X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=bacula%2Fsrc%2Fdird%2Fjobq.c;h=4ba28a5f7921028c29c730c5330107f20424a681;hb=10cfd798ced2d27f61ead2de6fe9b1bcc8e3468d;hp=eb4640efcd7e9c9ff55a19cc01c8316ad4fb13e6;hpb=823dfe14f07d783ade37293551900a64465ad0a9;p=bacula%2Fbacula diff --git a/bacula/src/dird/jobq.c b/bacula/src/dird/jobq.c old mode 100755 new mode 100644 index eb4640efcd..4ba28a5f79 --- a/bacula/src/dird/jobq.c +++ b/bacula/src/dird/jobq.c @@ -1,3 +1,21 @@ +/* + Bacula(R) - The Network Backup Solution + + Copyright (C) 2000-2017 Kern Sibbald + + The original author of Bacula is Kern Sibbald, with contributions + from many others, a complete list can be found in the file AUTHORS. + + You may use this file and others of this release according to the + license defined in the LICENSE file, which includes the Affero General + Public License, v3.0 ("AGPLv3") and some additional permissions and + terms pursuant to its AGPLv3 Section 7. + + This notice must be preserved when any source code is + conveyed and/or propagated. + + Bacula(R) is a registered trademark of Kern Sibbald. +*/ /* * Bacula job queue routines. * @@ -10,53 +28,41 @@ * * 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 * */ -/* - Copyright (C) 2000-2003 Kern Sibbald and John Walker - - This program is free software; you can redistribute it and/or - modify it under the terms of the GNU General Public License as - published by the Free Software Foundation; either version 2 of - the License, or (at your option) any later version. - - This program is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public - License along with this program; if not, write to the Free - Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, - MA 02111-1307, USA. - - */ #include "bacula.h" #include "dird.h" +extern JCR *jobs; /* Forward referenced functions */ -static void *jobq_server(void *arg); -static int start_server(jobq_t *jq); +extern "C" void *jobq_server(void *arg); +extern "C" void *sched_wait(void *arg); + +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_write_store(JCR *jcr); -/* +/* * 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)) { int stat; jobq_item_t *item = NULL; - + if ((stat = pthread_attr_init(&jq->attr)) != 0) { + berrno be; + 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) { @@ -64,24 +70,28 @@ int jobq_init(jobq_t *jq, int threads, void *(*engine)(void *arg)) return stat; } if ((stat = pthread_mutex_init(&jq->mutex, NULL)) != 0) { + berrno be; + 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.bstrerror(stat)); pthread_mutex_destroy(&jq->mutex); pthread_attr_destroy(&jq->attr); 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->valid = JOBQ_VALID; + 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); - jq->running_jobs = new dlist(item, &item->link); - jq->ready_jobs = new dlist(item, &item->link); + jq->waiting_jobs = New(dlist(item, &item->link)); + jq->running_jobs = New(dlist(item, &item->link)); + jq->ready_jobs = New(dlist(item, &item->link)); return 0; } @@ -89,48 +99,48 @@ int jobq_init(jobq_t *jq, int threads, void *(*engine)(void *arg)) * Destroy the job queue * * Returns: 0 on success - * errno on failure + * errno on failure */ int jobq_destroy(jobq_t *jq) { int stat, stat1, stat2; - if (jq->valid != JOBQ_VALID) { - return EINVAL; - } - if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) { - return stat; - } - 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) { - pthread_mutex_unlock(&jq->mutex); - return stat; - } - } - while (jq->num_workers > 0) { - if ((stat = pthread_cond_wait(&jq->work, &jq->mutex)) != 0) { - pthread_mutex_unlock(&jq->mutex); - return stat; - } - } - } - if ((stat = pthread_mutex_unlock(&jq->mutex)) != 0) { - return stat; - } - stat = pthread_mutex_destroy(&jq->mutex); - stat1 = pthread_cond_destroy(&jq->work); - stat2 = pthread_attr_destroy(&jq->attr); - delete jq->waiting_jobs; - delete jq->running_jobs; - delete jq->ready_jobs; - return (stat != 0 ? stat : (stat1 != 0 ? stat1 : stat2)); + if (jq->valid != JOBQ_VALID) { + return EINVAL; + } + P(jq->mutex); + 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; + Jmsg1(NULL, M_ERROR, 0, _("pthread_cond_broadcast: ERR=%s\n"), be.bstrerror(stat)); + V(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.bstrerror(stat)); + V(jq->mutex); + return stat; + } + } + } + V(jq->mutex); + stat = pthread_mutex_destroy(&jq->mutex); + stat1 = pthread_cond_destroy(&jq->work); + stat2 = pthread_attr_destroy(&jq->attr); + delete jq->waiting_jobs; + delete jq->running_jobs; + delete jq->ready_jobs; + return (stat != 0 ? stat : (stat1 != 0 ? stat1 : stat2)); } struct wait_pkt { @@ -143,44 +153,73 @@ struct wait_pkt { * this routine is only used for jobs started from the console * for which the user explicitly specified a start time. Otherwise * most jobs are put into the job queue only when their - * scheduled time arives. + * scheduled time arrives. */ -static void *sched_wait(void *arg) +extern "C" +void *sched_wait(void *arg) { JCR *jcr = ((wait_pkt *)arg)->jcr; jobq_t *jq = ((wait_pkt *)arg)->jq; - Dmsg0(100, "Enter sched_wait.\n"); + set_jcr_in_tsd(INVALID_JCR); + Dmsg0(2300, "Enter sched_wait.\n"); free(arg); time_t wtime = jcr->sched_time - time(NULL); + jcr->setJobStatus(JS_WaitStartTime); /* Wait until scheduled time arrives */ - if (wtime > 0 && verbose) { - Jmsg(jcr, M_INFO, 0, _("Job %s waiting %d seconds for scheduled start time.\n"), - jcr->Job, wtime); - set_jcr_job_status(jcr, JS_WaitStartTime); + if (wtime > 0) { + Jmsg(jcr, M_INFO, 0, _("Job %s waiting %d seconds for scheduled start time.\n"), + jcr->Job, wtime); } - /* Check every 30 seconds if canceled */ + /* Check every 30 seconds if canceled */ while (wtime > 0) { - Dmsg2(100, "Waiting on sched time, jobid=%d secs=%d\n", jcr->JobId, wtime); + Dmsg3(2300, "Waiting on sched time, jobid=%d secs=%d use=%d\n", + jcr->JobId, wtime, jcr->use_count()); if (wtime > 30) { - wtime = 30; + wtime = 30; } bmicrosleep(wtime, 0); if (job_canceled(jcr)) { - break; + break; } wtime = jcr->sched_time - time(NULL); } + Dmsg1(200, "resched use=%d\n", jcr->use_count()); jobq_add(jq, jcr); - Dmsg0(100, "Exit sched_wait\n"); + free_jcr(jcr); /* we are done with jcr */ + Dmsg0(2300, "Exit sched_wait\n"); return NULL; } +/* Procedure to update the client->NumConcurrentJobs */ +static void update_client_numconcurrentjobs(JCR *jcr, int val) +{ + int num; + if (!jcr->client) { + return; + } + + switch (jcr->getJobType()) + { + case JT_MIGRATE: + case JT_COPY: + case JT_ADMIN: + break; + case JT_BACKUP: + /* Fall through wanted */ + default: + if (jcr->no_client_used() || jcr->wasVirtualFull) { + break; + } + num = jcr->client->getNumConcurrentJobs(); + jcr->client->setNumConcurrentJobs(num + val); + break; + } +} /* * Add a job to the queue * jq is a queue that was created with jobq_init - * */ int jobq_add(jobq_t *jq, JCR *jcr) { @@ -190,105 +229,122 @@ int jobq_add(jobq_t *jq, JCR *jcr) time_t wtime = jcr->sched_time - time(NULL); pthread_t id; wait_pkt *sched_pkt; - - - Dmsg1(100, "jobq_add jobid=%d\n", jcr->JobId); + + if (!jcr->term_wait_inited) { + /* 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.bstrerror(stat)); + return stat; + } + jcr->term_wait_inited = true; + } + + Dmsg3(2300, "jobq_add jobid=%d jcr=0x%x use_count=%d\n", jcr->JobId, jcr, jcr->use_count()); if (jq->valid != JOBQ_VALID) { + Jmsg0(jcr, M_ERROR, 0, "Jobq_add queue not initialized.