[12] | 1 | /* $Id$ */ |
---|
[1] | 2 | /* |
---|
| 3 | * FedStage DRMAA for PBS Pro |
---|
| 4 | * Copyright (C) 2006-2009 FedStage Systems |
---|
| 5 | * |
---|
| 6 | * This program is free software: you can redistribute it and/or modify |
---|
| 7 | * it under the terms of the GNU General Public License as published by |
---|
| 8 | * the Free Software Foundation, either version 3 of the License, or |
---|
| 9 | * (at your option) any later version. |
---|
| 10 | * |
---|
| 11 | * This program is distributed in the hope that it will be useful, |
---|
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
---|
| 14 | * GNU General Public License for more details. |
---|
| 15 | * |
---|
| 16 | * You should have received a copy of the GNU General Public License |
---|
| 17 | * along with this program. If not, see <http://www.gnu.org/licenses/>. |
---|
| 18 | */ |
---|
| 19 | |
---|
| 20 | #ifdef HAVE_CONFIG_H |
---|
| 21 | # include <config.h> |
---|
| 22 | #endif |
---|
| 23 | |
---|
| 24 | #include <signal.h> |
---|
| 25 | #include <stdlib.h> |
---|
| 26 | #include <string.h> |
---|
| 27 | #include <unistd.h> |
---|
| 28 | |
---|
| 29 | #include <drmaa_utils/drmaa.h> |
---|
| 30 | #include <drmaa_utils/drmaa_util.h> |
---|
| 31 | #include <pbs_error.h> |
---|
| 32 | #include <pbs_ifl.h> |
---|
| 33 | |
---|
| 34 | #include <pbs_drmaa/job.h> |
---|
[7] | 35 | #include <pbs_drmaa/log_reader.h> |
---|
[1] | 36 | #include <pbs_drmaa/pbs_attrib.h> |
---|
| 37 | #include <pbs_drmaa/session.h> |
---|
| 38 | #include <pbs_drmaa/util.h> |
---|
| 39 | |
---|
| 40 | #ifndef lint |
---|
| 41 | static char rcsid[] |
---|
| 42 | # ifdef __GNUC__ |
---|
| 43 | __attribute__ ((unused)) |
---|
| 44 | # endif |
---|
[12] | 45 | = "$Id$"; |
---|
[1] | 46 | #endif |
---|
| 47 | |
---|
| 48 | |
---|
| 49 | static void |
---|
| 50 | pbsdrmaa_job_control( fsd_job_t *self, int action ); |
---|
| 51 | |
---|
| 52 | static void |
---|
| 53 | pbsdrmaa_job_update_status( fsd_job_t *self ); |
---|
| 54 | |
---|
| 55 | static void |
---|
| 56 | pbsdrmaa_job_on_missing( fsd_job_t *self ); |
---|
| 57 | |
---|
[7] | 58 | void |
---|
| 59 | pbsdrmaa_job_on_missing_standard( fsd_job_t *self ); |
---|
| 60 | |
---|
[1] | 61 | static void |
---|
| 62 | pbsdrmaa_job_update( fsd_job_t *self, struct batch_status* ); |
---|
| 63 | |
---|
| 64 | |
---|
| 65 | fsd_job_t * |
---|
| 66 | pbsdrmaa_job_new( char *job_id ) |
---|
| 67 | { |
---|
| 68 | pbsdrmaa_job_t *self = (pbsdrmaa_job_t*)fsd_job_new( job_id ); |
---|
| 69 | fsd_realloc( self, 1, pbsdrmaa_job_t ); |
---|
| 70 | self->super.control = pbsdrmaa_job_control; |
---|
| 71 | self->super.update_status = pbsdrmaa_job_update_status; |
---|
| 72 | self->super.on_missing = pbsdrmaa_job_on_missing; |
---|
| 73 | self->update = pbsdrmaa_job_update; |
---|
| 74 | return (fsd_job_t*)self; |
---|
| 75 | } |
---|
| 76 | |
---|
| 77 | |
---|
| 78 | static void |
---|
| 79 | pbsdrmaa_job_control( fsd_job_t *self, int action ) |
---|
| 80 | { |
---|
| 81 | volatile bool conn_lock = false; |
---|
