-/*
- * Bacula job queue routines.
- *
- * This code consists of three queues, the waiting_jobs
- * queue, where jobs are initially queued, the ready_jobs
- * queue, where jobs are placed when all the resources are
- * allocated and they can immediately be run, and the
- * running queue where jobs are placed when they are
- * running.
- *
- * Kern Sibbald, July MMIII
- *
- * Version $Id$
- *
- * This code was adapted from the Bacula workq, which was
- * adapted from "Programming with POSIX Threads", by
- * David R. Butenhof
- *
- */
/*
Bacula® - The Network Backup Solution
- Copyright (C) 2003-2007 Free Software Foundation Europe e.V.
+ Copyright (C) 2003-2008 Free Software Foundation Europe e.V.
The main author of Bacula is Kern Sibbald, with contributions from
many others, a complete list can be found in the file AUTHORS.
This program is Free Software; you can redistribute it and/or
modify it under the terms of version two of the GNU General Public
- License as published by the Free Software Foundation plus additions
- that are listed in the file LICENSE.
+ License as published by the Free Software Foundation and included
+ in the file LICENSE.
This program is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
(FSFE), Fiduciary Program, Sumatrastrasse 25, 8006 Zürich,
Switzerland, email:ftf@fsfeurope.org.
*/
+/*
+ * Bacula job queue routines.
+ *
+ * This code consists of three queues, the waiting_jobs
+ * queue, where jobs are initially queued, the ready_jobs
+ * queue, where jobs are placed when all the resources are
+ * allocated and they can immediately be run, and the
+ * running queue where jobs are placed when they are
+ * running.
+ *
+ * Kern Sibbald, July MMIII
+ *
+ * Version $Id$
+ *
+ * This code was adapted from the Bacula workq, which was
+ * adapted from "Programming with POSIX Threads", by
+ * David R. Butenhof
+ *
+ */
#include "bacula.h"
#include "dird.h"
static int start_server(jobq_t *jq);
static bool acquire_resources(JCR *jcr);
-
-
+static bool reschedule_job(JCR *jcr, jobq_t *jq, jobq_item_t *je);
+static void dec_read_store(JCR *jcr);
+static void dec_write_store(JCR *jcr);
/*
* Initialize a job queue
if ((stat = pthread_attr_init(&jq->attr)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_attr_init: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_attr_init: ERR=%s\n"), be.bstrerror(stat));
return stat;
}
if ((stat = pthread_attr_setdetachstate(&jq->attr, PTHREAD_CREATE_DETACHED)) != 0) {
}
if ((stat = pthread_mutex_init(&jq->mutex, NULL)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_init: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_init: ERR=%s\n"), be.bstrerror(stat));
pthread_attr_destroy(&jq->attr);
return stat;
}
if ((stat = pthread_cond_init(&jq->work, NULL)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_init: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_init: ERR=%s\n"), be.bstrerror(stat));
pthread_mutex_destroy(&jq->mutex);
pthread_attr_destroy(&jq->attr);
return stat;
}
if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_lock: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_lock: ERR=%s\n"), be.bstrerror(stat));
return stat;
}
jq->valid = 0; /* prevent any more operations */
if (jq->idle_workers) {
if ((stat = pthread_cond_broadcast(&jq->work)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_broadcast: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_broadcast: ERR=%s\n"), be.bstrerror(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;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_wait: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_wait: ERR=%s\n"), be.bstrerror(stat));
pthread_mutex_unlock(&jq->mutex);
return stat;
}
}
if ((stat = pthread_mutex_unlock(&jq->mutex)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_unlock: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_unlock: ERR=%s\n"), be.bstrerror(stat));
return stat;
}
stat = pthread_mutex_destroy(&jq->mutex);
/* Initialize termination condition variable */
if ((stat = pthread_cond_init(&jcr->term_wait, NULL)) != 0) {
berrno be;
- Jmsg1(jcr, M_FATAL, 0, _("Unable to init job cond variable: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(jcr, M_FATAL, 0, _("Unable to init job cond variable: ERR=%s\n"), be.bstrerror(stat));
return stat;
}
jcr->term_wait_inited = true;
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));
+ Jmsg1(jcr, M_ERROR, 0, _("pthread_thread_create: ERR=%s\n"), be.bstrerror(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));
+ Jmsg1(jcr, M_ERROR, 0, _("pthread_mutex_lock: ERR=%s\n"), be.bstrerror(stat));
free_jcr(jcr); /* release jcr */
return stat;
}
if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_lock: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_lock: ERR=%s\n"), be.bstrerror(stat));
return stat;
}
pthread_t id;
/*
- * if any threads are idle, wake one --
- * actually we do a broadcast because on /lib/tls
+ * 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_broadcast(&jq->work)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_signal: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_signal: ERR=%s\n"), be.bstrerror(stat));
return stat;
}
} else if (jq->num_workers < jq->max_workers) {
set_thread_concurrency(jq->max_workers + 1);
if ((stat = pthread_create(&id, &jq->attr, jobq_server, (void *)jq)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_create: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_create: ERR=%s\n"), be.bstrerror(stat));
return stat;
}
}
Dmsg0(2300, "Start jobq_server\n");
if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) {
berrno be;
- Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_lock: ERR=%s\n"), be.strerror(stat));
+ Jmsg1(NULL, M_ERROR, 0, _("pthread_mutex_lock: ERR=%s\n"), be.bstrerror(stat));
return NULL;
}
jq->num_workers++;
}
}
jq->running_jobs->append(je);
+ set_jcr_in_tsd(jcr);
Dmsg1(2300, "Took jobid=%d from ready and appended to run\n", jcr->JobId);
/* Release job queue lock */
* put into the ready queue.
