[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 | |
---|
| 61 | void |
---|
| 62 | pbsdrmaa_job_on_missing_log_based( fsd_job_t *self ); |
---|
| 63 | |
---|
[1] | 64 | static void |
---|
| 65 | pbsdrmaa_job_update( fsd_job_t *self, struct batch_status* ); |
---|
| 66 | |
---|
[21] | 67 | bool |
---|
| 68 | pbsdrmaa_job_update_status_accounting( fsd_job_t *self ); |
---|
[1] | 69 | |
---|
[21] | 70 | |
---|
[1] | 71 | fsd_job_t * |
---|
| 72 | pbsdrmaa_job_new( char *job_id ) |
---|
| 73 | { |
---|
| 74 | pbsdrmaa_job_t *self = (pbsdrmaa_job_t*)fsd_job_new( job_id ); |
---|
| 75 | fsd_realloc( self, 1, pbsdrmaa_job_t ); |
---|
| 76 | self->super.control = pbsdrmaa_job_control; |
---|
| 77 | self->super.update_status = pbsdrmaa_job_update_status; |
---|
| 78 | self->super.on_missing = pbsdrmaa_job_on_missing; |
---|
| 79 | self->update = pbsdrmaa_job_update; |
---|
| 80 | return (fsd_job_t*)self; |
---|
| 81 | } |
---|
| 82 | |
---|
| 83 | |
---|
| 84 | static void |
---|
| 85 | pbsdrmaa_job_control( fsd_job_t *self, int action ) |
---|
| 86 | { |
---|
| 87 | volatile bool conn_lock = false; |
---|
| 88 | pbsdrmaa_session_t *session = (pbsdrmaa_session_t*)self->session; |
---|
| 89 | const char *job_id = self->job_id; |
---|
| 90 | const char *apicall = NULL; |
---|
| 91 | int rc = PBSE_NONE; |
---|
| 92 | |
---|
| 93 | fsd_log_enter(( "({job_id=%s}, action=%d)", |
---|
| 94 | self->job_id, action )); |
---|
| 95 | |
---|
| 96 | TRY |
---|
| 97 | { |
---|
| 98 | int try_count; |
---|
| 99 | const int max_tries = 3; |
---|
| 100 | |
---|
| 101 | conn_lock = fsd_mutex_lock( &self->session->drm_connection_mutex ); |
---|
| 102 | |
---|
| 103 | /*TODO reconnect */ |
---|
| 104 | for( try_count=0; try_count < max_tries; try_count++ ) |
---|
| 105 | { |
---|
| 106 | switch( action ) |
---|
| 107 | { |
---|
| 108 | /* |
---|
| 109 | * We cannot know whether we did suspend job |
---|
| 110 | * in other way than remembering this inside DRMAA session. |
---|
| 111 | */ |
---|
| 112 | case DRMAA_CONTROL_SUSPEND: |
---|
| 113 | apicall = "pbs_sigjob"; |
---|
| 114 | rc = pbs_sigjob( session->pbs_conn, (char*)job_id, |
---|
| 115 | "SIGSTOP", NULL ); |
---|
[10] | 116 | fsd_log_info(("pbs_sigjob(%s, SIGSTOP) =%d", job_id, rc)); |
---|
[1] | 117 | if( rc == PBSE_NONE ) |
---|
| 118 | self->flags |= FSD_JOB_SUSPENDED; |
---|
| 119 | break; |
---|
| 120 | case DRMAA_CONTROL_RESUME: |
---|
| 121 | apicall = "pbs_sigjob"; |
---|
| 122 | rc = pbs_sigjob( session->pbs_conn, (char*)job_id, |
---|
| 123 | "SIGCONT", NULL ); |
---|
[10] | 124 | fsd_log_info(("pbs_sigjob(%s, SIGCONT) =%d", job_id, rc)); |
---|
[1] | 125 | if( rc == PBSE_NONE ) |
---|
| 126 | self->flags &= ~FSD_JOB_SUSPENDED; |
---|
| 127 | break; |
---|
| 128 | case DRMAA_CONTROL_HOLD: |
---|
| 129 | apicall = "pbs_holdjob"; |
---|
