2 * Bacula job queue routines.
4 * This code consists of three queues, the waiting_jobs
5 * queue, where jobs are initially queued, the ready_jobs
6 * queue, where jobs are placed when all the resources are
7 * allocated and they can immediately be run, and the
8 * running queue where jobs are placed when they are
11 * Kern Sibbald, July MMIII
15 * This code was adapted from the Bacula workq, which was
16 * adapted from "Programming with POSIX Threads", by
21 Copyright (C) 2000-2003 Kern Sibbald and John Walker
23 This program is free software; you can redistribute it and/or
24 modify it under the terms of the GNU General Public License as
25 published by the Free Software Foundation; either version 2 of
26 the License, or (at your option) any later version.
28 This program is distributed in the hope that it will be useful,
29 but WITHOUT ANY WARRANTY; without even the implied warranty of
30 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
31 General Public License for more details.
33 You should have received a copy of the GNU General Public
34 License along with this program; if not, write to the Free
35 Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
45 /* Forward referenced functions */
46 static void *jobq_server(void *arg);
47 static int start_server(jobq_t *jq);
50 * Initialize a job queue
52 * Returns: 0 on success
55 int jobq_init(jobq_t *jq, int threads, void *(*engine)(void *arg))
58 jobq_item_t *item = NULL;
60 if ((stat = pthread_attr_init(&jq->attr)) != 0) {
63 if ((stat = pthread_attr_setdetachstate(&jq->attr, PTHREAD_CREATE_DETACHED)) != 0) {
64 pthread_attr_destroy(&jq->attr);
67 if ((stat = pthread_mutex_init(&jq->mutex, NULL)) != 0) {
68 pthread_attr_destroy(&jq->attr);
71 if ((stat = pthread_cond_init(&jq->work, NULL)) != 0) {
72 pthread_mutex_destroy(&jq->mutex);
73 pthread_attr_destroy(&jq->attr);
77 jq->max_workers = threads; /* max threads to create */
78 jq->num_workers = 0; /* no threads yet */
79 jq->idle_workers = 0; /* no idle threads */
80 jq->engine = engine; /* routine to run */
81 jq->valid = JOBQ_VALID;
82 /* Initialize the job queues */
83 jq->waiting_jobs = new dlist(item, &item->link);
84 jq->running_jobs = new dlist(item, &item->link);
85 jq->ready_jobs = new dlist(item, &item->link);
90 * Destroy the job queue
92 * Returns: 0 on success
95 int jobq_destroy(jobq_t *jq)
97 int stat, stat1, stat2;
99 if (jq->valid != JOBQ_VALID) {
102 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
105 jq->valid = 0; /* prevent any more operations */
108 * If any threads are active, wake them
110 if (jq->num_workers > 0) {
112 if (jq->idle_workers) {
113 if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
114 pthread_mutex_unlock(&jq->mutex);
118 while (jq->num_workers > 0) {
119 if ((stat = pthread_cond_wait(&jq->work, &jq->mutex)) != 0) {
120 pthread_mutex_unlock(&jq->mutex);
125 if ((stat = pthread_mutex_unlock(&jq->mutex)) != 0) {
128 stat = pthread_mutex_destroy(&jq->mutex);
129 stat1 = pthread_cond_destroy(&jq->work);
130 stat2 = pthread_attr_destroy(&jq->attr);
131 delete jq->waiting_jobs;
132 delete jq->running_jobs;
133 delete jq->ready_jobs;
134 return (stat != 0 ? stat : (stat1 != 0 ? stat1 : stat2));
143 * Wait until schedule time arrives before starting. Normally
144 * this routine is only used for jobs started from the console
145 * for which the user explicitly specified a start time. Otherwise
146 * most jobs are put into the job queue only when their
147 * scheduled time arives.