| 82 | pbsdrmaa_session_t *session = (pbsdrmaa_session_t*)self->session; |
---|
| 83 | const char *job_id = self->job_id; |
---|
| 84 | const char *apicall = NULL; |
---|
| 85 | int rc = PBSE_NONE; |
---|
| 86 | |
---|
| 87 | fsd_log_enter(( "({job_id=%s}, action=%d)", |
---|
| 88 | self->job_id, action )); |
---|
| 89 | |
---|
| 90 | TRY |
---|
| 91 | { |
---|
[45] | 92 | int tries_left = session->max_retries_count; |
---|
| 93 | int sleep_time = 1; |
---|
[1] | 94 | |
---|
| 95 | conn_lock = fsd_mutex_lock( &self->session->drm_connection_mutex ); |
---|
| 96 | |
---|
| 97 | /*TODO reconnect */ |
---|
[45] | 98 | while ( true ) |
---|
[1] | 99 | { |
---|
| 100 | switch( action ) |
---|
| 101 | { |
---|
| 102 | /* |
---|
| 103 | * We cannot know whether we did suspend job |
---|
| 104 | * in other way than remembering this inside DRMAA session. |
---|
| 105 | */ |
---|
| 106 | case DRMAA_CONTROL_SUSPEND: |
---|
| 107 | apicall = "pbs_sigjob"; |
---|
| 108 | rc = pbs_sigjob( session->pbs_conn, (char*)job_id, |
---|
| 109 | "SIGSTOP", NULL ); |
---|
[10] | 110 | fsd_log_info(("pbs_sigjob(%s, SIGSTOP) =%d", job_id, rc)); |
---|
[1] | 111 | if( rc == PBSE_NONE ) |
---|
| 112 | self->flags |= FSD_JOB_SUSPENDED; |
---|
| 113 | break; |
---|
| 114 | case DRMAA_CONTROL_RESUME: |
---|
| 115 | apicall = "pbs_sigjob"; |
---|
| 116 | rc = pbs_sigjob( session->pbs_conn, (char*)job_id, |
---|
| 117 | "SIGCONT", NULL ); |
---|
[10] | 118 | fsd_log_info(("pbs_sigjob(%s, SIGCONT) =%d", job_id, rc)); |
---|
[1] | 119 | if( rc == PBSE_NONE ) |
---|
| 120 | self->flags &= ~FSD_JOB_SUSPENDED; |
---|
| 121 | break; |
---|
| 122 | case DRMAA_CONTROL_HOLD: |
---|
| 123 | apicall = "pbs_holdjob"; |
---|
| 124 | rc = pbs_holdjob( session->pbs_conn, (char*)job_id, |
---|
| 125 | USER_HOLD, NULL ); |
---|
[10] | 126 | fsd_log_info(("pbs_sigjob(%s, SIGHOLD) =%d", job_id, rc)); |
---|
[1] | 127 | if( rc == PBSE_NONE ) |
---|
| 128 | self->flags |= FSD_JOB_HOLD; |
---|
| 129 | break; |
---|
| 130 | case DRMAA_CONTROL_RELEASE: |
---|
| 131 | apicall = "pbs_rlsjob"; |
---|
| 132 | rc = pbs_rlsjob( session->pbs_conn, (char*)job_id, |
---|
| 133 | USER_HOLD, NULL ); |
---|
[10] | 134 | fsd_log_info(("pbs_rlsjob(%s) =%d", job_id, rc)); |
---|
[1] | 135 | if( rc == PBSE_NONE ) |
---|
| 136 | self->flags &= FSD_JOB_HOLD; |
---|
| 137 | break; |
---|
| 138 | case DRMAA_CONTROL_TERMINATE: |
---|
| 139 | apicall = "pbs_deljob"; |
---|
| 140 | rc = pbs_deljob( session->pbs_conn, (char*)job_id, NULL ); |
---|
[10] | 141 | fsd_log_info(("pbs_deljob(%s) =%d", job_id, rc)); |
---|
[1] | 142 | /* Torque: |
---|
| 143 | * deldelay=N -- delay between SIGTERM and SIGKILL (default 0) */ |
---|
| 144 | if( rc == PBSE_NONE ) |
---|
| 145 | { |
---|
| 146 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
| 147 | if( (self->flags & FSD_JOB_TERMINATED) == 0 ) |
---|
| 148 | self->flags |= FSD_JOB_TERMINATED | FSD_JOB_ABORTED; |
---|
| 149 | } |
---|
| 150 | break; |
---|
| 151 | } |
---|
| 152 | |
---|
[45] | 153 | retry_connect: |
---|
| 154 | if ( rc == PBSE_NONE ) |
---|
