* Initialize a job queue
*
* Returns: 0 on success
- * errno on failure
+ * errno on failure
*/
int jobq_init(jobq_t *jq, int threads, void *(*engine)(void *arg))
{
return stat;
}
jq->quit = false;
- jq->max_workers = threads; /* max threads to create */
- jq->num_workers = 0; /* no threads yet */
- jq->idle_workers = 0; /* no idle threads */
- jq->engine = engine; /* routine to run */
+ jq->max_workers = threads; /* max threads to create */
+ jq->num_workers = 0; /* no threads yet */
+ jq->idle_workers = 0; /* no idle threads */
+ jq->engine = engine; /* routine to run */
jq->valid = JOBQ_VALID;
/* Initialize the job queues */
jq->waiting_jobs = New(dlist(item, &item->link));
* Destroy the job queue
*
* Returns: 0 on success
- * errno on failure
+ * errno on failure
*/
int jobq_destroy(jobq_t *jq)
{
Jmsg1(NULL, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
return stat;
}
- jq->valid = 0; /* prevent any more operations */
+ jq->valid = 0; /* prevent any more operations */
/*
* If any threads are active, wake them
if (jq->num_workers > 0) {
jq->quit = true;
if (jq->idle_workers) {
- if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
- berrno be;
+ if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
+ berrno be;
Jmsg1(NULL, M_ERROR, 0, "pthread_cond_broadcast: ERR=%s\n", be.strerror(stat));
- pthread_mutex_unlock(&jq->mutex);
- return stat;
- }
+ pthread_mutex_unlock(&jq->mutex);
+ return stat;
+ }
}
while (jq->num_workers > 0) {
- if ((stat = pthread_cond_wait(&jq->work, &jq->mutex)) != 0) {
- berrno be;
+ if ((stat = pthread_cond_wait(&jq->work, &jq->mutex)) != 0) {
+ berrno be;
Jmsg1(NULL, M_ERROR, 0, "pthread_cond_wait: ERR=%s\n", be.strerror(stat));
- pthread_mutex_unlock(&jq->mutex);
- return stat;
- }
+ pthread_mutex_unlock(&jq->mutex);
+ return stat;
+ }
}
}
if ((stat = pthread_mutex_unlock(&jq->mutex)) != 0) {
/* Wait until scheduled time arrives */
if (wtime > 0) {
Jmsg(jcr, M_INFO, 0, _("Job %s waiting %d seconds for scheduled start time.\n"),
- jcr->Job, wtime);
+ jcr->Job, wtime);
}
/* Check every 30 seconds if canceled */
while (wtime > 0) {
Dmsg2(2300, "Waiting on sched time, jobid=%d secs=%d\n", jcr->JobId, wtime);
if (wtime > 30) {
- wtime = 30;
+ wtime = 30;
}
bmicrosleep(wtime, 0);
if (job_canceled(jcr)) {
- break;
+ break;
}
wtime = jcr->sched_time - time(NULL);
}
- P(jcr->mutex); /* lock jcr */
+ P(jcr->mutex); /* lock jcr */
jobq_add(jq, jcr);
V(jcr->mutex);
- free_jcr(jcr); /* we are done with jcr */
+ free_jcr(jcr); /* we are done with jcr */
Dmsg0(2300, "Exit sched_wait\n");
return NULL;
}
return EINVAL;
}
- jcr->use_count++; /* mark jcr in use by us */
+ jcr->use_count++; /* mark jcr in use by us */
Dmsg3(2300, "jobq_add jobid=%d jcr=0x%x use_count=%d\n", jcr->JobId, jcr, jcr->use_count);
if (!job_canceled(jcr) && wtime > 0) {
set_thread_concurrency(jq->max_workers + 2);
sched_pkt = (wait_pkt *)malloc(sizeof(wait_pkt));
sched_pkt->jcr = jcr;
sched_pkt->jq = jq;
-// jcr->use_count--; /* release our use of jcr */
- stat = pthread_create(&id, &jq->attr, sched_wait, (void *)sched_pkt);
- if (stat != 0) { /* thread not created */
- berrno be;
+ stat = pthread_create(&id, &jq->attr, sched_wait, (void *)sched_pkt);
+ if (stat != 0) { /* thread not created */
+ berrno be;
Jmsg1(jcr, M_ERROR, 0, "pthread_thread_create: ERR=%s\n", be.strerror(stat));
}
return stat;
if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
berrno be;
Jmsg1(jcr, M_ERROR, 0, "pthread_mutex_lock: ERR=%s\n", be.strerror(stat));
- jcr->use_count--; /* release jcr */
+ jcr->use_count--; /* release jcr */
return stat;
}
if ((item = (jobq_item_t *)malloc(sizeof(jobq_item_t))) == NULL) {
