/* Forward referenced subroutines */
static void *job_thread(void *arg);
-static char *edit_run_codes(JCR *jcr, char *omsg, char *imsg);
-static void release_resource_locks(JCR *jcr);
-static int acquire_resource_locks(JCR *jcr);
-#ifdef USE_SEMAPHORE
-static void backoff_resource_locks(JCR *jcr, int count);
-#endif
/* Exported subroutines */
-void run_job(JCR *jcr);
/* Imported subroutines */
extern int do_restore(JCR *jcr);
extern int do_verify(JCR *jcr);
-#ifdef USE_SEMAPHORE
-static semlock_t job_lock;
-static pthread_mutex_t mutex;
-static pthread_cond_t resource_wait;
-static int waiting = 0; /* count of waiting threads */
-#else
-/* Queue of jobs to be run */
-workq_t job_wq; /* our job work queue */
-#endif
+jobq_t job_queue;
void init_job_server(int max_workers)
{
int stat;
-#ifdef USE_SEMAPHORE
- if ((stat = sem_init(&job_lock, max_workers)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not init job lock: ERR=%s\n"), strerror(stat));
+ if ((stat = jobq_init(&job_queue, max_workers, job_thread)) != 0) {
+ Emsg1(M_ABORT, 0, _("Could not init job queue: ERR=%s\n"), strerror(stat));
}
- if ((stat = pthread_mutex_init(&mutex, NULL)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not init resource mutex: ERR=%s\n"), strerror(stat));
- }
- if ((stat = pthread_cond_init(&resource_wait, NULL)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not init resource wait: ERR=%s\n"), strerror(stat));
- }
-
-#else
- if ((stat = workq_init(&job_wq, max_workers, job_thread)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not init job work queue: ERR=%s\n"), strerror(stat));
- }
-#endif
return;
}
void run_job(JCR *jcr)
{
int stat, errstat;
-#ifdef USE_SEMAPHORE
- pthread_t tid;
-#else
- workq_ele_t *work_item;
-#endif
sm_check(__FILE__, __LINE__, True);
init_msg(jcr, jcr->messages);
set_jcr_job_status(jcr, JS_Created);
jcr->jr.SchedTime = jcr->sched_time;
jcr->jr.StartTime = jcr->start_time;
+ jcr->jr.EndTime = 0; /* perhaps rescheduled, clear it */
jcr->jr.Type = jcr->JobType;
jcr->jr.Level = jcr->JobLevel;
jcr->jr.JobStatus = jcr->JobStatus;
jcr->db=db_init_database(jcr, jcr->catalog->db_name, jcr->catalog->db_user,
jcr->catalog->db_password, jcr->catalog->db_address,
jcr->catalog->db_port, jcr->catalog->db_socket);
- if (!db_open_database(jcr, jcr->db)) {
- Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
+ if (!jcr->db || !db_open_database(jcr, jcr->db)) {
+ Jmsg(jcr, M_FATAL, 0, _("Could not open database \"%s\".\n"),
+ jcr->catalog->db_name);
+ if (jcr->db) {
+ Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
+ }
set_jcr_job_status(jcr, JS_ErrorTerminated);
free_jcr(jcr);
return;
jcr->JobId = jcr->jr.JobId;
ASSERT(jcr->jr.JobId > 0);
- Dmsg4(30, "Created job record JobId=%d Name=%s Type=%c Level=%c\n",
+ Dmsg4(50, "Created job record JobId=%d Name=%s Type=%c Level=%c\n",
jcr->JobId, jcr->Job, jcr->jr.Type, jcr->jr.Level);
Dmsg0(200, "Add jrc to work queue\n");
-#ifdef USE_SEMAPHORE
- if ((stat = pthread_create(&tid, NULL, job_thread, (void *)jcr)) != 0) {
- Emsg1(M_ABORT, 0, _("Unable to create job thread: ERR=%s\n"), strerror(stat));
- }
-#else
/* Queue the job to be run */
- if ((stat = workq_add(&job_wq, (void *)jcr, &work_item, 0)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not add job to work queue: ERR=%s\n"), strerror(stat));
+ if ((stat = jobq_add(&job_queue, jcr)) != 0) {
+ Emsg1(M_ABORT, 0, _("Could not add job queue: ERR=%s\n"), strerror(stat));
}
- jcr->work_item = work_item;
-#endif
- Dmsg0(200, "Done run_job()\n");
+ Dmsg0(100, "Done run_job()\n");
}
/*
- * This is the engine called by workq_add() when we were pulled
+ * This is the engine called by job_add() when we were pulled
* from the work queue.