*/
if (jcr->acquired_resource_locks) {
- if (jcr->rstore) {
- jcr->rstore->NumConcurrentJobs = 0;
- Dmsg1(200, "Dec rncj=%d\n", jcr->rstore->NumConcurrentJobs);
- }
- if (jcr->wstore) {
- jcr->wstore->NumConcurrentJobs--;
- Dmsg1(200, "Dec wncj=%d\n", jcr->wstore->NumConcurrentJobs);
- }
+ dec_read_store(jcr);
+ dec_write_store(jcr);
jcr->client->NumConcurrentJobs--;
jcr->job->NumConcurrentJobs--;
jcr->acquired_resource_locks = false;
}
- /*
- * Reschedule the job if necessary and requested
- */
- if (jcr->job->RescheduleOnError &&
- jcr->JobStatus != JS_Terminated &&
- jcr->JobStatus != JS_Canceled &&
- jcr->job->RescheduleTimes > 0 &&
- jcr->JobType == JT_BACKUP &&
- (jcr->job->RescheduleTimes == 0 ||
- jcr->reschedule_count < jcr->job->RescheduleTimes)) {
- char dt[50], dt2[50];
-
- /*
- * Reschedule this job by cleaning it up, but
- * reuse the same JobId if possible.
- */
- time_t now = time(NULL);
- jcr->reschedule_count++;
- jcr->sched_time = now + jcr->job->RescheduleInterval;
- bstrftime(dt, sizeof(dt), now);
- bstrftime(dt2, sizeof(dt2), jcr->sched_time);
- Dmsg4(2300, "Rescheduled Job %s to re-run in %d seconds.(now=%u,then=%u)\n", jcr->Job,
- (int)jcr->job->RescheduleInterval, now, jcr->sched_time);
- Jmsg(jcr, M_INFO, 0, _("Rescheduled Job %s at %s to re-run in %d seconds (%s).\n"),
- jcr->Job, dt, (int)jcr->job->RescheduleInterval, dt2);
- dird_free_jcr_pointers(jcr); /* partial cleanup old stuff */
- jcr->JobStatus = -1;
- set_jcr_job_status(jcr, JS_WaitStartTime);
- jcr->SDJobStatus = 0;
- if (jcr->JobBytes == 0) {
- Dmsg2(2300, "Requeue job=%d use=%d\n", jcr->JobId, jcr->use_count());
- V(jq->mutex);
- jobq_add(jq, jcr); /* queue the job to run again */
- P(jq->mutex);
- free_jcr(jcr); /* release jcr */
- 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->sched_time = jcr->sched_time;
- njcr->JobLevel = jcr->JobLevel;
- njcr->JobStatus = -1;
- set_jcr_job_status(njcr, jcr->JobStatus);
- if (jcr->rstore) {
- copy_rstorage(njcr, jcr->rstorage, _("previous Job"));
- } else {
- free_rstorage(njcr);
- }
- if (jcr->wstore) {
- copy_wstorage(njcr, jcr->wstorage, _("previous Job"));
- } else {
- free_wstorage(njcr);
- }
- njcr->messages = jcr->messages;
- Dmsg0(2300, "Call to run new job\n");
- V(jq->mutex);
- run_job(njcr); /* This creates a "new" job */
- free_jcr(njcr); /* release "new" jcr */
- P(jq->mutex);
- Dmsg0(2300, "Back from running new job.\n");
+ if (reschedule_job(jcr, jq, je)) {
+ continue; /* go look for more work */
}
+
/* 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 */
return NULL;
}
+/*
+ * Returns true if cleanup done and we should look for more work
+ */
+static bool reschedule_job(JCR *jcr, jobq_t *jq, jobq_item_t *je)
+{
+ /*
+ * Reschedule the job if necessary and requested
+ */
+ if (jcr->job->RescheduleOnError &&
+ jcr->JobStatus != JS_Terminated &&
+ jcr->JobStatus != JS_Canceled &&
+ jcr->JobType == JT_BACKUP &&
+ (jcr->job->RescheduleTimes == 0 ||
+ jcr->reschedule_count < jcr->job->RescheduleTimes)) {
+ char dt[50], dt2[50];
+
+ /*
+ * Reschedule this job by cleaning it up, but
+ * reuse the same JobId if possible.