\n"); return EINVAL; } + jcr->inc_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; - stat = pthread_create(&id, &jq->attr, sched_wait, (void *)sched_pkt); + 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.bstrerror(stat)); + } return stat; } - if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) { - return stat; - } + P(jq->mutex); if ((item = (jobq_item_t *)malloc(sizeof(jobq_item_t))) == NULL) { + free_jcr(jcr); /* release jcr */ return ENOMEM; } item->jcr = jcr; + /* While waiting in a queue this job is not attached to a thread */ + set_jcr_in_tsd(INVALID_JCR); if (job_canceled(jcr)) { /* Add job to ready queue so that it is canceled quickly */ jq->ready_jobs->prepend(item); - Dmsg1(100, "Prepended job=%d to ready queue\n", jcr->JobId); + Dmsg1(2300, "Prepended job=%d to ready queue\n", jcr->JobId); } else { /* Add this job to the wait queue in priority sorted order */ - for (li=NULL; (li=(jobq_item_t *)jq->waiting_jobs->next(li)); ) { - Dmsg2(100, "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); - Dmsg2(100, "insert_before jobid=%d before %d\n", - li->jcr->JobId, jcr->JobId); - inserted = true; - break; - } + 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); + Dmsg2(2300, "insert_before jobid=%d before waiting job=%d\n", + li->jcr->JobId, jcr->JobId); + inserted = true; + break; + } } /* If not jobs in wait queue, append it */ if (!inserted) { - jq->waiting_jobs->append(item); - Dmsg1(100, "Appended item jobid=%d\n", jcr->JobId); + jq->waiting_jobs->append(item); + Dmsg1(2300, "Appended item jobid=%d to waiting queue\n", jcr->JobId); } } /* Ensure that at least one server looks at the queue. */ stat = start_server(jq); - pthread_mutex_unlock(&jq->mutex); - Dmsg0(100, "Return jobq_add\n"); + V(jq->mutex); + Dmsg0(2300, "Return jobq_add\n"); return stat; } /* - * Remove a job from the job queue. Used only by cancel Console command. + * Remove a job from the job queue. Used only by cancel_job(). * jq is a queue that was created with jobq_init * work_item is an element of work * - * Note, it is "removed" by immediately calling a processing routine. - * if you want to cancel it, you need to provide some external means - * of doing so. + * Note, it is "removed" from the job queue. + * If you want to cancel it, you need to provide some external means + * of doing so (e.g. pthread_kill()). */ int jobq_remove(jobq_t *jq, JCR *jcr) { int stat; bool found = false; jobq_item_t *item; - - Dmsg1(100, "jobq_remove jobid=%d\n", jcr->JobId); + + Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x\n", jcr->JobId, jcr); if (jq->valid != JOBQ_VALID) { return EINVAL; } - if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) { - return stat; - } - - for (item=NULL; (item=(jobq_item_t *)jq->waiting_jobs->next(item)); ) { + P(jq->mutex); + foreach_dlist(item, jq->waiting_jobs) { if (jcr == item->jcr) { - found = true; - break; + found = true; + break; } } if (!found) { + V(jq->mutex); + Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x not in wait queue\n", jcr->JobId, jcr); return EINVAL; } /* Move item to be the first on the list */ jq->waiting_jobs->remove(item); jq->ready_jobs->prepend(item); - + Dmsg2(2300, "jobq_remove jobid=%d jcr=0x%x moved to ready queue\n", jcr->JobId, jcr); + stat = start_server(jq); - pthread_mutex_unlock(&jq->mutex); - Dmsg0(100, "Return jobq_remove\n"); + V(jq->mutex); + Dmsg0(2300, "Return jobq_remove\n"); return stat; } @@ -301,268 +357,550 @@ static int start_server(jobq_t *jq) 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(100, "Signal worker to wake up\n"); - if ((stat = pthread_cond_signal(&jq->work)) != 0) { - return stat; + 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.bstrerror(stat)); + return stat; } } else if (jq->num_workers < jq->max_workers) { - Dmsg0(100, "Create worker thread\n"); + Dmsg0(2300, "Create worker thread\n"); /* No idle threads so create a new one */ set_thread_concurrency(jq->max_workers + 1); + jq->num_workers++; if ((stat = pthread_create(&id, &jq->attr, jobq_server, (void *)jq)) != 