[1] | 155 | break; |
---|
[45] | 156 | else if (( rc == PBSE_INTERNAL || rc == PBSE_PROTOCOL || rc == PBSE_EXPIRED) && (tries_left--)) |
---|
[1] | 157 | { |
---|
[45] | 158 | if (rc == PBSE_PROTOCOL || rc == PBSE_EXPIRED) |
---|
| 159 | { |
---|
| 160 | if ( session->pbs_conn >= 0) |
---|
| 161 | pbs_disconnect( session->pbs_conn ); |
---|
| 162 | |
---|
| 163 | sleep( sleep_time++ ); |
---|
| 164 | |
---|
| 165 | session->pbs_conn = pbs_connect( session->super.contact ); |
---|
| 166 | |
---|
| 167 | if (session->pbs_conn < 0) |
---|
| 168 | goto retry_connect; |
---|
| 169 | |
---|
| 170 | fsd_log_info(( "pbs_connect(%s) =%d", session->super.contact, session->pbs_conn )); |
---|
| 171 | } |
---|
| 172 | else /* PBSE_INTERNAL */ |
---|
| 173 | { |
---|
| 174 | /* |
---|
| 175 | * In PBS Pro pbs_sigjob raises internal server error (PBSE_INTERNAL) |
---|
| 176 | * when job just changed its state to running. |
---|
| 177 | */ |
---|
| 178 | sleep( sleep_time++ ); |
---|
| 179 | } |
---|
| 180 | fsd_log_debug(( "repeating request (%d of %d)", tries_left, session->max_retries_count)); |
---|
[1] | 181 | } |
---|
| 182 | else |
---|
| 183 | pbsdrmaa_exc_raise_pbs( apicall ); |
---|
[45] | 184 | } /* end while */ |
---|
[1] | 185 | } |
---|
| 186 | FINALLY |
---|
| 187 | { |
---|
| 188 | if( conn_lock ) |
---|
| 189 | conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex ); |
---|
| 190 | } |
---|
| 191 | END_TRY |
---|
| 192 | |
---|
| 193 | fsd_log_return(("")); |
---|
| 194 | } |
---|
| 195 | |
---|
| 196 | |
---|
| 197 | void |
---|
| 198 | pbsdrmaa_job_update_status( fsd_job_t *self ) |
---|
| 199 | { |
---|
| 200 | volatile bool conn_lock = false; |
---|
| 201 | struct batch_status *volatile status = NULL; |
---|
| 202 | pbsdrmaa_session_t *session = (pbsdrmaa_session_t*)self->session; |
---|
[45] | 203 | int tries_left = session->max_retries_count; |
---|
| 204 | int sleep_time = 1; |
---|
[1] | 205 | |
---|
| 206 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
[21] | 207 | |
---|
[1] | 208 | TRY |
---|
| 209 | { |
---|
| 210 | conn_lock = fsd_mutex_lock( &self->session->drm_connection_mutex ); |
---|
| 211 | retry: |
---|
[16] | 212 | |
---|
[1] | 213 | #ifdef PBS_PROFESSIONAL |
---|
| 214 | status = pbs_statjob( session->pbs_conn, self->job_id, NULL, NULL ); |
---|
| 215 | #else |
---|
| 216 | status = pbs_statjob( session->pbs_conn, self->job_id, session->status_attrl, NULL ); |
---|
| 217 | #endif |
---|
[9] | 218 | fsd_log_info(( "pbs_statjob(fd=%d, job_id=%s, attribs={...}) =%p", |
---|
[1] | 219 | session->pbs_conn, self->job_id, (void*)status )); |
---|
| 220 | if( status == NULL ) |
---|
| 221 | { |
---|
[21] | 222 | |
---|
[25] | 223 | #ifndef PBS_PROFESSIONAL |
---|
[40] | 224 | if ( pbs_errno != PBSE_UNKJOBID ) |
---|
| 225 | fsd_log_error(("pbs_statjob error: %d, %s, %s", pbs_errno, pbse_to_txt(pbs_errno), pbs_strerror(pbs_errno))); |
---|
| 226 | else |
---|
| 227 | fsd_log_debug(("pbs_statjob error: %d, %s, %s", pbs_errno, pbse_to_txt(pbs_errno), pbs_strerror(pbs_errno))); |
---|
