" |
---|
| 2724 | if [[ ! -f "${localin[$i]}" ]] |
---|
| 2725 | then |
---|
| 2726 | if [[ $link_local_input = true ]] |
---|
| 2727 | then |
---|
| 2728 | printf " ln failed, using cp...\n" |
---|
| 2729 | fi |
---|
| 2730 | cp ${absnamein[$i]} ${localin[$i]} |
---|
| 2731 | fi |
---|
[1] | 2732 | fi |
---|
| 2733 | fi |
---|
| 2734 | fi |
---|
| 2735 | |
---|
| 2736 | done |
---|
| 2737 | if (( i != 0 )) |
---|
| 2738 | then |
---|
[1289] | 2739 | printf "\n$dashes\n *** all INPUT-files provided \n" |
---|
[1] | 2740 | fi |
---|
| 2741 | |
---|
| 2742 | |
---|
[1289] | 2743 | # EXECUTE INPUT-COMMANDS GIVEN IN THE CONFIGURATION FILE |
---|
[1] | 2744 | (( i = 0 )) |
---|
| 2745 | while (( i < iic )) |
---|
| 2746 | do |
---|
| 2747 | (( i = i + 1 )) |
---|
| 2748 | if (( i == 1 )) |
---|
| 2749 | then |
---|
[1289] | 2750 | printf "\n\n *** execution of INPUT-commands:\n$dashes" |
---|
[1] | 2751 | fi |
---|
| 2752 | printf "\n >>> ${in_command[$i]}" |
---|
| 2753 | eval ${in_command[$i]} |
---|
| 2754 | if (( i == iic )) |
---|
| 2755 | then |
---|
[1289] | 2756 | printf "\n$dashes\n" |
---|
[1] | 2757 | fi |
---|
| 2758 | done |
---|
| 2759 | |
---|
| 2760 | |
---|
[1289] | 2761 | # SET THE REMAINING CPU-TIME |
---|
[1] | 2762 | cpurest=${cpumax}. |
---|
| 2763 | |
---|
| 2764 | |
---|
[260] | 2765 | # START DVR STREAMING SERVER |
---|
| 2766 | if [[ $(echo $package_list | grep -c dvrp_graphics) != 0 ]] |
---|
| 2767 | then |
---|
| 2768 | if [[ "$dvr_server" != "" ]] |
---|
| 2769 | then |
---|
| 2770 | |
---|
| 2771 | printf "\n\n *** preparing the dvr streaming server configuration file" |
---|
| 2772 | |
---|
[1289] | 2773 | # CHECK, IF A DVR SERVER IS ALREADY RUNNING |
---|
[260] | 2774 | running_dvrserver_id=`echo $(ps -edaf | grep .dvrserver.config | grep -v grep) | cut -d" " -f2` |
---|
| 2775 | if [[ "$running_dvrserver_id" != "" ]] |
---|
| 2776 | then |
---|
| 2777 | |
---|
[261] | 2778 | printf "\n\n +++ WARNING: A dvr server with id=$running_dvrserver_id is already running!" |
---|
[260] | 2779 | printf "\n This server is used instead starting a new one!" |
---|
[261] | 2780 | printf "\n If required, script \"process_dvr_output\" has to be run manually." |
---|
[260] | 2781 | |
---|
| 2782 | else |
---|
| 2783 | |
---|
| 2784 | # COPY CONFIGURATION FILE FOR STREAMING SERVER FROM REPOSITORY TO HERE |
---|
| 2785 | if [[ -f ${PALM_BIN}/.dvrserver.config ]] |
---|
| 2786 | then |
---|
| 2787 | cp ${PALM_BIN}/.dvrserver.config . |
---|
| 2788 | |
---|
[1289] | 2789 | # ENTERING THE BASEDIR, UID AND GID INTO THIS FILE |
---|
[260] | 2790 | user_id=`id -u` |
---|
| 2791 | group_id=`id -g` |
---|
[1289] | 2792 | |
---|
| 2793 | # "&" IS REQUIRED AS A SEPERATOR, BECAUSE TEMPDIR CONTAINS "/" |
---|
[260] | 2794 | sed "s&<replace by dvr data directory>&${TEMPDIR}&g" .dvrserver.config > .dvrserver.1 |
---|
| 2795 | sed "s/<replace by user id>/$user_id/g" .dvrserver.1 > .dvrserver.2 |
---|
| 2796 | sed "s/<replace by group id>/$group_id/g" .dvrserver.2 > .dvrserver.3 |
---|
| 2797 | mv .dvrserver.3 .dvrserver.config |
---|
| 2798 | rm .dvrserver.1 .dvrserver.2 |
---|
| 2799 | |
---|
[1289] | 2800 | # START DVR SERVER IN BACKGROUND, GET HIS ID AND PRINT ON TERMINAL |
---|
[260] | 2801 | $dvr_server .dvrserver.config >> DVR_LOGFILE 2>&1 & |
---|
| 2802 | dvrserver_id=`echo $(ps -edaf | grep .dvrserver.config) | cut -d" " -f2` |
---|
| 2803 | printf "\n *** streaming server with id=$dvrserver_id is started in background" |
---|
| 2804 | local_dvrserver_running=.TRUE. |
---|
| 2805 | else |
---|
| 2806 | printf "\n +++ missing file \".dvrserver.config\" in directory:" |
---|
| 2807 | printf "\n \"$PALM_BIN\" " |
---|
| 2808 | locat=dvr |
---|
| 2809 | exit |
---|
| 2810 | fi |
---|
| 2811 | |
---|
| 2812 | fi |
---|
| 2813 | |
---|
| 2814 | else |
---|
| 2815 | printf "\n\n --- INFORMATIVE: no dvr streaming server will be started" |
---|
| 2816 | fi |
---|
| 2817 | fi |
---|
| 2818 | |
---|
| 2819 | |
---|
[1289] | 2820 | # CREATE THE NAMELIST-FILE WITH VALUES OF ENVIRONMENT-VARIABLES REQUIRED BY PALM |
---|
| 2821 | # (FILE ENVPAR WILL BE READ BY PALM) |
---|
[260] | 2822 | cat > ENVPAR << %%END%% |
---|
| 2823 | &envpar run_identifier = '$fname', host = '$localhost', |
---|
| 2824 | write_binary = '$write_binary', tasks_per_node = $tasks_per_node, |
---|
[759] | 2825 | maximum_parallel_io_streams = $maximum_parallel_io_streams, |
---|
[260] | 2826 | maximum_cpu_time_allowed = ${cpumax}., |
---|
| 2827 | revision = '$global_revision', |
---|
[1402] | 2828 | local_dvrserver_running = $local_dvrserver_running, |
---|
| 2829 | batch_job = $batch_job / |
---|
[260] | 2830 | |
---|
| 2831 | %%END%% |
---|
| 2832 | |
---|
| 2833 | |
---|
[1289] | 2834 | # STARTING THE EXECUTABLE |
---|
| 2835 | printf "\n\n *** execution starts in directory\n \"`pwd`\"\n$dashes\n" |
---|
[1] | 2836 | PATH=$PATH:$TEMPDIR |
---|
| 2837 | |
---|
[1779] | 2838 | |
---|
| 2839 | if [[ $execute_command != "none" ]] |
---|
[359] | 2840 | then |
---|
| 2841 | |
---|
[1779] | 2842 | printf "\n +++ branch still not realized" |
---|
| 2843 | locat=execution |
---|
| 2844 | exit |
---|
| 2845 | |
---|
[693] | 2846 | else |
---|
| 2847 | |
---|
[1779] | 2848 | # MPI DEBUG OPTION (ARGUMENT CHECKING, SLOWS DOWN EXECUTION DUE TO INCREASED LATENCY) |
---|
| 2849 | if [[ "$mpi_debug" = true ]] |
---|
[1549] | 2850 | then |
---|
[1779] | 2851 | export MPI_CHECK_ARGS=1 |
---|
| 2852 | printf "\n +++ MPI_CHECK_ARGS=$MPI_CHECK_ARGS" |
---|
[1549] | 2853 | fi |
---|
| 2854 | |
---|
[1779] | 2855 | if [[ "$totalview" = true ]] |
---|
[1] | 2856 | then |
---|
[1779] | 2857 | printf "\n *** totalview debugger will be used" |
---|
| 2858 | tv_opt="-tv" |
---|
| 2859 | else |
---|
| 2860 | tv_opt="" |
---|
[1] | 2861 | fi |
---|
[1779] | 2862 | |
---|
| 2863 | if [[ "$cond1" = debug || "$cond2" = debug ]] |
---|
[1] | 2864 | then |
---|
| 2865 | |
---|
[1779] | 2866 | #Interactive ALLINEA DEBUG seesion |
---|
| 2867 | if [[ "$ENVIRONMENT" != BATCH ]] |
---|
| 2868 | then |
---|
| 2869 | if [[ $host = lccrayb || $host = lccrayh ]] |
---|
| 2870 | then |
---|
| 2871 | if [[ "$allinea" = true ]] |
---|
| 2872 | then |
---|
| 2873 | echo "--- aprun -n $ii -N $tasks_per_node a.out < runfile_atmos" |
---|
[1] | 2874 | |
---|
[1779] | 2875 | ddt aprun -n $ii -N $tasks_per_node a.out |
---|
| 2876 | wait |
---|
| 2877 | fi |
---|
| 2878 | fi |
---|
| 2879 | fi |
---|
[1] | 2880 | |
---|
[1779] | 2881 | if [[ $localhost = ibmh ]] |
---|
| 2882 | then |
---|
[1] | 2883 | |
---|
[1779] | 2884 | # SETUP THE IBM MPI ENVIRONMENT |
---|
| 2885 | export MP_SHARED_MEMORY=yes |
---|
| 2886 | export AIXTHREADS_SCOPE=S |
---|
| 2887 | export OMP_NUM_THREADS=$threads_per_task |
---|
| 2888 | export AUTHSTATE=files |
---|
| 2889 | export XLFRTEOPTS="nlwidth=132:err_recovery=no" # RECORD-LENGTH OF NAMELIST-OUTPUT |
---|
[1] | 2890 | |
---|
[1779] | 2891 | # FOLLOWING OPTIONS ARE MANDATORY FOR TOTALVIEW |
---|
| 2892 | export MP_ADAPTER_USE=shared |
---|
| 2893 | export MP_CPU_USE=multiple |
---|
| 2894 | export MP_TIMEOUT=1200 |
---|
[622] | 2895 | |
---|
[1779] | 2896 | unset MP_TASK_AFFINITY |
---|
[1295] | 2897 | |
---|
[1779] | 2898 | if [[ "$LOADLBATCH" = yes ]] |
---|
| 2899 | then |
---|
| 2900 | totalview poe a.out |
---|
| 2901 | else |
---|
| 2902 | echo totalview poe -a a.out -procs $numprocs -rmpool 0 -nodes 1 |
---|
| 2903 | export TVDSVRLAUNCHCMD=ssh |
---|
| 2904 | totalview poe -a a.out -procs $numprocs -rmpool 0 -nodes 1 |
---|
| 2905 | fi |
---|
[622] | 2906 | |
---|
[1779] | 2907 | fi # END DEBUG MODE |
---|
[1289] | 2908 | |
---|
[1779] | 2909 | else |
---|
| 2910 | |
---|
| 2911 | # NORMAL EXECUTION |
---|
| 2912 | if [[ -n $numprocs ]] |
---|
[1] | 2913 | then |
---|
[1289] | 2914 | |
---|
[1779] | 2915 | # RUNNING THE PROGRAM ON PARALLEL MACHINES |
---|
| 2916 | if [[ $(echo $host | cut -c1-3) = ibm ]] |
---|
[312] | 2917 | then |
---|
[1] | 2918 | |
---|
[1779] | 2919 | # SETUP THE IBM MPI ENVIRONMENT |
---|
| 2920 | if [[ $host != ibmh && $host != ibmkisti ]] |
---|
[696] | 2921 | then |
---|
[1779] | 2922 | export MP_SHARED_MEMORY=yes |
---|
| 2923 | export AIXTHREAD_SCOPE=S |
---|
[696] | 2924 | export OMP_NUM_THREADS=$threads_per_task |
---|
[1779] | 2925 | export XLSMPOPTS="spins=0:yields=0:stack=20000000" |
---|
| 2926 | export AUTHSTATE=files |
---|
| 2927 | export XLFRTEOPTS="nlwidth=132:err_recovery=no" # RECORD-LENGTH OF NAMELIST-OUTPUT |
---|
| 2928 | # export MP_PRINTENV=yes |
---|
| 2929 | |
---|
| 2930 | # TUNING-VARIABLES TO IMPROVE COMMUNICATION SPEED |
---|
| 2931 | # DO NOT SHOW SIGNIFICANT EFFECTS (SEP 04, FEDERATION-SWITCH) |
---|
| 2932 | export MP_WAIT_MODE=poll |
---|
| 2933 | [[ $node_usage = not_shared ]] && export MP_SINGLE_THREAD=yes |
---|
[696] | 2934 | fi |
---|
[693] | 2935 | |
---|
[1779] | 2936 | if [[ $host = ibmkisti ]] |
---|
[1] | 2937 | then |
---|
[1779] | 2938 | export LANG=en_US |
---|
| 2939 | export MP_SHARED_MEMORY=yes |
---|
| 2940 | if [[ $threads_per_task = 1 ]] |
---|
[343] | 2941 | then |
---|
[1779] | 2942 | export MP_SINGLE_THREAD=yes |
---|
| 2943 | export MEMORY_AFFINITY=MCM |
---|
[343] | 2944 | else |
---|
[1779] | 2945 | export OMP_NUM_THREADS=$threads_per_task |
---|
[343] | 2946 | fi |
---|
[1] | 2947 | fi |
---|
[1779] | 2948 | |
---|
| 2949 | if [[ "$LOADLBATCH" = yes ]] |
---|
[1] | 2950 | then |
---|
[1779] | 2951 | printf "\n--- Control: OMP_NUM_THREADS = \"$OMP_NUM_THREADS\" \n" |
---|
| 2952 | if [[ "$cond1" = hpmcount || "$cond2" = hpmcount ]] |
---|
[1] | 2953 | then |
---|
[1779] | 2954 | /opt/optibm/HPM_2_4_1/bin/hpmcount a.out |
---|
[1] | 2955 | else |
---|
[1779] | 2956 | if [[ $run_coupled_model = false ]] |
---|
| 2957 | then |
---|
| 2958 | if [[ "$ocean_file_appendix" = true ]] |
---|
| 2959 | then |
---|
| 2960 | echo "precursor_ocean" > runfile_atmos |
---|
| 2961 | else |
---|
| 2962 | echo "precursor_atmos" > runfile_atmos |
---|
| 2963 | fi |
---|
| 2964 | else |
---|
| 2965 | (( iia = $numprocs_atmos / $threads_per_task )) |
---|
| 2966 | (( iio = $numprocs_ocean / $threads_per_task )) |
---|
| 2967 | printf "\n coupled run ($iia atmosphere, $iio ocean)" |
---|
| 2968 | printf "\n using $coupled_mode coupling" |
---|
| 2969 | printf "\n\n" |
---|
| 2970 | echo "coupled_run $iia $iio" > runfile_atmos |
---|
| 2971 | fi |
---|
| 2972 | poe ./a.out < runfile_atmos |
---|
[1] | 2973 | fi |
---|
[1779] | 2974 | else |
---|
| 2975 | if [[ $localhost = ibmh || $localhost = ibms ]] |
---|
[331] | 2976 | then |
---|
[1779] | 2977 | poe a.out -procs $numprocs -nodes 1 -rmpool 0 |
---|
| 2978 | elif [[ $localhost = ibmkisti || $localhost = ibmku ]] |
---|
| 2979 | then |
---|
| 2980 | if [[ -f $hostfile ]] |
---|
[331] | 2981 | then |
---|