| 130 | rc = pbs_holdjob( session->pbs_conn, (char*)job_id, |
---|
| 131 | USER_HOLD, NULL ); |
---|
[10] | 132 | fsd_log_info(("pbs_sigjob(%s, SIGHOLD) =%d", job_id, rc)); |
---|
[1] | 133 | if( rc == PBSE_NONE ) |
---|
| 134 | self->flags |= FSD_JOB_HOLD; |
---|
| 135 | break; |
---|
| 136 | case DRMAA_CONTROL_RELEASE: |
---|
| 137 | apicall = "pbs_rlsjob"; |
---|
| 138 | rc = pbs_rlsjob( session->pbs_conn, (char*)job_id, |
---|
| 139 | USER_HOLD, NULL ); |
---|
[10] | 140 | fsd_log_info(("pbs_rlsjob(%s) =%d", job_id, rc)); |
---|
[1] | 141 | if( rc == PBSE_NONE ) |
---|
| 142 | self->flags &= FSD_JOB_HOLD; |
---|
| 143 | break; |
---|
| 144 | case DRMAA_CONTROL_TERMINATE: |
---|
| 145 | apicall = "pbs_deljob"; |
---|
| 146 | rc = pbs_deljob( session->pbs_conn, (char*)job_id, NULL ); |
---|
[10] | 147 | fsd_log_info(("pbs_deljob(%s) =%d", job_id, rc)); |
---|
[1] | 148 | /* Torque: |
---|
| 149 | * deldelay=N -- delay between SIGTERM and SIGKILL (default 0) */ |
---|
| 150 | if( rc == PBSE_NONE ) |
---|
| 151 | { |
---|
| 152 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
| 153 | if( (self->flags & FSD_JOB_TERMINATED) == 0 ) |
---|
| 154 | self->flags |= FSD_JOB_TERMINATED | FSD_JOB_ABORTED; |
---|
| 155 | } |
---|
| 156 | break; |
---|
| 157 | } |
---|
| 158 | |
---|
| 159 | if( rc == PBSE_NONE ) |
---|
| 160 | break; |
---|
| 161 | else if( rc == PBSE_INTERNAL ) |
---|
| 162 | { |
---|
| 163 | /* |
---|
| 164 | * In PBS Pro pbs_sigjob raises internal server error (PBSE_INTERNAL) |
---|
| 165 | * when job just changed its state to running. |
---|
| 166 | */ |
---|
| 167 | fsd_log_debug(( "repeating request (%d of %d)", |
---|
| 168 | try_count+2, max_tries )); |
---|
| 169 | sleep( 1 ); |
---|
| 170 | } |
---|
| 171 | else |
---|
| 172 | pbsdrmaa_exc_raise_pbs( apicall ); |
---|
| 173 | } /* end for */ |
---|
| 174 | } |
---|
| 175 | FINALLY |
---|
| 176 | { |
---|
| 177 | if( conn_lock ) |
---|
| 178 | conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex ); |
---|
| 179 | } |
---|
| 180 | END_TRY |
---|
| 181 | |
---|
| 182 | fsd_log_return(("")); |
---|
| 183 | } |
---|
| 184 | |
---|
| 185 | |
---|
| 186 | void |
---|
| 187 | pbsdrmaa_job_update_status( fsd_job_t *self ) |
---|
| 188 | { |
---|
| 189 | volatile bool conn_lock = false; |
---|
| 190 | struct batch_status *volatile status = NULL; |
---|
| 191 | pbsdrmaa_session_t *session = (pbsdrmaa_session_t*)self->session; |
---|
| 192 | |
---|
| 193 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
[21] | 194 | |
---|
[1] | 195 | TRY |
---|
| 196 | { |
---|
| 197 | conn_lock = fsd_mutex_lock( &self->session->drm_connection_mutex ); |
---|
| 198 | retry: |
---|
[16] | 199 | |
---|
[1] | 200 | #ifdef PBS_PROFESSIONAL |
---|
| 201 | status = pbs_statjob( session->pbs_conn, self->job_id, NULL, NULL ); |
---|
| 202 | #else |
---|