[25] | 228 | #else |
---|
| 229 | # ifndef PBS_PROFESSIONAL_NO_LOG |
---|
[40] | 230 | if ( pbs_errno != PBSE_UNKJOBID ) |
---|
| 231 | fsd_log_error(("pbs_statjob error: %d, %s", pbs_errno, pbse_to_txt(pbs_errno))); |
---|
| 232 | else |
---|
| 233 | fsd_log_debug(("pbs_statjob error: %d, %s", pbs_errno, pbse_to_txt(pbs_errno))); |
---|
[25] | 234 | # else |
---|
[40] | 235 | if ( pbs_errno != PBSE_UNKJOBID ) |
---|
| 236 | fsd_log_error(("pbs_statjob error: %d", pbs_errno)); |
---|
| 237 | else |
---|
| 238 | fsd_log_debug(("pbs_statjob error: %d", pbs_errno)); |
---|
[25] | 239 | # endif |
---|
| 240 | #endif |
---|
[21] | 241 | |
---|
[25] | 242 | switch( pbs_errno ) |
---|
| 243 | { |
---|
| 244 | case PBSE_UNKJOBID: |
---|
| 245 | break; |
---|
| 246 | case PBSE_PROTOCOL: |
---|
| 247 | case PBSE_EXPIRED: |
---|
| 248 | if ( session->pbs_conn >= 0 ) |
---|
| 249 | pbs_disconnect( session->pbs_conn ); |
---|
[45] | 250 | retry_connect: |
---|
| 251 | sleep(sleep_time++); |
---|
[25] | 252 | session->pbs_conn = pbs_connect( session->super.contact ); |
---|
| 253 | if( session->pbs_conn < 0 ) |
---|
[45] | 254 | { |
---|
| 255 | if (tries_left--) |
---|
| 256 | goto retry_connect; |
---|
| 257 | else |
---|
| 258 | pbsdrmaa_exc_raise_pbs( "pbs_connect" ); |
---|
| 259 | } |
---|
[25] | 260 | else |
---|
| 261 | { |
---|
| 262 | goto retry; |
---|
| 263 | } |
---|
| 264 | default: |
---|
| 265 | pbsdrmaa_exc_raise_pbs( "pbs_statjob" ); |
---|
| 266 | break; |
---|
| 267 | case 0: /* ? */ |
---|
| 268 | fsd_exc_raise_code( FSD_ERRNO_INTERNAL_ERROR ); |
---|
| 269 | break; |
---|
[1] | 270 | } |
---|
[25] | 271 | |
---|
[1] | 272 | } |
---|
| 273 | |
---|
| 274 | conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex ); |
---|
| 275 | |
---|
[25] | 276 | |
---|
[1] | 277 | if( status != NULL ) |
---|
| 278 | { |
---|
| 279 | ((pbsdrmaa_job_t*)self)->update( self, status ); |
---|
| 280 | } |
---|
[7] | 281 | else if( self->state < DRMAA_PS_DONE ) |
---|
[25] | 282 | { |
---|
[1] | 283 | self->on_missing( self ); |
---|
[25] | 284 | } |
---|
[1] | 285 | } |
---|
| 286 | FINALLY |
---|
| 287 | { |
---|
| 288 | if( conn_lock ) |
---|
| 289 | conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex ); |
---|
| 290 | if( status != NULL ) |
---|
| 291 | pbs_statfree( status ); |
---|
| 292 | } |
---|
| 293 | END_TRY |
---|
| 294 | |
---|
| 295 | fsd_log_return(("")); |
---|
| 296 | } |
---|
| 297 | |
---|
| 298 | |
---|
| 299 | void |
---|
| 300 | pbsdrmaa_job_update( fsd_job_t *self, struct batch_status *status ) |
---|
| 301 | { |
---|
| 302 | struct attrl *attribs = status->attribs; |
---|
| 303 | struct attrl *i = NULL; |
---|
| 304 | char pbs_state = 0; |
---|
| 305 | int exit_status = -2; |
---|
| 306 | const char *cpu_usage = NULL; |
---|
| 307 | const char *mem_usage = NULL; |
---|
| 308 | const char *vmem_usage = NULL; |
---|
| 309 | const char *walltime = NULL; |
---|
| 310 | long unsigned int modify_time = 0; |
---|
| 311 | |
---|
| 312 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
| 313 | #ifdef DEBUGGING |