0) { - return stat; + berrno be; + jq->num_workers--; + Jmsg1(NULL, M_ERROR, 0, _("pthread_create: ERR=%s\n"), be.bstrerror(stat)); + return stat; } } return stat; } -/* +/* * This is the worker thread that serves the job queue. - * When all the resources are acquired for the job, + * When all the resources are acquired for the job, * it will call the user's engine. */ -static void *jobq_server(void *arg) +extern "C" +void *jobq_server(void *arg) { 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; + bool timedout = false; bool work = true; - Dmsg0(100, "Start jobq_server\n"); - if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) { - return NULL; - } - jq->num_workers++; + set_jcr_in_tsd(INVALID_JCR); + Dmsg0(2300, "Start jobq_server\n"); + P(jq->mutex); for (;;) { struct timeval tv; struct timezone tz; - Dmsg0(100, "Top of for loop\n"); - timedout = false; - Dmsg0(100, "gettimeofday()\n"); - gettimeofday(&tv, &tz); - timeout.tv_nsec = 0; - timeout.tv_sec = tv.tv_sec + 4; - - while (!work && !jq->quit) { - /* - * Wait 4 seconds, then if no more work, exit - */ - Dmsg0(200, "pthread_cond_timedwait()\n"); - stat = pthread_cond_timedwait(&jq->work, &jq->mutex, &timeout); - Dmsg1(100, "timedwait=%d\n", stat); - if (stat == ETIMEDOUT) { - timedout = true; - break; - } else if (stat != 0) { - /* This shouldn't happen */ - Dmsg0(100, "This shouldn't happen\n"); - jq->num_workers--; - pthread_mutex_unlock(&jq->mutex); - return NULL; - } - } - /* + 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 + */ + Dmsg0(2300, "pthread_cond_timedwait()\n"); + stat = pthread_cond_timedwait(&jq->work, &jq->mutex, &timeout); + if (stat == ETIMEDOUT) { + Dmsg0(2300, "timedwait timedout.\n"); + timedout = true; + break; + } else if (stat != 0) { + /* This shouldn't happen */ + Dmsg0(2300, "This shouldn't happen\n"); + jq->num_workers--; + V(jq->mutex); + return NULL; + } + break; + } + } + /* * If anything is in the ready queue, run it */ - Dmsg0(100, "Checking ready queue.\n"); + 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()) { - Dmsg0(100, "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); - Dmsg1(100, "Took jobid=%d from ready and appended to run\n", jcr->JobId); - if ((stat = pthread_mutex_unlock(&jq->mutex)) != 0) { - jq->num_workers--; - return NULL; - } + 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--; + V(jq->mutex); + return NULL; + } + } + jq->running_jobs->append(je); + + /* Attach jcr to this thread while we run the job */ + jcr->my_thread_id = pthread_self(); + jcr->set_killable(true); + set_jcr_in_tsd(jcr); + Dmsg1(2300, "Took jobid=%d from ready and appended to run\n", jcr->JobId); + + /* Release job queue lock */ + V(jq->mutex); + /* Call user's routine here */ - Dmsg1(100, "Calling user engine for jobid=%d\n", jcr->JobId); - jq->engine(je->jcr); - Dmsg1(100, "Back from user engine jobid=%d.\n", jcr->JobId); - if ((stat = pthread_mutex_lock(&jq->mutex)) != 0) { - jq->num_workers--; - free(je); /* release job entry */ - return NULL; - } - Dmsg0(200, "Done lock mutex\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--; - } - - if (jcr->job->RescheduleOnError && - jcr->JobStatus != JS_Terminated && - jcr->JobStatus != JS_Canceled && - jcr->job->RescheduleTimes > 0 && - jcr->reschedule_count < jcr->job->RescheduleTimes) { - - /* - * 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(100, "Rescheduled Job %s to re-run in %d seconds.\n", jcr->Job, - (int)jcr->job->RescheduleInterval); - jcr->JobStatus = JS_Created; /* force new status */ - dird_free_jcr(jcr); /* partial cleanup old stuff */ - if (jcr->JobBytes == 0) { - Dmsg1(100, "Requeue job=%d\n", jcr->JobId); - 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; - njcr->pool = jcr->pool; - njcr->store = jcr->store; - njcr->messages = jcr->messages; - Dmsg0(100, "Call to run new job\n"); - V(jq->mutex); - run_job(njcr); /* This creates a "new" job */ - P(jq->mutex); - Dmsg0(100, "Back from running new job.