| 203 | status = pbs_statjob( session->pbs_conn, self->job_id, session->status_attrl, NULL ); |
---|
| 204 | #endif |
---|
[9] | 205 | fsd_log_info(( "pbs_statjob(fd=%d, job_id=%s, attribs={...}) =%p", |
---|
[1] | 206 | session->pbs_conn, self->job_id, (void*)status )); |
---|
| 207 | if( status == NULL ) |
---|
| 208 | { |
---|
[21] | 209 | |
---|
[25] | 210 | #ifndef PBS_PROFESSIONAL |
---|
| 211 | fsd_log_error(("pbs_statjob error: %d, %s, %s", pbs_errno, pbse_to_txt(pbs_errno), pbs_strerror(pbs_errno))); |
---|
| 212 | #else |
---|
| 213 | # ifndef PBS_PROFESSIONAL_NO_LOG |
---|
| 214 | fsd_log_error(("pbs_statjob error: %d, %s", pbs_errno, pbse_to_txt(pbs_errno))); |
---|
| 215 | # else |
---|
| 216 | fsd_log_error(("pbs_statjob error: %d", pbs_errno)); |
---|
| 217 | # endif |
---|
| 218 | #endif |
---|
[21] | 219 | |
---|
[25] | 220 | switch( pbs_errno ) |
---|
| 221 | { |
---|
| 222 | case PBSE_UNKJOBID: |
---|
| 223 | break; |
---|
| 224 | case PBSE_PROTOCOL: |
---|
| 225 | case PBSE_EXPIRED: |
---|
| 226 | if ( session->pbs_conn >= 0 ) |
---|
| 227 | pbs_disconnect( session->pbs_conn ); |
---|
| 228 | sleep(1); |
---|
| 229 | session->pbs_conn = pbs_connect( session->super.contact ); |
---|
| 230 | if( session->pbs_conn < 0 ) |
---|
| 231 | pbsdrmaa_exc_raise_pbs( "pbs_connect" ); |
---|
| 232 | else |
---|
| 233 | { |
---|
| 234 | fsd_log_error(("retry:")); |
---|
| 235 | goto retry; |
---|
| 236 | } |
---|
| 237 | default: |
---|
| 238 | pbsdrmaa_exc_raise_pbs( "pbs_statjob" ); |
---|
| 239 | break; |
---|
| 240 | case 0: /* ? */ |
---|
| 241 | fsd_exc_raise_code( FSD_ERRNO_INTERNAL_ERROR ); |
---|
| 242 | break; |
---|
[1] | 243 | } |
---|
[25] | 244 | |
---|
[1] | 245 | } |
---|
| 246 | |
---|
| 247 | conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex ); |
---|
| 248 | |
---|
[25] | 249 | |
---|
[1] | 250 | if( status != NULL ) |
---|
| 251 | { |
---|
| 252 | ((pbsdrmaa_job_t*)self)->update( self, status ); |
---|
| 253 | } |
---|
[7] | 254 | else if( self->state < DRMAA_PS_DONE ) |
---|
[25] | 255 | { |
---|
| 256 | #ifndef PBS_PROFESSIONAL |
---|
| 257 | /*best effort call*/ |
---|
| 258 | if (pbsdrmaa_job_update_status_accounting(self) == false) |
---|
| 259 | self->on_missing( self ); |
---|
| 260 | #else |
---|
[1] | 261 | self->on_missing( self ); |
---|
[25] | 262 | #endif |
---|
| 263 | } |
---|
[1] | 264 | } |
---|
| 265 | FINALLY |
---|
| 266 | { |
---|
| 267 | if( conn_lock ) |
---|
| 268 | conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex ); |
---|
| 269 | if( status != NULL ) |
---|
| 270 | pbs_statfree( status ); |
---|
| 271 | } |
---|
| 272 | END_TRY |
---|
| 273 | |
---|
| 274 | fsd_log_return(("")); |
---|
| 275 | } |
---|
| 276 | |
---|
| 277 | |
---|
| 278 | void |
---|
| 279 | pbsdrmaa_job_update( fsd_job_t *self, struct batch_status *status ) |