149 static void *sched_wait(void *arg)
151 JCR *jcr = ((wait_pkt *)arg)->jcr;
152 jobq_t *jq = ((wait_pkt *)arg)->jq;
154 Dmsg0(100, "Enter sched_wait.\n");
156 time_t wtime = jcr->sched_time - time(NULL);
157 /* Wait until scheduled time arrives */
158 if (wtime > 0 && verbose) {
159 Jmsg(jcr, M_INFO, 0, _("Job %s waiting %d seconds for scheduled start time.\n"),
161 set_jcr_job_status(jcr, JS_WaitStartTime);
163 /* Check every 30 seconds if canceled */
165 Dmsg2(100, "Waiting on sched time, jobid=%d secs=%d\n", jcr->JobId, wtime);
169 bmicrosleep(wtime, 0);
170 if (job_canceled(jcr)) {
173 wtime = jcr->sched_time - time(NULL);
176 Dmsg0(100, "Exit sched_wait\n");
182 * Add a job to the queue
183 * jq is a queue that was created with jobq_init
186 int jobq_add(jobq_t *jq, JCR *jcr)
189 jobq_item_t *item, *li;
190 bool inserted = false;
191 time_t wtime = jcr->sched_time - time(NULL);
196 Dmsg1(100, "jobq_add jobid=%d\n", jcr->JobId);
197 if (jq->valid != JOBQ_VALID) {
201 if (!job_canceled(jcr) && wtime > 0) {
202 set_thread_concurrency(jq->max_workers + 2);
203 sched_pkt = (wait_pkt *)malloc(sizeof(wait_pkt));
204 sched_pkt->jcr = jcr;
206 stat = pthread_create(&id, &jq->attr, sched_wait, (void *)sched_pkt);
210 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
214 if ((item = (jobq_item_t *)malloc(sizeof(jobq_item_t))) == NULL) {
219 if (job_canceled(jcr)) {
220 /* Add job to ready queue so that it is canceled quickly */
221 jq->ready_jobs->prepend(item);
222 Dmsg1(100, "Prepended job=%d to ready queue\n", jcr->JobId);
224 /* Add this job to the wait queue in priority sorted order */
225 for (li=NULL; (li=(jobq_item_t *)jq->waiting_jobs->next(li)); ) {
226 Dmsg2(100, "waiting item jobid=%d priority=%d\n",
227 li->jcr->JobId, li->jcr->JobPriority);
228 if (li->jcr->JobPriority > jcr->JobPriority) {
229 jq->waiting_jobs->insert_before(item, li);
230 Dmsg2(100, "insert_before jobid=%d before %d\n",
231 li->jcr->JobId, jcr->JobId);
236 /* If not jobs in wait queue, append it */
238 jq->waiting_jobs->append(item);
239 Dmsg1(100, "Appended item jobid=%d\n", jcr->JobId);
243 /* Ensure that at least one server looks at the queue. */
244 stat = start_server(jq);
247 pthread_mutex_unlock(&jq->mutex);
249 Dmsg0(100, "Return jobq_add\n");
254 * Remove a job from the job queue. Used only by cancel Console command.
255 * jq is a queue that was created with jobq_init
256 * work_item is an element of work
258 * Note, it is "removed" by immediately calling a processing routine.
259 * if you want to cancel it, you need to provide some external means
262 int jobq_remove(jobq_t *jq, JCR *jcr)
268 Dmsg1(100, "jobq_remove jobid=%d\n", jcr->JobId);
269 if (jq->valid != JOBQ_VALID) {
273 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
277 for (item=NULL; (item=(jobq_item_t *)jq->waiting_jobs->next(item)); ) {
278 if (jcr == item->jcr) {
287 /* Move item to be the first on the list */
288 jq->waiting_jobs->remove(item);
289 jq->ready_jobs->prepend(item);
291 stat = start_server(jq);
295 pthread_mutex_unlock(&jq->mutex);
296 Dmsg0(100, "Return jobq_remove\n");
302 * Start the server thread if it isn't already running
304 static int start_server(jobq_t *jq)
309 /* if any threads are idle, wake one */
310 if (jq->idle_workers > 0) {
311 Dmsg0(100, "Signal worker to wake up\n");
312 if ((stat = pthread_cond_signal(&jq->work)) != 0) {
313 pthread_mutex_unlock(&jq->mutex);
316 } else if (jq->num_workers < jq->max_workers) {
317 Dmsg0(100, "Create worker thread\n");
318 /* No idle threads so create a new one */
319 set_thread_concurrency(jq->max_workers + 1);
320 if ((stat = pthread_create(&id, &jq->attr, jobq_server, (void *)jq)) != 0) {
321 pthread_mutex_unlock(&jq->mutex);
331 * This is the worker thread that serves the job queue.