- jcr->use_count--; /* release jcr */
+ jcr->use_count--; /* release jcr */
return ENOMEM;
}
item->jcr = jcr;
/* Add this job to the wait queue in priority sorted order */
foreach_dlist(li, jq->waiting_jobs) {
Dmsg2(2300, "waiting item jobid=%d priority=%d\n",
- li->jcr->JobId, li->jcr->JobPriority);
- if (li->jcr->JobPriority > jcr->JobPriority) {
- jq->waiting_jobs->insert_before(item, li);
+ li->jcr->JobId, li->jcr->JobPriority);
+ if (li->jcr->JobPriority > jcr->JobPriority) {
+ jq->waiting_jobs->insert_before(item, li);
Dmsg2(2300, "insert_before jobid=%d before waiting job=%d\n",
- li->jcr->JobId, jcr->JobId);
- inserted = true;
- break;
- }
+ li->jcr->JobId, jcr->JobId);
+ inserted = true;
+ break;
+ }
}
/* If not jobs in wait queue, append it */
if (!inserted) {
- jq->waiting_jobs->append(item);
+ jq->waiting_jobs->append(item);
Dmsg1(2300, "Appended item jobid=%d to waiting queue\n", jcr->JobId);
}
}
foreach_dlist(item, jq->waiting_jobs) {
if (jcr == item->jcr) {
- found = true;
- break;
+ found = true;
+ break;
}
}
if (!found) {
int stat = 0;
pthread_t id;
- /* if any threads are idle, wake one */
+ /*
+ * if any threads are idle, wake one --
+ * actually we do a broadcast because on /lib/tls
+ * these signals seem to get lost from time to time.
+ */
if (jq->idle_workers > 0) {
Dmsg0(2300, "Signal worker to wake up\n");
- if ((stat = pthread_cond_signal(&jq->work)) != 0) {
- berrno be;
+ if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
+ berrno be;
Jmsg1(NULL, M_ERROR, 0, "pthread_cond_signal: ERR=%s\n", be.strerror(stat));
- return stat;
+ return stat;
}
} else if (jq->num_workers < jq->max_workers) {
Dmsg0(2300, "Create worker thread\n");
/* No idle threads so create a new one */
set_thread_concurrency(jq->max_workers + 1);
if ((stat = pthread_create(&id, &jq->attr, jobq_server, (void *)jq)) != 0) {
- berrno be;
+ berrno be;
Jmsg1(NULL, M_ERROR, 0, "pthread_create: ERR=%s\n", be.strerror(stat));
- return stat;
+ return stat;
}
}
return stat;
{
struct timespec timeout;
jobq_t *jq = (jobq_t *)arg;
- jobq_item_t *je; /* job entry in queue */
+ jobq_item_t *je; /* job entry in queue */
int stat;
bool timedout = false;
bool work = true;
Dmsg0(2300, "Top of for loop\n");
if (!work && !jq->quit) {
- gettimeofday(&tv, &tz);
- timeout.tv_nsec = 0;
- timeout.tv_sec = tv.tv_sec + 4;
-
- while (!jq->quit) {
- /*
- * Wait 4 seconds, then if no more work, exit
- */
+ gettimeofday(&tv, &tz);
+ timeout.tv_nsec = 0;
+ timeout.tv_sec = tv.tv_sec + 4;
+
+ while (!jq->quit) {
+ /*
+ * Wait 4 seconds, then if no more work, exit
+ */
Dmsg0(2300, "pthread_cond_timedwait()\n");
- stat = pthread_cond_timedwait(&jq->work, &jq->mutex, &timeout);
- if (stat == ETIMEDOUT) {
+ stat = pthread_cond_timedwait(&jq->work, &jq->mutex, &timeout);
+ if (stat == ETIMEDOUT) {
Dmsg0(2300, "timedwait timedout.\n");
- timedout = true;
- break;
- } else if (stat != 0) {
+ timedout = true;
+ break;
+ } else if (stat != 0) {
/* This shouldn't happen */
Dmsg0(2300, "This shouldn't happen\n");
- jq->num_workers--;
- pthread_mutex_unlock(&jq->mutex);
- return NULL;
- }
- break;
- }
+ jq->num_workers--;
+ pthread_mutex_unlock(&jq->mutex);
+ return NULL;
+ }
+ break;
+ }
}
/*
* If anything is in the ready queue, run it
*/
Dmsg0(2300, "Checking ready queue.\n");
while (!jq->ready_jobs->empty() && !jq->quit) {
- JCR *jcr;
- je = (jobq_item_t *)jq->ready_jobs->first();
- jcr = je->jcr;
- jq->ready_jobs->remove(je);
- if (!jq->ready_jobs->empty()) {
+ JCR *jcr;
+ je = (jobq_item_t *)jq->ready_jobs->first();
+ jcr = je->jcr;
+ jq->ready_jobs->remove(je);