---|
| 280 | { |
---|
| 281 | struct attrl *attribs = status->attribs; |
---|
| 282 | struct attrl *i = NULL; |
---|
| 283 | char pbs_state = 0; |
---|
| 284 | int exit_status = -2; |
---|
| 285 | const char *cpu_usage = NULL; |
---|
| 286 | const char *mem_usage = NULL; |
---|
| 287 | const char *vmem_usage = NULL; |
---|
| 288 | const char *walltime = NULL; |
---|
| 289 | long unsigned int modify_time = 0; |
---|
| 290 | |
---|
| 291 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
| 292 | #ifdef DEBUGGING |
---|
| 293 | pbsdrmaa_dump_attrl( attribs, NULL ); |
---|
| 294 | #endif |
---|
| 295 | fsd_assert( !strcmp( self->job_id, status->name ) ); |
---|
| 296 | |
---|
| 297 | for( i = attribs; i != NULL; i = i->next ) |
---|
| 298 | { |
---|
| 299 | int attr; |
---|
| 300 | attr = pbsdrmaa_pbs_attrib_by_name( i->name ); |
---|
| 301 | switch( attr ) |
---|
| 302 | { |
---|
| 303 | case PBSDRMAA_ATTR_JOB_STATE: |
---|
| 304 | pbs_state = i->value[0]; |
---|
| 305 | break; |
---|
| 306 | case PBSDRMAA_ATTR_EXIT_STATUS: |
---|
| 307 | exit_status = atoi( i->value ); |
---|
| 308 | break; |
---|
| 309 | case PBSDRMAA_ATTR_RESOURCES_USED: |
---|
| 310 | if( !strcmp( i->resource, "cput" ) ) |
---|
| 311 | cpu_usage = i->value; |
---|
| 312 | else if( !strcmp( i->resource, "mem" ) ) |
---|
| 313 | mem_usage = i->value; |
---|
| 314 | else if( !strcmp( i->resource, "vmem" ) ) |
---|
| 315 | vmem_usage = i->value; |
---|
| 316 | else if( !strcmp( i->resource, "walltime" ) ) |
---|
| 317 | walltime = i->value; |
---|
| 318 | break; |
---|
| 319 | case PBSDRMAA_ATTR_QUEUE: |
---|
| 320 | if (!self->queue) |
---|
| 321 | self->queue = fsd_strdup(i->value); |
---|
| 322 | break; |
---|
| 323 | case PBSDRMAA_ATTR_ACCOUNT_NAME: |
---|
| 324 | if (!self->project) |
---|
| 325 | self->project = fsd_strdup(i->value); |
---|
| 326 | break; |
---|
[26] | 327 | #ifndef PBS_PROFESSIONAL |
---|
[1] | 328 | case PBSDRMAA_ATTR_EXECUTION_HOST: |
---|
| 329 | if (!self->execution_hosts) { |
---|
| 330 | fsd_log_debug(("execution_hosts = %s", i->value)); |
---|
| 331 | self->execution_hosts = fsd_strdup(i->value); |
---|
| 332 | } |
---|
| 333 | break; |
---|
[26] | 334 | #else |
---|
| 335 | case PBSDRMAA_ATTR_EXECUTION_VNODE: |
---|
| 336 | if (!self->execution_hosts) { |
---|
| 337 | fsd_log_debug(("execution_hosts = %s", i->value)); |
---|
| 338 | self->execution_hosts = fsd_strdup(i->value); |
---|
| 339 | } |
---|
| 340 | break; |
---|
| 341 | #endif |
---|
[1] | 342 | case PBSDRMAA_ATTR_START_TIME: |
---|
| 343 | { |
---|
| 344 | long unsigned int start_time; |
---|
| 345 | if (self->start_time == 0 && sscanf(i->value, "%lu", &start_time) == 1) |
---|
| 346 | self->start_time = start_time; |
---|
| 347 | break; |
---|
| 348 | } |
---|
| 349 | case PBSDRMAA_ATTR_MTIME: |
---|