---|
| 314 | pbsdrmaa_dump_attrl( attribs, NULL ); |
---|
| 315 | #endif |
---|
| 316 | fsd_assert( !strcmp( self->job_id, status->name ) ); |
---|
| 317 | |
---|
| 318 | for( i = attribs; i != NULL; i = i->next ) |
---|
| 319 | { |
---|
| 320 | int attr; |
---|
| 321 | attr = pbsdrmaa_pbs_attrib_by_name( i->name ); |
---|
| 322 | switch( attr ) |
---|
| 323 | { |
---|
| 324 | case PBSDRMAA_ATTR_JOB_STATE: |
---|
| 325 | pbs_state = i->value[0]; |
---|
| 326 | break; |
---|
| 327 | case PBSDRMAA_ATTR_EXIT_STATUS: |
---|
[29] | 328 | exit_status = fsd_atoi( i->value ); |
---|
[1] | 329 | break; |
---|
| 330 | case PBSDRMAA_ATTR_RESOURCES_USED: |
---|
| 331 | if( !strcmp( i->resource, "cput" ) ) |
---|
| 332 | cpu_usage = i->value; |
---|
| 333 | else if( !strcmp( i->resource, "mem" ) ) |
---|
| 334 | mem_usage = i->value; |
---|
| 335 | else if( !strcmp( i->resource, "vmem" ) ) |
---|
| 336 | vmem_usage = i->value; |
---|
| 337 | else if( !strcmp( i->resource, "walltime" ) ) |
---|
| 338 | walltime = i->value; |
---|
| 339 | break; |
---|
| 340 | case PBSDRMAA_ATTR_QUEUE: |
---|
| 341 | if (!self->queue) |
---|
| 342 | self->queue = fsd_strdup(i->value); |
---|
| 343 | break; |
---|
| 344 | case PBSDRMAA_ATTR_ACCOUNT_NAME: |
---|
| 345 | if (!self->project) |
---|
| 346 | self->project = fsd_strdup(i->value); |
---|
| 347 | break; |
---|
[26] | 348 | #ifndef PBS_PROFESSIONAL |
---|
[1] | 349 | case PBSDRMAA_ATTR_EXECUTION_HOST: |
---|
| 350 | if (!self->execution_hosts) { |
---|
| 351 | fsd_log_debug(("execution_hosts = %s", i->value)); |
---|
| 352 | self->execution_hosts = fsd_strdup(i->value); |
---|
| 353 | } |
---|
| 354 | break; |
---|
[26] | 355 | #else |
---|
| 356 | case PBSDRMAA_ATTR_EXECUTION_VNODE: |
---|
| 357 | if (!self->execution_hosts) { |
---|
| 358 | fsd_log_debug(("execution_hosts = %s", i->value)); |
---|
| 359 | self->execution_hosts = fsd_strdup(i->value); |
---|
| 360 | } |
---|
| 361 | break; |
---|
| 362 | #endif |
---|
[1] | 363 | case PBSDRMAA_ATTR_START_TIME: |
---|
| 364 | { |
---|
| 365 | long unsigned int start_time; |
---|
| 366 | if (self->start_time == 0 && sscanf(i->value, "%lu", &start_time) == 1) |
---|
| 367 | self->start_time = start_time; |
---|
| 368 | break; |
---|
| 369 | } |
---|
| 370 | case PBSDRMAA_ATTR_MTIME: |
---|
| 371 | if (sscanf(i->value, "%lu", &modify_time) != 1) |
---|
| 372 | modify_time = 0; |
---|
| 373 | break; |
---|
| 374 | } |
---|
| 375 | } |
---|
| 376 | |
---|
| 377 | if( pbs_state ) |
---|
| 378 | fsd_log_debug(( "pbs_state: %c", pbs_state )); |
---|
| 379 | |
---|
| 380 | if( exit_status != -2 ) |
---|
| 381 | { |
---|
| 382 | fsd_log_debug(( "exit_status: %d", exit_status )); |
---|
| 383 | self->exit_status = exit_status; |
---|
| 384 | } |
---|
| 385 | if(pbs_state){ |
---|
| 386 | switch( pbs_state ) |
---|
| 387 | { |
---|
| 388 | case 'C': /* Job is completed after having run. */ |
---|
| 389 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
| 390 | self->flags |= FSD_JOB_TERMINATED; |
---|
| 391 | if (exit_status != -2) { /* has exit code */ |