[1779] | 2982 | cp $hostfile hostfile |
---|
[331] | 2983 | else |
---|
[1779] | 2984 | (( ii = 1 )) |
---|
| 2985 | while (( ii <= $numprocs )) |
---|
| 2986 | do |
---|
| 2987 | echo $localhost_realname >> hostfile |
---|
| 2988 | (( ii = ii + 1 )) |
---|
| 2989 | done |
---|
[331] | 2990 | fi |
---|
[1779] | 2991 | export MP_HOSTFILE=hostfile |
---|
| 2992 | if [[ $run_coupled_model = false ]] |
---|
| 2993 | then |
---|
| 2994 | if [[ "$ocean_file_appendix" = true ]] |
---|
| 2995 | then |
---|
| 2996 | echo "precursor_ocean" > runfile_atmos |
---|
| 2997 | else |
---|
| 2998 | echo "precursor_atmos" > runfile_atmos |
---|
| 2999 | fi |
---|
| 3000 | else |
---|
| 3001 | (( iia = $numprocs_atmos / $threads_per_task )) |
---|
| 3002 | (( iio = $numprocs_ocean / $threads_per_task )) |
---|
| 3003 | printf "\n coupled run ($iia atmosphere, $iio ocean)" |
---|
| 3004 | printf "\n using $coupled_mode coupling" |
---|
| 3005 | printf "\n\n" |
---|
| 3006 | echo "coupled_run $iia $iio" > runfile_atmos |
---|
| 3007 | fi |
---|
[331] | 3008 | |
---|
[1779] | 3009 | poe ./a.out -procs $numprocs < runfile_atmos |
---|
[1289] | 3010 | |
---|
[1] | 3011 | else |
---|
[1779] | 3012 | if [[ "$host_file" = "" ]] |
---|
| 3013 | then |
---|
| 3014 | printf "\n +++ no hostfile given in configuration file" |
---|
| 3015 | locat=config_file |
---|
| 3016 | exit |
---|
| 3017 | else |
---|
| 3018 | eval host_file=$host_file |
---|
| 3019 | fi |
---|
| 3020 | export MP_HOSTFILE=$host_file |
---|
| 3021 | poe a.out -procs $numprocs -tasks_per_node $numprocs |
---|
[1] | 3022 | fi |
---|
| 3023 | fi |
---|
[1779] | 3024 | |
---|
| 3025 | elif [[ $host = nech ]] # running on NEC machines |
---|
[1] | 3026 | then |
---|
| 3027 | |
---|
[1779] | 3028 | (( ii = nodes )) |
---|
| 3029 | if [[ $ii = 1 ]] |
---|
| 3030 | then |
---|
| 3031 | export F_ERRCNT=0 # acceptable number of errors before program is stopped |
---|
| 3032 | export MPIPROGINF=YES |
---|
| 3033 | # export F_TRACE=YES|FMT1|FMT2 # output of ftrace informations to job protocol |
---|
| 3034 | echo "*** execution on single node with mpirun" |
---|
| 3035 | mpirun -np $numprocs ./a.out |
---|
| 3036 | else |
---|
| 3037 | (( i = 0 )) |
---|
| 3038 | while (( i < ii )) |
---|
| 3039 | do |
---|
| 3040 | echo "-h $i -p $tasks_per_node -e ./mpi_exec_shell" >> multinode_config |
---|
| 3041 | (( i = i + 1 )) |
---|
| 3042 | done |
---|
[1] | 3043 | |
---|
[1779] | 3044 | echo "#!/bin/sh" > mpi_exec_shell |
---|
| 3045 | echo " " >> mpi_exec_shell |
---|
| 3046 | echo "set -u" >> mpi_exec_shell |
---|
| 3047 | echo "F_ERRCNT=0" >> mpi_exec_shell |
---|
| 3048 | echo "MPIPROGINV=YES" >> mpi_exec_shell |
---|
| 3049 | echo "OMP_NUM_THREADS=$threads_per_task" >> mpi_exec_shell |
---|
| 3050 | echo "cpurest=$cpurest" >> mpi_exec_shell |
---|
| 3051 | echo "fname=$fname" >> mpi_exec_shell |
---|
| 3052 | echo "localhost=$localhost" >> mpi_exec_shell |
---|
| 3053 | echo "return_address=$return_address" >> mpi_exec_shell |
---|
| 3054 | echo "return_username=$return_username" >> mpi_exec_shell |
---|
| 3055 | echo "tasks_per_node=$tasks_per_node" >> mpi_exec_shell |
---|
| 3056 | echo "write_binary=$write_binary" >> mpi_exec_shell |
---|
| 3057 | echo "use_seperate_pe_for_dvrp_output=$use_seperate_pe_for_dvrp_output" >> mpi_exec_shell |
---|
| 3058 | echo " " >> mpi_exec_shell |
---|
| 3059 | echo "export F_ERRCNT" >> mpi_exec_shell |
---|
| 3060 | echo "export MPIPROGINV" >> mpi_exec_shell |
---|
| 3061 | echo "export OMP_NUM_THREADS" >> mpi_exec_shell |
---|
| 3062 | echo "export cpurest" >> mpi_exec_shell |
---|
| 3063 | echo "export fname" >> mpi_exec_shell |
---|
| 3064 | echo "export localhost" >> mpi_exec_shell |
---|
| 3065 | echo "export return_address" >> mpi_exec_shell |
---|
| 3066 | echo "export return_username" >> mpi_exec_shell |
---|
| 3067 | echo "export tasks_per_node" >> mpi_exec_shell |
---|
| 3068 | echo "export write_binary" >> mpi_exec_shell |
---|
| 3069 | echo "export use_seperate_pe_for_dvrp_output" >> mpi_exec_shell |
---|
| 3070 | echo " " >> mpi_exec_shell |
---|
| 3071 | echo "exec ./a.out" >> mpi_exec_shell |
---|
[1] | 3072 | |
---|
[1779] | 3073 | chmod u+x mpi_exec_shell |
---|
| 3074 | export MPIPROGINF=YES |
---|
| 3075 | mpirun -f multinode_config & |
---|
| 3076 | wait |
---|
[1] | 3077 | |
---|
[1779] | 3078 | fi |
---|
[1622] | 3079 | |
---|
[1779] | 3080 | elif [[ $(echo $host | cut -c1-2) = lc && $host != lckyoto && $host != lctit ]] |
---|
[1] | 3081 | then |
---|
[1779] | 3082 | |
---|
| 3083 | # COPY HOSTFILE FROM SOURCE DIRECTORY OR CREATE IT, IF IT |
---|
| 3084 | # DOES NOT EXIST |
---|
[1866] | 3085 | if [[ $host != lcbullhh && $host != lccrayb && $host != lccrayf && $host != lccrayh && $host != lckyuh && $host != lckyut && $host != lcocean ]] |
---|
[179] | 3086 | then |
---|
[1779] | 3087 | if [[ -f $hostfile ]] |
---|
[1598] | 3088 | then |
---|
[1779] | 3089 | cp $hostfile hostfile |
---|
| 3090 | (( ii = $numprocs / $threads_per_task )) |
---|
| 3091 | [[ $ii = 0 ]] && (( ii = 1 )) |
---|
| 3092 | else |
---|
| 3093 | (( ii = 1 )) |
---|
| 3094 | while (( ii <= $numprocs / $threads_per_task )) |
---|
| 3095 | do |
---|
| 3096 | echo $localhost_realname >> hostfile |
---|
| 3097 | (( ii = ii + 1 )) |
---|
| 3098 | done |
---|
| 3099 | if (( $numprocs / $threads_per_task == 0 )) |
---|
| 3100 | then |
---|
| 3101 | echo $localhost_realname >> hostfile |
---|
| 3102 | fi |
---|
[1598] | 3103 | fi |
---|
[1779] | 3104 | eval zeile=\"`head -n $ii hostfile`\" |
---|
| 3105 | printf "\n *** running on: $zeile" |
---|
[179] | 3106 | fi |
---|
| 3107 | |
---|
[1779] | 3108 | (( ii = $numprocs / $threads_per_task )) |
---|
| 3109 | [[ $ii = 0 ]] && (( ii = 1 )) |
---|
| 3110 | export OMP_NUM_THREADS=$threads_per_task |
---|
[1289] | 3111 | |
---|
[1779] | 3112 | if [[ $threads_per_task != 1 ]] |
---|
[291] | 3113 | then |
---|
[1779] | 3114 | # INCREASE STACK SIZE TO UNLIMITED, BECAUSE OTHERWISE LARGE RUNS |
---|
| 3115 | # MAY ABORT |
---|
| 3116 | ulimit -s unlimited |
---|
| 3117 | printf "\n threads per task: $threads_per_task stacksize: unlimited" |
---|
[291] | 3118 | fi |
---|
[1779] | 3119 | if [[ $run_coupled_model = false ]] |
---|
[1224] | 3120 | then |
---|
[1779] | 3121 | if [[ "$ocean_file_appendix" = true ]] |
---|
| 3122 | then |
---|
| 3123 | echo "precursor_ocean" > runfile_atmos |
---|
| 3124 | else |
---|
| 3125 | echo "precursor_atmos" > runfile_atmos |
---|
| 3126 | fi |
---|
| 3127 | printf "\n\n" |
---|
[1622] | 3128 | |
---|
[1779] | 3129 | if [[ $host = lccrayb || $host = lccrayh ]] |
---|
| 3130 | then |
---|
| 3131 | echo "--- aprun -n $ii -N $tasks_per_node a.out < runfile_atmos" |
---|
| 3132 | aprun -n $ii -N $tasks_per_node a.out < runfile_atmos |
---|
| 3133 | elif [[ $host = lcbullhh ]] |
---|
| 3134 | then |
---|
| 3135 | export OMPI_MCA_pml=cm |
---|
| 3136 | export OMPI_MCA_mtl=mxm |
---|
| 3137 | export OMPI_MCA_coll=^ghc |
---|
| 3138 | export OMPI_MCA_mtl_mxm_np=0 |
---|
| 3139 | export MXM_RDMA_PORTS=mlx5_0:1 |
---|
| 3140 | export MXM_LOG_LEVEL=ERROR |
---|
| 3141 | export OMP_NUM_THREADS=$threads_per_task |
---|
| 3142 | export KMP_AFFINITY=verbose,granularity=core,compact,1 |
---|
| 3143 | export KMP_STACKSIZE=64m |
---|
[1622] | 3144 | |
---|
[1779] | 3145 | srun --nodes=$nodes --ntasks-per-node=$tasks_per_node ./a.out < runfile_atmos |
---|
[1622] | 3146 | |
---|
[1779] | 3147 | elif [[ $host = lccrayf ]] |
---|
| 3148 | then |
---|
| 3149 | aprun -j1 -n $ii -N $tasks_per_node -m ${memory}M a.out < runfile_atmos |
---|
| 3150 | elif [[ $host = lcxe6 || $host = lcxt5m ]] |
---|
| 3151 | then |
---|
| 3152 | aprun -n $ii -N $tasks_per_node a.out < runfile_atmos |
---|
| 3153 | elif [[ $host = lcflow ]] |
---|
| 3154 | then |
---|
| 3155 | mpirun -np $ii a.out < runfile_atmos |
---|
[1866] | 3156 | elif [[ $host = lcocean ]] |
---|
| 3157 | then |
---|
| 3158 | mpirun a.out $ROPTS < runfile_atmos |
---|
[1779] | 3159 | elif [[ $host = lcsb ]] |
---|
| 3160 | then |
---|
| 3161 | mpirun_rsh -hostfile $PBS_NODEFILE -np `cat $PBS_NODEFILE | wc -l` a.out < runfile_atmos |
---|
| 3162 | elif [[ $host = lckiaps ]] |
---|
| 3163 | then |
---|
| 3164 | mpirun -np $ii -f $PBS_NODEFILE a.out < runfile_atmos |
---|
| 3165 | elif [[ $host = lckyu* ]] |
---|
| 3166 | then |
---|
| 3167 | mpiexec -n $ii --stdin runfile_atmos ./a.out |
---|
| 3168 | else |
---|
| 3169 | mpiexec -machinefile hostfile -n $ii a.out < runfile_atmos |
---|
| 3170 | fi |
---|
| 3171 | |
---|
[164] | 3172 | else |
---|
| 3173 | |
---|
[1779] | 3174 | # CURRENTLY THERE IS NO FULL MPI-2 SUPPORT ON ICE AND XT4 |
---|
| 3175 | (( iia = $numprocs_atmos / $threads_per_task )) |
---|
| 3176 | (( iio = $numprocs_ocean / $threads_per_task )) |
---|
| 3177 | printf "\n coupled run ($iia atmosphere, $iio ocean)" |
---|
| 3178 | printf "\n using $coupled_mode coupling" |
---|
| 3179 | printf "\n\n" |
---|
[164] | 3180 | |
---|
[1779] | 3181 | if [[ $coupled_mode = "mpi2" ]] |
---|
[206] | 3182 | then |
---|
[1779] | 3183 | echo "atmosphere_to_ocean $iia $iio" > runfile_atmos |
---|
| 3184 | echo "ocean_to_atmosphere $iia $iio" > runfile_ocean |
---|
[197] | 3185 | |
---|
[1779] | 3186 | if [[ $host = lccrayf || $host = lcxe6 || $host = lcxt5m ]] |
---|
| 3187 | then |
---|
[206] | 3188 | |
---|
[1779] | 3189 | aprun -n $iia -N $tasks_per_node a.out < runfile_atmos & |
---|
| 3190 | aprun -n $iio -N $tasks_per_node a.out < runfile_ocean & |
---|
| 3191 | |
---|
| 3192 | else |
---|
| 3193 | # WORKAROUND BECAUSE mpiexec WITH -env option IS NOT AVAILABLE ON SOME SYSTEMS |
---|
| 3194 | mpiexec -machinefile hostfile -n $iia a.out < runfile_atmos & |
---|
| 3195 | mpiexec -machinefile hostfile -n $iio a.out < runfile_ocean & |
---|
| 3196 | # mpiexec -machinefile hostfile -n $iia -env coupling_mode atmosphere_to_ocean a.out & |
---|
| 3197 | # mpiexec -machinefile hostfile -n $iio -env coupling_mode ocean_to_atmosphere a.out & |
---|
| 3198 | fi |
---|
| 3199 | wait |
---|
| 3200 | |
---|
[206] | 3201 | else |
---|
| 3202 | |
---|
[1779] | 3203 | echo "coupled_run $iia $iio" > runfile_atmos |
---|
[164] | 3204 | |
---|
[1779] | 3205 | if [[ $host = lccrayf || $host = lcxe6 || $host = lcxt5m ]] |
---|
| 3206 | then |
---|
[206] | 3207 | |
---|
[1779] | 3208 | aprun -n $ii -N $tasks_per_node a.out < runfile_atmos |
---|
[206] | 3209 | |
---|
[1779] | 3210 | elif [[ $host = lck || $host = lckordi ]] |
---|
| 3211 | then |
---|
[206] | 3212 | |
---|
[1779] | 3213 | mpiexec -n $ii ./a.out < runfile_atmos & |
---|
[1108] | 3214 | |
---|
[1779] | 3215 | elif [[ $host = lckyu* ]] |
---|
| 3216 | then |
---|
[1108] | 3217 | |
---|
[1779] | 3218 | mpiexec -n $ii --stdin runfile_atmos ./a.out |
---|
[1108] | 3219 | |
---|
