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) 2003-2005 Kern Sibbald
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 extern "C" void *jobq_server(void *arg);
47 extern "C" void *sched_wait(void *arg);
49 static int start_server(jobq_t *jq);
50 static bool acquire_resources(JCR *jcr);
55 * Initialize a job queue
57 * Returns: 0 on success
60 int jobq_init(jobq_t *jq, int threads, void *(*engine)(void *arg))
63 jobq_item_t *item = NULL;
65 if ((stat = pthread_attr_init(&jq->attr)) != 0) {
67 Jmsg1(NULL, M_ERROR, 0, "pthread_attr_init: ERR=%s\n", be.strerror(stat));
70 if ((stat = pthread_attr_setdetachstate(&jq->attr, PTHREAD_CREATE_DETACHED)) != 0) {
71 pthread_attr_destroy(&jq->attr);
74 if ((stat = pthread_mutex_init(&jq->mutex, NULL)) != 0) {
76 Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_init: ERR=%s\n", be.strerror(stat));
77 pthread_attr_destroy(&jq->attr);
80 if ((stat = pthread_cond_init(&jq->work, NULL)) != 0) {
82 Jmsg1(NULL, M_ERROR, 0, "pthread_cond_init: ERR=%s\n", be.strerror(stat));
83 pthread_mutex_destroy(&jq->mutex);
84 pthread_attr_destroy(&jq->attr);
88 jq->max_workers = threads; /* max threads to create */
89 jq->num_workers = 0; /* no threads yet */
90 jq->idle_workers = 0; /* no idle threads */
91 jq->engine = engine; /* routine to run */
92 jq->valid = JOBQ_VALID;
93 /* Initialize the job queues */
94 jq->waiting_jobs = New(dlist(item, &item->link));
95 jq->running_jobs = New(dlist(item, &item->link));
96 jq->ready_jobs = New(dlist(item, &item->link));
101 * Destroy the job queue
103 * Returns: 0 on success
106 int jobq_destroy(jobq_t *jq)
108 int stat, stat1, stat2;
110 if (jq->valid != JOBQ_VALID) {
113 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
115 Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
118 jq->valid = 0; /* prevent any more operations */
121 * If any threads are active, wake them
123 if (jq->num_workers > 0) {
125 if (jq->idle_workers) {
126 if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
128 Jmsg1(NULL, M_ERROR, 0, "pthread_cond_broadcast: ERR=%s\n", be.strerror(stat));
129 pthread_mutex_unlock(&jq->mutex);
133 while (jq->num_workers > 0) {
134 if ((stat = pthread_cond_wait(&jq->work, &jq->mutex)) != 0) {
136 Jmsg1(NULL, M_ERROR, 0, "pthread_cond_wait: ERR=%s\n", be.strerror(stat));
137 pthread_mutex_unlock(&jq->mutex);
142 if ((stat = pthread_mutex_unlock(&jq->mutex)) != 0) {
144 Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_unlock: ERR=%s\n", be.strerror(stat));
147 stat = pthread_mutex_destroy(&jq->mutex);
148 stat1 = pthread_cond_destroy(&jq->work);
149 stat2 = pthread_attr_destroy(&jq->attr);
150 delete jq->waiting_jobs;
151 delete jq->running_jobs;
152 delete jq->ready_jobs;
153 return (stat != 0 ? stat : (stat1 != 0 ? stat1 : stat2));
162 * Wait until schedule time arrives before starting. Normally
163 * this routine is only used for jobs started from the console
164 * for which the user explicitly specified a start time. Otherwise
165 * most jobs are put into the job queue only when their
166 * scheduled time arives.