+ */
+ time_t now = time(NULL);
+ jcr->reschedule_count++;
+ jcr->sched_time = now + jcr->job->RescheduleInterval;
+ bstrftime(dt, sizeof(dt), now);
+ bstrftime(dt2, sizeof(dt2), jcr->sched_time);
+ Dmsg4(2300, "Rescheduled Job %s to re-run in %d seconds.(now=%u,then=%u)\n", jcr->Job,
+ (int)jcr->job->RescheduleInterval, now, jcr->sched_time);
+ Jmsg(jcr, M_INFO, 0, _("Rescheduled Job %s at %s to re-run in %d seconds (%s).\n"),
+ jcr->Job, dt, (int)jcr->job->RescheduleInterval, dt2);
+ dird_free_jcr_pointers(jcr); /* partial cleanup old stuff */
+ jcr->JobStatus = -1;
+ set_jcr_job_status(jcr, JS_WaitStartTime);
+ jcr->SDJobStatus = 0;
+ if (!allow_duplicate_job(jcr)) {
+ return false;
+ }
+ if (jcr->JobBytes == 0) {
+ Dmsg2(2300, "Requeue job=%d use=%d\n", jcr->JobId, jcr->use_count());
+ V(jq->mutex);
+ jobq_add(jq, jcr); /* queue the job to run again */
+ P(jq->mutex);
+ free_jcr(jcr); /* release jcr */
+ free(je); /* free the job entry */
+ return true; /* we already cleaned up */
+ }
+ /*
+ * 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->sched_time = jcr->sched_time;
+ njcr->JobLevel = jcr->JobLevel;
+ njcr->pool = jcr->pool;
+ njcr->run_pool_override = jcr->run_pool_override;
+ njcr->full_pool = jcr->full_pool;
+ njcr->run_full_pool_override = jcr->run_full_pool_override;
+ njcr->inc_pool = jcr->inc_pool;
+ njcr->run_inc_pool_override = jcr->run_inc_pool_override;
+ njcr->diff_pool = jcr->diff_pool;
+ njcr->JobStatus = -1;
+ set_jcr_job_status(njcr, jcr->JobStatus);
+ if (jcr->rstore) {
+ copy_rstorage(njcr, jcr->rstorage, _("previous Job"));
+ } else {
+ free_rstorage(njcr);
+ }
+ if (jcr->wstore) {
+ copy_wstorage(njcr, jcr->wstorage, _("previous Job"));
+ } else {
+ free_wstorage(njcr);
+ }
+ njcr->messages = jcr->messages;
+ njcr->spool_data = jcr->spool_data;
+ njcr->write_part_after_job = jcr->write_part_after_job;
+ Dmsg0(2300, "Call to run new job\n");
+ V(jq->mutex);
+ run_job(njcr); /* This creates a "new" job */
+ free_jcr(njcr); /* release "new" jcr */
+ P(jq->mutex);
+ Dmsg0(2300, "Back from running new job.\n");
+ }
+ return false;
+}
+
/*
* See if we can acquire all the necessary resources for the job (JCR)
*
bool skip_this_jcr = false;
jcr->acquired_resource_locks = false;
+ if (jcr->rstore == jcr->wstore) { /* deadlock */
+ Jmsg(jcr, M_FATAL, 0, _("Job canceled. Attempt to read and write same device.\n"
+ " Read storage \"%s\" (From %s) -- Write storage \"%s\" (From %s)\n"),
+ jcr->rstore->name(), jcr->rstore_source, jcr->wstore->name(), jcr->wstore_source);
+ set_jcr_job_status(jcr, JS_Canceled);
+ return false;
+ }
if (jcr->rstore) {
Dmsg1(200, "Rstore=%s\n", jcr->rstore->name());
- /*
- * Let only one Restore/verify job run at a time regardless
- * of MaxConcurrentJobs.