\n"); - } - /* Clean up and release old jcr */ - if (jcr->db) { - Dmsg0(100, "Close DB\n"); - db_close_database(jcr, jcr->db); - jcr->db = NULL; - } - Dmsg1(100, "====== Termination job=%d\n", jcr->JobId); - free_jcr(jcr); - free(je); /* release job entry */ + Dmsg3(2300, "Calling user engine for jobid=%d use=%d stat=%c\n", jcr->JobId, + jcr->use_count(), jcr->JobStatus); + jq->engine(je->jcr); + + /* Job finished detach from thread */ + remove_jcr_from_tsd(je->jcr); + je->jcr->set_killable(false); + + Dmsg2(2300, "Back from user engine jobid=%d use=%d.\n", jcr->JobId, + jcr->use_count()); + + /* 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) { + int num; + dec_read_store(jcr); + dec_write_store(jcr); + update_client_numconcurrentjobs(jcr, -1); + num = jcr->job->getNumConcurrentJobs() - 1; + jcr->job->setNumConcurrentJobs(num); + jcr->acquired_resource_locks = false; + } + + if (reschedule_job(jcr, jq, je)) { + continue; /* go look for more work */ + } + + /* Clean up and release old jcr */ + 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 */ } /* * If any job in the wait queue can be run, * move it to the ready queue */ - Dmsg0(100, "Done check ready, now check wait queue.\n"); - while (!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; - Dmsg1(100, "Set Run pri=%d\n", Priority); - } else { - Priority = je->jcr->JobPriority; - Dmsg1(100, "Set 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); - Dmsg3(100, "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 (jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) { - jcr->store->NumConcurrentJobs++; - } else { - set_jcr_job_status(jcr, JS_WaitStoreRes); - je = jn; - continue; - } - if (jcr->client->NumConcurrentJobs < jcr->client->MaxConcurrentJobs) { - jcr->client->NumConcurrentJobs++; - } else { - jcr->store->NumConcurrentJobs--; - set_jcr_job_status(jcr, JS_WaitClientRes); - je = jn; - continue; - } - if (jcr->job->NumConcurrentJobs < jcr->job->MaxConcurrentJobs) { - jcr->job->NumConcurrentJobs++; - } else { - jcr->store->NumConcurrentJobs--; - jcr->client->NumConcurrentJobs--; - set_jcr_job_status(jcr, JS_WaitJobRes); - je = jn; - 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(100, "moved JobId=%d from wait to ready queue\n", je->jcr->JobId); - je = jn; - } /* end for loop */ - break; - } /* end while loop */ - Dmsg0(100, "Done checking wait queue.\n"); + Dmsg0(2300, "Done check ready, now check wait queue.\n"); + if (!jq->waiting_jobs->empty() && !jq->quit) { + int Priority; + bool running_allow_mix = false; + 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); + running_allow_mix = true; + for ( ; re; ) { + Dmsg2(2300, "JobId %d is also running with %s\n", + re->jcr->JobId, + re->jcr->job->allow_mixed_priority ? "mix" : "no mix"); + if (!re->jcr->job->allow_mixed_priority) { + running_allow_mix = false; + break; + } + re = (jobq_item_t *)jq->running_jobs->next(re); + } + Dmsg1(2300, "The running job(s) %s mixing priorities.\n", + running_allow_mix ? "allow" : "don't allow"); + } 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); + + Dmsg4(2300, "Examining Job=%d JobPri=%d want Pri=%d (%s)\n", + jcr->JobId, jcr->JobPriority, Priority, + jcr->job->allow_mixed_priority ? "mix" : "no mix"); + + /* Take only jobs of correct Priority */ + if (!(jcr->JobPriority == Priority + || (jcr->JobPriority < Priority && + jcr->job->allow_mixed_priority && running_allow_mix))) { + jcr->setJobStatus(JS_WaitPriority); + break; + } + + if (!acquire_resources(jcr)) { + /* If resource conflict, job is canceled */ + if (!job_canceled(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. Note, we may also get here if the + * job was canceled. Once it is "run", it will quickly + * terminate. + */ + 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 */ + + } /* end if */ + + Dmsg0(2300, "Done checking wait queue.