169 void *sched_wait(void *arg)
171 JCR *jcr = ((wait_pkt *)arg)->jcr;
172 jobq_t *jq = ((wait_pkt *)arg)->jq;
174 Dmsg0(2300, "Enter sched_wait.\n");
176 time_t wtime = jcr->sched_time - time(NULL);
177 set_jcr_job_status(jcr, JS_WaitStartTime);
178 /* Wait until scheduled time arrives */
180 Jmsg(jcr, M_INFO, 0, _("Job %s waiting %d seconds for scheduled start time.\n"),
183 /* Check every 30 seconds if canceled */
185 Dmsg2(2300, "Waiting on sched time, jobid=%d secs=%d\n", jcr->JobId, wtime);
189 bmicrosleep(wtime, 0);
190 if (job_canceled(jcr)) {
193 wtime = jcr->sched_time - time(NULL);
195 P(jcr->mutex); /* lock jcr */
198 free_jcr(jcr); /* we are done with jcr */
199 Dmsg0(2300, "Exit sched_wait\n");
204 * Add a job to the queue
205 * jq is a queue that was created with jobq_init
207 * On entry jcr->mutex must be locked.
210 int jobq_add(jobq_t *jq, JCR *jcr)
213 jobq_item_t *item, *li;
214 bool inserted = false;
215 time_t wtime = jcr->sched_time - time(NULL);
219 Dmsg3(2300, "jobq_add jobid=%d jcr=0x%x use_count=%d\n", jcr->JobId, jcr, jcr->use_count);
220 if (jq->valid != JOBQ_VALID) {
221 Jmsg0(jcr, M_ERROR, 0, "Jobq_add queue not initialized.\n");
225 jcr->use_count++; /* mark jcr in use by us */
226 Dmsg3(2300, "jobq_add jobid=%d jcr=0x%x use_count=%d\n", jcr->JobId, jcr, jcr->use_count);
227 if (!job_canceled(jcr) && wtime > 0) {
228 set_thread_concurrency(jq->max_workers + 2);
229 sched_pkt = (wait_pkt *)malloc(sizeof(wait_pkt));
230 sched_pkt->jcr = jcr;
232 stat = pthread_create(&id, &jq->attr, sched_wait, (void *)sched_pkt);
233 if (stat != 0) { /* thread not created */
235 Jmsg1(jcr, M_ERROR, 0, "pthread_thread_create: ERR=%s\n", be.strerror(stat));
240 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
242 Jmsg1(jcr, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
243 jcr->use_count--; /* release jcr */
247 if ((item = (jobq_item_t *)malloc(sizeof(jobq_item_t))) == NULL) {
248 jcr->use_count--; /* release jcr */
253 if (job_canceled(jcr)) {
254 /* Add job to ready queue so that it is canceled quickly */
255 jq->ready_jobs->prepend(item);
256 Dmsg1(2300, "Prepended job=%d to ready queue\n", jcr->JobId);
258 /* Add this job to the wait queue in priority sorted order */
259 foreach_dlist(li, jq->waiting_jobs) {
260 Dmsg2(2300, "waiting item jobid=%d priority=%d\n",
261 li->jcr->JobId, li->jcr->JobPriority);
262 if (li->jcr->JobPriority > jcr->JobPriority) {
263 jq->waiting_jobs->insert_before(item, li);
264 Dmsg2(2300, "insert_before jobid=%d before waiting job=%d\n",
265 li->jcr->JobId, jcr->JobId);
270 /* If not jobs in wait queue, append it */
272 jq->waiting_jobs->append(item);
273 Dmsg1(2300, "Appended item jobid=%d to waiting queue\n", jcr->JobId);
277 /* Ensure that at least one server looks at the queue. */
278 stat = start_server(jq);
280 pthread_mutex_unlock(&jq->mutex);
281 Dmsg0(2300, "Return jobq_add\n");
286 * Remove a job from the job queue. Used only by cancel_job().
287 * jq is a queue that was created with jobq_init
288 * work_item is an element of work
290 * Note, it is "removed" from the job queue.
291 * If you want to cancel it, you need to provide some external means
292 * of doing so (e.g. pthread_kill()).