[1779] | 3220 | elif [[ $host = lcmuk ]] |
---|
| 3221 | then |
---|
[1108] | 3222 | |
---|
[1779] | 3223 | mpiexec -machinefile hostfile -n $ii a.out < runfile_atmos |
---|
[1604] | 3224 | |
---|
[1779] | 3225 | fi |
---|
| 3226 | wait |
---|
| 3227 | fi |
---|
[1604] | 3228 | |
---|
[164] | 3229 | fi |
---|
[102] | 3230 | |
---|
[1779] | 3231 | elif [[ $host = lckyoto ]] |
---|
[440] | 3232 | then |
---|
[1779] | 3233 | set -xv |
---|
| 3234 | export P4_RSHCOMMAND=plesh |
---|
| 3235 | echo " P4_RSHCOMMAND = $P4_RSHCOMMAND" |
---|
| 3236 | if [[ "$ENVIRONMENT" = BATCH ]] |
---|
[505] | 3237 | then |
---|
[1779] | 3238 | if [[ "$cond2" = fujitsu ]] |
---|
| 3239 | then |
---|
| 3240 | mpiexec -n $numprocs ./a.out # for fujitsu-compiler |
---|
| 3241 | elif [[ "cond2" = pgi ]] |
---|
| 3242 | then |
---|
| 3243 | mpirun -np $numprocs -machinefile ${QSUB_NODEINF} ./a.out |
---|
| 3244 | else |
---|
| 3245 | mpirun_rsh -np $numprocs -hostfile ${QSUB_NODEINF} MV2_USE_SRQ=0 ./a.out || /bin/true |
---|
| 3246 | fi |
---|
[505] | 3247 | else |
---|
[1779] | 3248 | if [[ "$cond2" = "" ]] |
---|
| 3249 | then |
---|
| 3250 | mpiruni_rsh -np $numprocs ./a.out # for intel |
---|
| 3251 | else |
---|
| 3252 | mpirun -np $numprocs ./a.out |
---|
| 3253 | fi |
---|
[505] | 3254 | fi |
---|
[1779] | 3255 | set +xv |
---|
| 3256 | |
---|
| 3257 | elif [[ $host = lctit ]] |
---|
| 3258 | then |
---|
| 3259 | export OMP_NUM_THREADS=$threads_per_task |
---|
| 3260 | echo "OMP_NUM_THREADS=$OMP_NUM_THREADS" |
---|
| 3261 | if [[ "$threads_per_task" != 1 ]] |
---|
[505] | 3262 | then |
---|
[1779] | 3263 | export MV2_ENABLE_AFFINITY=0 |
---|
[505] | 3264 | fi |
---|
[1779] | 3265 | echo "----- PBS_NODEFILE content:" |
---|
| 3266 | cat $PBS_NODEFILE |
---|
| 3267 | echo "-----" |
---|
| 3268 | (( ii = $numprocs / $threads_per_task )) |
---|
| 3269 | echo "mpirun -np $ii -hostfile $PBS_NODEFILE ./a.out" |
---|
| 3270 | mpirun -np $ii -hostfile $PBS_NODEFILE ./a.out |
---|
[1289] | 3271 | |
---|
[1779] | 3272 | else |
---|
| 3273 | mpprun -n $numprocs a.out |
---|
[678] | 3274 | fi |
---|
[1] | 3275 | else |
---|
[1779] | 3276 | a.out |
---|
[1] | 3277 | fi |
---|
[1289] | 3278 | |
---|
[1779] | 3279 | fi # end normal (non-debug) execution |
---|
| 3280 | |
---|
| 3281 | fi # end explicit execute_command or host-specific execute actions |
---|
| 3282 | |
---|
[1289] | 3283 | if [[ $? != 0 ]] |
---|
[1] | 3284 | then |
---|
| 3285 | |
---|
[1289] | 3286 | # ABORT IN CASE OF RUNTIME ERRORS |
---|
[1] | 3287 | printf "\n +++ runtime error occured" |
---|
| 3288 | locat=execution |
---|
| 3289 | exit |
---|
[1289] | 3290 | |
---|
[1] | 3291 | else |
---|
[260] | 3292 | |
---|
[1289] | 3293 | printf "\n$dashes\n *** execution finished \n" |
---|
| 3294 | |
---|
| 3295 | # STOP THE DVR STREAMING SERVER AND PROCESS THE DVR OUTPUT IN ORDER |
---|
| 3296 | # TO CREAT DVRS- AND HTML-FILES CONTAINING ALL STREAMS |
---|
[260] | 3297 | if [[ "$dvrserver_id" != "" ]] |
---|
| 3298 | then |
---|
| 3299 | kill $dvrserver_id |
---|
| 3300 | printf "\n *** dvr server with id=$dvrserver_id has been stopped" |
---|
[261] | 3301 | |
---|
[1289] | 3302 | # IF THERE IS A DIRECTORY, DATA HAVE BEEN OUTPUT BY THE |
---|
| 3303 | # STREAMING SERVER. OTHERWISE, USER HAS CHOSEN dvrp_output=local |
---|
[262] | 3304 | if [[ -d DATA_DVR ]] |
---|
| 3305 | then |
---|
[261] | 3306 | |
---|
[1289] | 3307 | # ADD THE CURRENT DVR CONFIGURATION FILE TO THE DVR OUTPUT |
---|
| 3308 | # DIRECTORY |
---|
[262] | 3309 | cp .dvrserver.config DATA_DVR |
---|
| 3310 | |
---|
[1289] | 3311 | # PROCESS THE DVR OUTPUT (OPTION -s FOR GENERATING |
---|
| 3312 | # SEQUENCE MODE DATA TOO) |
---|
[262] | 3313 | process_dvr_output -d DATA_DVR -f $fname -s |
---|
| 3314 | |
---|
| 3315 | else |
---|
| 3316 | |
---|
[1289] | 3317 | # PROCESS THE LOCAL OUTPUT |
---|
[262] | 3318 | process_dvr_output -l -d DATA_DVR -f $fname |
---|
| 3319 | |
---|
| 3320 | fi |
---|
| 3321 | |
---|
| 3322 | elif [[ $(echo $package_list | grep -c dvrp_graphics) != 0 ]] |
---|
| 3323 | then |
---|
| 3324 | |
---|
[1289] | 3325 | # PROCESS DVR OUTPUT GENERATD IN LOCAL MODE (dvrp_output=local) |
---|
[262] | 3326 | process_dvr_output -l -d DATA_DVR -f $fname |
---|
| 3327 | |
---|
[260] | 3328 | fi |
---|
[1] | 3329 | fi |
---|
| 3330 | |
---|
| 3331 | |
---|
| 3332 | |
---|
[1289] | 3333 | # CALL OF combine_plot_fields IN ORDER TO MERGE SINGLE FILES WRITTEN |
---|
| 3334 | # BY EACH CORE INTO ONE FILE |
---|
[920] | 3335 | if [[ ! -f ${PALM_BIN}/combine_plot_fields${block}.x ]] |
---|
| 3336 | then |
---|
| 3337 | printf "\n\n\n +++ WARNING: no combine_plot_fields found for given block \"$cond1 $cond2\"" |
---|
| 3338 | printf "\n 2d- and/or 3d-data may be incomplete!" |
---|
| 3339 | printf "\n Run \"mbuild -u -h $localhost\" to generate utilities for this block.\n" |
---|
| 3340 | elif [[ "$combine_plot_fields" == true ]] |
---|
| 3341 | then |
---|
[1270] | 3342 | if [[ $localhost = lccrayh || $localhost = lccrayb ]] |
---|
| 3343 | then |
---|
| 3344 | printf "\n\n\n *** post-processing: now executing \"aprun -n 1 -N 1 combine_plot_fields${block}.x\" ..." |
---|
[1779] | 3345 | aprun -n 1 -N 1 combine_plot_fields${block}.x |
---|
[1270] | 3346 | else |
---|
| 3347 | printf "\n\n\n *** post-processing: now executing \"combine_plot_fields${block}.x\" ..." |
---|
| 3348 | combine_plot_fields${block}.x |
---|
| 3349 | fi |
---|
[920] | 3350 | else |
---|
[1289] | 3351 | |
---|
| 3352 | # TEMPORARY SOLUTION TO SKIP combine_plot_fields. THIS IS REQUIRED IN CASE OF HUGE AMOUNT OF |
---|
| 3353 | # DATA OUTPUT. TO DO: EXTEND THIS BRANCH BY CREATING A BATCH JOB for combine_plot_fields. |
---|
| 3354 | # ??? koennen wir das streichen ??? |
---|
[920] | 3355 | printf "\n\n\n *** post-processing: skipping combine_plot_fields (-Z option set) ..." |
---|
| 3356 | fi |
---|
[503] | 3357 | |
---|
| 3358 | |
---|
| 3359 | |
---|
[1289] | 3360 | # EXECUTE OUTPUT-COMMANDS GIVEN IN THE CONFIGURATION FILE |
---|
[1] | 3361 | (( i = 0 )) |
---|
| 3362 | while (( i < ioc )) |
---|
| 3363 | do |
---|
| 3364 | (( i = i + 1 )) |
---|
| 3365 | if (( i == 1 )) |
---|
| 3366 | then |
---|
[1289] | 3367 | printf "\n\n *** execution of OUTPUT-commands:\n$dashes" |
---|
[1] | 3368 | fi |
---|
| 3369 | printf "\n >>> ${out_command[$i]}" |
---|
| 3370 | eval ${out_command[$i]} |
---|
| 3371 | if (( i == ioc )) |
---|
| 3372 | then |
---|
[1289] | 3373 | printf "\n$dashes\n" |
---|
[1] | 3374 | fi |
---|
| 3375 | done |
---|
| 3376 | |
---|
| 3377 | |
---|
[1289] | 3378 | # IN TRACE-MODE PRINT CONTENTS OF THE CURRENT (TEMPORARY) WORKING DIRECTORY |
---|
[1] | 3379 | if [[ $do_trace = true ]] |
---|
| 3380 | then |
---|
| 3381 | printf "\n\n" |
---|
| 3382 | ls -al |
---|
| 3383 | fi |
---|
| 3384 | |
---|
| 3385 | |
---|
| 3386 | |
---|
[1289] | 3387 | # COPY LOCAL OUTPUT-FILES TO THEIR PERMANENT DESTINATIONS |
---|
[1] | 3388 | (( i = 0 )) |
---|
| 3389 | while (( i < iout )) |
---|
| 3390 | do |
---|
| 3391 | (( i = i + 1 )) |
---|
| 3392 | if (( i == 1 )) |
---|
| 3393 | then |
---|
[1289] | 3394 | printf "\n\n *** saving OUTPUT-files:\n$dashes" |
---|
[1] | 3395 | fi |
---|
| 3396 | |
---|
[1289] | 3397 | # CHECK FOR SINGLE FILE (SERIAL RUN) OR DIRECTORY (ONE FILE PER CORE FOR PARELLEL EXECUTION) |
---|
[1] | 3398 | files_for_pes=false; filetyp=file |
---|
| 3399 | if [[ "${actionout[$i]}" = pe && -n $numprocs ]] |
---|
| 3400 | then |
---|
| 3401 | files_for_pes=true; filetyp=directory |
---|
| 3402 | actionout[$i]="" |
---|
| 3403 | elif [[ "${actionout[$i]}" = pe && ! -n $numprocs ]] |
---|
| 3404 | then |
---|
| 3405 | actionout[$i]="" |
---|
| 3406 | elif [[ "${actionout[$i]}" = arpe && -n $numprocs ]] |
---|
| 3407 | then |
---|
| 3408 | files_for_pes=true; filetyp=directory |
---|
| 3409 | actionout[$i]="ar" |
---|
| 3410 | elif [[ "${actionout[$i]}" = arpe && ! -n $numprocs ]] |
---|
| 3411 | then |
---|
| 3412 | actionout[$i]="ar" |
---|
| 3413 | elif [[ "${actionout[$i]}" = flpe && -n $numprocs ]] |
---|
| 3414 | then |
---|
| 3415 | files_for_pes=true; filetyp=directory |
---|
| 3416 | actionout[$i]="fl" |
---|
| 3417 | elif [[ "${actionout[$i]}" = flpe && ! -n $numprocs ]] |
---|
| 3418 | then |
---|
| 3419 | actionout[$i]="fl" |
---|
| 3420 | elif [[ "${actionout[$i]}" = trpe && -n $numprocs ]] |
---|
| 3421 | then |
---|
| 3422 | files_for_pes=true; filetyp=directory |
---|
| 3423 | actionout[$i]="tr" |
---|
| 3424 | elif [[ "${actionout[$i]}" = trpe && ! -n $numprocs ]] |
---|
| 3425 | then |
---|
| 3426 | actionout[$i]="tr" |
---|
| 3427 | fi |
---|
| 3428 | |
---|
| 3429 | if [[ ! -f ${localout[$i]} && $files_for_pes = false ]] |
---|
| 3430 | then |
---|
| 3431 | printf "\n +++ temporary OUTPUT-file ${localout[$i]} does not exist\n" |
---|
[260] | 3432 | elif [[ ! -d ${localout[$i]} && $files_for_pes = true ]] |
---|
[1] | 3433 | then |
---|
| 3434 | printf "\n +++ temporary OUTPUT-file ${localout[$i]}/.... does not exist\n" |
---|
| 3435 | else |
---|
| 3436 | |
---|
| 3437 | |
---|
[1289] | 3438 | # COPY VIA SCP TO LOCAL HOST (ALWAYS IN BINARY MODE USING batch_scp option -m) |
---|
| 3439 | # IF TARGET DIRECTORY DOES NOT EXISTS, TRY TO CREATE IT |
---|
[1] | 3440 | if [[ "${actionout[$i]}" = tr ]] |
---|
| 3441 | then |
---|
| 3442 | if [[ $localhost != $fromhost ]] |
---|
| 3443 | then |
---|
| 3444 | if [[ $files_for_pes = false ]] |
---|
| 3445 | then |
---|
| 3446 | cps="" |
---|
| 3447 | cst="" |
---|
| 3448 | else |
---|
| 3449 | cps=-c |
---|
| 3450 | cst="/" |
---|
| 3451 | fi |
---|
[622] | 3452 | if [[ $localhost = nech ]] |
---|
[1] | 3453 | then |
---|
| 3454 | |
---|
[1289] | 3455 | # TRANSFER IN SEPERATE JOB |
---|
| 3456 | # FIRST COPY FILE TO TEMPORY DATA DIRECTORY |
---|
[1] | 3457 | [[ ! -d $tmp_data_catalog/TRANSFER ]] && mkdir -p $tmp_data_catalog/TRANSFER |
---|
| 3458 | file_to_transfer=${fname}_${localout[$i]}_to_transfer_$kennung |
---|
| 3459 | if [[ $files_for_pes = false ]] |
---|
| 3460 | then |
---|
| 3461 | ln -f ${localout[$i]} $tmp_data_catalog/TRANSFER/$file_to_transfer |
---|
| 3462 | else |
---|
| 3463 | mkdir $tmp_data_catalog/TRANSFER/$file_to_transfer |
---|
| 3464 | ln ${localout[$i]}/* $tmp_data_catalog/TRANSFER/$file_to_transfer |
---|
| 3465 | fi |
---|
| 3466 | |
---|
| 3467 | echo "set -x" > transfer_${localout[$i]} |
---|
| 3468 | echo "cd $tmp_data_catalog/TRANSFER" >> transfer_${localout[$i]} |
---|
[82] | 3469 | |
---|
| 3470 | printf "\n >>> OUTPUT: ${localout[$i]}$cst by SCP in seperate job to" |
---|
| 3471 | printf "\n ${pathout[$i]}/${localhost}_${fname}${endout[$i]}$cst" |
---|
| 3472 | printf "\n or higher cycle\n" |
---|
[1468] | 3473 | echo "batch_scp $PORTOPT $cps -b -m -u $return_username $return_address $file_to_transfer \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" >> transfer_${localout[$i]} |