- */
- if (jcr->rstore->NumConcurrentJobs == 0) {
- jcr->rstore->NumConcurrentJobs = 1;
- Dmsg0(200, "Set rncj=1\n");
+ if (jcr->rstore->NumConcurrentJobs < jcr->rstore->MaxConcurrentJobs) {
+ jcr->rstore->NumConcurrentReadJobs++;
+ jcr->rstore->NumConcurrentJobs++;
+ Dmsg1(200, "Inc rncj=%d\n", jcr->rstore->NumConcurrentJobs);
} else {
Dmsg1(200, "Fail rncj=%d\n", jcr->rstore->NumConcurrentJobs);
set_jcr_job_status(jcr, JS_WaitStoreRes);
if (jcr->wstore) {
Dmsg1(200, "Wstore=%s\n", jcr->wstore->name());
- if (jcr->rstore == jcr->wstore) { /* deadlock */
- jcr->rstore->NumConcurrentJobs = 0; /* back out rstore */
- Jmsg(jcr, M_FATAL, 0, _("Job canceled. Attempt to read and write same device.\n"
- " Read storage \"%s\" (From %s) -- Write storage \"%s\" (From %s)\n"),
- jcr->rstore->name(), jcr->rstore_source, jcr->wstore->name(), jcr->wstore_source);
- set_jcr_job_status(jcr, JS_Canceled);
- return false;
- }
- if (jcr->wstore->NumConcurrentJobs == 0 &&
- jcr->wstore->NumConcurrentJobs < jcr->wstore->MaxConcurrentJobs) {
- /* Simple case, first job */
- jcr->wstore->NumConcurrentJobs = 1;
- Dmsg0(200, "Set wncj=1\n");
- } else if (jcr->wstore->NumConcurrentJobs < jcr->wstore->MaxConcurrentJobs) {
+ if (jcr->wstore->NumConcurrentJobs < jcr->wstore->MaxConcurrentJobs) {
jcr->wstore->NumConcurrentJobs++;
Dmsg1(200, "Inc wncj=%d\n", jcr->wstore->NumConcurrentJobs);
} else if (jcr->rstore) {
- jcr->rstore->NumConcurrentJobs = 0; /* back out rstore */
- Dmsg1(200, "Fail wncj=%d\n", jcr->wstore->NumConcurrentJobs);
+ dec_read_store(jcr);
skip_this_jcr = true;
} else {
Dmsg1(200, "Fail wncj=%d\n", jcr->wstore->NumConcurrentJobs);
jcr->client->NumConcurrentJobs++;
} else {
/* Back out previous locks */
- if (jcr->wstore) {
- jcr->wstore->NumConcurrentJobs--;
- Dmsg1(200, "Dec wncj=%d\n", jcr->wstore->NumConcurrentJobs);
- }
- if (jcr->rstore) {
- jcr->rstore->NumConcurrentJobs = 0;
- Dmsg1(200, "Dec rncj=%d\n", jcr->rstore->NumConcurrentJobs);
- }
+ dec_write_store(jcr);
+ dec_read_store(jcr);
set_jcr_job_status(jcr, JS_WaitClientRes);
return false;
}
jcr->job->NumConcurrentJobs++;
} else {
/* Back out previous locks */
- if (jcr->wstore) {
- jcr->wstore->NumConcurrentJobs--;
- Dmsg1(200, "Dec wncj=%d\n", jcr->wstore->NumConcurrentJobs);
- }
- if (jcr->rstore) {
- jcr->rstore->NumConcurrentJobs = 0;
- Dmsg1(200, "Dec rncj=%d\n", jcr->rstore->NumConcurrentJobs);
- }
+ dec_write_store(jcr);
+ dec_read_store(jcr);
jcr->client->NumConcurrentJobs--;
set_jcr_job_status(jcr, JS_WaitJobRes);
return false;
jcr->acquired_resource_locks = true;
return true;
}
+
+static void dec_read_store(JCR *jcr)
+{
+ if (jcr->rstore) {
+ jcr->rstore->NumConcurrentReadJobs--; /* back out rstore */
+ jcr->rstore->NumConcurrentJobs--; /* back out rstore */
+ Dmsg1(200, "Dec rncj=%d\n", jcr->rstore->NumConcurrentJobs);
+ ASSERT(jcr->rstore->NumConcurrentReadJobs >= 0);
+ ASSERT(jcr->rstore->NumConcurrentJobs >= 0);
+ }
+}
+
+static void dec_write_store(JCR *jcr)
+{
+ if (jcr->wstore) {
+ jcr->wstore->NumConcurrentJobs--;
+ Dmsg1(200, "Dec wncj=%d\n", jcr->wstore->NumConcurrentJobs);
+ ASSERT(jcr->wstore->NumConcurrentJobs >= 0);
+ }
+}
projects / bacula / bacula / blobdiff