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 // jcr->use_count--; /* release our use of jcr */
233 stat = pthread_create(&id, &jq->attr, sched_wait, (void *)sched_pkt);
234 if (stat != 0) { /* thread not created */
236 Jmsg1(jcr, M_ERROR, 0, "pthread_thread_create: ERR=%s\n", be.strerror(stat));
241 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
243 Jmsg1(jcr, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
244 jcr->use_count--; /* release jcr */
248 if ((item = (jobq_item_t *)malloc(sizeof(jobq_item_t))) == NULL) {
249 jcr->use_count--; /* release jcr */
254 if (job_canceled(jcr)) {
255 /* Add job to ready queue so that it is canceled quickly */
256 jq->ready_jobs->prepend(item);
257 Dmsg1(2300, "Prepended job=%d to ready queue\n", jcr->JobId);
259 /* Add this job to the wait queue in priority sorted order */
260 foreach_dlist(li, jq->waiting_jobs) {
261 Dmsg2(2300, "waiting item jobid=%d priority=%d\n",
262 li->jcr->JobId, li->jcr->JobPriority);
263 if (li->jcr->JobPriority > jcr->JobPriority) {
264 jq->waiting_jobs->insert_before(item, li);
265 Dmsg2(2300, "insert_before jobid=%d before waiting job=%d\n",
266 li->jcr->JobId, jcr->JobId);
271 /* If not jobs in wait queue, append it */
273 jq->waiting_jobs->append(item);
274 Dmsg1(2300, "Appended item jobid=%d to waiting queue\n", jcr->JobId);
278 /* Ensure that at least one server looks at the queue. */
279 stat = start_server(jq);
281 pthread_mutex_unlock(&jq->mutex);
282 Dmsg0(2300, "Return jobq_add\n");
287 * Remove a job from the job queue. Used only by cancel_job().
288 * jq is a queue that was created with jobq_init
289 * work_item is an element of work
291 * Note, it is "removed" from the job queue.
292 * If you want to cancel it, you need to provide some external means
293 * of doing so (e.g. pthread_kill()).
295 int jobq_remove(jobq_t *jq, JCR *jcr)
301 Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x\n", jcr->JobId, jcr);
302 if (jq->valid != JOBQ_VALID) {
306 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
308 Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
312 foreach_dlist(item, jq->waiting_jobs) {
313 if (jcr == item->jcr) {
319 pthread_mutex_unlock(&jq->mutex);
320 Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x not in wait queue\n", jcr->JobId, jcr);
324 /* Move item to be the first on the list */
325 jq->waiting_jobs->remove(item);
326 jq->ready_jobs->prepend(item);
327 Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x moved to ready queue\n", jcr->JobId, jcr);
329 stat = start_server(jq);
331 pthread_mutex_unlock(&jq->mutex);
332 Dmsg0(2300, "Return jobq_remove\n");
338 * Start the server thread if it isn't already running
340 static int start_server(jobq_t *jq)
346 * if any threads are idle, wake one --
347 * actually we do a broadcast because on /lib/tls
348 * these signals seem to get lost from time to time.
350 if (jq->idle_workers > 0) {
351 Dmsg0(2300, "Signal worker to wake up\n");
352 if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
354 Jmsg1(NULL, M_ERROR, 0, "pthread_cond_signal: ERR=%s\n", be.strerror(stat));
357 } else if (jq->num_workers < jq->max_workers) {
358 Dmsg0(2300, "Create worker thread\n");
359 /* No idle threads so create a new one */
360 set_thread_concurrency(jq->max_workers + 1);
361 if ((stat = pthread_create(&id, &jq->attr, jobq_server, (void *)jq)) != 0) {
363 Jmsg1(NULL, M_ERROR, 0, "pthread_create: ERR=%s\n", be.strerror(stat));
372 * This is the worker thread that serves the job queue.
373 * When all the resources are acquired for the job,
374 * it will call the user's engine.