---|
[82] | 3474 | |
---|
[1] | 3475 | echo "[[ \$? = 0 ]] && rm $file_to_transfer" >> transfer_${localout[$i]} |
---|
| 3476 | |
---|
| 3477 | if [[ $localhost = nech ]] |
---|
| 3478 | then |
---|
[1094] | 3479 | subjob -d -c /pf/b/$usern/job_queue -v -q pp -X 0 -m 1000 -t 900 $PORTOPT transfer_${localout[$i]} |
---|
[1] | 3480 | else |
---|
| 3481 | if [[ "$LOGNAME" = b323013 ]] |
---|
| 3482 | then |
---|
[1094] | 3483 | subjob -v -q c1 -X 0 -m 1000 -t 900 -c $job_catalog $PORTOPT transfer_${localout[$i]} |
---|
[1] | 3484 | else |
---|
[1094] | 3485 | subjob -d -v -q c1 -X 0 -m 1000 -t 900 -c $job_catalog $PORTOPT transfer_${localout[$i]} |
---|
[1] | 3486 | fi |
---|
| 3487 | fi |
---|
| 3488 | |
---|
| 3489 | else |
---|
| 3490 | |
---|
[1289] | 3491 | # TRANSFER WITHIN THIS JOB |
---|
[1] | 3492 | transfer_failed=false |
---|
[82] | 3493 | printf "\n >>> OUTPUT: ${localout[$i]}$cst by SCP to" |
---|
| 3494 | printf "\n ${pathout[$i]}/${localhost}_${fname}${endout[$i]}$cst\n" |
---|
[1255] | 3495 | if [[ $localhost = lccrayb ]] |
---|
| 3496 | then |
---|
[1468] | 3497 | ssh $usern@blogin1 ". \\$HOME/.profile; cd $TEMPDIR; batch_scp $PORTOPT $cps -b -m -u $return_username $return_address ${localout[$i]} \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" |
---|
[1274] | 3498 | elif [[ $localhost = lccrayh ]] |
---|
| 3499 | then |
---|
[1468] | 3500 | ssh $usern@hlogin1 ". \\$HOME/.profile; cd $TEMPDIR; batch_scp $PORTOPT $cps -b -m -u $return_username $return_address ${localout[$i]} \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" |
---|
[1620] | 3501 | elif [[ $localhost = lcbullhh ]] |
---|
| 3502 | then |
---|
| 3503 | ssh $usern@mlogin101 ". \\$HOME/.profile; cd $TEMPDIR; batch_scp $PORTOPT $cps -b -m -u $return_username $return_address ${localout[$i]} \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" |
---|
[1468] | 3504 | elif [[ $localhost = lcxe6 ]] |
---|
| 3505 | then |
---|
| 3506 | ssh $usern@hexagon ". \\$HOME/.profile; cd $TEMPDIR; batch_scp $PORTOPT $cps -b -m -u $return_username $return_address ${localout[$i]} \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" |
---|
[1255] | 3507 | else |
---|
[1468] | 3508 | batch_scp $PORTOPT $cps -b -m -u $return_username $return_address ${localout[$i]} "${pathout[$i]}" ${localhost}_${fname}${endout[$i]} ${extout[$i]} |
---|
[1255] | 3509 | fi |
---|
[82] | 3510 | [[ $? != 0 ]] && transfer_failed=true |
---|
[1] | 3511 | |
---|
[1289] | 3512 | # IF TRANSFER FAILED, CREATE BACKUP COPY ON THIS MACHINE |
---|
[1] | 3513 | if [[ $transfer_failed = true ]] |
---|
| 3514 | then |
---|
[1289] | 3515 | printf " +++ transfer failed. Trying to save a copy on this host under:\n" |
---|
[1] | 3516 | printf " ${pathout[$i]}/${localhost}_${fname}${endout[$i]}_$kennung\n" |
---|
| 3517 | |
---|
[1289] | 3518 | # FIRST CHECK, IF DIRECTORY EXISTS, AND CREATE IT, IF NECESSARY |
---|
[1] | 3519 | eval local_catalog=${pathout[$i]} |
---|
| 3520 | if [[ ! -d $local_catalog ]] |
---|
| 3521 | then |
---|
| 3522 | printf " *** local directory does not exist. Trying to create:\n" |
---|
| 3523 | printf " $local_catalog \n" |
---|
| 3524 | mkdir -p $local_catalog |
---|
| 3525 | fi |
---|
| 3526 | eval cp ${localout[$i]} ${pathout[$i]}/${localhost}_${fname}${endout[$i]}_$kennung |
---|
| 3527 | transfer_problems=true |
---|
| 3528 | fi |
---|
| 3529 | fi |
---|
| 3530 | else |
---|
| 3531 | |
---|
[1289] | 3532 | # UNSET actionout. DUE TO THIS SETTING, FILE WILL LATER JUST BE COPIED ON THIS MACHINE |
---|
[1] | 3533 | actionout[$i]="" |
---|
| 3534 | fi |
---|
| 3535 | fi |
---|
| 3536 | |
---|
| 3537 | |
---|
[1289] | 3538 | # APPEND VIA SCP TO LOCAL HOST (ALWAYS IN BINARY MODE USING batch_scp option -m) |
---|
| 3539 | # IF TARGET DIRECTORY DOES NOT EXISTS, TRY TO CREATE IT |
---|
[1] | 3540 | if [[ "${actionout[$i]}" = tra ]] |
---|
| 3541 | then |
---|
| 3542 | if [[ $localhost != $fromhost ]] |
---|
| 3543 | then |
---|
[622] | 3544 | if [[ $localhost = ibmh || $localhost = nech ]] |
---|
[1] | 3545 | then |
---|
| 3546 | |
---|
[1289] | 3547 | # TRANSFER IN SEPERATE JOB |
---|
| 3548 | # FIRST COPY FILE TO TEMPORY DATA DIRECTORY |
---|
[1] | 3549 | [[ ! -d $tmp_data_catalog/TRANSFER ]] && mkdir -p $tmp_data_catalog/TRANSFER |
---|
| 3550 | file_to_transfer=${fname}_${localout[$i]}_to_transfer_$kennung |
---|
| 3551 | ln -f ${localout[$i]} $tmp_data_catalog/TRANSFER/$file_to_transfer |
---|
| 3552 | |
---|
| 3553 | echo "set -x" > transfer_${localout[$i]} |
---|
| 3554 | echo "cd $tmp_data_catalog/TRANSFER" >> transfer_${localout[$i]} |
---|
[82] | 3555 | |
---|
| 3556 | printf "\n >>> OUTPUT: ${localout[$i]} append by SCP in seperate job to" |
---|
| 3557 | printf "\n ${pathout[$i]}/${localhost}_${fname}${endout[$i]}" |
---|
| 3558 | printf "\n or higher cycle\n" |
---|
[1468] | 3559 | echo "batch_scp $PORTOPT -A -b -m -u $return_username $return_address $file_to_transfer \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" >> transfer_${localout[$i]} |
---|
[82] | 3560 | |
---|
[1] | 3561 | echo "[[ \$? = 0 ]] && rm $file_to_transfer" >> transfer_${localout[$i]} |
---|
| 3562 | |
---|
| 3563 | if [[ $localhost = nech ]] |
---|
| 3564 | then |
---|
[1094] | 3565 | subjob -d -c /pf/b/$usern/job_queue -v -q pp -X 0 -m 1000 -t 900 $PORTOPT transfer_${localout[$i]} |
---|
[1] | 3566 | else |
---|
| 3567 | if [[ $LOGNAME = b323013 ]] |
---|
| 3568 | then |
---|
[1094] | 3569 | subjob -v -q c1 -X 0 -m 1000 -t 900 -c $job_catalog $PORTOPT transfer_${localout[$i]} |
---|
[1] | 3570 | else |
---|
[1094] | 3571 | subjob -d -v -q c1 -X 0 -m 1000 -t 900 -c $job_catalog $PORTOPT transfer_${localout[$i]} |
---|
[1] | 3572 | fi |
---|
| 3573 | fi |
---|
| 3574 | |
---|
| 3575 | else |
---|
| 3576 | |
---|
[1289] | 3577 | # TRANSFER WITHIN THIS JOB |
---|
[1] | 3578 | transfer_failed=false |
---|
[82] | 3579 | printf "\n >>> OUTPUT: ${localout[$i]} append by SCP to" |
---|
| 3580 | printf "\n ${pathout[$i]}/${localhost}_${fname}${endout[$i]}\n" |
---|
[1255] | 3581 | if [[ $localhost = lccrayb ]] |
---|
| 3582 | then |
---|
[1468] | 3583 | ssh $usern@blogin1 ". \\$HOME/.profile; cd $TEMPDIR; batch_scp $PORTOPT -A -b -m -u $return_username $return_address ${localout[$i]} \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" |
---|
[1274] | 3584 | elif [[ $localhost = lccrayh ]] |
---|
| 3585 | then |
---|
[1468] | 3586 | ssh $usern@hlogin1 ". \\$HOME/.profile; cd $TEMPDIR; batch_scp $PORTOPT -A -b -m -u $return_username $return_address ${localout[$i]} \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" |
---|
| 3587 | elif [[ $localhost = lcxe6 ]] |
---|
| 3588 | then |
---|
| 3589 | ssh $usern@hexagon ". \\$HOME/.profile; cd $TEMPDIR; batch_scp $PORTOPT -A -b -m -u $return_username $return_address ${localout[$i]} \"${pathout[$i]}\" ${localhost}_${fname}${endout[$i]} ${extout[$i]}" |
---|
[1255] | 3590 | else |
---|
[1468] | 3591 | batch_scp $PORTOPT -A -b -m -u $return_username $return_address ${localout[$i]} "${pathout[$i]}" ${localhost}_${fname}${endout[$i]} ${extout[$i]} |
---|
[1255] | 3592 | fi |
---|
[82] | 3593 | [[ $? != 0 ]] && transfer_failed=true |
---|
[1] | 3594 | |
---|
[1289] | 3595 | # IF TRANSFER FAILED, CREATE BACKUP COPY ON THIS MACHINE |
---|
[1] | 3596 | if [[ $transfer_failed = true ]] |
---|
| 3597 | then |
---|
[1289] | 3598 | printf " +++ transfer failed. Trying to save a copy on this host under:\n" |
---|
[1] | 3599 | printf " ${pathout[$i]}/${localhost}_${fname}${endout[$i]}_$kennung\n" |
---|
| 3600 | |
---|
[1289] | 3601 | # FIRST CHECK, IF DIRECTORY EXISTS, AND CREATE IT, IF NECESSARY |
---|
[1] | 3602 | eval local_catalog=${pathout[$i]} |
---|
| 3603 | if [[ ! -d $local_catalog ]] |
---|
| 3604 | then |
---|
| 3605 | printf " *** local directory does not exist. Trying to create:\n" |
---|
| 3606 | printf " $local_catalog \n" |
---|
| 3607 | mkdir -p $local_catalog |
---|
| 3608 | fi |
---|
| 3609 | |
---|
| 3610 | eval cp ${localout[$i]} ${pathout[$i]}/${localhost}_${fname}${endout[$i]}_$kennung |
---|
| 3611 | transfer_problems=true |
---|
| 3612 | fi |
---|
| 3613 | fi |
---|
| 3614 | else |
---|
| 3615 | |
---|
[1289] | 3616 | # RESET actionout. DUE TO THIS SETTING, FILE WILL LATER JUST BE APPENDED ON THIS MACHINE |
---|
[1] | 3617 | actionout[$i]=a |
---|
| 3618 | fi |
---|
| 3619 | fi |
---|
| 3620 | |
---|
| 3621 | |
---|
[1289] | 3622 | # OUTPUT-FILE FOR A RESTART-RUN. |
---|
| 3623 | # FILE IS MOVED TO THE TEMPORARY DATA DIRECTORY USING link COMMAND. |
---|
| 3624 | # OPTION -f IS USED TO OVERWRITE EXISTING FILES. |
---|
| 3625 | # AS LAST ACTION, ARCHIVING IS INITIATED. |
---|
[1] | 3626 | if [[ "${actionout[$i]}" = fl ]] |
---|
| 3627 | then |
---|
| 3628 | [[ ! -d $tmp_data_catalog ]] && mkdir -p $tmp_data_catalog |
---|
| 3629 | chmod g+rx $tmp_data_catalog |
---|
| 3630 | if [[ $files_for_pes = false ]] |
---|
| 3631 | then |
---|
| 3632 | printf "\n >>> OUTPUT: ${localout[$i]} to" |
---|
| 3633 | printf "\n $tmp_data_catalog/${frelout[$i]} (temporary data catalog)\n" |
---|
| 3634 | ln -f ${localout[$i]} $tmp_data_catalog/${frelout[$i]} |
---|
| 3635 | else |
---|
| 3636 | printf "\n >>> OUTPUT: ${localout[$i]}/.... to" |
---|
| 3637 | printf "\n $tmp_data_catalog/${frelout[$i]} (temporary data catalog)\n" |
---|
| 3638 | mkdir $tmp_data_catalog/${frelout[$i]} |
---|
[149] | 3639 | cd ${localout[$i]} |
---|
| 3640 | for file in $(ls *) |
---|
| 3641 | do |
---|
| 3642 | ln -f $file $tmp_data_catalog/${frelout[$i]} |
---|
| 3643 | done |
---|
| 3644 | cd $TEMPDIR |
---|
[1] | 3645 | fi |
---|
| 3646 | |
---|
| 3647 | |
---|
[1289] | 3648 | # CREATE AND SUBMIT ARCHIVING-JOB |
---|
[1] | 3649 | if [[ $store_on_archive_system = true ]] |
---|
| 3650 | then |
---|
| 3651 | |
---|
| 3652 | if [[ $archive_system = asterix ]] |
---|
| 3653 | then |
---|
| 3654 | echo "cd $tmp_data_catalog" >> archive_${frelout[$i]} |
---|
| 3655 | if [[ $files_for_pes = false ]] |
---|
| 3656 | then |
---|
| 3657 | echo "stageout ${frelout[$i]} > STAGE_OUTPUT${i}_$kennung" >> archive_${frelout[$i]} |
---|
| 3658 | else |
---|
| 3659 | echo "stageout -t ${frelout[$i]} > STAGE_OUTPUT${i}_$kennung" >> archive_${frelout[$i]} |
---|
| 3660 | fi |
---|
| 3661 | echo "cat STAGE_OUTPUT${i}_$kennung" >> archive_${frelout[$i]} |
---|
| 3662 | echo "if [[ \$(grep -c \"st.msg:150\" STAGE_OUTPUT${i}_$kennung) != 0 ]]" >> archive_${frelout[$i]} |
---|
| 3663 | echo "then" >> archive_${frelout[$i]} |
---|
| 3664 | echo " do_stageout=false" >> archive_${frelout[$i]} |
---|
| 3665 | echo "else" >> archive_${frelout[$i]} |
---|
| 3666 | echo " echo \" +++ $filetyp ${frelout[$i]} could not be stored on archive-system\" " >> archive_${frelout[$i]} |