+ if (!jq->ready_jobs->empty()) {
Dmsg0(2300, "ready queue not empty start server\n");
- if (start_server(jq) != 0) {
- jq->num_workers--;
- pthread_mutex_unlock(&jq->mutex);
- return NULL;
- }
- }
- jq->running_jobs->append(je);
+ if (start_server(jq) != 0) {
+ jq->num_workers--;
+ pthread_mutex_unlock(&jq->mutex);
+ return NULL;
+ }
+ }
+ jq->running_jobs->append(je);
Dmsg1(2300, "Took jobid=%d from ready and appended to run\n", jcr->JobId);
- /* Release job queue lock */
- V(jq->mutex);
+ /* Release job queue lock */
+ V(jq->mutex);
/* Call user's routine here */
Dmsg1(2300, "Calling user engine for jobid=%d\n", jcr->JobId);
- jq->engine(je->jcr);
+ jq->engine(je->jcr);
Dmsg1(2300, "Back from user engine jobid=%d.\n", jcr->JobId);
- /* Reacquire job queue lock */
- P(jq->mutex);
+ /* Reacquire job queue lock */
+ P(jq->mutex);
Dmsg0(200, "Done lock mutex after running job. Release locks.\n");
- jq->running_jobs->remove(je);
- /*
- * Release locks if acquired. Note, they will not have
- * been acquired for jobs canceled before they were
- * put into the ready queue.
- */
- if (jcr->acquired_resource_locks) {
- jcr->store->NumConcurrentJobs--;
- jcr->client->NumConcurrentJobs--;
- jcr->job->NumConcurrentJobs--;
- }
-
- /*
- * Reschedule the job if necessary and requested
- */
- if (jcr->job->RescheduleOnError &&
- jcr->JobStatus != JS_Terminated &&
- jcr->JobStatus != JS_Canceled &&
- jcr->job->RescheduleTimes > 0 &&
- jcr->reschedule_count < jcr->job->RescheduleTimes) {
- char dt[50];
-
- /*
- * Reschedule this job by cleaning it up, but
- * reuse the same JobId if possible.
- */
- jcr->reschedule_count++;
- jcr->sched_time = time(NULL) + jcr->job->RescheduleInterval;
+ jq->running_jobs->remove(je);
+ /*
+ * Release locks if acquired. Note, they will not have
+ * been acquired for jobs canceled before they were
+ * put into the ready queue.
+ */
+ if (jcr->acquired_resource_locks) {
+ jcr->store->NumConcurrentJobs--;
+ jcr->client->NumConcurrentJobs--;
+ jcr->job->NumConcurrentJobs--;
+ }
+
+ /*
+ * Reschedule the job if necessary and requested
+ */
+ if (jcr->job->RescheduleOnError &&
+ jcr->JobStatus != JS_Terminated &&
+ jcr->JobStatus != JS_Canceled &&
+ jcr->job->RescheduleTimes > 0 &&
+ jcr->reschedule_count < jcr->job->RescheduleTimes) {
+ char dt[50];
+
+ /*
+ * Reschedule this job by cleaning it up, but
+ * reuse the same JobId if possible.
+ */
+ jcr->reschedule_count++;
+ jcr->sched_time = time(NULL) + jcr->job->RescheduleInterval;
Dmsg2(2300, "Rescheduled Job %s to re-run in %d seconds.\n", jcr->Job,
- (int)jcr->job->RescheduleInterval);
- bstrftime(dt, sizeof(dt), time(NULL));
+ (int)jcr->job->RescheduleInterval);
+ bstrftime(dt, sizeof(dt), time(NULL));
Jmsg(jcr, M_INFO, 0, _("Rescheduled Job %s at %s to re-run in %d seconds.\n"),
- jcr->Job, dt, (int)jcr->job->RescheduleInterval);
- dird_free_jcr(jcr); /* partial cleanup old stuff */
- jcr->JobStatus = JS_WaitStartTime;
- jcr->SDJobStatus = 0;
- if (jcr->JobBytes == 0) {
+ jcr->Job, dt, (int)jcr->job->RescheduleInterval);
+ dird_free_jcr_pointers(jcr); /* partial cleanup old stuff */
+ jcr->JobStatus = JS_WaitStartTime;
+ jcr->SDJobStatus = 0;
+ if (jcr->JobBytes == 0) {
Dmsg1(2300, "Requeue job=%d\n", jcr->JobId);
- jcr->JobStatus = JS_WaitStartTime;
- V(jq->mutex);
- jobq_add(jq, jcr); /* queue the job to run again */
- P(jq->mutex);
- free(je); /* free the job entry */
- continue; /* look for another job to run */
- }
- /*
- * Something was actually backed up, so we cannot reuse
- * the old JobId or there will be database record
- * conflicts. We now create a new job, copying the
- * appropriate fields.