332 * When all the resources are acquired for the job,
333 * it will call the user's engine.
335 static void *jobq_server(void *arg)
337 struct timespec timeout;
338 jobq_t *jq = (jobq_t *)arg;
339 jobq_item_t *je; /* job entry in queue */
344 Dmsg0(100, "Start jobq_server\n");
345 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
353 Dmsg0(100, "Top of for loop\n");
355 Dmsg0(100, "gettimeofday()\n");
356 gettimeofday(&tv, &tz);
358 timeout.tv_sec = tv.tv_sec + 4;
360 while (!work && !jq->quit) {
362 * Wait 4 seconds, then if no more work, exit
364 Dmsg0(200, "pthread_cond_timedwait()\n");
365 stat = pthread_cond_timedwait(&jq->work, &jq->mutex, &timeout);
366 Dmsg1(100, "timedwait=%d\n", stat);
367 if (stat == ETIMEDOUT) {
370 } else if (stat != 0) {
371 /* This shouldn't happen */
372 Dmsg0(100, "This shouldn't happen\n");
374 pthread_mutex_unlock(&jq->mutex);
379 * If anything is in the ready queue, run it
381 Dmsg0(100, "Checking ready queue.\n");
382 while (!jq->ready_jobs->empty() && !jq->quit) {
384 je = (jobq_item_t *)jq->ready_jobs->first();
386 jq->ready_jobs->remove(je);
387 if (!jq->ready_jobs->empty()) {
388 Dmsg0(100, "ready queue not empty start server\n");
389 if (start_server(jq) != 0) {
393 jq->running_jobs->append(je);
394 Dmsg1(100, "Took jobid=%d from ready and appended to run\n", jcr->JobId);
395 if ((stat = pthread_mutex_unlock(&jq->mutex)) != 0) {
398 /* Call user's routine here */
399 Dmsg1(100, "Calling user engine for jobid=%d\n", jcr->JobId);
401 Dmsg1(100, "Back from user engine jobid=%d.\n", jcr->JobId);
402 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
403 free(je); /* release job entry */
406 Dmsg0(200, "Done lock mutex\n");
407 jq->running_jobs->remove(je);
409 * Release locks if acquired. Note, they will not have
410 * been acquired for jobs canceled before they were
411 * put into the ready queue.
413 if (jcr->acquired_resource_locks) {
414 jcr->store->NumConcurrentJobs--;
415 jcr->client->NumConcurrentJobs--;
416 jcr->job->NumConcurrentJobs--;
419 if (jcr->job->RescheduleOnError &&
420 jcr->JobStatus != JS_Terminated &&
421 jcr->JobStatus != JS_Canceled &&
422 jcr->job->RescheduleTimes > 0 &&
423 jcr->reschedule_count < jcr->job->RescheduleTimes) {
426 * Reschedule this job by cleaning it up, but
427 * reuse the same JobId if possible.
429 jcr->reschedule_count++;
430 jcr->sched_time = time(NULL) + jcr->job->RescheduleInterval;
431 Dmsg2(100, "Rescheduled Job %s to re-run in %d seconds.\n", jcr->Job,
432 (int)jcr->job->RescheduleInterval);
433 jcr->JobStatus = JS_Created; /* force new status */
434 dird_free_jcr(jcr); /* partial cleanup old stuff */
435 if (jcr->JobBytes == 0) {
436 Dmsg1(100, "Requeue job=%d\n", jcr->JobId);
438 jobq_add(jq, jcr); /* queue the job to run again */
440 free(je); /* free the job entry */
441 continue; /* look for another job to run */
444 * Something was actually backed up, so we cannot reuse
445 * the old JobId or there will be database record
446 * conflicts. We now create a new job, copying the
447 * appropriate fields.