---|
| 3667 | echo " cat /dev/null > ~/job_queue/ARCHIVE_ERROR_$fname" >> archive_${frelout[$i]} |
---|
| 3668 | echo " cat STAGE_OUTPUT${i}_$kennung > ~/job_queue/archive_${frelout[$i]}_error" >> archive_${frelout[$i]} |
---|
| 3669 | echo " echo \" *** $filetyp ${frelout[$i]} will be copied to \$WORK as backup\" " >> archive_${frelout[$i]} |
---|
| 3670 | if [[ $files_for_pes = false ]] |
---|
| 3671 | then |
---|
| 3672 | echo " cp ${frelout[$i]} \$WORK" >> archive_${frelout[$i]} |
---|
| 3673 | else |
---|
| 3674 | echo " cp -r ${frelout[$i]} \$WORK/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3675 | fi |
---|
| 3676 | echo " echo \" *** $filetyp ${frelout[$i]} saved\" " >> archive_${frelout[$i]} |
---|
| 3677 | echo "fi" >> archive_${frelout[$i]} |
---|
| 3678 | echo "rm STAGE_OUTPUT${i}_$kennung" >> archive_${frelout[$i]} |
---|
| 3679 | elif [[ $archive_system = DMF ]] |
---|
| 3680 | then |
---|
| 3681 | echo "cd $tmp_data_catalog" >> archive_${frelout[$i]} |
---|
| 3682 | if [[ $files_for_pes = false ]] |
---|
| 3683 | then |
---|
| 3684 | printf "\n +++ archiving of single files impossible with $archive_system !\n" |
---|
| 3685 | locat=DMF |
---|
| 3686 | exit |
---|
| 3687 | else |
---|
[1289] | 3688 | # ??? nicht benötigte Archiv-Systeme entfernen ??? |
---|
[1] | 3689 | # FUER RECHNER IN JUELICH. DORT KOENNTEN AUCH |
---|
| 3690 | # EINZELNE DATEIEN GESICHERT WERDEN (SPAETER KORR.) |
---|
| 3691 | echo "rm -rf \$ARCHIVE/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3692 | echo "cp -r ${frelout[$i]} \$ARCHIVE" >> archive_${frelout[$i]} |
---|
| 3693 | fi |
---|
| 3694 | elif [[ $archive_system = tivoli ]] |
---|
| 3695 | then |
---|
| 3696 | echo "cd $tmp_data_catalog" >> archive_${frelout[$i]} |
---|
| 3697 | if [[ $files_for_pes = false ]] |
---|
| 3698 | then |
---|
[1289] | 3699 | # REMOVE EXISTING OLD FILES FROM THE ARCHIVE |
---|
[1] | 3700 | echo "set -x" >> archive_${frelout[$i]} |
---|
| 3701 | echo "rm -rf \$PERM/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3702 | echo "cp ${frelout[$i]} \$PERM/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3703 | else |
---|
| 3704 | |
---|
| 3705 | echo "set -x" >> archive_${frelout[$i]} |
---|
| 3706 | echo "rm -rf \$PERM/${frelout[$i]}/*" >> archive_${frelout[$i]} |
---|
| 3707 | echo "[[ ! -d \$PERM/${frelout[$i]} ]] && mkdir $PERM/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3708 | cd $tmp_data_catalog |
---|
| 3709 | all_files=`ls -1 ${frelout[$i]}/*` |
---|
| 3710 | cd - > /dev/null |
---|
| 3711 | (( inode = 0 )) |
---|
| 3712 | (( tp1 = tasks_per_node + 1 )) |
---|
| 3713 | while (( inode < nodes )) |
---|
| 3714 | do |
---|
| 3715 | files=`echo $all_files | cut -d" " -f1-$tasks_per_node` |
---|
| 3716 | all_files=`echo $all_files | cut -d" " -f$tp1-` |
---|
[1289] | 3717 | echo "tar cvf \$PERM/${frelout[$i]}/${frelout[$i]}.node_$inode.tar $files" >> archive_${frelout[$i]} |
---|
[1] | 3718 | (( inode = inode + 1 )) |
---|
| 3719 | done |
---|
| 3720 | fi |
---|
| 3721 | elif [[ $archive_system = ut ]] |
---|
| 3722 | then |
---|
| 3723 | echo "cd $tmp_data_catalog" >> archive_${frelout[$i]} |
---|
| 3724 | if [[ $files_for_pes = false ]] |
---|
| 3725 | then |
---|
[1289] | 3726 | # REMOVE EXISTING OLD FILES FROM THE ARCHIVE |
---|
[1] | 3727 | echo "set -x" >> archive_${frelout[$i]} |
---|
| 3728 | echo "rm -rf \$UT/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3729 | echo "cp ${frelout[$i]} \$UT/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3730 | else |
---|
| 3731 | |
---|
| 3732 | echo "set -x" >> archive_${frelout[$i]} |
---|
| 3733 | echo "rm -rf \$UT/${frelout[$i]}/*" >> archive_${frelout[$i]} |
---|
| 3734 | echo "[[ ! -d \$UT/${frelout[$i]} ]] && mkdir $UT/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3735 | cd $tmp_data_catalog |
---|
| 3736 | all_files=`ls -1 ${frelout[$i]}/*` |
---|
| 3737 | cd - > /dev/null |
---|
| 3738 | (( inode = 0 )) |
---|
| 3739 | (( tp1 = tasks_per_node + 1 )) |
---|
| 3740 | while (( inode < nodes )) |
---|
| 3741 | do |
---|
| 3742 | files=`echo $all_files | cut -d" " -f1-$tasks_per_node` |
---|
| 3743 | all_files=`echo $all_files | cut -d" " -f$tp1-` |
---|
| 3744 | echo "tar cvf \$UT/${frelout[$i]}/${frelout[$i]}.node_$inode.tar $files" >> archive_${frelout[$i]} |
---|
| 3745 | (( inode = inode + 1 )) |
---|
| 3746 | done |
---|
| 3747 | fi |
---|
| 3748 | elif [[ $archive_system = none ]] |
---|
| 3749 | then |
---|
| 3750 | printf " +++ archiving on $localhost not available!\n" |
---|
| 3751 | fi |
---|
| 3752 | |
---|
| 3753 | if [[ $archive_system != none ]] |
---|
| 3754 | then |
---|
[1289] | 3755 | if [[ $localhost = nech ]] |
---|
[1] | 3756 | then |
---|
[1094] | 3757 | subjob -d -c /pf/b/$usern/job_queue -v -q pp -X 0 -m 1000 -t 7200 $PORTOPT archive_${frelout[$i]} |
---|
[1] | 3758 | fi |
---|
| 3759 | printf " Archiving of $tmp_data_catalog/${frelout[$i]} initiated (batch job submitted)\n" |
---|
| 3760 | fi |
---|
| 3761 | else |
---|
[399] | 3762 | printf " +++ caution: option -A is switched off. No archiving on $archive_system!\n" |
---|
[1] | 3763 | fi |
---|
| 3764 | |
---|
| 3765 | |
---|
[1289] | 3766 | # CREATE AN EMPTY DIRECTORY IN THE USERS PERMANENT DIRECTORY, |
---|
| 3767 | # IN ORDER TO NOTE THE RESPECTIVE CYCLE NUMBER ON THE ARCHIVE SYSTEM |
---|
| 3768 | # RESTART-JOBS ARE USING THESE EMPTY-DIRECTORIES TO GET THE CURRENT HIGHEST |
---|
| 3769 | # CYCLE NUMBER ON THE ARCHIVE SYSTEM (IN CASE THAT INFORMATIONS ARE DIFFICULT TO |
---|
| 3770 | # TO ACCESS FROM THE ARCHIVE-SYSTEM DIRECTLY)) |
---|
[1] | 3771 | if [[ $files_for_pes = false ]] |
---|
| 3772 | then |
---|
| 3773 | cat /dev/null > ${pathout[$i]} |
---|
| 3774 | else |
---|
| 3775 | mkdir -p ${pathout[$i]} |
---|
| 3776 | fi |
---|
| 3777 | |
---|
| 3778 | fi |
---|
| 3779 | |
---|
| 3780 | |
---|
[1289] | 3781 | # COPY FROM THIS HOST TO THE ARCHIVE-SYSTEM |
---|
| 3782 | # IF ANY ARCHIVING FAILS, AN ERROR-FLAG-FILE IS SET |
---|
| 3783 | # THIS FILE REMAINS TO BE SET, EVEN IF ARCHIVING OF FURTHER FILES IS SUCCESSFULL |
---|
[1] | 3784 | if [[ "${actionout[$i]}" = ar ]] |
---|
| 3785 | then |
---|
| 3786 | if [[ $files_for_pes = false ]] |
---|
| 3787 | then |
---|
| 3788 | printf "\n >>> OUTPUT: ${localout[$i]} to" |
---|
| 3789 | printf "\n ${pathout[$i]}" |
---|
| 3790 | printf "\n File will be copied to archive-system ($archive_system) !\n" |
---|
| 3791 | else |
---|
| 3792 | printf "\n >>> OUTPUT: ${localout[$i]}/_.... to" |
---|
| 3793 | printf "\n ${pathout[$i]}" |
---|
| 3794 | printf "\n Directory will be copied to archive-system ($archive_system) !\n" |
---|
| 3795 | fi |
---|
| 3796 | mv ${localout[$i]} ${frelout[$i]} |
---|
| 3797 | |
---|
| 3798 | file_saved=false |
---|
| 3799 | |
---|
| 3800 | if [[ $archive_system = asterix ]] |
---|
| 3801 | then |
---|
| 3802 | do_stageout=true |
---|
| 3803 | (( stageout_anz = 0 )) |
---|
| 3804 | while [[ $do_stageout = true ]] |
---|
| 3805 | do |
---|
| 3806 | if [[ $files_for_pes = false ]] |
---|
| 3807 | then |
---|
| 3808 | stageout ${frelout[$i]} > STAGE_OUTPUT |
---|
| 3809 | else |
---|
| 3810 | stageout -t ${frelout[$i]} > STAGE_OUTPUT |
---|
| 3811 | fi |
---|
| 3812 | cat STAGE_OUTPUT |
---|
| 3813 | if [[ $(grep -c "st.msg:150" STAGE_OUTPUT) != 0 ]] |
---|
| 3814 | then |
---|
| 3815 | file_saved=true |
---|
| 3816 | do_stageout=false |
---|
| 3817 | else |
---|
| 3818 | if [[ $files_for_pes = false ]] |
---|
| 3819 | then |
---|
| 3820 | printf "\n +++ file ${frelout[$i]} could not be saved on archive-system" |
---|
| 3821 | else |
---|
| 3822 | printf "\n +++ directory ${frelout[$i]} could not be saved on archive-system" |
---|
| 3823 | fi |
---|
| 3824 | (( stageout_anz = stageout_anz + 1 )) |
---|
| 3825 | if (( stageout_anz == 10 )) |
---|
| 3826 | then |
---|
| 3827 | printf "\n +++ stoped after 10 unsuccessful tries!" |
---|
| 3828 | archive_save=false |
---|
| 3829 | do_stageout=false |
---|
| 3830 | else |
---|
| 3831 | printf "\n *** new try to store on archive after 15 min:" |
---|
| 3832 | sleep 900 |
---|
| 3833 | fi |
---|
| 3834 | fi |
---|
| 3835 | done |
---|
| 3836 | elif [[ $archive_system = DMF ]] |
---|
| 3837 | then |
---|
| 3838 | if [[ $files_for_pes = false ]] |
---|
| 3839 | then |
---|
| 3840 | printf "\n +++ archiving of single files impossible on $archive_system!\n" |
---|
| 3841 | locat=DMF |
---|
| 3842 | exit |
---|
| 3843 | else |
---|
| 3844 | rm -rf $ARCHIVE/${frelout[$i]} |
---|
| 3845 | cp -r ${frelout[$i]} $ARCHIVE |
---|
| 3846 | fi |
---|
| 3847 | file_saved=true |
---|
| 3848 | elif [[ $archive_system = tivoli ]] |
---|
| 3849 | then |
---|
[1289] | 3850 | # ARCHIVING ONLY POSSIBLE VIA BATCH-JOB |
---|
| 3851 | # IN THE MEANTIME, FILE IS STORED IN THE TEMPORARY DATA DIRECTORY, |
---|
| 3852 | # BECAUSE MRUN'S CURRENT TEMPORARY WORKING DIRECTORY MAY ALREADY BE DELETED |
---|
| 3853 | # WHEN THE ARCHIVE-JOB IS EXECUTED |
---|
[1] | 3854 | [[ ! -d $tmp_data_catalog ]] && mkdir -p $tmp_data_catalog |
---|
| 3855 | chmod g+rx $tmp_data_catalog |
---|
| 3856 | if [[ $files_for_pes = false ]] |
---|
| 3857 | then |
---|
| 3858 | ln -f ${frelout[$i]} $tmp_data_catalog/${frelout[$i]} |
---|
| 3859 | else |
---|
| 3860 | mkdir $tmp_data_catalog/${frelout[$i]} |
---|
| 3861 | ln -f ${frelout[$i]}/* $tmp_data_catalog/${frelout[$i]} |
---|
| 3862 | fi |
---|
| 3863 | |
---|
[1289] | 3864 | # GENERATE AND SUBMIT BATCH-JOB |
---|
| 3865 | # FILE HAS TO BE DELETED FROM THE TEMPORARY DATA DIRECTORY |
---|
| 3866 | # DELETE OLD EXISTING FILES FROM THE ARCHIVE |
---|
[1] | 3867 | echo "cd $tmp_data_catalog" > archive_${frelout[$i]} |
---|
| 3868 | if [[ $files_for_pes = false ]] |
---|
| 3869 | then |
---|
| 3870 | echo "rm -rf \$PERM/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3871 | echo "cp ${frelout[$i]} \$PERM/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3872 | echo "rm -rf ${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3873 | else |
---|
| 3874 | echo "rm -rf \$PERM/${frelout[$i]}.tar" >> archive_${frelout[$i]} |
---|
| 3875 | echo "tar cvf \$PERM/${frelout[$i]}.tar ${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3876 | echo "rm -rf ${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3877 | fi |
---|
| 3878 | |
---|
[1094] | 3879 | subjob -v -d -q cdata -X 0 -m 1000 -t 43200 -c $job_catalog $PORTOPT archive_${frelout[$i]} |
---|
[1] | 3880 | printf " Archiving of $tmp_data_catalog/${frelout[$i]} initiated (batch job submitted)\n" |
---|
| 3881 | file_saved=true |
---|
| 3882 | |
---|
| 3883 | elif [[ $archive_system = ut ]] |
---|
| 3884 | then |
---|
[1289] | 3885 | # ARCHIVING ONLY POSSIBLE VIA BATCH-JOB |
---|
| 3886 | # IN THE MEANTIME, FILE IS STORED IN THE TEMPORARY DATA DIRECTORY, |
---|