294 int jobq_remove(jobq_t *jq, JCR *jcr)
300 Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x\n", jcr->JobId, jcr);
301 if (jq->valid != JOBQ_VALID) {
305 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
307 Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
311 foreach_dlist(item, jq->waiting_jobs) {
312 if (jcr == item->jcr) {
318 pthread_mutex_unlock(&jq->mutex);
319 Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x not in wait queue\n", jcr->JobId, jcr);
323 /* Move item to be the first on the list */
324 jq->waiting_jobs->remove(item);
325 jq->ready_jobs->prepend(item);
326 Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x moved to ready queue\n", jcr->JobId, jcr);
328 stat = start_server(jq);
330 pthread_mutex_unlock(&jq->mutex);
331 Dmsg0(2300, "Return jobq_remove\n");
337 * Start the server thread if it isn't already running
339 static int start_server(jobq_t *jq)
345 * if any threads are idle, wake one --
346 * actually we do a broadcast because on /lib/tls
347 * these signals seem to get lost from time to time.
349 if (jq->idle_workers > 0) {
350 Dmsg0(2300, "Signal worker to wake up\n");
351 if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
353 Jmsg1(NULL, M_ERROR, 0, "pthread_cond_signal: ERR=%s\n", be.strerror(stat));
356 } else if (jq->num_workers < jq->max_workers) {
357 Dmsg0(2300, "Create worker thread\n");
358 /* No idle threads so create a new one */
359 set_thread_concurrency(jq->max_workers + 1);
360 if ((stat = pthread_create(&id, &jq->attr, jobq_server, (void *)jq)) != 0) {
362 Jmsg1(NULL, M_ERROR, 0, "pthread_create: ERR=%s\n", be.strerror(stat));
371 * This is the worker thread that serves the job queue.
372 * When all the resources are acquired for the job,
373 * it will call the user's engine.
376 void *jobq_server(void *arg)
378 struct timespec timeout;
379 jobq_t *jq = (jobq_t *)arg;
380 jobq_item_t *je; /* job entry in queue */
382 bool timedout = false;
385 Dmsg0(2300, "Start jobq_server\n");
386 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
388 Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
397 Dmsg0(2300, "Top of for loop\n");
398 if (!work && !jq->quit) {
399 gettimeofday(&tv, &tz);
401 timeout.tv_sec = tv.tv_sec + 4;
405 * Wait 4 seconds, then if no more work, exit
407 Dmsg0(2300, "pthread_cond_timedwait()\n");
408 stat = pthread_cond_timedwait(&jq->work, &jq->mutex, &timeout);
409 if (stat == ETIMEDOUT) {
410 Dmsg0(2300, "timedwait timedout.\n");
413 } else if (stat != 0) {
414 /* This shouldn't happen */
415 Dmsg0(2300, "This shouldn't happen\n");
417 pthread_mutex_unlock(&jq->mutex);
424 * If anything is in the ready queue, run it
426 Dmsg0(2300, "Checking ready queue.\n");
427 while (!jq->ready_jobs->empty() && !jq->quit) {
429 je = (jobq_item_t *)jq->ready_jobs->first();
431 jq->ready_jobs->remove(je);
432 if (!jq->ready_jobs->empty()) {
433 Dmsg0(2300, "ready queue not empty start server\n");
434 if (start_server(jq) != 0) {
436 pthread_mutex_unlock(&jq->mutex);
440 jq->running_jobs->append(je);
441 Dmsg1(2300, "Took jobid=%d from ready and appended to run\n", jcr->JobId);
443 /* Release job queue lock */
446 /* Call user's routine here */
447 Dmsg1(2300, "Calling user engine for jobid=%d\n", jcr->JobId);
450 Dmsg1(2300, "Back from user engine jobid=%d.\n", jcr->JobId);
452 /* Reacquire job queue lock */
454 Dmsg0(200, "Done lock mutex after running job. Release locks.\n");
455 jq->running_jobs->remove(je);
457 * Release locks if acquired. Note, they will not have
458 * been acquired for jobs canceled before they were
459 * put into the ready queue.
