X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=bacula%2Fsrc%2Fdird%2Fjobq.c;h=af8605268d300db1aa10e27c3d8ae6d333921def;hb=495460e08127544c76c0037fe252c40308dbab15;hp=4d79e77a3d2d67d271afda8394233aeb8009b7b7;hpb=fe3cc7407f557f6881075dd6e546fef71a5afdee;p=bacula%2Fbacula diff --git a/bacula/src/dird/jobq.c b/bacula/src/dird/jobq.c old mode 100755 new mode 100644 index 4d79e77a3d..af8605268d --- a/bacula/src/dird/jobq.c +++ b/bacula/src/dird/jobq.c @@ -1,3 +1,30 @@ +/* + Bacula® - The Network Backup Solution + + 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 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 + 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., 51 Franklin Street, Fifth Floor, Boston, MA + 02110-1301, USA. + + Bacula® is a registered trademark of John Walker. + The licensor of Bacula is the Free Software Foundation Europe + (FSFE), Fiduciary Program, Sumatrastrasse 25, 8006 Zürich, + Switzerland, email:ftf@fsfeurope.org. +*/ /* * Bacula job queue routines. * @@ -17,20 +44,6 @@ * David R. Butenhof * */ -/* - Copyright (C) 2003-2006 Kern Sibbald - - This program is free software; you can redistribute it and/or - modify it under the terms of the GNU General Public License - version 2 as amended with additional clauses defined in the - file LICENSE in the main source directory. - - 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 - the file LICENSE for additional details. - - */ #include "bacula.h" #include "dird.h" @@ -43,8 +56,9 @@ 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_read_store(JCR *jcr); +static void dec_write_store(JCR *jcr); /* * Initialize a job queue @@ -59,7 +73,7 @@ int jobq_init(jobq_t *jq, int threads, void *(*engine)(void *arg)) 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) { @@ -68,13 +82,13 @@ int jobq_init(jobq_t *jq, int threads, void *(*engine)(void *arg)) } 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; @@ -107,7 +121,7 @@ int jobq_destroy(jobq_t *jq) } 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 */ @@ -120,7 +134,7 @@ int jobq_destroy(jobq_t *jq) 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; } @@ -128,7 +142,7 @@ int jobq_destroy(jobq_t *jq) 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; } @@ -136,7 +150,7 @@ int jobq_destroy(jobq_t *jq) } 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); @@ -208,6 +222,16 @@ int jobq_add(jobq_t *jq, JCR *jcr) pthread_t id; wait_pkt *sched_pkt; + 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"); @@ -224,14 +248,14 @@ int jobq_add(jobq_t *jq, JCR *jcr) 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; } @@ -296,7 +320,7 @@ int jobq_remove(jobq_t *jq, JCR *jcr) 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; } @@ -334,15 +358,15 @@ static int start_server(jobq_t *jq) 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) { @@ -351,7 +375,7 @@ static int start_server(jobq_t *jq) 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; } } @@ -377,7 +401,7 @@ void *jobq_server(void *arg) 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++; @@ -430,6 +454,7 @@ void *jobq_server(void *arg) } } 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 */ @@ -453,71 +478,18 @@ void *jobq_server(void *arg) * put into the ready queue. */ if (jcr->acquired_resource_locks) { - jcr->store->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->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)); - 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_pointers(jcr); /* partial cleanup old stuff */ - jcr->JobStatus = JS_WaitStartTime; - jcr->SDJobStatus = 0; - if (jcr->JobBytes == 0) { - Dmsg2(2300, "Requeue job=%d use=%d\n", jcr->JobId, jcr->use_count()); - jcr->JobStatus = JS_WaitStartTime; - 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->JobLevel = jcr->JobLevel; - njcr->JobStatus = jcr->JobStatus; - copy_storage(njcr, jcr->storage); - 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 */ @@ -560,14 +532,19 @@ void *jobq_server(void *arg) } if (!acquire_resources(jcr)) { - je = jn; /* point to next waiting job */ - continue; + /* 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 + /* + * 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. */ - 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); @@ -624,6 +601,94 @@ void *jobq_server(void *arg) 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) * @@ -634,26 +699,39 @@ static bool acquire_resources(JCR *jcr) { bool skip_this_jcr = false; - if (jcr->JobType == JT_RESTORE || jcr->JobType == JT_VERIFY) { - /* - * Let only one Restore/verify job run at a time regardless - * of MaxConcurrentJobs. - */ - if (jcr->store->NumConcurrentJobs == 0) { - jcr->store->NumConcurrentJobs = 1; + 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()); + 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); return false; } - /* We are not doing a Restore or Verify */ - } else if (jcr->store->NumConcurrentJobs == 0 && - jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) { - /* Simple case, first job */ - jcr->store->NumConcurrentJobs = 1; - } else if (jcr->store->NumConcurrentJobs < jcr->store->MaxConcurrentJobs) { - jcr->store->NumConcurrentJobs++; - } else { - skip_this_jcr = true; + } + + if (jcr->wstore) { + Dmsg1(200, "Wstore=%s\n", jcr->wstore->name()); + if (jcr->wstore->NumConcurrentJobs < jcr->wstore->MaxConcurrentJobs) { + jcr->wstore->NumConcurrentJobs++; + Dmsg1(200, "Inc wncj=%d\n", jcr->wstore->NumConcurrentJobs); + } else if (jcr->rstore) { + dec_read_store(jcr); + skip_this_jcr = true; + } else { + Dmsg1(200, "Fail wncj=%d\n", jcr->wstore->NumConcurrentJobs); + skip_this_jcr = true; + } } if (skip_this_jcr) { set_jcr_job_status(jcr, JS_WaitStoreRes); @@ -664,7 +742,8 @@ static bool acquire_resources(JCR *jcr) jcr->client->NumConcurrentJobs++; } else { /* Back out previous locks */ - jcr->store->NumConcurrentJobs--; + dec_write_store(jcr); + dec_read_store(jcr); set_jcr_job_status(jcr, JS_WaitClientRes); return false; } @@ -672,12 +751,33 @@ static bool acquire_resources(JCR *jcr) jcr->job->NumConcurrentJobs++; } else { /* Back out previous locks */ - jcr->store->NumConcurrentJobs--; + dec_write_store(jcr); + dec_read_store(jcr); jcr->client->NumConcurrentJobs--; set_jcr_job_status(jcr, JS_WaitJobRes); return false; } - /* Check actual device availability */ - /* ***FIXME****/ + + 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); + } +}