| 3887 | # BECAUSE MRUN'S CURRENT TEMPORARY WORKING DIRECTORY MAY ALREADY BE DELETED |
---|
| 3888 | # WHEN THE ARCHIVE-JOB IS EXECUTED |
---|
[1] | 3889 | [[ ! -d $tmp_data_catalog ]] && mkdir -p $tmp_data_catalog |
---|
| 3890 | chmod g+rx $tmp_data_catalog |
---|
| 3891 | if [[ $files_for_pes = false ]] |
---|
| 3892 | then |
---|
| 3893 | ln -f ${frelout[$i]} $tmp_data_catalog/${frelout[$i]} |
---|
| 3894 | else |
---|
| 3895 | mkdir $tmp_data_catalog/${frelout[$i]} |
---|
| 3896 | ln -f ${frelout[$i]}/* $tmp_data_catalog/${frelout[$i]} |
---|
| 3897 | fi |
---|
| 3898 | |
---|
[1289] | 3899 | # GENERATE AND SUBMIT BATCH-JOB |
---|
| 3900 | # FILE HAS TO BE DELETED FROM THE TEMPORARY DATA DIRECTORY |
---|
| 3901 | # DELETE OLD EXISTING FILES FROM THE ARCHIVE |
---|
[1] | 3902 | echo "cd $tmp_data_catalog" > archive_${frelout[$i]} |
---|
| 3903 | if [[ $files_for_pes = false ]] |
---|
| 3904 | then |
---|
| 3905 | echo "rm -rf \$UT/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3906 | echo "cp ${frelout[$i]} \$UT/${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3907 | echo "rm -rf ${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3908 | else |
---|
| 3909 | echo "rm -rf \$UT/${frelout[$i]}.tar" >> archive_${frelout[$i]} |
---|
| 3910 | echo "tar cvf \$UT/${frelout[$i]}.tar ${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3911 | echo "rm -rf ${frelout[$i]}" >> archive_${frelout[$i]} |
---|
| 3912 | fi |
---|
| 3913 | |
---|
[1094] | 3914 | subjob -v -c /pf/b/$usern/job_queue -d -q pp -X 0 -m 1000 -t 7200 $PORTOPT archive_${frelout[$i]} |
---|
[1] | 3915 | printf " Archiving of $tmp_data_catalog/${frelout[$i]} initiated (batch job submitted)\n" |
---|
| 3916 | file_saved=true |
---|
| 3917 | |
---|
| 3918 | else |
---|
| 3919 | printf "\n +++ archive_system=\"$archive_system\" archiving impossible!" |
---|
| 3920 | archive_save=false |
---|
| 3921 | fi |
---|
[1289] | 3922 | |
---|
| 3923 | # CREATE EMPTY FLAG-FILE OR -DIRECTORY |
---|
| 3924 | # IN ORDER TO NOTE THE RESPECTIVE CYCLE NUMBER ON THE ARCHIVE SYSTEM |
---|
| 3925 | # RESTART-JOBS ARE USING THESE EMPTY-DIRECTORIES TO GET THE CURRENT HIGHEST |
---|
| 3926 | # CYCLE NUMBER ON THE ARCHIVE SYSTEM (IN CASE THAT INFORMATIONS ARE DIFFICULT TO |
---|
| 3927 | # TO ACCESS FROM THE ARCHIVE-SYSTEM DIRECTLY)) |
---|
[1] | 3928 | if [[ $file_saved = true ]] |
---|
| 3929 | then |
---|
| 3930 | if [[ $files_for_pes = false ]] |
---|
| 3931 | then |
---|
| 3932 | cat /dev/null > ${pathout[$i]} |
---|
| 3933 | else |
---|
| 3934 | mkdir -p ${pathout[$i]} |
---|
| 3935 | fi |
---|
| 3936 | fi |
---|
[1289] | 3937 | |
---|
[1] | 3938 | fi |
---|
| 3939 | |
---|
[1289] | 3940 | # APPEND ON THIS MACHINE |
---|
[1] | 3941 | if [[ "${actionout[$i]}" = "a" ]] |
---|
| 3942 | then |
---|
[741] | 3943 | if [[ "${extout[$i]}" != " " && "${extout[$i]}" != "" ]] |
---|
| 3944 | then |
---|
| 3945 | printf "\n >>> OUTPUT: ${localout[$i]} append to ${pathout[$i]}.${extout[$i]}\n" |
---|
| 3946 | cat ${localout[$i]} >> ${pathout[$i]}.${extout[$i]} |
---|
| 3947 | else |
---|
| 3948 | printf "\n >>> OUTPUT: ${localout[$i]} append to ${pathout[$i]}\n" |
---|
| 3949 | cat ${localout[$i]} >> ${pathout[$i]} |
---|
| 3950 | fi |
---|
[1] | 3951 | fi |
---|
| 3952 | |
---|
[1289] | 3953 | # COPY ON THIS MACHINE |
---|
| 3954 | # COPY HAS TO BE USED, BECAUSE MOVE DOES NOT WORK IF FILE-ORIGIN AND TARGET ARE |
---|
| 3955 | # ON DIFFERENT FILE-SYSTEMS |
---|
[1] | 3956 | if [[ "${actionout[$i]}" = "" && $files_for_pes = false ]] |
---|
| 3957 | then |
---|
| 3958 | |
---|
[1289] | 3959 | # COPY IN CASE OF RUNS ON SINGLE CORES |
---|
[5] | 3960 | if [[ "${extout[$i]}" != " " && "${extout[$i]}" != "" ]] |
---|
[1] | 3961 | then |
---|
| 3962 | printf "\n >>> OUTPUT: ${localout[$i]} to ${pathout[$i]}.${extout[$i]}\n" |
---|
[149] | 3963 | if [[ $link_local_output = true ]] |
---|
| 3964 | then |
---|
| 3965 | printf " using ln -f\n" |
---|
| 3966 | ln -f ${localout[$i]} ${pathout[$i]}.${extout[$i]} |
---|
| 3967 | fi |
---|
| 3968 | # If "ln -f" fails of if "$link_local_output = false" do a normal "cp" |
---|
| 3969 | if [[ ! -f "${pathout[$i]}.${extout[$i]}" ]] |
---|
| 3970 | then |
---|
| 3971 | if [[ $link_local_output = true ]] |
---|
| 3972 | then |
---|
| 3973 | printf " ln failed, using cp...\n" |
---|
| 3974 | fi |
---|
| 3975 | cp ${localout[$i]} ${pathout[$i]}.${extout[$i]} |
---|
| 3976 | fi |
---|
[1] | 3977 | else |
---|
| 3978 | printf "\n >>> OUTPUT: ${localout[$i]} to ${pathout[$i]}\n" |
---|
[149] | 3979 | if [[ $link_local_output = true ]] |
---|
| 3980 | then |
---|
| 3981 | printf " using ln -f\n" |
---|
| 3982 | ln -f ${localout[$i]} ${pathout[$i]} |
---|
| 3983 | fi |
---|
| 3984 | # If "ln -f" fails of if "$link_local_output = false" do a normal "cp" |
---|
| 3985 | if [[ ! -f "${pathout[$i]}" ]] |
---|
| 3986 | then |
---|
| 3987 | if [[ $link_local_output = true ]] |
---|
| 3988 | then |
---|
| 3989 | printf " ln failed, using cp...\n" |
---|
| 3990 | fi |
---|
| 3991 | cp ${localout[$i]} ${pathout[$i]} |
---|
| 3992 | fi |
---|
[1] | 3993 | fi |
---|
| 3994 | |
---|
| 3995 | elif [[ "${actionout[$i]}" = "" && $files_for_pes = true ]] |
---|
| 3996 | then |
---|
| 3997 | |
---|
[1289] | 3998 | # FILES FROM THE DIFFERENT CORES ARE MOVED WITH ln-COMMAND TO THE PERMANENT DIRECTORY |
---|
| 3999 | # AS A FIRST STEP, THE PERMANENT DIRECTORY IS CREATED |
---|
[1] | 4000 | printf "\n >>> OUTPUT: ${localout[$i]}/_.... to ${pathout[$i]}\n" |
---|
[149] | 4001 | if [[ $link_local_output = true ]] |
---|
| 4002 | then |
---|
| 4003 | printf " using ln -f\n" |
---|
| 4004 | mkdir ${pathout[$i]} |
---|
| 4005 | cd ${localout[$i]} |
---|
| 4006 | for file in $(ls *) |
---|
| 4007 | do |
---|
| 4008 | ln -f $file ${pathout[$i]} |
---|
| 4009 | done |
---|
| 4010 | cd $TEMPDIR |
---|
| 4011 | fi |
---|
[1289] | 4012 | |
---|
| 4013 | # IF "ln -f" HAS FAILED OR IF "$link_local_output = false" DO A NORMAL COPY "cp -r" |
---|
[149] | 4014 | if [[ ! -f "${pathout[$i]}/_0000" ]] |
---|
| 4015 | then |
---|
| 4016 | if [[ $link_local_output = true ]] |
---|
| 4017 | then |
---|
| 4018 | printf " ln failed for .../_0000, using cp...\n" |
---|
| 4019 | fi |
---|
| 4020 | cp -r ${localout[$i]} ${pathout[$i]} |
---|
| 4021 | fi |
---|
[1] | 4022 | |
---|
| 4023 | fi |
---|
| 4024 | fi |
---|
| 4025 | done |
---|
| 4026 | if (( i != 0 )) |
---|
| 4027 | then |
---|
| 4028 | if [[ $transfer_problems = true ]] |
---|
| 4029 | then |
---|
[1289] | 4030 | printf "\n$dashes\n *** OUTPUT-files saved" |
---|
[1] | 4031 | printf "\n +++ WARNING: some data transfers failed! \n" |
---|
| 4032 | else |
---|
[1289] | 4033 | printf "\n$dashes\n *** all OUTPUT-files saved \n" |
---|
[1] | 4034 | fi |
---|
| 4035 | fi |
---|
| 4036 | |
---|
| 4037 | |
---|
[1289] | 4038 | # IF REQUIRED, START A RESTART-JOB |
---|
| 4039 | # FILE CONTINUE_RUN MUST HAVE BEEN CREATED BY THE EXECUTABLE (PALM) |
---|
[1] | 4040 | if [[ -f CONTINUE_RUN ]] |
---|
| 4041 | then |
---|
| 4042 | |
---|
| 4043 | if [[ $archive_save = true ]] |
---|
| 4044 | then |
---|
| 4045 | |
---|
[1289] | 4046 | # ADD RESTART-OPTIONS TO THE MRUN-CALL (IF THEY ARE NOT USED ALREADY): |
---|
| 4047 | # -C TELLS MRUN THAT IT IS A RESTART-RUN |
---|
| 4048 | # -v SILENT MODE WITHOUT INTERACTIVE QUERIES |
---|
| 4049 | # -n BATCH-MODE (IMPORTANT ONLY IN CASE OF BATCH JOBS ON THE LOCAL MACHINE) |
---|
[1] | 4050 | [[ $(echo $mc | grep -c "\-C") = 0 ]] && mc="$mc -C" |
---|
| 4051 | [[ $(echo $mc | grep -c "\-v") = 0 ]] && mc="$mc -v" |
---|
| 4052 | [[ $(echo $mc | grep -c "\-b") = 0 ]] && mc="$mc -b" |
---|
| 4053 | if [[ $(echo $mc | grep -c "#") != 0 ]] |
---|
| 4054 | then |
---|
| 4055 | mc=`echo $mc | sed 's/#/f/g'` |
---|
| 4056 | fi |
---|
| 4057 | |
---|
| 4058 | |
---|
[1289] | 4059 | # START THE RESTART-JOB |
---|
[1468] | 4060 | printf "\n\n *** initiating restart-run on \"$return_address\" using command:\n" |
---|
[1] | 4061 | echo " $mc" |
---|
[1289] | 4062 | printf "\n$dashes\n" |
---|
[1] | 4063 | if [[ $localhost != $fromhost ]] |
---|
| 4064 | then |
---|
| 4065 | |
---|
[1620] | 4066 | if [[ $localhost = lcbullhh || $localhost = lccrayb || $localhost = lccrayh || $localhost = nech || $localhost = ibmh || $localhost = ibmkisti || $localhost = ibmku || $localhost = ibms || $localhost = lcflow || $localhost = lckyu* || $localhost = lcxe6 ]] |
---|
[1] | 4067 | then |
---|
| 4068 | echo "*** ssh will be used to initiate restart-runs!" |
---|
[1468] | 4069 | echo " return_address=\"$return_address\" " |
---|
[1] | 4070 | echo " return_username=\"$return_username\" " |
---|
[1468] | 4071 | if [[ $(echo $return_address | grep -c "130.75.105") = 1 ]] |
---|
[1] | 4072 | then |
---|
[1190] | 4073 | if [[ $localhost = ibmh ]] |
---|
| 4074 | then |
---|
[1468] | 4075 | ssh $SSH_PORTOPT $usern@136.172.40.15 "ssh $SSH_PORTOPT $return_address -l $return_username \". \\\$HOME/.profile; module load intel-compiler hdf5 netcdf; PATH=\\\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
[1620] | 4076 | elif [[ $localhost = lcbullhh ]] |
---|
| 4077 | then |
---|
| 4078 | ssh $SSH_PORTOPT $usern@mlogin101 "ssh $SSH_PORTOPT $return_address -l $return_username \". \\\$HOME/.profile; module load intel-compiler hdf5 netcdf; PATH=\\\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
[1255] | 4079 | elif [[ $localhost = lccrayb ]] |
---|
| 4080 | then |
---|
[1468] | 4081 | ssh $usern@blogin1 "ssh $SSH_PORTOPT $return_address -l $return_username \". \\\$HOME/.profile; module load intel-compiler hdf5 netcdf; PATH=\\\$PATH:$LOCAL_MRUN_PATH; export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
[1274] | 4082 | elif [[ $localhost = lccrayh ]] |
---|
| 4083 | then |
---|
[1468] | 4084 | ssh $usern@hlogin1 "ssh $SSH_PORTOPT $return_address -l $return_username \". \\\$HOME/.profile; module load intel-compiler hdf5 netcdf; PATH=\\\$PATH:$LOCAL_MRUN_PATH; export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
| 4085 | elif [[ $localhost = lcxe6 ]] |
---|
| 4086 | then |
---|
| 4087 | ssh $usern@hexagon "ssh $SSH_PORTOPT $return_address -l $return_username \". \\\$HOME/.profile; module load intel-compiler hdf5 netcdf; PATH=\\\$PATH:$LOCAL_MRUN_PATH; export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
[1190] | 4088 | else |
---|
[1468] | 4089 | ssh $SSH_PORTOPT $return_address -l $return_username ". \\\$HOME/.profile; module load intel-compiler hdf5 netcdf; PATH=\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc " |
---|
[1190] | 4090 | fi |
---|
[1] | 4091 | else |
---|
[693] | 4092 | if [[ $localhost = ibmkisti ]] |
---|
| 4093 | then |
---|
[1468] | 4094 | ssh $SSH_PORTOPT $usern@gaiad "ssh $SSH_PORTOPT $return_address -l $return_username \"PATH=\\\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