377 void *jobq_server(void *arg)
379 struct timespec timeout;
380 jobq_t *jq = (jobq_t *)arg;
381 jobq_item_t *je; /* job entry in queue */
383 bool timedout = false;
386 Dmsg0(2300, "Start jobq_server\n");
387 if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
389 Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
398 Dmsg0(2300, "Top of for loop\n");
399 if (!work && !jq->quit) {
400 gettimeofday(&tv, &tz);
402 timeout.tv_sec = tv.tv_sec + 4;
406 * Wait 4 seconds, then if no more work, exit
408 Dmsg0(2300, "pthread_cond_timedwait()\n");
409 stat = pthread_cond_timedwait(&jq->work, &jq->mutex, &timeout);
410 if (stat == ETIMEDOUT) {
411 Dmsg0(2300, "timedwait timedout.\n");
414 } else if (stat != 0) {
415 /* This shouldn't happen */
416 Dmsg0(2300, "This shouldn't happen\n");
418 pthread_mutex_unlock(&jq->mutex);
425 * If anything is in the ready queue, run it
427 Dmsg0(2300, "Checking ready queue.\n");
428 while (!jq->ready_jobs->empty() && !jq->quit) {
430 je = (jobq_item_t *)jq->ready_jobs->first();
432 jq->ready_jobs->remove(je);
433 if (!jq->ready_jobs->empty()) {
434 Dmsg0(2300, "ready queue not empty start server\n");
435 if (start_server(jq) != 0) {
437 pthread_mutex_unlock(&jq->mutex);
441 jq->running_jobs->append(je);
442 Dmsg1(2300, "Took jobid=%d from ready and appended to run\n", jcr->JobId);
444 /* Release job queue lock */
447 /* Call user's routine here */
448 Dmsg1(2300, "Calling user engine for jobid=%d\n", jcr->JobId);
451 Dmsg1(2300, "Back from user engine jobid=%d.\n", jcr->JobId);
453 /* Reacquire job queue lock */
455 Dmsg0(200, "Done lock mutex after running job. Release locks.\n");
456 jq->running_jobs->remove(je);
458 * Release locks if acquired. Note, they will not have
459 * been acquired for jobs canceled before they were
460 * put into the ready queue.
462 if (jcr->acquired_resource_locks) {
463 jcr->store->NumConcurrentJobs--;
464 jcr->client->NumConcurrentJobs--;
465 jcr->job->NumConcurrentJobs--;
469 * Reschedule the job if necessary and requested
471 if (jcr->job->RescheduleOnError &&
472 jcr->JobStatus != JS_Terminated &&
473 jcr->JobStatus != JS_Canceled &&
474 jcr->job->RescheduleTimes > 0 &&
475 jcr->reschedule_count < jcr->job->RescheduleTimes) {
479 * Reschedule this job by cleaning it up, but
480 * reuse the same JobId if possible.
482 jcr->reschedule_count++;
483 jcr->sched_time = time(NULL) + jcr->job->RescheduleInterval;
484 Dmsg2(2300, "Rescheduled Job %s to re-run in %d seconds.\n", jcr->Job,
485 (int)jcr->job->RescheduleInterval);
486 bstrftime(dt, sizeof(dt), time(NULL));
487 Jmsg(jcr, M_INFO, 0, _("Rescheduled Job %s at %s to re-run in %d seconds.\n"),
488 jcr->Job, dt, (int)jcr->job->RescheduleInterval);
489 dird_free_jcr_pointers(jcr); /* partial cleanup old stuff */
490 jcr->JobStatus = JS_WaitStartTime;
491 jcr->SDJobStatus = 0;
492 if (jcr->JobBytes == 0) {
493 Dmsg1(2300, "Requeue job=%d\n", jcr->JobId);
494 jcr->JobStatus = JS_WaitStartTime;
496 jobq_add(jq, jcr); /* queue the job to run again */
498 free(je); /* free the job entry */
499 continue; /* look for another job to run */
502 * Something was actually backed up, so we cannot reuse
503 * the old JobId or there will be database record
504 * conflicts. We now create a new job, copying the
505 * appropriate fields.
507 JCR *njcr = new_jcr(sizeof(JCR), dird_free_jcr);
508 set_jcr_defaults(njcr, jcr->job);
509 njcr->reschedule_count = jcr->reschedule_count;
510 njcr->JobLevel = jcr->JobLevel;
511 njcr->JobStatus = jcr->JobStatus;
512 copy_storage(njcr, jcr);
513 njcr->messages = jcr->messages;
514 Dmsg0(2300, "Call to run new job\n");
516 run_job(njcr); /* This creates a "new" job */
517 free_jcr(njcr); /* release "new" jcr */
519 Dmsg0(2300, "Back from running new job.\n");
521 /* Clean up and release old jcr */
523 db_close_database(jcr, jcr->db);
526 Dmsg2(2300, "====== Termination job=%d use_cnt=%d\n", jcr->JobId, jcr->use_count);
527 jcr->SDJobStatus = 0;
528 V(jq->mutex); /* release internal lock */
530 free(je); /* release job entry */
531 P(jq->mutex); /* reacquire job queue lock */
534 * If any job in the wait queue can be run,
535 * move it to the ready queue
537 Dmsg0(2300, "Done check ready, now check wait queue.\n");
538 if (!jq->waiting_jobs->empty() && !jq->quit) {
540 je = (jobq_item_t *)jq->waiting_jobs->first();
541 jobq_item_t *re = (jobq_item_t *)jq->running_jobs->first();
543 Priority = re->jcr->JobPriority;
544 Dmsg2(2300, "JobId %d is running. Look for pri=%d\n", re->jcr->JobId, Priority);
546 Priority = je->jcr->JobPriority;
547 Dmsg1(2300, "No job running. Look for Job pri=%d\n", Priority);
550 * Walk down the list of waiting jobs and attempt
551 * to acquire the resources it needs.