- * At this point, we are running in our own thread
+ * At this point, we are running in our own thread and all
+ * necessary resources are allocated -- see jobq.c
*/
static void *job_thread(void *arg)
{
sm_check(__FILE__, __LINE__, True);
for ( ;; ) {
- if (!acquire_resource_locks(jcr)) {
- set_jcr_job_status(jcr, JS_Canceled);
- }
Dmsg0(200, "=====Start Job=========\n");
jcr->start_time = time(NULL); /* set the real start time */
BPIPE *bpipe;
char line[MAXSTRING];
- before = edit_run_codes(jcr, before, jcr->job->RunBeforeJob);
+ before = edit_job_codes(jcr, before, jcr->job->RunBeforeJob, "");
bpipe = open_bpipe(before, 0, "r");
free_pool_memory(before);
while (fgets(line, sizeof(line), bpipe->rfd)) {
}
}
switch (jcr->JobType) {
- case JT_BACKUP:
- do_backup(jcr);
- if (jcr->JobStatus == JS_Terminated) {
- do_autoprune(jcr);
- }
- break;
- case JT_VERIFY:
- do_verify(jcr);
- if (jcr->JobStatus == JS_Terminated) {
- do_autoprune(jcr);
- }
- break;
- case JT_RESTORE:
- do_restore(jcr);
- if (jcr->JobStatus == JS_Terminated) {
- do_autoprune(jcr);
- }
- break;
- case JT_ADMIN:
- do_admin(jcr);
- if (jcr->JobStatus == JS_Terminated) {
- do_autoprune(jcr);
- }
- break;
- default:
- Pmsg1(0, "Unimplemented job type: %d\n", jcr->JobType);
- break;
+ case JT_BACKUP:
+ do_backup(jcr);
+ if (jcr->JobStatus == JS_Terminated) {
+ do_autoprune(jcr);
+ }
+ break;
+ case JT_VERIFY:
+ do_verify(jcr);
+ if (jcr->JobStatus == JS_Terminated) {
+ do_autoprune(jcr);
+ }
+ break;
+ case JT_RESTORE:
+ do_restore(jcr);
+ if (jcr->JobStatus == JS_Terminated) {
+ do_autoprune(jcr);
+ }
+ break;
+ case JT_ADMIN:
+ do_admin(jcr);
+ if (jcr->JobStatus == JS_Terminated) {
+ do_autoprune(jcr);
}
- if (jcr->job->RunAfterJob) {
+ break;
+ default:
+ Pmsg1(0, "Unimplemented job type: %d\n", jcr->JobType);
+ break;
+ }
+ if ((jcr->job->RunAfterJob && jcr->JobStatus == JS_Terminated) ||
+ (jcr->job->RunAfterFailedJob && jcr->JobStatus != JS_Terminated)) {
POOLMEM *after = get_pool_memory(PM_FNAME);
int status;
BPIPE *bpipe;
char line[MAXSTRING];
- after = edit_run_codes(jcr, after, jcr->job->RunAfterJob);
+ if (jcr->JobStatus == JS_Terminated) {
+ after = edit_job_codes(jcr, after, jcr->job->RunAfterJob, "");
+ } else {
+ after = edit_job_codes(jcr, after, jcr->job->RunAfterFailedJob, "");
+ }
bpipe = open_bpipe(after, 0, "r");
free_pool_memory(after);
while (fgets(line, sizeof(line), bpipe->rfd)) {
}
status = close_bpipe(bpipe);
if (status != 0) {
- Jmsg(jcr, M_FATAL, 0, _("RunAfterJob returned non-zero status=%d\n"),
- status);
+ if (jcr->JobStatus == JS_Terminated) {
+ Jmsg(jcr, M_FATAL, 0, _("RunAfterJob returned non-zero status=%d\n"),
+ status);
+ } else {
+ Jmsg(jcr, M_FATAL, 0, _("RunAfterFailedJob returned non-zero status=%d\n"),
+ status);
+ }
set_jcr_job_status(jcr, JS_FatalError);
update_job_end_record(jcr);
}
}
}
bail_out:
- release_resource_locks(jcr);
- 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) {
- continue; /* reschedule the job */
- }
- /*
- * 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;
- run_job(njcr);
- }
break;
}
- if (jcr->db) {
- Dmsg0(200, "Close DB\n");
- db_close_database(jcr, jcr->db);
- jcr->db = NULL;
- }
- free_jcr(jcr);
Dmsg0(50, "======== End Job ==========\n");
sm_check(__FILE__, __LINE__, True);
return NULL;
}
-/*
- * Acquire the resources needed. These locks limit the
- * number of jobs by each resource. We have limits on
- * Jobs, Clients, Storage, and total jobs.