---|
| 392 | if( self->exit_status == 0) |
---|
| 393 | self->state = DRMAA_PS_DONE; |
---|
| 394 | else |
---|
| 395 | self->state = DRMAA_PS_FAILED; |
---|
| 396 | } else { |
---|
| 397 | self->state = DRMAA_PS_FAILED; |
---|
| 398 | self->exit_status = -1; |
---|
| 399 | } |
---|
[22] | 400 | if (modify_time != 0) |
---|
| 401 | self->end_time = modify_time; /* take last modify time as end time */ |
---|
| 402 | else |
---|
| 403 | self->end_time = time(NULL); |
---|
| 404 | |
---|
| 405 | if (self->start_time == 0) |
---|
| 406 | self->start_time = self->end_time; |
---|
| 407 | |
---|
[1] | 408 | break; |
---|
| 409 | case 'E': /* Job is exiting after having run. - MM: ignore exiting state (transient state) - outputs might have not been transfered yet, |
---|
| 410 | MM2: mark job as running if current job status is undetermined - fix "ps after job was ripped" */ |
---|
| 411 | if (self->state == DRMAA_PS_UNDETERMINED) |
---|
| 412 | self->state = DRMAA_PS_RUNNING; |
---|
| 413 | break; |
---|
| 414 | case 'H': /* Job is held. */ |
---|
| 415 | self->state = DRMAA_PS_USER_ON_HOLD; |
---|
| 416 | self->flags |= FSD_JOB_HOLD; |
---|
| 417 | break; |
---|
| 418 | case 'Q': /* Job is queued, eligible to run or routed. */ |
---|
| 419 | case 'W': /* Job is waiting for its execution time to be reached. */ |
---|
| 420 | self->state = DRMAA_PS_QUEUED_ACTIVE; |
---|
| 421 | self->flags &= ~FSD_JOB_HOLD; |
---|
| 422 | break; |
---|
| 423 | case 'R': /* Job is running. */ |
---|
| 424 | case 'T': /* Job is being moved to new location (?). */ |
---|
| 425 | { |
---|
| 426 | if( self->flags & FSD_JOB_SUSPENDED ) |
---|
| 427 | self->state = DRMAA_PS_USER_SUSPENDED; |
---|
| 428 | else |
---|
| 429 | self->state = DRMAA_PS_RUNNING; |
---|
| 430 | break; |
---|
| 431 | } |
---|
| 432 | case 'S': /* (Unicos only) job is suspend. */ |
---|
| 433 | self->state = DRMAA_PS_SYSTEM_SUSPENDED; |
---|
| 434 | break; |
---|
| 435 | case 0: default: |
---|
| 436 | self->state = DRMAA_PS_UNDETERMINED; |
---|
| 437 | break; |
---|
| 438 | } |
---|
| 439 | } |
---|
| 440 | fsd_log_debug(( "job_ps: %s", drmaa_job_ps_to_str(self->state) )); |
---|
| 441 | |
---|
| 442 | { |
---|
| 443 | int hours, minutes, seconds; |
---|
| 444 | long mem; |
---|
[12] | 445 | if( cpu_usage && sscanf( cpu_usage, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 ) |
---|
[1] | 446 | { |
---|
| 447 | self->cpu_usage = 60*( 60*hours + minutes ) + seconds; |
---|
| 448 | fsd_log_debug(( "cpu_usage: %s=%lds", cpu_usage, self->cpu_usage )); |
---|
| 449 | } |
---|
| 450 | if( mem_usage && sscanf( mem_usage, "%ldkb", &mem ) == 1 ) |
---|
| 451 | { |
---|
| 452 | self->mem_usage = 1024*mem; |
---|
| 453 | fsd_log_debug(( "mem_usage: %s=%ldB", mem_usage, self->mem_usage )); |
---|
| 454 | } |
---|
| 455 | if( vmem_usage && sscanf( vmem_usage, "%ldkb", &mem ) == 1 ) |
---|
| 456 | { |
---|
| 457 | self->vmem_usage = 1024*mem; |
---|
| 458 | fsd_log_debug(( "vmem_usage: %s=%ldB", vmem_usage, self->vmem_usage )); |
---|
| 459 | } |
---|