449 JCR *njcr = new_jcr(sizeof(JCR), dird_free_jcr);
450 set_jcr_defaults(njcr, jcr->job);
451 njcr->reschedule_count = jcr->reschedule_count;
452 njcr->JobLevel = jcr->JobLevel;
453 njcr->JobStatus = jcr->JobStatus;
454 njcr->pool = jcr->pool;
455 njcr->store = jcr->store;
456 njcr->messages = jcr->messages;
457 Dmsg0(100, "Call to run new job\n");
459 run_job(njcr); /* This creates a "new" job */
461 Dmsg0(100, "Back from running new job.\n");
463 /* Clean up and release old jcr */
465 Dmsg0(100, "Close DB\n");
466 db_close_database(jcr, jcr->db);
469 Dmsg1(100, "====== Termination job=%d\n", jcr->JobId);
471 free(je); /* release job entry */
474 * If any job in the wait queue can be run,
475 * move it to the ready queue
477 Dmsg0(100, "Done check ready, now check wait queue.\n");
478 while (!jq->waiting_jobs->empty() && !jq->quit) {
480 je = (jobq_item_t *)jq->waiting_jobs->first();
481 jobq_item_t *re = (jobq_item_t *)jq->running_jobs->first();
483 Priority = re->jcr->JobPriority;
484 Dmsg1(100, "Set Run pri=%d\n", Priority);
486 Priority = je->jcr->JobPriority;
487 Dmsg1(100, "Set Job pri=%d\n", Priority);
490 * Walk down the list of waiting jobs and attempt
491 * to acquire the resources it needs.
494 /* je is current job item on the queue, jn is the next one */
496 jobq_item_t *jn = (jobq_item_t *)jq->waiting_jobs->next(je);
497 Dmsg3(100, "Examining Job=%d JobPri=%d want Pri=%d\n",
498 jcr->JobId, jcr->JobPriority, Priority);
499 /* Take only jobs of correct Priority */
500 if (jcr->JobPriority != Priority) {
501 set_jcr_job_status(jcr, JS_WaitPriority);
504 if (jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {
505 jcr->store->NumConcurrentJobs++;
507 set_jcr_job_status(jcr, JS_WaitStoreRes);
511 if (jcr->client->NumConcurrentJobs < jcr->client->MaxConcurrentJobs) {
512 jcr->client->NumConcurrentJobs++;
514 jcr->store->NumConcurrentJobs--;
515 set_jcr_job_status(jcr, JS_WaitClientRes);
519 if (jcr->job->NumConcurrentJobs < jcr->job->MaxConcurrentJobs) {
520 jcr->job->NumConcurrentJobs++;
522 jcr->store->NumConcurrentJobs--;
523 jcr->client->NumConcurrentJobs--;
524 set_jcr_job_status(jcr, JS_WaitJobRes);
528 /* Got all locks, now remove it from wait queue and append it
531 jcr->acquired_resource_locks = true;
532 jq->waiting_jobs->remove(je);
533 jq->ready_jobs->append(je);
534 Dmsg1(100, "moved JobId=%d from wait to ready queue\n", je->jcr->JobId);
538 } /* end while loop */
539 Dmsg0(100, "Done checking wait queue.\n");
541 * If no more ready work and we are asked to quit, then do it
543 if (jq->ready_jobs->empty() && jq->quit) {
545 if (jq->num_workers == 0) {
546 Dmsg0(100, "Wake up destroy routine\n");
547 /* Wake up destroy routine if he is waiting */
548 pthread_cond_broadcast(&jq->work);
552 Dmsg0(100, "Check for work request\n");
554 * If no more work requests, and we waited long enough, quit
556 Dmsg2(100, "timedout=%d read empty=%d\n", timedout,
557 jq->ready_jobs->empty());
558 if (jq->ready_jobs->empty() && timedout) {
559 Dmsg0(100, "break big loop\n");
563 Dmsg0(100, "Loop again\n");
565 } /* end of big for loop */
567 Dmsg0(200, "unlock mutex\n");
568 pthread_mutex_unlock(&jq->mutex);
569 Dmsg0(100, "End jobq_server\n");
573 #endif /* JOB_QUEUE */