1 | /* $Id$ */ |
---|
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 | #include <sys/stat.h> |
---|
29 | |
---|
30 | #include <drmaa_utils/drmaa.h> |
---|
31 | #include <drmaa_utils/drmaa_util.h> |
---|
32 | #include <pbs_error.h> |
---|
33 | #include <pbs_ifl.h> |
---|
34 | |
---|
35 | #include <pbs_drmaa/job.h> |
---|
36 | #include <pbs_drmaa/log_reader.h> |
---|
37 | #include <pbs_drmaa/pbs_attrib.h> |
---|
38 | #include <pbs_drmaa/session.h> |
---|
39 | #include <pbs_drmaa/util.h> |
---|
40 | |
---|
41 | #ifndef lint |
---|
42 | static char rcsid[] |
---|
43 | # ifdef __GNUC__ |
---|
44 | __attribute__ ((unused)) |
---|
45 | # endif |
---|
46 | = "$Id$"; |
---|
47 | #endif |
---|
48 | |
---|
49 | |
---|
50 | static void pbsdrmaa_job_control( fsd_job_t *self, int action ); |
---|
51 | |
---|
52 | static void pbsdrmaa_job_update_status( fsd_job_t *self ); |
---|
53 | |
---|
54 | static void pbsdrmaa_job_on_missing( fsd_job_t *self ); |
---|
55 | |
---|
56 | static void pbsdrmaa_job_on_missing_standard( fsd_job_t *self ); |
---|
57 | |
---|
58 | static void pbsdrmaa_job_update( fsd_job_t *self, struct batch_status* ); |
---|
59 | |
---|
60 | static int pbsdrmaa_job_read_exit_status( const char *job_id, const char *job_state_dir_prefix); |
---|
61 | |
---|
62 | fsd_job_t * |
---|
63 | pbsdrmaa_job_new( char *job_id ) |
---|
64 | { |
---|
65 | pbsdrmaa_job_t *self = (pbsdrmaa_job_t*)fsd_job_new( job_id ); |
---|
66 | fsd_realloc( self, 1, pbsdrmaa_job_t ); |
---|
67 | self->super.control = pbsdrmaa_job_control; |
---|
68 | self->super.update_status = pbsdrmaa_job_update_status; |
---|
69 | self->super.on_missing = pbsdrmaa_job_on_missing; |
---|
70 | self->missing_time = 0; |
---|
71 | self->update = pbsdrmaa_job_update; |
---|
72 | return (fsd_job_t*)self; |
---|
73 | } |
---|
74 | |
---|
75 | |
---|
76 | static void |
---|
77 | pbsdrmaa_job_control( fsd_job_t *self, int action ) |
---|
78 | { |
---|
79 | pbsdrmaa_session_t *session = (pbsdrmaa_session_t*)self->session; |
---|
80 | const char *job_id = self->job_id; |
---|
81 | |
---|
82 | fsd_log_enter(( "({job_id=%s}, action=%d)", self->job_id, action )); |
---|
83 | |
---|
84 | switch( action ) |
---|
85 | { |
---|
86 | /* |
---|
87 | * We cannot know whether we did suspend job |
---|
88 | * in other way than remembering this inside DRMAA session. |
---|
89 | */ |
---|
90 | case DRMAA_CONTROL_SUSPEND: |
---|
91 | session->pbs_connection->sigjob( session->pbs_connection, (char*)job_id, "SIGSTOP"); |
---|
92 | self->flags |= FSD_JOB_SUSPENDED; |
---|
93 | break; |
---|
94 | case DRMAA_CONTROL_RESUME: |
---|
95 | session->pbs_connection->sigjob( session->pbs_connection, (char*)job_id, "SIGCONT"); |
---|
96 | self->flags &= ~FSD_JOB_SUSPENDED; |
---|
97 | break; |
---|
98 | case DRMAA_CONTROL_HOLD: |
---|
99 | session->pbs_connection->holdjob( session->pbs_connection, (char*)job_id ); |
---|
100 | self->flags |= FSD_JOB_HOLD; |
---|
101 | break; |
---|
102 | case DRMAA_CONTROL_RELEASE: |
---|
103 | session->pbs_connection->rlsjob( session->pbs_connection, (char*)job_id ); |
---|
104 | self->flags &= ~FSD_JOB_HOLD; |
---|
105 | break; |
---|
106 | case DRMAA_CONTROL_TERMINATE: |
---|