- */
-static int acquire_resource_locks(JCR *jcr)
-{
- time_t now = time(NULL);
- time_t wtime = jcr->sched_time - now;
-
- /* 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);
- }
- /* Check every 30 seconds if canceled */
- while (wtime > 0) {
- Dmsg2(100, "Waiting on sched time, jobid=%d secs=%d\n", jcr->JobId, wtime);
- if (wtime > 30) {
- wtime = 30;
- }
- bmicrosleep(wtime, 0);
- if (job_canceled(jcr)) {
- return 0;
- }
- wtime = jcr->sched_time - time(NULL);
- }
-
-
-#ifdef USE_SEMAPHORE
- int stat;
-
- /* Initialize semaphores */
- if (jcr->store->sem.valid != SEMLOCK_VALID) {
- if ((stat = sem_init(&jcr->store->sem, jcr->store->MaxConcurrentJobs)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not init Storage semaphore: ERR=%s\n"), strerror(stat));
- }
- }
- if (jcr->client->sem.valid != SEMLOCK_VALID) {
- if ((stat = sem_init(&jcr->client->sem, jcr->client->MaxConcurrentJobs)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not init Client semaphore: ERR=%s\n"), strerror(stat));
- }
- }
- if (jcr->job->sem.valid != SEMLOCK_VALID) {
- if ((stat = sem_init(&jcr->job->sem, jcr->job->MaxConcurrentJobs)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not init Job semaphore: ERR=%s\n"), strerror(stat));
- }
- }
-
- for ( ;; ) {
- /* Acquire semaphore */
- set_jcr_job_status(jcr, JS_WaitJobRes);
- if ((stat = sem_lock(&jcr->job->sem)) != 0) {
- Emsg1(M_ABORT, 0, _("Could not acquire Job max jobs lock: ERR=%s\n"), strerror(stat));
- }
- set_jcr_job_status(jcr, JS_WaitClientRes);
- if ((stat = sem_trylock(&jcr->client->sem)) != 0) {
- if (stat == EBUSY) {
- backoff_resource_locks(jcr, 1);
- goto wait;
- } else {
- Emsg1(M_ABORT, 0, _("Could not acquire Client max jobs lock: ERR=%s\n"), strerror(stat));
- }
- }
- set_jcr_job_status(jcr, JS_WaitStoreRes);
- if ((stat = sem_trylock(&jcr->store->sem)) != 0) {
- if (stat == EBUSY) {
- backoff_resource_locks(jcr, 2);
- goto wait;
- } else {
- Emsg1(M_ABORT, 0, _("Could not acquire Storage max jobs lock: ERR=%s\n"), strerror(stat));
- }
- }
- set_jcr_job_status(jcr, JS_WaitMaxJobs);
- if ((stat = sem_trylock(&job_lock)) != 0) {
- if (stat == EBUSY) {
- backoff_resource_locks(jcr, 3);
- goto wait;
- } else {
- Emsg1(M_ABORT, 0, _("Could not acquire max jobs lock: ERR=%s\n"), strerror(stat));
- }
- }
- break;
-
-wait:
- if (job_canceled(jcr)) {
- return 0;
- }
- P(mutex);
- /*
- * Wait for a resource to be released either by backoff or
- * by a job terminating.
- */
- waiting++;
- pthread_cond_wait(&resource_wait, &mutex);
- waiting--;
- V(mutex);
- /* Try again */
- }
- jcr->acquired_resource_locks = 1;
-#endif
- return 1;
-}
-
-#ifdef USE_SEMAPHORE
-/*
- * We could not get all the resource locks because
- * too many jobs are running, so release any locks
- * we did acquire, giving others a chance to use them
- * while we wait.