\n"); /* * 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) { - Dmsg0(100, "Wake up destroy routine\n"); - /* Wake up destroy routine if he is waiting */ - pthread_cond_broadcast(&jq->work); - } - break; + 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; } - Dmsg0(100, "Check for work request\n"); - /* + Dmsg0(2300, "Check for work request\n"); + /* * If no more work requests, and we waited long enough, quit */ - Dmsg2(100, "timedout=%d read empty=%d\n", timedout, - jq->ready_jobs->empty()); + Dmsg2(2300, "timedout=%d read empty=%d\n", timedout, + jq->ready_jobs->empty()); if (jq->ready_jobs->empty() && timedout) { - Dmsg0(100, "break big loop\n"); - jq->num_workers--; - break; + Dmsg0(2300, "break big loop\n"); + jq->num_workers--; + break; } - Dmsg0(100, "Loop again\n"); - work = false; + + 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(); + } + Dmsg1(2300, "Loop again. work=%d\n", work); } /* end of big for loop */ Dmsg0(200, "unlock mutex\n"); - pthread_mutex_unlock(&jq->mutex); - Dmsg0(100, "End jobq_server\n"); + V(jq->mutex); + Dmsg0(2300, "End jobq_server\n"); 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) +{ + bool resched = false; + /* + * Reschedule the job if requested and possible + */ + /* Basic condition is that more reschedule times remain */ + if (jcr->job->RescheduleTimes == 0 || + jcr->reschedule_count < jcr->job->RescheduleTimes) { + + /* Check for incomplete jobs */ + if (jcr->is_incomplete()) { + resched = (jcr->RescheduleIncompleteJobs && jcr->is_JobType(JT_BACKUP) && + !(jcr->HasBase||jcr->is_JobLevel(L_BASE))); + } else { + /* Check for failed jobs */ + resched = (jcr->job->RescheduleOnError && + !jcr->is_JobStatus(JS_Terminated) && + !jcr->is_JobStatus(JS_Canceled) && + jcr->is_JobType(JT_BACKUP)); + } + } + if (resched) { + char dt[50], dt2[50]; + + /* + * Reschedule this job by cleaning it up, but + * reuse the same JobId if possible. + */ + jcr->rerunning = jcr->is_incomplete(); /* save incomplete status */ + 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; + jcr->setJobStatus(JS_WaitStartTime); + jcr->SDJobStatus = 0; + jcr->JobErrors = 0; + if (!allow_duplicate_job(jcr)) { + return false; + } + /* Only jobs with no output or Incomplete jobs can run on same JCR */ + if (jcr->JobBytes == 0 || jcr->rerunning) { + Dmsg2(2300, "Requeue job=%d use=%d\n", jcr->JobId, jcr->use_count()); + V(jq->mutex); + /* + * Special test here since a Virtual Full gets marked + * as a Full, so we look at the resource record + */ + if (jcr->wasVirtualFull) { + jcr->setJobLevel(L_VIRTUAL_FULL); + } + /* + * When we are using the same jcr then make sure to reset + * RealEndTime back to zero. + */ + jcr->jr.RealEndTime = 0; + 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); + /* + * Eliminate the new job_end_push, then copy the one from + * the old job, and set the old one to be empty. + */ + void *v; + lock_jobs(); /* protect ourself from reload_config() */ + LockRes(); + foreach_alist(v, (&jcr->job_end_push)) { + njcr->job_end_push.append(v); + } + jcr->job_end_push.destroy(); + jcr->job_end_push.init(1, false); + UnlockRes(); + unlock_jobs(); + + njcr->reschedule_count = jcr->reschedule_count; + njcr->sched_time = jcr->sched_time; + njcr->initial_sched_time = jcr->initial_sched_time; + /* + * Special test here since a Virtual Full gets marked + * as a Full, so we look at the resource record + */ + if (jcr->wasVirtualFull) { + njcr->setJobLevel(L_VIRTUAL_FULL); + } else { + njcr->setJobLevel(jcr->getJobLevel()); + } + njcr->pool = jcr->pool; + njcr->run_pool_override = jcr->run_pool_override; + njcr->next_pool = jcr->next_pool; + njcr->run_next_pool_override = jcr->run_next_pool_override; + njcr->full_pool = jcr->full_pool; + njcr->vfull_pool = jcr->vfull_pool; + njcr->run_full_pool_override = jcr->run_full_pool_override; + njcr->run_vfull_pool_override = jcr->run_vfull_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; + njcr->setJobStatus(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) + * + * Returns: true if successful + * false if resource failure + */ +static bool acquire_resources(JCR *jcr) +{ + bool skip_this_jcr = false; + + jcr->acquired_resource_locks = false; +/* + * Turning this code off is likely to cause some deadlocks, + * but we do not really have enough information here to + * know if this is really a deadlock (it may be a dual drive + * autochanger), and in principle, the SD reservation system + * should detect these deadlocks, so push the work off on it. + */ +#ifdef xxx + if (jcr->rstore && jcr->rstore == jcr->wstore) { /* possible 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); + jcr->setJobStatus(JS_Canceled); + return false; + } +#endif + if (jcr->rstore) { + Dmsg1(200, "Rstore=%s\n", jcr->rstore->name()); + if (!inc_read_store(jcr)) { + Dmsg1(200, "Fail rncj=%d\n", jcr->rstore->getNumConcurrentJobs()); + jcr->setJobStatus(JS_WaitStoreRes); + return false; + } + } + + if (jcr->wstore) { + Dmsg1(200, "Wstore=%s\n", jcr->wstore->name()); + int num = jcr->wstore->getNumConcurrentJobs(); + if (num < jcr->wstore->MaxConcurrentJobs) { + Dmsg1(200, "Inc wncj=%d\n", num + 1); + jcr->wstore->setNumConcurrentJobs(num + 1); + } else if (jcr->rstore) { + dec_read_store(jcr); + skip_this_jcr = true; + } else { + Dmsg1(200, "Fail wncj=%d\n", num); + skip_this_jcr = true; + } + } + if (skip_this_jcr) { + jcr->setJobStatus(JS_WaitStoreRes); + return false; + } + + if (jcr->client) { + if (jcr->client->getNumConcurrentJobs() < jcr->client->MaxConcurrentJobs) { + update_client_numconcurrentjobs(jcr, 1); + } else { + /* Back out previous locks */ + dec_write_store(jcr); + dec_read_store(jcr); + jcr->setJobStatus(JS_WaitClientRes); + return false; + } + } + if (jcr->job->getNumConcurrentJobs() < jcr->job->MaxConcurrentJobs) { + int num; + num = jcr->job->getNumConcurrentJobs() + 1; + jcr->job->setNumConcurrentJobs(num); + } else { + /* Back out previous locks */ + dec_write_store(jcr); + dec_read_store(jcr); + update_client_numconcurrentjobs(jcr, -1); + jcr->setJobStatus(JS_WaitJobRes); + return false; + } + + jcr->acquired_resource_locks = true; + return true; +} + +static pthread_mutex_t rstore_mutex = PTHREAD_MUTEX_INITIALIZER; + +/* + * Note: inc_read_store() and dec_read_store() are + * called from select_rstore() in src/dird/restore.c + */ +bool inc_read_store(JCR *jcr) +{ + P(rstore_mutex); + int num = jcr->rstore->getNumConcurrentJobs(); + int numread = jcr->rstore->getNumConcurrentReadJobs(); + int maxread = jcr->rstore->MaxConcurrentReadJobs; + if (num < jcr->rstore->MaxConcurrentJobs && + (jcr->getJobType() == JT_RESTORE || + numread == 0 || + maxread == 0 || /* No limit set */ + numread < maxread)) /* Below the limit */ + { + num++; + numread++; + jcr->rstore->setNumConcurrentReadJobs(numread); + jcr->rstore->setNumConcurrentJobs(num); + Dmsg1(200, "Inc rncj=%d\n", num); + V(rstore_mutex); + return true; + } + V(rstore_mutex); + return false; +} + +void dec_read_store(JCR *jcr) +{ + if (jcr->rstore) { + P(rstore_mutex); + int numread = jcr->rstore->getNumConcurrentReadJobs() - 1; + int num = jcr->rstore->getNumConcurrentJobs() - 1; + jcr->rstore->setNumConcurrentReadJobs(numread); + jcr->rstore->setNumConcurrentJobs(num); + Dmsg1(200, "Dec rncj=%d\n", num); + V(rstore_mutex); + } +} + +static void dec_write_store(JCR *jcr) +{ + if (jcr->wstore) { + int num = jcr->wstore->getNumConcurrentJobs() - 1; + Dmsg1(200, "Dec wncj=%d\n", num); + jcr->wstore->setNumConcurrentJobs(num); + } +}