107 | session->pbs_connection->deljob( session->pbs_connection, (char*)job_id ); |
---|
108 | /* TODO: make deldelay configurable ???: |
---|
109 | * deldelay=N -- delay between SIGTERM and SIGKILL (default 0) */ |
---|
110 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
111 | if( (self->flags & FSD_JOB_TERMINATED) == 0 ) |
---|
112 | self->flags |= FSD_JOB_TERMINATED | FSD_JOB_ABORTED; |
---|
113 | break; |
---|
114 | } |
---|
115 | |
---|
116 | fsd_log_return(("")); |
---|
117 | } |
---|
118 | |
---|
119 | |
---|
120 | void |
---|
121 | pbsdrmaa_job_update_status( fsd_job_t *self ) |
---|
122 | { |
---|
123 | struct batch_status *volatile status = NULL; |
---|
124 | pbsdrmaa_session_t *session = (pbsdrmaa_session_t*)self->session; |
---|
125 | |
---|
126 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
127 | |
---|
128 | TRY |
---|
129 | { |
---|
130 | |
---|
131 | #ifdef PBS_PROFESSIONAL |
---|
132 | status = session->pbs_connection->statjob( session->pbs_connection, self->job_id, NULL); |
---|
133 | #else |
---|
134 | status = session->pbs_connection->statjob( session->pbs_connection, self->job_id, session->status_attrl); |
---|
135 | #endif |
---|
136 | |
---|
137 | if( status != NULL ) |
---|
138 | { |
---|
139 | ((pbsdrmaa_job_t*)self)->update( self, status ); |
---|
140 | } |
---|
141 | else if( self->state < DRMAA_PS_DONE ) |
---|
142 | { |
---|
143 | self->on_missing( self ); |
---|
144 | } |
---|
145 | } |
---|
146 | FINALLY |
---|
147 | { |
---|
148 | if( status != NULL ) |
---|
149 | session->pbs_connection->statjob_free( session->pbs_connection, status ); |
---|
150 | } |
---|
151 | END_TRY |
---|
152 | |
---|
153 | fsd_log_return(("")); |
---|
154 | } |
---|
155 | |
---|
156 | |
---|
157 | void |
---|
158 | pbsdrmaa_job_update( fsd_job_t *self, struct batch_status *status ) |
---|
159 | { |
---|
160 | struct attrl *attribs = status->attribs; |
---|
161 | struct attrl *i = NULL; |
---|
162 | char pbs_state = 0; |
---|
163 | int exit_status = -101; |
---|
164 | const char *cpu_usage = NULL; |
---|
165 | const char *mem_usage = NULL; |
---|
166 | const char *vmem_usage = NULL; |
---|
167 | const char *walltime = NULL; |
---|
168 | long unsigned int modify_time = 0; |
---|
169 | |
---|
170 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
171 | #ifdef DEBUGGING |
---|
172 | pbsdrmaa_dump_attrl( attribs, NULL ); |
---|
173 | #endif |
---|
174 | fsd_assert( !strcmp( self->job_id, status->name ) ); |
---|
175 | |
---|
176 | for( i = attribs; i != NULL; i = i->next ) |
---|
177 | { |
---|
178 | int attr; |
---|
179 | attr = pbsdrmaa_pbs_attrib_by_name( i->name ); |
---|
180 | switch( attr ) |
---|
181 | { |
---|
182 | case PBSDRMAA_ATTR_JOB_STATE: |
---|
183 | pbs_state = i->value[0]; |
---|
184 | break; |
---|
185 | case PBSDRMAA_ATTR_EXIT_STATUS: |
---|
186 | exit_status = fsd_atoi( i->value ); |
---|
187 | break; |
---|
188 | case PBSDRMAA_ATTR_RESOURCES_USED: |
---|
189 | if( !strcmp( i->resource, "cput" ) ) |
---|
190 | cpu_usage = i->value; |
---|
191 | else if( !strcmp( i->resource, "mem" ) ) |
---|
192 | mem_usage = i->value; |
---|
193 | else if( !strcmp( i->resource, "vmem" ) ) |
---|
194 | vmem_usage = i->value; |
---|
195 | else if( !strcmp( i->resource, "walltime" ) ) |