- */
- JCR *njcr = new_jcr(sizeof(JCR), dird_free_jcr);
- set_jcr_defaults(njcr, jcr->job);
- njcr->reschedule_count = jcr->reschedule_count;
- njcr->JobLevel = jcr->JobLevel;
- njcr->JobStatus = jcr->JobStatus;
- copy_storage(njcr, jcr);
- njcr->messages = jcr->messages;
+ jcr->JobStatus = JS_WaitStartTime;
+ V(jq->mutex);
+ jobq_add(jq, jcr); /* queue the job to run again */
+ P(jq->mutex);
+ free(je); /* free the job entry */
+ continue; /* look for another job to run */
+ }
+ /*
+ * Something was actually backed up, so we cannot reuse
+ * the old JobId or there will be database record
+ * conflicts. We now create a new job, copying the
+ * appropriate fields.
+ */
+ JCR *njcr = new_jcr(sizeof(JCR), dird_free_jcr);
+ set_jcr_defaults(njcr, jcr->job);
+ njcr->reschedule_count = jcr->reschedule_count;
+ njcr->JobLevel = jcr->JobLevel;
+ njcr->JobStatus = jcr->JobStatus;
+ copy_storage(njcr, jcr);
+ njcr->messages = jcr->messages;
Dmsg0(2300, "Call to run new job\n");
- V(jq->mutex);
+ V(jq->mutex);
run_job(njcr); /* This creates a "new" job */
free_jcr(njcr); /* release "new" jcr */
- P(jq->mutex);
+ P(jq->mutex);
Dmsg0(2300, "Back from running new job.\n");
- }
- /* Clean up and release old jcr */
- if (jcr->db) {
- db_close_database(jcr, jcr->db);
- jcr->db = NULL;
- }
+ }
+ /* Clean up and release old jcr */
+ if (jcr->db) {
+ db_close_database(jcr, jcr->db);
+ jcr->db = NULL;
+ }
Dmsg2(2300, "====== Termination job=%d use_cnt=%d\n", jcr->JobId, jcr->use_count);
- jcr->SDJobStatus = 0;
- V(jq->mutex); /* release internal lock */
- free_jcr(jcr);
- free(je); /* release job entry */
- P(jq->mutex); /* reacquire job queue lock */
+ jcr->SDJobStatus = 0;
+ V(jq->mutex); /* release internal lock */
+ free_jcr(jcr);
+ free(je); /* release job entry */
+ P(jq->mutex); /* reacquire job queue lock */
}
/*
* If any job in the wait queue can be run,
*/
Dmsg0(2300, "Done check ready, now check wait queue.\n");
if (!jq->waiting_jobs->empty() && !jq->quit) {
- int Priority;
- je = (jobq_item_t *)jq->waiting_jobs->first();
- jobq_item_t *re = (jobq_item_t *)jq->running_jobs->first();
- if (re) {
- Priority = re->jcr->JobPriority;
+ int Priority;
+ je = (jobq_item_t *)jq->waiting_jobs->first();
+ jobq_item_t *re = (jobq_item_t *)jq->running_jobs->first();
+ if (re) {
+ Priority = re->jcr->JobPriority;
Dmsg2(2300, "JobId %d is running. Look for pri=%d\n", re->jcr->JobId, Priority);
- } else {
- Priority = je->jcr->JobPriority;
+ } else {
+ Priority = je->jcr->JobPriority;
Dmsg1(2300, "No job running. Look for Job pri=%d\n", Priority);
- }
- /*
- * Walk down the list of waiting jobs and attempt
- * to acquire the resources it needs.
- */
- for ( ; je; ) {
- /* je is current job item on the queue, jn is the next one */
- JCR *jcr = je->jcr;
- jobq_item_t *jn = (jobq_item_t *)jq->waiting_jobs->next(je);
+ }
+ /*
+ * Walk down the list of waiting jobs and attempt
+ * to acquire the resources it needs.