[1255] | 4095 | elif [[ $localhost = lcflow ]] |
---|
| 4096 | then |
---|
[1468] | 4097 | /usr/bin/ssh $SSH_PORTOPT $return_address -l $return_username "PATH=\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc " |
---|
[1866] | 4098 | elif [[ $localhost = lcocean ]] |
---|
| 4099 | then |
---|
| 4100 | /usr/bin/ssh $SSH_PORTOPT $return_address -l $return_username "PATH=\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc " |
---|
[1255] | 4101 | elif [[ $localhost = lccrayb ]] |
---|
| 4102 | then |
---|
[1468] | 4103 | ssh $usern@blogin1 "ssh $SSH_PORTOPT $return_address -l $return_username \"PATH=\\\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
[1274] | 4104 | elif [[ $localhost = lccrayh ]] |
---|
| 4105 | then |
---|
[1468] | 4106 | ssh $usern@hlogin1 "ssh $SSH_PORTOPT $return_address -l $return_username \"PATH=\\\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc\" " |
---|
[693] | 4107 | else |
---|
[1468] | 4108 | ssh $SSH_PORTOPT $return_address -l $return_username "PATH=\$PATH:$LOCAL_MRUN_PATH;export PALM_BIN=$LOCAL_MRUN_PATH;cd $LOCAL_PWD; $mc " |
---|
[693] | 4109 | fi |
---|
[1] | 4110 | fi |
---|
| 4111 | else |
---|
| 4112 | printf "\n +++ no restart mechanism available for host \"$localhost\" " |
---|
| 4113 | locat=restart; exit |
---|
| 4114 | fi |
---|
| 4115 | |
---|
[1289] | 4116 | # WAIT TO ALLOW THE RESTART-JOB TO BE QUEUED, BEFORE THE CURRENT JOB IS FINISHED |
---|
[399] | 4117 | if [[ $queue = special1q ]] |
---|
| 4118 | then |
---|
| 4119 | sleep 120 |
---|
| 4120 | else |
---|
| 4121 | sleep 30 |
---|
| 4122 | fi |
---|
[1] | 4123 | |
---|
| 4124 | else |
---|
| 4125 | |
---|
[1289] | 4126 | # JOBS RUNNING (AND STARTED) ON THE LOCAL MACHINE CAN DIRECTLY CALL MRUN (WITHOUT |
---|
| 4127 | # USING SSH) |
---|
[1] | 4128 | cd $LOCAL_PWD |
---|
[1289] | 4129 | if [[ $localhost = lckyuh ]] |
---|
[108] | 4130 | then |
---|
[1103] | 4131 | printf "\n +++ WARNING: no restart mechanism available for host \"$localhost\" " |
---|
| 4132 | printf "\n please restart job manually using command:\n" |
---|
| 4133 | printf "\n \"$mc\" " |
---|
[108] | 4134 | else |
---|
[1289] | 4135 | eval $mc # THE ' MUST BE EVALUATED |
---|
[108] | 4136 | fi |
---|
[1] | 4137 | cd - > /dev/null |
---|
| 4138 | fi |
---|
[1103] | 4139 | if [[ $localhost = lckyuh ]] |
---|
| 4140 | then |
---|
[1289] | 4141 | printf "\n$dashes\n *** restart-run to be initiated manually\n" |
---|
[1103] | 4142 | else |
---|
[1289] | 4143 | printf "\n$dashes\n *** restart-run initiated \n" |
---|
[1103] | 4144 | fi |
---|
[1] | 4145 | |
---|
| 4146 | |
---|
[1289] | 4147 | # DELETE INPUT-(RESTART)FILES, WHICH HAVE BEEN FETCHED FROM THE TEMPORARY DATA |
---|
| 4148 | # DIRECTORY, BACAUSE THEY ARE NOT REQUIRED BY THE RESTART-JOB. |
---|
| 4149 | # THIS IS DONE IN ORDER TO AVOID EXCEEDING DISC QUOTAS OR DISC SPACE (RESTART-FILES |
---|
| 4150 | # MAY BE VERY HUGE) |
---|
[1] | 4151 | (( i = 0 )) |
---|
| 4152 | while (( i < iin )) |
---|
| 4153 | do |
---|
| 4154 | (( i = i + 1 )) |
---|
| 4155 | if [[ "${got_tmp[$i]}" = true && $keep_data_from_previous_run = false ]] |
---|
| 4156 | then |
---|
| 4157 | rm -r $tmp_data_catalog/${frelin[$i]} |
---|
| 4158 | fi |
---|
| 4159 | done |
---|
| 4160 | |
---|
| 4161 | else |
---|
| 4162 | |
---|
| 4163 | printf "\n +++ no restart-run possible, since errors occured" |
---|
| 4164 | printf "\n during the archive process" |
---|
| 4165 | fi |
---|
| 4166 | |
---|
| 4167 | fi |
---|
| 4168 | |
---|
| 4169 | |
---|
| 4170 | |
---|
[1289] | 4171 | # SEND EMAIL NOTIFICATION ABOUT THE FINISHED RUN |
---|
[352] | 4172 | if [[ "$email_notification" != "none" ]] |
---|
[1] | 4173 | then |
---|
| 4174 | |
---|
| 4175 | if [[ $localhost != $fromhost ]] |
---|
| 4176 | then |
---|
| 4177 | if [[ -f CONTINUE_RUN ]] |
---|
| 4178 | then |
---|
| 4179 | echo "PALM restart run necessary" > email_text |
---|
| 4180 | echo "description header of actual run:" >> email_text |
---|
| 4181 | cat CONTINUE_RUN >> email_text |
---|
| 4182 | echo "mrun-command to restart:" >> email_text |
---|
| 4183 | echo "$mc" >> email_text |
---|
| 4184 | else |
---|
| 4185 | echo "PALM run with base filename \"$fname\" on host \"$localhost\" finished" > email_text |
---|
| 4186 | fi |
---|
| 4187 | mail $email_notification < email_text |
---|
| 4188 | printf "\n *** email notification sent to \"$email_notification\" " |
---|
| 4189 | fi |
---|
| 4190 | fi |
---|
| 4191 | |
---|
| 4192 | |
---|
| 4193 | |
---|
[1289] | 4194 | # ALL ACTIONS FINISHED, TEMPORARY WORKING-DIRECTORY CAN BE DELETED |
---|
[1] | 4195 | cd $HOME |
---|
| 4196 | [[ $delete_temporary_catalog = true ]] && rm -rf $TEMPDIR |
---|
| 4197 | |
---|
| 4198 | else |
---|
| 4199 | |
---|
[1289] | 4200 | # PREPARING ACTIONS, |
---|
| 4201 | # IF A BATCH-JOB IS TO BE GENERATED AND STARTED ON A REMOTE-MACHINE GERECHNET |
---|
[1] | 4202 | |
---|
[1289] | 4203 | # BUILD THE MRUN-COMMAND TO BE CALLED IN THE BATCH-JOB ON THE REMOTE-MACHINE |
---|
[1468] | 4204 | mrun_com="$mrun_script_name -a $afname -c $config_file -d $fname -h $host -H $fromhost -m $memory -t $cpumax -q $queue -R $return_address -U $return_username -u $remote_username" |
---|
[1] | 4205 | [[ "$cpp_opts" != "" ]] && mrun_com=${mrun_com}" -D \"$cpp_opts\"" |
---|
[79] | 4206 | [[ "$global_revision" != "" ]] && mrun_com=${mrun_com}" -G \"$global_revision\"" |
---|
[123] | 4207 | [[ $group_number != none ]] && mrun_com=${mrun_com}" -g $group_number" |
---|
[1] | 4208 | [[ $do_compile = true ]] && mrun_com=${mrun_com}" -s \"$source_list\"" |
---|
| 4209 | [[ "$input_list" != "" ]] && mrun_com=${mrun_com}" -i \"$input_list\"" |
---|
| 4210 | [[ $ignore_archive_error = true ]] && mrun_com=${mrun_com}" -I" |
---|
| 4211 | [[ $keep_data_from_previous_run = true ]] && mrun_com=${mrun_com}" -k" |
---|
| 4212 | [[ "$additional_conditions" != "" ]] && mrun_com=${mrun_com}" -K \"$additional_conditions\"" |
---|
| 4213 | [[ "$output_list" != "" ]] && mrun_com=${mrun_com}" -o \"$output_list\"" |
---|
| 4214 | [[ "$read_from_config" = false ]] && mrun_com=${mrun_com}" -S" |
---|
| 4215 | [[ $do_trace = true ]] && mrun_com=${mrun_com}" -x" |
---|
| 4216 | [[ "$numprocs" != "" ]] && mrun_com=${mrun_com}" -X $numprocs" |
---|
| 4217 | if [[ $use_openmp = true ]] |
---|
| 4218 | then |
---|
[622] | 4219 | mrun_com=${mrun_com}" -O $threads_per_task" |
---|
[1] | 4220 | fi |
---|
[759] | 4221 | [[ "$tasks_per_node" != "" ]] && mrun_com=${mrun_com}" -T $tasks_per_node" |
---|
[397] | 4222 | [[ $store_on_archive_system = true ]] && mrun_com=${mrun_com}" -A" |
---|
[1] | 4223 | [[ $package_list != "" ]] && mrun_com=${mrun_com}" -p \"$package_list\"" |
---|
| 4224 | [[ $return_password != "" ]] && mrun_com=${mrun_com}" -P $return_password" |
---|
| 4225 | [[ $delete_temporary_catalog = false ]] && mrun_com=${mrun_com}" -B" |
---|
[127] | 4226 | [[ $node_usage != default && "$(echo $node_usage | cut -c1-3)" != "sla" && $node_usage != novice ]] && mrun_com=${mrun_com}" -n $node_usage" |
---|
[291] | 4227 | [[ "$ocean_file_appendix" = true ]] && mrun_com=${mrun_com}" -y" |
---|
[206] | 4228 | [[ $run_coupled_model = true ]] && mrun_com=${mrun_com}" -Y \"$coupled_dist\"" |
---|
[920] | 4229 | [[ "$check_namelist_files" = false ]] && mrun_com=${mrun_com}" -z" |
---|
| 4230 | [[ "$combine_plot_fields" = false ]] && mrun_com=${mrun_com}" -Z" |
---|
[759] | 4231 | [[ "$max_par_io_str" != "" ]] && mrun_com=${mrun_com}" -w $max_par_io_str" |
---|
[1] | 4232 | if [[ $do_remote = true ]] |
---|
| 4233 | then |
---|
| 4234 | printf "\n>>>> MRUN-command on execution host:\n>>>> $mrun_com \n" |
---|
| 4235 | fi |
---|
| 4236 | |
---|
| 4237 | |
---|
[1289] | 4238 | # CREATE EXECUTABLE FOR BATCH JOB |
---|
[1199] | 4239 | if [[ $create_executable_for_batch = true && $restart_run != true ]] |
---|
| 4240 | then |
---|
| 4241 | |
---|
| 4242 | printf "\n *** creating the executable for batch job\n" |
---|
| 4243 | |
---|
[1289] | 4244 | # METHOD ONLY WORKS FOR BATCH JOBS ON LOCAL HOSTS |
---|
[1199] | 4245 | if [[ $host != $localhost ]] |
---|
| 4246 | then |
---|
| 4247 | printf "\n +++ creation of executables is only allowed for batch jobs on local hosts." |
---|
| 4248 | printf "\n Please set create_executable_for_batch = false in the config-file.\n" |
---|
| 4249 | locat=create_executable; exit |
---|
| 4250 | fi |
---|
| 4251 | |
---|
| 4252 | mkdir $working_directory/SOURCES_FOR_RUN_${fname}/TMPDIR_FOR_CREATING_EXECUTABLE |
---|
| 4253 | cd $working_directory/SOURCES_FOR_RUN_${fname}/TMPDIR_FOR_CREATING_EXECUTABLE |
---|
| 4254 | |
---|
| 4255 | cp $make_depository . |
---|
| 4256 | tar -xf $make_depository > /dev/null 2>&1 |
---|
| 4257 | cp ../* . > /dev/null 2>&1 |
---|
| 4258 | |
---|
| 4259 | make $mopts -f Makefile PROG=a.out F90=$compiler_name COPT="$cpp_options" F90FLAGS="$fopts" LDFLAGS="$lopts" |
---|
| 4260 | |
---|
| 4261 | if [[ $? != 0 || "$compile_error" = true || "$module_compile_error" = true ]] |
---|
| 4262 | then |
---|
| 4263 | printf "\n +++ error occured while compiling or linking" |
---|
| 4264 | locat=compile |
---|
| 4265 | exit |
---|
| 4266 | fi |
---|
| 4267 | |
---|
| 4268 | mv a.out .. |
---|
| 4269 | cd - > /dev/null 2>&1 |
---|
| 4270 | rm -rf $working_directory/SOURCES_FOR_RUN_${fname}/TMPDIR_FOR_CREATING_EXECUTABLE |
---|
| 4271 | |
---|
| 4272 | printf " *** executable created in \"$working_directory/SOURCES_FOR_RUN_${fname}\"\n " |
---|
| 4273 | |
---|
| 4274 | fi |
---|
| 4275 | |
---|
| 4276 | |
---|
[1289] | 4277 | # BUILD THE JOB-SCRIPTS ON FILE jobfile |
---|
[1] | 4278 | jobfile=jobfile.$RANDOM |
---|
| 4279 | |
---|
| 4280 | |
---|
[1289] | 4281 | # CREATE TEMPORARY DIRECTORY AND SWITCH TO IT |
---|
[1] | 4282 | echo "mkdir $TEMPDIR" >> $jobfile |
---|
| 4283 | echo "cd $TEMPDIR" >> $jobfile |
---|
| 4284 | |
---|
| 4285 | |
---|
[1289] | 4286 | # ACTIVATE ERROR-TRACEBACK |
---|
[1] | 4287 | if [[ $do_trace = true ]] |
---|
| 4288 | then |
---|
| 4289 | echo "set -x" >> $jobfile |
---|
| 4290 | else |
---|
| 4291 | echo "set +vx" >> $jobfile |
---|
| 4292 | fi |
---|
| 4293 | |
---|
| 4294 | |
---|
[1289] | 4295 | # INITIALIZE THE ENVIRONMENT AND LOAD MODULES |
---|
[1043] | 4296 | if [[ "$init_cmds" != "" ]] |
---|
| 4297 | then |
---|
| 4298 | echo "$init_cmds" >> $jobfile |
---|
| 4299 | fi |
---|
| 4300 | if [[ "$module_calls" != "" ]] |
---|
| 4301 | then |
---|
| 4302 | echo "$module_calls" >> $jobfile |
---|
| 4303 | fi |
---|
| 4304 | |
---|
| 4305 | |
---|
[1289] | 4306 | # PROVIDE SOURCE-CODE FILES, MRUN-SCRIPT AND CONFIGURATION-FILE FOR THE JOB |
---|
| 4307 | # then-CLAUSE: FILES ARE COLLECTED IN THE SOURCES_FOR_RUN_... DIRECTORY ON THE LOCAL HOST, |
---|
| 4308 | # WHICH IS THEN FETCHED FROM THE BATCH-JOB BY USING cp/scp |