---|
196 | walltime = i->value; |
---|
197 | break; |
---|
198 | case PBSDRMAA_ATTR_QUEUE: |
---|
199 | if (!self->queue) |
---|
200 | self->queue = fsd_strdup(i->value); |
---|
201 | break; |
---|
202 | case PBSDRMAA_ATTR_ACCOUNT_NAME: |
---|
203 | if (!self->project) |
---|
204 | self->project = fsd_strdup(i->value); |
---|
205 | break; |
---|
206 | #ifndef PBS_PROFESSIONAL |
---|
207 | case PBSDRMAA_ATTR_EXECUTION_HOST: |
---|
208 | if (!self->execution_hosts) { |
---|
209 | fsd_log_debug(("execution_hosts = %s", i->value)); |
---|
210 | self->execution_hosts = fsd_strdup(i->value); |
---|
211 | } |
---|
212 | break; |
---|
213 | #else |
---|
214 | case PBSDRMAA_ATTR_EXECUTION_VNODE: |
---|
215 | if (!self->execution_hosts) { |
---|
216 | fsd_log_debug(("execution_hosts = %s", i->value)); |
---|
217 | self->execution_hosts = fsd_strdup(i->value); |
---|
218 | } |
---|
219 | break; |
---|
220 | #endif |
---|
221 | case PBSDRMAA_ATTR_START_TIME: |
---|
222 | { |
---|
223 | long unsigned int start_time; |
---|
224 | if (self->start_time == 0 && sscanf(i->value, "%lu", &start_time) == 1) |
---|
225 | self->start_time = start_time; |
---|
226 | break; |
---|
227 | } |
---|
228 | case PBSDRMAA_ATTR_MTIME: |
---|
229 | if (sscanf(i->value, "%lu", &modify_time) != 1) |
---|
230 | modify_time = 0; |
---|
231 | break; |
---|
232 | } |
---|
233 | } |
---|
234 | |
---|
235 | if( pbs_state ) |
---|
236 | fsd_log_debug(( "pbs_state: %c", pbs_state )); |
---|
237 | |
---|
238 | if( exit_status != -101 ) |
---|
239 | { |
---|
240 | fsd_log_debug(( "exit_status: %d", exit_status )); |
---|
241 | self->exit_status = exit_status; |
---|
242 | |
---|
243 | if (self->exit_status < 0) |
---|
244 | { |
---|
245 | self->exit_status = -1; |
---|
246 | fsd_log_error(("ExitStatus = %d, probably system problem, report job %s to the local administrator", exit_status, self->job_id)); |
---|
247 | } |
---|
248 | |
---|
249 | } |
---|
250 | if(pbs_state){ |
---|
251 | switch( pbs_state ) |
---|
252 | { |
---|
253 | case 'C': /* Job is completed after having run. */ |
---|
254 | #ifdef PBS_PROFESSIONAL |
---|
255 | case 'F': /* Job is completed after having run. */ |
---|
256 | #endif |
---|
257 | self->flags &= FSD_JOB_TERMINATED_MASK; |
---|
258 | self->flags |= FSD_JOB_TERMINATED; |
---|
259 | if (exit_status != -101) { /* has exit code */ |
---|
260 | if( self->exit_status == 0) |
---|
261 | self->state = DRMAA_PS_DONE; |
---|
262 | else |
---|
263 | self->state = DRMAA_PS_FAILED; |
---|
264 | } else { |
---|
265 | self->state = DRMAA_PS_FAILED; |
---|
266 | self->exit_status = -1; |
---|
267 | } |
---|
268 | if (modify_time != 0) |
---|
269 | self->end_time = modify_time; /* take last modify time as end time */ |
---|
270 | else |
---|
271 | self->end_time = time(NULL); |
---|
272 | |
---|
273 | if (self->start_time == 0) |
---|
274 | self->start_time = self->end_time; |
---|
275 | |
---|
276 | break; |
---|
277 | case 'E': /* Job is exiting after having run. - MM: ignore exiting state (transient state) - outputs might have not been transfered yet, |
---|
278 | MM2: mark job as running if current job status is undetermined - fix "ps after job was ripped" */ |
---|