| 350 | if (sscanf(i->value, "%lu", &modify_time) != 1) |
---|
| 351 | modify_time = 0; |
---|
| 352 | break; |
---|
| 353 | } |
---|
| 354 | } |
---|
| 355 | |
---|
| 356 | if( pbs_state ) |
---|
| 357 | fsd_log_debug(( "pbs_state: %c", pbs_state )); |
---|
| 358 | |
---|
| 359 | if( exit_status != -2 ) |
---|
| 360 | { |
---|
| 361 | fsd_log_debug(( "exit_status: %d", exit_status )); |
---|
| 362 | self->exit_status = exit_status; |
---|
| 363 | } |
---|
| 364 | if(pbs_state){ |
---|
| 365 | switch( pbs_state ) |
---|
| 366 | { |
---|
| 367 | case 'C': /* Job is completed after having run. */ |
---|
| 368 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
| 369 | self->flags |= FSD_JOB_TERMINATED; |
---|
| 370 | if (exit_status != -2) { /* has exit code */ |
---|
| 371 | if( self->exit_status == 0) |
---|
| 372 | self->state = DRMAA_PS_DONE; |
---|
| 373 | else |
---|
| 374 | self->state = DRMAA_PS_FAILED; |
---|
| 375 | } else { |
---|
| 376 | self->state = DRMAA_PS_FAILED; |
---|
| 377 | self->exit_status = -1; |
---|
| 378 | } |
---|
[22] | 379 | if (modify_time != 0) |
---|
| 380 | self->end_time = modify_time; /* take last modify time as end time */ |
---|
| 381 | else |
---|
| 382 | self->end_time = time(NULL); |
---|
| 383 | |
---|
| 384 | if (self->start_time == 0) |
---|
| 385 | self->start_time = self->end_time; |
---|
| 386 | |
---|
[1] | 387 | break; |
---|
| 388 | case 'E': /* Job is exiting after having run. - MM: ignore exiting state (transient state) - outputs might have not been transfered yet, |
---|
| 389 | MM2: mark job as running if current job status is undetermined - fix "ps after job was ripped" */ |
---|
| 390 | if (self->state == DRMAA_PS_UNDETERMINED) |
---|
| 391 | self->state = DRMAA_PS_RUNNING; |
---|
| 392 | break; |
---|
| 393 | case 'H': /* Job is held. */ |
---|
| 394 | self->state = DRMAA_PS_USER_ON_HOLD; |
---|
| 395 | self->flags |= FSD_JOB_HOLD; |
---|
| 396 | break; |
---|
| 397 | case 'Q': /* Job is queued, eligible to run or routed. */ |
---|
| 398 | case 'W': /* Job is waiting for its execution time to be reached. */ |
---|
| 399 | self->state = DRMAA_PS_QUEUED_ACTIVE; |
---|
| 400 | self->flags &= ~FSD_JOB_HOLD; |
---|
| 401 | break; |
---|
| 402 | case 'R': /* Job is running. */ |
---|
| 403 | case 'T': /* Job is being moved to new location (?). */ |
---|
| 404 | { |
---|
| 405 | if( self->flags & FSD_JOB_SUSPENDED ) |
---|
| 406 | self->state = DRMAA_PS_USER_SUSPENDED; |
---|
| 407 | else |
---|
| 408 | self->state = DRMAA_PS_RUNNING; |
---|
| 409 | break; |
---|
| 410 | } |
---|
| 411 | case 'S': /* (Unicos only) job is suspend. */ |
---|
| 412 | self->state = DRMAA_PS_SYSTEM_SUSPENDED; |
---|
| 413 | break; |
---|
| 414 | case 0: default: |
---|
| 415 | self->state = DRMAA_PS_UNDETERMINED; |
---|
| 416 | break; |
---|
| 417 | } |
---|
| 418 | } |
---|