---|
| 4309 | # THE SOURCE-CODE FILES ALREADY EXIST IN THIS DIRECTORY |
---|
| 4310 | # else-CLAUSE: FILE-CONTENTS ARE PUT INTO THE JOB-FILE AS HERE-DOCUMENTS |
---|
| 4311 | # THIS MAY CREATE A QUITE LARGE JOB-FILE, WHICH CAN CAUSE PROBLEMS WITH SOME |
---|
| 4312 | # QUEUEING-SYSTEMS |
---|
[1866] | 4313 | if [[ $host = ibmkisti || $host = lcbullhh || $host = lccrayb || $host = lccrayf || $host = lccrayh || $host = lcocean ]] |
---|
[1] | 4314 | then |
---|
| 4315 | |
---|
[1289] | 4316 | # COPY CONFIGURATION-FILE AND MRUN-SCRIPT INTO THE SOURCES_FOR_RUN... DIRECTORY |
---|
[191] | 4317 | if [[ $restart_run != true ]] |
---|
| 4318 | then |
---|
| 4319 | cp $config_file $working_directory/SOURCES_FOR_RUN_$fname |
---|
| 4320 | cp ${PALM_BIN}/$mrun_script_name $working_directory/SOURCES_FOR_RUN_$fname |
---|
| 4321 | fi |
---|
[1] | 4322 | |
---|
[1289] | 4323 | # COPY THE SOURCES_FOR_RUN... DIRECTORY FROM THE LOCAL HOST TO THE JOB VIA scp |
---|
| 4324 | # (then-CLAUSE: JOBS ON THE LOCAL HOST CAN JUST USE cp) |
---|
[193] | 4325 | echo "set -x" >> $jobfile |
---|
[693] | 4326 | if [[ $host = $localhost ]] |
---|
| 4327 | then |
---|
[1199] | 4328 | # DUE TO UNKNOWN REASONS, COPY WITH cp COMMAND CREATES CORRUPT |
---|
| 4329 | # FILES ON CRAY XC30 SYSTEMS (CSC HELSINKI), rsync IS USED INSTEAD |
---|
[1281] | 4330 | if [[ $(echo $host | cut -c1-6) = lccray ]] |
---|
| 4331 | then |
---|
| 4332 | echo "rsync -av -t $working_directory/SOURCES_FOR_RUN_$fname ." >> $jobfile |
---|
| 4333 | else |
---|
| 4334 | echo "cp -r $working_directory/SOURCES_FOR_RUN_$fname ." >> $jobfile |
---|
| 4335 | fi |
---|
[1199] | 4336 | |
---|
[693] | 4337 | else |
---|
[1255] | 4338 | if [[ $host = ibmkisti ]] |
---|
[693] | 4339 | then |
---|
[1289] | 4340 | # ON KISTI'S IBM FIREWALL IS ONLY OPENED ON INTERACTIVE NODE |
---|
[693] | 4341 | echo "localdir=\`pwd\`" >> $jobfile |
---|
[1468] | 4342 | echo "ssh $SSH_PORTOPT $remote_username@gaiad \"cd \$localdir; scp $PORTOPT -r $return_username@$return_address:$working_directory/SOURCES_FOR_RUN_$fname .\" " >> $jobfile |
---|
[1620] | 4343 | elif [[ $host = lcbullhh ]] |
---|
| 4344 | then |
---|
| 4345 | echo "localdir=\`pwd\`" >> $jobfile |
---|
| 4346 | echo "ssh $SSH_PORTOPT $remote_username@mlogin101 \"cd \$localdir; scp $PORTOPT -r $return_username@$return_address:$working_directory/SOURCES_FOR_RUN_$fname .\" " >> $jobfile |
---|
| 4347 | |
---|
[1255] | 4348 | elif [[ $host = lccrayb ]] |
---|
| 4349 | then |
---|
| 4350 | echo "localdir=\`pwd\`" >> $jobfile |
---|
[1468] | 4351 | echo "ssh $SSH_PORTOPT $remote_username@blogin1 \"cd \$localdir; scp $PORTOPT -r $return_username@$return_address:$working_directory/SOURCES_FOR_RUN_$fname .\" " >> $jobfile |
---|
[1274] | 4352 | elif [[ $host = lccrayh ]] |
---|
| 4353 | then |
---|
| 4354 | echo "localdir=\`pwd\`" >> $jobfile |
---|
[1468] | 4355 | echo "ssh $SSH_PORTOPT $remote_username@hlogin1 \"cd \$localdir; scp $PORTOPT -r $return_username@$return_address:$working_directory/SOURCES_FOR_RUN_$fname .\" " >> $jobfile |
---|
[1274] | 4356 | |
---|
[1255] | 4357 | else |
---|
[1468] | 4358 | echo "scp $PORTOPT -r $return_username@$return_address:$working_directory/SOURCES_FOR_RUN_$fname ." >> $jobfile |
---|
[693] | 4359 | fi |
---|
| 4360 | fi |
---|
[193] | 4361 | echo "export SOURCES_COMPLETE=true" >> $jobfile |
---|
[1] | 4362 | |
---|
[1289] | 4363 | # MOVE MRUN-SCRIPT AND CONFIGURATION-FILE FROM THE SOURCES_FOR_RUN... DIRECTORY TO THE |
---|
| 4364 | # WORKING DIRECTORY OF THE JOB |
---|
[193] | 4365 | echo "mv SOURCES_FOR_RUN_$fname/$config_file . " >> $jobfile |
---|
| 4366 | echo "mv SOURCES_FOR_RUN_$fname/$mrun_script_name . " >> $jobfile |
---|
[1199] | 4367 | echo "chmod u+rwx $mrun_script_name" >> $jobfile |
---|
[193] | 4368 | echo "execute_mrun=true" >> $jobfile |
---|
| 4369 | echo " " >> $jobfile |
---|
| 4370 | |
---|
[191] | 4371 | else |
---|
| 4372 | |
---|
[1289] | 4373 | # PROVIDE SOURCE-CODE FILES AND MAKEFILE AS HERE DOCUMENT |
---|
[191] | 4374 | if [[ $do_compile = true ]] |
---|
| 4375 | then |
---|
| 4376 | |
---|
| 4377 | source_catalog=SOURCES_FOR_RUN_$fname |
---|
| 4378 | |
---|
[1289] | 4379 | # CREATE SOURCES_FOR_RUN... DIRECTORY TO STORE THE SOURCE CODE FILES AND THE MAKEFILE |
---|
[191] | 4380 | echo "mkdir SOURCES_FOR_RUN_$fname" >> $jobfile |
---|
| 4381 | echo "export SOURCES_COMPLETE=true" >> $jobfile |
---|
| 4382 | echo "cd SOURCES_FOR_RUN_$fname" >> $jobfile |
---|
| 4383 | |
---|
| 4384 | for filename in $source_list |
---|
| 4385 | do |
---|
[1289] | 4386 | # BACKSLASH IS USED FOR MASKING |
---|
[191] | 4387 | echo "cat > $filename << \"%END%\"" >> $jobfile |
---|
| 4388 | cat $source_catalog/$filename >> $jobfile |
---|
| 4389 | echo " " >> $jobfile |
---|
| 4390 | echo "%END%" >> $jobfile |
---|
| 4391 | echo " " >> $jobfile |
---|
| 4392 | done |
---|
| 4393 | |
---|
[1289] | 4394 | # BACKSLASH IS USED FOR MASKING |
---|
[191] | 4395 | echo "cat > Makefile << \"%END%\"" >> $jobfile |
---|
| 4396 | cat $source_catalog/Makefile >> $jobfile |
---|
[1] | 4397 | echo " " >> $jobfile |
---|
| 4398 | echo "%END%" >> $jobfile |
---|
| 4399 | echo " " >> $jobfile |
---|
| 4400 | |
---|
[191] | 4401 | echo "cd - > /dev/null" >> $jobfile |
---|
[1] | 4402 | |
---|
[191] | 4403 | fi |
---|
[1] | 4404 | |
---|
| 4405 | |
---|
[1289] | 4406 | # PROVIDE THE CONFIGURATION-FILE AS HERE-DOCUMENT |
---|
| 4407 | # BACKSLASH IS USED FOR MASKING |
---|
| 4408 | # LINES WITH #$ IN THE CONFIGURATION-FILE, COMING FROM THE SVN KEYWORD SUBSTITUTION, |
---|
| 4409 | # ARE REMOVED FROM THE FILE IN ORDER TO AVOID PROBLEMS WITH THE SGE BATCH SYSTEM |
---|
[191] | 4410 | echo "cat > $config_file << \"%END%\"" >> $jobfile |
---|
[1101] | 4411 | if [[ $host = lckyuh ]] |
---|
| 4412 | then |
---|
[1289] | 4413 | # NO CROSS-COMPILER ON COMPUTE NODE |
---|
[1101] | 4414 | sed 's/frtpx/frt/g' $config_file >> $jobfile |
---|
| 4415 | else |
---|
| 4416 | sed 's/#$.*//g' $config_file >> $jobfile |
---|
| 4417 | fi |
---|
[191] | 4418 | echo "%END%" >> $jobfile |
---|
| 4419 | echo " " >> $jobfile |
---|
[1] | 4420 | |
---|
| 4421 | |
---|
[1289] | 4422 | # PROVIDE THE MRUN-SCRIPTS AS HERE-DOCUMENT |
---|
| 4423 | # BACKSLASH IS USED FOR MASKING |
---|
[1101] | 4424 | echo "cat > $mrun_script_name <<\"%END%\"" >> $jobfile |
---|
[1103] | 4425 | if [[ $host = lckyuh ]] |
---|
| 4426 | then |
---|
| 4427 | sed 's/\/bin\/ksh/\/bin\/bash/g' ${PALM_BIN}/$mrun_script_name >> $jobfile |
---|
| 4428 | else |
---|
| 4429 | cat ${PALM_BIN}/$mrun_script_name >> $jobfile |
---|
| 4430 | fi |
---|
[1101] | 4431 | echo "%END%" >> $jobfile |
---|
[191] | 4432 | echo "chmod u+x $mrun_script_name" >> $jobfile |
---|
| 4433 | echo "execute_mrun=true" >> $jobfile |
---|
| 4434 | echo " " >> $jobfile |
---|
[1] | 4435 | |
---|
| 4436 | fi |
---|
| 4437 | |
---|
[191] | 4438 | |
---|
[1289] | 4439 | # GET REQUIRED INPUT-FILES BY SCP OR BY SENDING THEM WITH THE JOB AS HERE-DOCUMENT |
---|
| 4440 | # PUT THESE FILES INTO THE USER'S RESPECTIVE PERMANENT DIRECTORIES ON THE REMOTE-HOST |
---|
| 4441 | # IF THE DIRECTORIES DO NOT EXIST, TRY TO CREATE THEM |
---|
[1] | 4442 | if [[ $do_remote = true ]] |
---|
| 4443 | then |
---|
| 4444 | (( i = 0 )) |
---|
| 4445 | while (( i < iin )) |
---|
| 4446 | do |
---|
| 4447 | (( i = i + 1 )) |
---|
| 4448 | echo "[[ ! -d ${pathin[$i]} ]] && mkdir -p ${pathin[$i]}" >> $jobfile |
---|
| 4449 | if [[ "${transin[$i]}" = job ]] |
---|
| 4450 | then |
---|
| 4451 | echo "cat > ${remotepathin[$i]} <<\"%END%\"" >> $jobfile |
---|
| 4452 | eval cat ${pathin[$i]}/${frelin[$i]} >> $jobfile |
---|
| 4453 | echo " " >> $jobfile |
---|
| 4454 | echo "%END%" >> $jobfile |
---|
| 4455 | else |
---|
[1468] | 4456 | echo "batch_scp $PORTOPT -b -o -g -s -u $return_username $return_address ${remotepathin[$i]} \"${pathin[$i]}\" ${frelin[$i]}" >> $jobfile |
---|
[1] | 4457 | fi |
---|
| 4458 | |
---|
[1289] | 4459 | # CHECK, IF FILE COULD BE CREATED |
---|
[1] | 4460 | echo "if [[ \$? = 1 ]]" >> $jobfile |
---|
| 4461 | echo "then" >> $jobfile |
---|
| 4462 | echo " echo \" \" " >> $jobfile |
---|
| 4463 | echo " echo \"+++ file ${remotepathin[$i]} could not be created\" " >> $jobfile |
---|
| 4464 | echo " echo \" please check, if directory exists on $host!\" " >> $jobfile |
---|
| 4465 | echo " echo \"+++ MRUN will not be continued\" " >> $jobfile |
---|
| 4466 | echo " execute_mrun=false" >> $jobfile |
---|
| 4467 | echo "fi" >> $jobfile |
---|
| 4468 | done |
---|
| 4469 | fi |
---|
| 4470 | |
---|
[1289] | 4471 | # PROVIDE NAME OF THE CURRENT WORKING-DIRECTORY ON THE LOCAL MACHINE (FROM WHERE THE JOB IS |
---|
| 4472 | # STARTED) BY SETTING AN ENVIRONMENT-VARIABLE. THIS INFORMATION IS USED IN THE JOB BY MRUN |
---|
| 4473 | # IN CASE THAT RESTART-RUNS HAVE TO BE GENERATED |
---|
[1] | 4474 | echo "LOCAL_PWD=$working_directory" >> $jobfile |
---|
| 4475 | echo "export LOCAL_PWD" >> $jobfile |
---|
| 4476 | |
---|
[1289] | 4477 | # PROVIDE THE PATH OF THE LOCAL MRUN-SCRIPT FOR THE SAME REASON |
---|
[22] | 4478 | echo "LOCAL_MRUN_PATH=$PALM_BIN" >> $jobfile |
---|
[1] | 4479 | echo "export LOCAL_MRUN_PATH" >> $jobfile |
---|
| 4480 | |
---|
[1289] | 4481 | # lcflow ALSO REQUIRES TO PROVIDE PATH FOR THE PALM-SCRIPTS |
---|
[1547] | 4482 | # if [[ $host = lcflow || $localhost = lcflow ]] |
---|
| 4483 | # then |
---|
[1099] | 4484 | echo "export PALM_BIN=$PALM_BIN" | sed -e 's:'$HOME':$HOME:' >> $jobfile |
---|
| 4485 | echo "export PATH=\$PATH:\$PALM_BIN" >> $jobfile |
---|
[1547] | 4486 | # fi |
---|
[1] | 4487 | |
---|
[1289] | 4488 | # CALL MRUN WITHIN THE JOB (SETTING QUEUE IS A WORKAROUND FOR ibmkisti) |
---|
| 4489 | # AS FINAL ACTION, REMOVE THE TEMPORARY DIRECTORY CREATED AT THE BEGINNING OF THE JOB |
---|
[1101] | 4490 | echo "set -x" >> $jobfile |
---|
| 4491 | echo "queue=$queue" >> $jobfile |
---|
[1] | 4492 | echo "[[ \$execute_mrun = true ]] && ./$mrun_com" >> $jobfile |
---|
[1101] | 4493 | echo 'ls -al; echo `pwd`' >> $jobfile |
---|
| 4494 | echo "cd \$HOME" >> $jobfile |
---|
| 4495 | echo "rm -rf $TEMPDIR" >> $jobfile |
---|
[1] | 4496 | |
---|
| 4497 | |
---|
| 4498 | |
---|
| 4499 | |
---|
[1289] | 4500 | # START THE JOB USING SUBJOB-COMMAND |
---|
[1] | 4501 | if [[ $silent = false ]] |
---|
| 4502 | then |
---|
| 4503 | printf "\n " |
---|
| 4504 | else |
---|
| 4505 | printf "\n\n" |
---|
| 4506 | fi |
---|
| 4507 | |
---|
[1094] | 4508 | subjob $job_on_file -h $host -u $remote_username -g $group_number -q $queue -m $memory -N $node_usage -t $cpumax $XOPT $TOPT $OOPT -n $fname -v -c $job_catalog -e $email_notification $PORTOPT $jobfile |
---|
[1] | 4509 | rm -rf $jobfile |
---|
| 4510 | |
---|
| 4511 | |
---|
[1289] | 4512 | fi # END OF REMOTE-PART |
---|