- */
-static void backoff_resource_locks(JCR *jcr, int count)
-{
- P(mutex);
- switch (count) {
- case 3:
- sem_unlock(&jcr->store->sem);
- /* Fall through wanted */
- case 2:
- sem_unlock(&jcr->client->sem);
- /* Fall through wanted */
- case 1:
- sem_unlock(&jcr->job->sem);
- break;
- }
- /*
- * Since we released a lock, if there are any threads
- * waiting, wake them up so that they can try again.
- */
- if (waiting > 0) {
- pthread_cond_broadcast(&resource_wait);
- }
- V(mutex);
-}
-#endif
-
-/*
- * This is called at the end of the job to release
- * any resource limits on the number of jobs. If
- * there are any other jobs waiting, we wake them
- * up so that they can try again.
- */
-static void release_resource_locks(JCR *jcr)
-{
- if (!jcr->acquired_resource_locks) {
- return; /* Job canceled, no locks acquired */
- }
-#ifdef USE_SEMAPHORE
- P(mutex);
- sem_unlock(&jcr->store->sem);
- sem_unlock(&jcr->client->sem);
- sem_unlock(&jcr->job->sem);
- sem_unlock(&job_lock);
- if (waiting > 0) {
- pthread_cond_broadcast(&resource_wait);
- }
- jcr->acquired_resource_locks = 0;
- V(mutex);
-#endif
-}
/*
* Get or create a Client record for this Job
free_pool_memory(jcr->client_uname);
jcr->client_uname = NULL;
}
+ pthread_cond_destroy(&jcr->term_wait);
Dmsg0(200, "End dird free_jcr\n");
}
{
jcr->job = job;
jcr->JobType = job->JobType;
- jcr->JobLevel = job->level;
+ switch (jcr->JobType) {
+ case JT_ADMIN:
+ case JT_RESTORE:
+ jcr->JobLevel = L_NONE;
+ break;
+ default:
+ jcr->JobLevel = job->level;
+ break;
+ }
+ jcr->JobPriority = job->Priority;
jcr->store = job->storage;
jcr->client = job->client;
if (!jcr->client_name) {
jcr->messages = job->messages;
if (jcr->RestoreBootstrap) {
free(jcr->RestoreBootstrap);
+ jcr->RestoreBootstrap = NULL;
}
/* This can be overridden by Console program */
if (job->RestoreBootstrap) {
break;
case JT_RESTORE:
case JT_ADMIN:
- jcr->JobLevel = L_FULL;
+ jcr->JobLevel = L_NONE;
break;
default:
break;
}
}
}
-
-/*
- * Edit codes into Run command
- * %% = %
- * %c = Client's name
- * %d = Director's name
- * %i = JobId
- * %e = Job Exit
- * %j = Job
- * %l = Job Level
- * %n = Job name
- * %t = Job type
- *
- * omsg = edited output message
- * imsg = input string containing edit codes (%x)
- *
- */
-static char *edit_run_codes(JCR *jcr, char *omsg, char *imsg)
-{
- char *p;
- const char *str;
- char add[20];
-
- *omsg = 0;
- Dmsg1(200, "edit_run_codes: %s\n", imsg);
- for (p=imsg; *p; p++) {
- if (*p == '%') {
- switch (*++p) {
- case '%':
- str = "%";
- break;
- case 'c':
- str = jcr->client_name;
- if (!str) {
- str = "";
- }
- break;
- case 'd':
- str = my_name;
- break;
- case 'e':
- str = job_status_to_str(jcr->JobStatus);
- break;
- case 'i':
- sprintf(add, "%d", jcr->JobId);
- str = add;
- break;
- case 'j': /* Job */
- str = jcr->Job;
- break;
- case 'l':
- str = job_level_to_str(jcr->JobLevel);
- break;
- case 'n':
- str = jcr->job->hdr.name;
- break;
- case 't':
- str = job_type_to_str(jcr->JobType);
- break;
- default:
- add[0] = '%';
- add[1] = *p;
- add[2] = 0;
- str = add;
- break;
- }
- } else {
- add[0] = *p;
- add[1] = 0;
- str = add;
- }
- Dmsg1(200, "add_str %s\n", str);
- pm_strcat(&omsg, (char *)str);
- Dmsg1(200, "omsg=%s\n", omsg);
- }
- return omsg;
-}
projects / bacula / bacula / blobdiff