| 419 | fsd_log_debug(( "job_ps: %s", drmaa_job_ps_to_str(self->state) )); |
---|
| 420 | |
---|
| 421 | { |
---|
| 422 | int hours, minutes, seconds; |
---|
| 423 | long mem; |
---|
[12] | 424 | if( cpu_usage && sscanf( cpu_usage, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 ) |
---|
[1] | 425 | { |
---|
| 426 | self->cpu_usage = 60*( 60*hours + minutes ) + seconds; |
---|
| 427 | fsd_log_debug(( "cpu_usage: %s=%lds", cpu_usage, self->cpu_usage )); |
---|
| 428 | } |
---|
| 429 | if( mem_usage && sscanf( mem_usage, "%ldkb", &mem ) == 1 ) |
---|
| 430 | { |
---|
| 431 | self->mem_usage = 1024*mem; |
---|
| 432 | fsd_log_debug(( "mem_usage: %s=%ldB", mem_usage, self->mem_usage )); |
---|
| 433 | } |
---|
| 434 | if( vmem_usage && sscanf( vmem_usage, "%ldkb", &mem ) == 1 ) |
---|
| 435 | { |
---|
| 436 | self->vmem_usage = 1024*mem; |
---|
| 437 | fsd_log_debug(( "vmem_usage: %s=%ldB", vmem_usage, self->vmem_usage )); |
---|
| 438 | } |
---|
[12] | 439 | if( walltime && sscanf( walltime, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 ) |
---|
[1] | 440 | { |
---|
| 441 | self->walltime = 60*( 60*hours + minutes ) + seconds; |
---|
| 442 | fsd_log_debug(( "walltime: %s=%lds", walltime, self->walltime )); |
---|
| 443 | } |
---|
| 444 | } |
---|
| 445 | } |
---|
| 446 | |
---|
| 447 | void |
---|
| 448 | pbsdrmaa_job_on_missing( fsd_job_t *self ) |
---|
| 449 | { |
---|
[7] | 450 | pbsdrmaa_session_t *pbssession = (pbsdrmaa_session_t*)self->session; |
---|
[12] | 451 | |
---|
[17] | 452 | if( pbssession->pbs_home == NULL || pbssession->super.wait_thread_started ) |
---|
[7] | 453 | pbsdrmaa_job_on_missing_standard( self ); |
---|
| 454 | else |
---|
| 455 | pbsdrmaa_job_on_missing_log_based( self ); |
---|
| 456 | } |
---|
| 457 | |
---|
| 458 | void |
---|
| 459 | pbsdrmaa_job_on_missing_standard( fsd_job_t *self ) |
---|
| 460 | { |
---|
[1] | 461 | fsd_drmaa_session_t *session = self->session; |
---|
[7] | 462 | |
---|
[1] | 463 | unsigned missing_mask = 0; |
---|
| 464 | |
---|
| 465 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
[12] | 466 | fsd_log_warning(( "Job %s missing from DRM queue", self->job_id )); |
---|
[1] | 467 | |
---|
| 468 | switch( session->missing_jobs ) |
---|
[7] | 469 | { |
---|
[1] | 470 | case FSD_REVEAL_MISSING_JOBS: missing_mask = 0; break; |
---|
| 471 | case FSD_IGNORE_MISSING_JOBS: missing_mask = 0x73; break; |
---|
| 472 | case FSD_IGNORE_QUEUED_MISSING_JOBS: missing_mask = 0x13; break; |
---|
[7] | 473 | } |
---|
[1] | 474 | fsd_log_debug(( "last job_ps: %s (0x%02x); mask: 0x%02x", |
---|
| 475 | drmaa_job_ps_to_str(self->state), self->state, missing_mask )); |
---|
| 476 | |
---|
| 477 | if( self->state < DRMAA_PS_DONE |
---|
| 478 | && (self->state & ~missing_mask) ) |
---|
| 479 | fsd_exc_raise_fmt( |
---|
[17] | 480 | FSD_DRMAA_ERRNO_INVALID_JOB, |
---|