461 if (jcr->acquired_resource_locks) {
462 jcr->store->NumConcurrentJobs--;
463 jcr->client->NumConcurrentJobs--;
464 jcr->job->NumConcurrentJobs--;
468 * Reschedule the job if necessary and requested
470 if (jcr->job->RescheduleOnError &&
471 jcr->JobStatus != JS_Terminated &&
472 jcr->JobStatus != JS_Canceled &&
473 jcr->job->RescheduleTimes > 0 &&
474 jcr->reschedule_count < jcr->job->RescheduleTimes) {
478 * Reschedule this job by cleaning it up, but
479 * reuse the same JobId if possible.
481 jcr->reschedule_count++;
482 jcr->sched_time = time(NULL) + jcr->job->RescheduleInterval;
483 Dmsg2(2300, "Rescheduled Job %s to re-run in %d seconds.\n", jcr->Job,
484 (int)jcr->job->RescheduleInterval);
485 bstrftime(dt, sizeof(dt), time(NULL));
486 Jmsg(jcr, M_INFO, 0, _("Rescheduled Job %s at %s to re-run in %d seconds.\n"),
487 jcr->Job, dt, (int)jcr->job->RescheduleInterval);
488 dird_free_jcr_pointers(jcr); /* partial cleanup old stuff */
489 jcr->JobStatus = JS_WaitStartTime;
490 jcr->SDJobStatus = 0;
491 if (jcr->JobBytes == 0) {
492 Dmsg1(2300, "Requeue job=%d\n", jcr->JobId);
493 jcr->JobStatus = JS_WaitStartTime;
495 jobq_add(jq, jcr); /* queue the job to run again */
497 free(je); /* free the job entry */
498 continue; /* look for another job to run */
501 * Something was actually backed up, so we cannot reuse
502 * the old JobId or there will be database record
503 * conflicts. We now create a new job, copying the
504 * appropriate fields.
506 JCR *njcr = new_jcr(sizeof(JCR), dird_free_jcr);
507 set_jcr_defaults(njcr, jcr->job);
508 njcr->reschedule_count = jcr->reschedule_count;
509 njcr->JobLevel = jcr->JobLevel;
510 njcr->JobStatus = jcr->JobStatus;
511 copy_storage(njcr, jcr);
512 njcr->messages = jcr->messages;
513 Dmsg0(2300, "Call to run new job\n");
515 run_job(njcr); /* This creates a "new" job */
516 free_jcr(njcr); /* release "new" jcr */
518 Dmsg0(2300, "Back from running new job.\n");
520 /* Clean up and release old jcr */
522 db_close_database(jcr, jcr->db);
525 Dmsg2(2300, "====== Termination job=%d use_cnt=%d\n", jcr->JobId, jcr->use_count);
526 jcr->SDJobStatus = 0;
527 V(jq->mutex); /* release internal lock */
529 free(je); /* release job entry */
530 P(jq->mutex); /* reacquire job queue lock */
533 * If any job in the wait queue can be run,
534 * move it to the ready queue
536 Dmsg0(2300, "Done check ready, now check wait queue.\n");
537 if (!jq->waiting_jobs->empty() && !jq->quit) {
539 je = (jobq_item_t *)jq->waiting_jobs->first();
540 jobq_item_t *re = (jobq_item_t *)jq->running_jobs->first();
542 Priority = re->jcr->JobPriority;
543 Dmsg2(2300, "JobId %d is running. Look for pri=%d\n", re->jcr->JobId, Priority);
545 Priority = je->jcr->JobPriority;
546 Dmsg1(2300, "No job running. Look for Job pri=%d\n", Priority);
549 * Walk down the list of waiting jobs and attempt
550 * to acquire the resources it needs.