[12] | 460 | if( walltime && sscanf( walltime, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 ) |
---|
[1] | 461 | { |
---|
| 462 | self->walltime = 60*( 60*hours + minutes ) + seconds; |
---|
| 463 | fsd_log_debug(( "walltime: %s=%lds", walltime, self->walltime )); |
---|
| 464 | } |
---|
| 465 | } |
---|
| 466 | } |
---|
| 467 | |
---|
| 468 | void |
---|
| 469 | pbsdrmaa_job_on_missing( fsd_job_t *self ) |
---|
| 470 | { |
---|
[7] | 471 | pbsdrmaa_session_t *pbssession = (pbsdrmaa_session_t*)self->session; |
---|
[12] | 472 | |
---|
[33] | 473 | if( pbssession->pbs_home != NULL && pbssession->super.wait_thread_started ) |
---|
[40] | 474 | fsd_log_info(("Job on missing but WT is running. Skipping...")); /* TODO: try to provide implementation that uses accounting/server log files */ |
---|
[33] | 475 | else |
---|
[7] | 476 | pbsdrmaa_job_on_missing_standard( self ); |
---|
| 477 | } |
---|
| 478 | |
---|
| 479 | void |
---|
| 480 | pbsdrmaa_job_on_missing_standard( fsd_job_t *self ) |
---|
| 481 | { |
---|
[1] | 482 | fsd_drmaa_session_t *session = self->session; |
---|
[7] | 483 | |
---|
[1] | 484 | unsigned missing_mask = 0; |
---|
| 485 | |
---|
| 486 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
[12] | 487 | fsd_log_warning(( "Job %s missing from DRM queue", self->job_id )); |
---|
[1] | 488 | |
---|
| 489 | switch( session->missing_jobs ) |
---|
[7] | 490 | { |
---|
[1] | 491 | case FSD_REVEAL_MISSING_JOBS: missing_mask = 0; break; |
---|
| 492 | case FSD_IGNORE_MISSING_JOBS: missing_mask = 0x73; break; |
---|
| 493 | case FSD_IGNORE_QUEUED_MISSING_JOBS: missing_mask = 0x13; break; |
---|
[7] | 494 | } |
---|
[40] | 495 | fsd_log_info(( "last job_ps: %s (0x%02x); mask: 0x%02x", |
---|
[1] | 496 | drmaa_job_ps_to_str(self->state), self->state, missing_mask )); |
---|
| 497 | |
---|
| 498 | if( self->state < DRMAA_PS_DONE |
---|
| 499 | && (self->state & ~missing_mask) ) |
---|
| 500 | fsd_exc_raise_fmt( |
---|
[17] | 501 | FSD_DRMAA_ERRNO_INVALID_JOB, |
---|
[1] | 502 | "self %s missing from queue", self->job_id |
---|
| 503 | ); |
---|
| 504 | |
---|
| 505 | if( (self->flags & FSD_JOB_TERMINATED_MASK) == 0 ) |
---|
[7] | 506 | { |
---|
[1] | 507 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
| 508 | self->flags |= FSD_JOB_TERMINATED; |
---|
[7] | 509 | } |
---|
[1] | 510 | |
---|
| 511 | if( (self->flags & FSD_JOB_ABORTED) == 0 |
---|
| 512 | && session->missing_jobs == FSD_IGNORE_MISSING_JOBS ) |
---|
[7] | 513 | { /* assume everthing was ok */ |
---|
[1] | 514 | self->state = DRMAA_PS_DONE; |
---|
| 515 | self->exit_status = 0; |
---|
[7] | 516 | } |
---|
[1] | 517 | else |
---|
[7] | 518 | { /* job aborted */ |
---|
[1] | 519 | self->state = DRMAA_PS_FAILED; |
---|
| 520 | self->exit_status = -1; |
---|
[7] | 521 | } |
---|
[1] | 522 | |
---|
[12] | 523 | fsd_cond_broadcast( &self->status_cond); |
---|
| 524 | |
---|
[1] | 525 | fsd_log_return(( "; job_ps=%s, exit_status=%d", |
---|
| 526 | drmaa_job_ps_to_str(self->state), self->exit_status )); |
---|
[7] | 527 | } |
---|
| 528 | |
---|