554 /* je is current job item on the queue, jn is the next one */
556 jobq_item_t *jn = (jobq_item_t *)jq->waiting_jobs->next(je);
558 Dmsg3(2300, "Examining Job=%d JobPri=%d want Pri=%d\n",
559 jcr->JobId, jcr->JobPriority, Priority);
561 /* Take only jobs of correct Priority */
562 if (jcr->JobPriority != Priority) {
563 set_jcr_job_status(jcr, JS_WaitPriority);
567 if (!acquire_resources(jcr)) {
568 je = jn; /* point to next waiting job */
572 /* Got all locks, now remove it from wait queue and append it
575 jcr->acquired_resource_locks = true;
576 jq->waiting_jobs->remove(je);
577 jq->ready_jobs->append(je);
578 Dmsg1(2300, "moved JobId=%d from wait to ready queue\n", je->jcr->JobId);
579 je = jn; /* Point to next waiting job */
584 Dmsg0(2300, "Done checking wait queue.\n");
586 * If no more ready work and we are asked to quit, then do it
588 if (jq->ready_jobs->empty() && jq->quit) {
590 if (jq->num_workers == 0) {
591 Dmsg0(2300, "Wake up destroy routine\n");
592 /* Wake up destroy routine if he is waiting */
593 pthread_cond_broadcast(&jq->work);
597 Dmsg0(2300, "Check for work request\n");
599 * If no more work requests, and we waited long enough, quit
601 Dmsg2(2300, "timedout=%d read empty=%d\n", timedout,
602 jq->ready_jobs->empty());
603 if (jq->ready_jobs->empty() && timedout) {
604 Dmsg0(2300, "break big loop\n");
609 work = !jq->ready_jobs->empty() || !jq->waiting_jobs->empty();
612 * If a job is waiting on a Resource, don't consume all
613 * the CPU time looping looking for work, and even more
614 * important, release the lock so that a job that has
615 * terminated can give us the resource.
618 bmicrosleep(2, 0); /* pause for 2 seconds */
620 /* Recompute work as something may have changed in last 2 secs */
621 work = !jq->ready_jobs->empty() || !jq->waiting_jobs->empty();
623 Dmsg1(2300, "Loop again. work=%d\n", work);
624 } /* end of big for loop */
626 Dmsg0(200, "unlock mutex\n");
628 Dmsg0(2300, "End jobq_server\n");
633 * See if we can acquire all the necessary resources for the job (JCR)
635 * Returns: true if successful
636 * false if resource failure
638 static bool acquire_resources(JCR *jcr)
640 bool skip_this_jcr = false;
642 if (jcr->JobType == JT_RESTORE || jcr->JobType == JT_VERIFY) {
644 * Let only one Restore/verify job run at a time regardless
645 * of MaxConcurrentJobs.
647 if (jcr->store->NumConcurrentJobs == 0) {
648 jcr->store->NumConcurrentJobs = 1;
650 set_jcr_job_status(jcr, JS_WaitStoreRes);
653 /* We are not doing a Restore or Verify */
654 } else if (jcr->store->NumConcurrentJobs == 0 &&
655 jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {
656 /* Simple case, first job */
657 jcr->store->NumConcurrentJobs = 1;
658 } else if (jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {
659 jcr->store->NumConcurrentJobs++;
661 skip_this_jcr = true;
664 set_jcr_job_status(jcr, JS_WaitStoreRes);
668 if (jcr->client->NumConcurrentJobs < jcr->client->MaxConcurrentJobs) {
669 jcr->client->NumConcurrentJobs++;
671 /* Back out previous locks */
672 jcr->store->NumConcurrentJobs--;
673 set_jcr_job_status(jcr, JS_WaitClientRes);
676 if (jcr->job->NumConcurrentJobs < jcr->job->MaxConcurrentJobs) {
677 jcr->job->NumConcurrentJobs++;
679 /* Back out previous locks */
680 jcr->store->NumConcurrentJobs--;
681 jcr->client->NumConcurrentJobs--;
682 set_jcr_job_status(jcr, JS_WaitJobRes);
685 /* Check actual device availability */