553 /* je is current job item on the queue, jn is the next one */
555 jobq_item_t *jn = (jobq_item_t *)jq->waiting_jobs->next(je);
557 Dmsg3(2300, "Examining Job=%d JobPri=%d want Pri=%d\n",
558 jcr->JobId, jcr->JobPriority, Priority);
560 /* Take only jobs of correct Priority */
561 if (jcr->JobPriority != Priority) {
562 set_jcr_job_status(jcr, JS_WaitPriority);
566 if (!acquire_resources(jcr)) {
567 je = jn; /* point to next waiting job */
571 /* Got all locks, now remove it from wait queue and append it
574 jcr->acquired_resource_locks = true;
575 jq->waiting_jobs->remove(je);
576 jq->ready_jobs->append(je);
577 Dmsg1(2300, "moved JobId=%d from wait to ready queue\n", je->jcr->JobId);
578 je = jn; /* Point to next waiting job */
583 Dmsg0(2300, "Done checking wait queue.\n");
585 * If no more ready work and we are asked to quit, then do it
587 if (jq->ready_jobs->empty() && jq->quit) {
589 if (jq->num_workers == 0) {
590 Dmsg0(2300, "Wake up destroy routine\n");
591 /* Wake up destroy routine if he is waiting */
592 pthread_cond_broadcast(&jq->work);
596 Dmsg0(2300, "Check for work request\n");
598 * If no more work requests, and we waited long enough, quit
600 Dmsg2(2300, "timedout=%d read empty=%d\n", timedout,
601 jq->ready_jobs->empty());
602 if (jq->ready_jobs->empty() && timedout) {
603 Dmsg0(2300, "break big loop\n");
608 work = !jq->ready_jobs->empty() || !jq->waiting_jobs->empty();
611 * If a job is waiting on a Resource, don't consume all
612 * the CPU time looping looking for work, and even more
613 * important, release the lock so that a job that has
614 * terminated can give us the resource.
617 bmicrosleep(2, 0); /* pause for 2 seconds */
619 /* Recompute work as something may have changed in last 2 secs */
620 work = !jq->ready_jobs->empty() || !jq->waiting_jobs->empty();
622 Dmsg1(2300, "Loop again. work=%d\n", work);
623 } /* end of big for loop */
625 Dmsg0(200, "unlock mutex\n");
627 Dmsg0(2300, "End jobq_server\n");
632 * See if we can acquire all the necessary resources for the job (JCR)
634 * Returns: true if successful
635 * false if resource failure
637 static bool acquire_resources(JCR *jcr)
639 bool skip_this_jcr = false;
641 if (jcr->JobType == JT_RESTORE || jcr->JobType == JT_VERIFY) {
643 * Let only one Restore/verify job run at a time regardless
644 * of MaxConcurrentJobs.
646 if (jcr->store->NumConcurrentJobs == 0) {
647 jcr->store->NumConcurrentJobs = 1;
649 set_jcr_job_status(jcr, JS_WaitStoreRes);
652 /* We are not doing a Restore or Verify */
653 } else if (jcr->store->NumConcurrentJobs == 0 &&
654 jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {
655 /* Simple case, first job */
656 jcr->store->NumConcurrentJobs = 1;
657 } else if (jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {
658 jcr->store->NumConcurrentJobs++;
660 skip_this_jcr = true;
663 set_jcr_job_status(jcr, JS_WaitStoreRes);
667 if (jcr->client->NumConcurrentJobs < jcr->client->MaxConcurrentJobs) {
668 jcr->client->NumConcurrentJobs++;
670 /* Back out previous locks */
671 jcr->store->NumConcurrentJobs--;
672 set_jcr_job_status(jcr, JS_WaitClientRes);
675 if (jcr->job->NumConcurrentJobs < jcr->job->MaxConcurrentJobs) {
676 jcr->job->NumConcurrentJobs++;
678 /* Back out previous locks */
679 jcr->store->NumConcurrentJobs--;
680 jcr->client->NumConcurrentJobs--;
681 set_jcr_job_status(jcr, JS_WaitJobRes);
684 /* Check actual device availability */