+ */
+ for ( ; je; ) {
+ /* je is current job item on the queue, jn is the next one */
+ JCR *jcr = je->jcr;
+ jobq_item_t *jn = (jobq_item_t *)jq->waiting_jobs->next(je);
Dmsg3(2300, "Examining Job=%d JobPri=%d want Pri=%d\n",
- jcr->JobId, jcr->JobPriority, Priority);
-
- /* Take only jobs of correct Priority */
- if (jcr->JobPriority != Priority) {
- set_jcr_job_status(jcr, JS_WaitPriority);
- break;
- }
-
- if (!acquire_resources(jcr)) {
- je = jn; /* point to next waiting job */
- continue;
- }
-
- /* Got all locks, now remove it from wait queue and append it
- * to the ready queue
- */
- jcr->acquired_resource_locks = true;
- jq->waiting_jobs->remove(je);
- jq->ready_jobs->append(je);
+ jcr->JobId, jcr->JobPriority, Priority);
+
+ /* Take only jobs of correct Priority */
+ if (jcr->JobPriority != Priority) {
+ set_jcr_job_status(jcr, JS_WaitPriority);
+ break;
+ }
+
+ if (!acquire_resources(jcr)) {
+ je = jn; /* point to next waiting job */
+ continue;
+ }
+
+ /* Got all locks, now remove it from wait queue and append it
+ * to the ready queue
+ */
+ jcr->acquired_resource_locks = true;
+ jq->waiting_jobs->remove(je);
+ jq->ready_jobs->append(je);
Dmsg1(2300, "moved JobId=%d from wait to ready queue\n", je->jcr->JobId);
- je = jn; /* Point to next waiting job */
- } /* end for loop */
+ je = jn; /* Point to next waiting job */
+ } /* end for loop */
} /* end if */
* If no more ready work and we are asked to quit, then do it
*/
if (jq->ready_jobs->empty() && jq->quit) {
- jq->num_workers--;
- if (jq->num_workers == 0) {
+ jq->num_workers--;
+ if (jq->num_workers == 0) {
Dmsg0(2300, "Wake up destroy routine\n");
- /* Wake up destroy routine if he is waiting */
- pthread_cond_broadcast(&jq->work);
- }
- break;
+ /* Wake up destroy routine if he is waiting */
+ pthread_cond_broadcast(&jq->work);
+ }
+ break;
}
Dmsg0(2300, "Check for work request\n");
/*
* If no more work requests, and we waited long enough, quit
*/
Dmsg2(2300, "timedout=%d read empty=%d\n", timedout,
- jq->ready_jobs->empty());
+ jq->ready_jobs->empty());
if (jq->ready_jobs->empty() && timedout) {
Dmsg0(2300, "break big loop\n");
- jq->num_workers--;
- break;
+ jq->num_workers--;
+ break;
}
work = !jq->ready_jobs->empty() || !jq->waiting_jobs->empty();
if (work) {
- /*
+ /*
* If a job is waiting on a Resource, don't consume all
- * the CPU time looping looking for work, and even more
- * important, release the lock so that a job that has
- * terminated can give us the resource.
- */
- V(jq->mutex);
- bmicrosleep(2, 0); /* pause for 2 seconds */
- P(jq->mutex);
- /* Recompute work as something may have changed in last 2 secs */
- work = !jq->ready_jobs->empty() || !jq->waiting_jobs->empty();
+ * the CPU time looping looking for work, and even more
+ * important, release the lock so that a job that has
+ * terminated can give us the resource.
+ */
+ V(jq->mutex);
+ bmicrosleep(2, 0); /* pause for 2 seconds */
+ P(jq->mutex);
+ /* Recompute work as something may have changed in last 2 secs */
+ work = !jq->ready_jobs->empty() || !jq->waiting_jobs->empty();
}
Dmsg1(2300, "Loop again. work=%d\n", work);
} /* end of big for loop */
* See if we can acquire all the necessary resources for the job (JCR)
*
* Returns: true if successful
- * false if resource failure
+ * false if resource failure
*/
static bool acquire_resources(JCR *jcr)
{
* of MaxConcurrentJobs.
*/
if (jcr->store->NumConcurrentJobs == 0) {
- jcr->store->NumConcurrentJobs = 1;
+ jcr->store->NumConcurrentJobs = 1;
} else {
- set_jcr_job_status(jcr, JS_WaitStoreRes);
- return false;
+ set_jcr_job_status(jcr, JS_WaitStoreRes);
+ return false;
}
/* We are not doing a Restore or Verify */
} else if (jcr->store->NumConcurrentJobs == 0 &&
- jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {
+ jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {
/* Simple case, first job */
jcr->store->NumConcurrentJobs = 1;
} else if (jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) {