[1] | 481 | "self %s missing from queue", self->job_id |
---|
| 482 | ); |
---|
| 483 | |
---|
| 484 | if( (self->flags & FSD_JOB_TERMINATED_MASK) == 0 ) |
---|
[7] | 485 | { |
---|
[1] | 486 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
| 487 | self->flags |= FSD_JOB_TERMINATED; |
---|
[7] | 488 | } |
---|
[1] | 489 | |
---|
| 490 | if( (self->flags & FSD_JOB_ABORTED) == 0 |
---|
| 491 | && session->missing_jobs == FSD_IGNORE_MISSING_JOBS ) |
---|
[7] | 492 | { /* assume everthing was ok */ |
---|
[1] | 493 | self->state = DRMAA_PS_DONE; |
---|
| 494 | self->exit_status = 0; |
---|
[7] | 495 | } |
---|
[1] | 496 | else |
---|
[7] | 497 | { /* job aborted */ |
---|
[1] | 498 | self->state = DRMAA_PS_FAILED; |
---|
| 499 | self->exit_status = -1; |
---|
[7] | 500 | } |
---|
[1] | 501 | |
---|
[12] | 502 | fsd_cond_broadcast( &self->status_cond); |
---|
| 503 | |
---|
[1] | 504 | fsd_log_return(( "; job_ps=%s, exit_status=%d", |
---|
| 505 | drmaa_job_ps_to_str(self->state), self->exit_status )); |
---|
[7] | 506 | } |
---|
| 507 | |
---|
| 508 | void |
---|
| 509 | pbsdrmaa_job_on_missing_log_based( fsd_job_t *self ) |
---|
| 510 | { |
---|
| 511 | fsd_drmaa_session_t *session = self->session; |
---|
| 512 | pbsdrmaa_log_reader_t *log_reader = NULL; |
---|
| 513 | |
---|
| 514 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
[12] | 515 | fsd_log_info(( "Job %s missing from DRM queue", self->job_id )); |
---|
[7] | 516 | |
---|
| 517 | TRY |
---|
| 518 | { |
---|
| 519 | log_reader = pbsdrmaa_log_reader_new( session, self); |
---|
| 520 | log_reader->read_log( log_reader ); |
---|
[1] | 521 | } |
---|
[7] | 522 | FINALLY |
---|
| 523 | { |
---|
| 524 | pbsdrmaa_log_reader_destroy( log_reader ); |
---|
| 525 | } |
---|
| 526 | END_TRY |
---|
| 527 | |
---|
| 528 | fsd_log_return(( "; job_ps=%s, exit_status=%d", |
---|
| 529 | drmaa_job_ps_to_str(self->state), self->exit_status )); |
---|
[1] | 530 | } |
---|
[21] | 531 | |
---|
| 532 | bool |
---|
| 533 | pbsdrmaa_job_update_status_accounting( fsd_job_t *self ) |
---|
| 534 | { |
---|
| 535 | fsd_drmaa_session_t *session = self->session; |
---|
| 536 | pbsdrmaa_log_reader_t *log_reader = NULL; |
---|
| 537 | bool res = false; |
---|
| 538 | |
---|
| 539 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
| 540 | fsd_log_info(( "Reading job %s info from accounting file", self->job_id )); |
---|
| 541 | |
---|
| 542 | TRY |
---|
| 543 | { |
---|
| 544 | log_reader = pbsdrmaa_log_reader_accounting_new( session, self); |
---|
| 545 | bool res = log_reader->read_log( log_reader ); |
---|
| 546 | } |
---|
| 547 | FINALLY |
---|
| 548 | { |
---|
| 549 | pbsdrmaa_log_reader_destroy( log_reader ); |
---|
| 550 | } |
---|
| 551 | END_TRY |
---|
| 552 | |
---|
| 553 | fsd_log_return(("")); |
---|
| 554 | return res; |
---|
| 555 | } |
---|