279 | if (self->state == DRMAA_PS_UNDETERMINED) |
---|
280 | self->state = DRMAA_PS_RUNNING; |
---|
281 | break; |
---|
282 | case 'H': /* Job is held. */ |
---|
283 | self->state = DRMAA_PS_USER_ON_HOLD; |
---|
284 | self->flags |= FSD_JOB_HOLD; |
---|
285 | break; |
---|
286 | case 'Q': /* Job is queued, eligible to run or routed. */ |
---|
287 | case 'W': /* Job is waiting for its execution time to be reached. */ |
---|
288 | self->state = DRMAA_PS_QUEUED_ACTIVE; |
---|
289 | self->flags &= ~FSD_JOB_HOLD; |
---|
290 | break; |
---|
291 | case 'R': /* Job is running. */ |
---|
292 | case 'T': /* Job is being moved to new location (?). */ |
---|
293 | { |
---|
294 | if( self->flags & FSD_JOB_SUSPENDED ) |
---|
295 | self->state = DRMAA_PS_USER_SUSPENDED; |
---|
296 | else |
---|
297 | self->state = DRMAA_PS_RUNNING; |
---|
298 | break; |
---|
299 | } |
---|
300 | case 'S': /* (Unicos only) job is suspend. */ |
---|
301 | self->state = DRMAA_PS_SYSTEM_SUSPENDED; |
---|
302 | break; |
---|
303 | case 0: default: |
---|
304 | self->state = DRMAA_PS_UNDETERMINED; |
---|
305 | break; |
---|
306 | } |
---|
307 | } |
---|
308 | fsd_log_debug(( "job_ps: %s", drmaa_job_ps_to_str(self->state) )); |
---|
309 | |
---|
310 | { |
---|
311 | int hours, minutes, seconds; |
---|
312 | long mem; |
---|
313 | if( cpu_usage && sscanf( cpu_usage, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 ) |
---|
314 | { |
---|
315 | self->cpu_usage = 60*( 60*hours + minutes ) + seconds; |
---|
316 | fsd_log_debug(( "cpu_usage: %s=%lds", cpu_usage, self->cpu_usage )); |
---|
317 | } |
---|
318 | if( mem_usage && sscanf( mem_usage, "%ldkb", &mem ) == 1 ) |
---|
319 | { |
---|
320 | self->mem_usage = 1024*mem; |
---|
321 | fsd_log_debug(( "mem_usage: %s=%ldB", mem_usage, self->mem_usage )); |
---|
322 | } |
---|
323 | if( vmem_usage && sscanf( vmem_usage, "%ldkb", &mem ) == 1 ) |
---|
324 | { |
---|
325 | self->vmem_usage = 1024*mem; |
---|
326 | fsd_log_debug(( "vmem_usage: %s=%ldB", vmem_usage, self->vmem_usage )); |
---|
327 | } |
---|
328 | if( walltime && sscanf( walltime, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 ) |
---|
329 | { |
---|
330 | self->walltime = 60*( 60*hours + minutes ) + seconds; |
---|
331 | fsd_log_debug(( "walltime: %s=%lds", walltime, self->walltime )); |
---|
332 | } |
---|
333 | } |
---|
334 | } |
---|
335 | |
---|
336 | void |
---|
337 | pbsdrmaa_job_on_missing( fsd_job_t *self ) |
---|
338 | { |
---|
339 | pbsdrmaa_session_t *pbssession = (pbsdrmaa_session_t*)self->session; |
---|
340 | pbsdrmaa_job_t *pbsself = (pbsdrmaa_job_t *)self; |
---|
341 | |
---|
342 | if (!pbsself->missing_time) |
---|
343 | { |
---|
344 | pbsself->missing_time = time(NULL); |
---|
345 | } |
---|
346 | |
---|
347 | fsd_log_info(("pbsdrmaa_job_on_missing: pbs_home=%s, wait_thread_started=%d, submit_time=%d, missing_time=%d", |
---|
348 | pbssession->pbs_home, |
---|
349 | pbssession->super.wait_thread_started, |
---|
350 | (int)self->submit_time, |
---|
351 | (int)pbsself->missing_time)); |
---|
352 | |
---|
353 | #define DRMAA_MAX_MISSING_TIME (30) |
---|
354 | if( pbssession->pbs_home != NULL && pbssession->super.wait_thread_started && self->submit_time && (time(NULL) - pbsself->missing_time < DRMAA_MAX_MISSING_TIME)) |
---|
355 | fsd_log_info(("Job on missing but WT is running. Skipping...")); /* TODO: try to provide implementation that uses accounting/server log files */ |
---|
356 | else |
---|
357 | pbsdrmaa_job_on_missing_standard( self ); |
---|
358 | } |
---|
359 | |
---|
360 | void |
---|
361 | pbsdrmaa_job_on_missing_standard( fsd_job_t *self ) |
---|
362 | { |
---|
363 | fsd_drmaa_session_t *session = self->session; |
---|
364 | pbsdrmaa_session_t *pbssession = (pbsdrmaa_session_t *)session; |
---|
365 | int exit_status = -1; |
---|
366 | |
---|
367 | fsd_log_enter(( "({job_id=%s})", self->job_id )); |
---|
368 | fsd_log_warning(( "Job %s missing from DRM queue", self->job_id )); |
---|
369 | |
---|
370 | fsd_log_info(( "job_on_missing: last job_ps: %s (0x%02x)", drmaa_job_ps_to_str(self->state), self->state)); |
---|
371 | |
---|
372 | if( (exit_status = pbsdrmaa_job_read_exit_status(self->job_id, pbssession->job_exit_status_file_prefix)) == 0 ) |
---|
373 | { |
---|
374 | self->state = DRMAA_PS_DONE; |
---|
375 | self->exit_status = exit_status; |
---|
376 | } |
---|
377 | else |
---|
378 | { |
---|
379 | self->state = DRMAA_PS_FAILED; |
---|
380 | self->exit_status = exit_status; |
---|
381 | } |
---|
382 | fsd_log_info(("job_on_missing evaluation result: state=%d exit_status=%d", self->state, self->exit_status)); |
---|
383 | |
---|
384 | fsd_cond_broadcast( &self->status_cond); |
---|
385 | fsd_cond_broadcast( &self->session->wait_condition ); |
---|
386 | |
---|
387 | fsd_log_return(( "; job_ps=%s, exit_status=%d", drmaa_job_ps_to_str(self->state), self->exit_status )); |
---|
388 | } |
---|
389 | |
---|
390 | int |
---|
391 | pbsdrmaa_job_read_exit_status( const char *job_id, const char *job_state_dir_prefix) |
---|
392 | { |
---|
393 | char *status_file = NULL, *start_file = NULL; |
---|
394 | FILE *fhandle = NULL; |
---|
395 | int exit_status = -1; |
---|
396 | |
---|
397 | fsd_log_enter(("({job_id=%s, job_state_dir_prefix=%s})", job_id, job_state_dir_prefix)); |
---|
398 | |
---|
399 | status_file = fsd_asprintf("%s/%s.exitcode", job_state_dir_prefix, job_id); |
---|
400 | start_file = fsd_asprintf("%s/%s.started", job_state_dir_prefix, job_id); |
---|
401 | |
---|
402 | if ((fhandle = fopen(status_file, "r")) == NULL) |
---|
403 | { |
---|
404 | struct stat tmpstat; |
---|
405 | |
---|
406 | fsd_log_error(("Failed to open job status file: %s", status_file)); |
---|
407 | if (stat(start_file, &tmpstat) == 0 && (tmpstat.st_mode & S_IFREG)) |
---|
408 | { |
---|
409 | exit_status = 143; /* SIGTERM */ |
---|
410 | fsd_log_info(("But start file exist %s. Assuming that job was killed (exit_status=%d).", start_file, exit_status)); |
---|
411 | } |
---|
412 | else |
---|
413 | { |
---|
414 | fsd_log_error(("Start file not found: %s", start_file)); |
---|
415 | } |
---|
416 | |
---|
417 | |
---|
418 | } |
---|
419 | else |
---|
420 | { |
---|
421 | (void)fscanf(fhandle, "%d", &exit_status); /*on error exit_status == -1 */ |
---|
422 | fclose(fhandle); |
---|
423 | } |
---|
424 | |
---|
425 | fsd_free(status_file); |
---|
426 | fsd_free(start_file); |
---|
427 | |
---|
428 | return exit_status; |
---|
429 | } |
---|
430 | |
---|