static void jcr_timeout_check(watchdog_t *self)
{
JCR *jcr;
BSOCK *bs;
time_t timer_start;
Dmsg0(dbglvl, "Start JCR timeout checks\n");
/* Walk through all JCRs checking if any one is
* blocked for more than specified max time.
*/
foreach_jcr(jcr) {
Dmsg2(dbglvl, "jcr_timeout_check JobId=%u jcr=0x%x\n", jcr->JobId, jcr);
if (jcr->JobId == 0) {
continue;
}
bs = jcr->store_bsock;
if (bs) {
timer_start = bs->timer_start;
if (timer_start && (watchdog_time - timer_start) > bs->timeout) {
bs->timer_start = 0; /* turn off timer */
bs->set_timed_out();
Qmsg(jcr, M_ERROR, 0, _(
"Watchdog sending kill after %d secs to thread stalled reading Storage daemon.\n"),
watchdog_time - timer_start);
pthread_kill(jcr->my_thread_id, TIMEOUT_SIGNAL);
}
}
bs = jcr->file_bsock;
if (bs) {
timer_start = bs->timer_start;
if (timer_start && (watchdog_time - timer_start) > bs->timeout) {
bs->timer_start = 0; /* turn off timer */
bs->set_timed_out();
Qmsg(jcr, M_ERROR, 0, _(
"Watchdog sending kill after %d secs to thread stalled reading File daemon.\n"),
watchdog_time - timer_start);
pthread_kill(jcr->my_thread_id, TIMEOUT_SIGNAL);
}
}
bs = jcr->dir_bsock;
if (bs) {
timer_start = bs->timer_start;
if (timer_start && (watchdog_time - timer_start) > bs->timeout) {
bs->timer_start = 0; /* turn off timer */
bs->set_timed_out();
Qmsg(jcr, M_ERROR, 0, _(
"Watchdog sending kill after %d secs to thread stalled reading Director.\n"),
watchdog_time - timer_start);
pthread_kill(jcr->my_thread_id, TIMEOUT_SIGNAL);
}
}
}
endeach_jcr(jcr);
Dmsg0(dbglvl, "Finished JCR timeout checks\n");
}
/*
* Given a JobId, find the JCR
*
* Returns: jcr on success
* NULL on failure
*/
JCR *get_jcr_by_id(uint32_t JobId)
{
JCR *jcr;
foreach_jcr(jcr) {
if (jcr->JobId == JobId) {
jcr->inc_use_count();
Dmsg3(dbglvl, "Inc get_jcr jid=%u use_count=%d Job=%s\n",
jcr->JobId, jcr->use_count(), jcr->Job);
break;
}
}
endeach_jcr(jcr);
return jcr;
}
/*
* Given a SessionId and SessionTime, find the JCR
*
* Returns: jcr on success
* NULL on failure
*/
JCR *get_jcr_by_session(uint32_t SessionId, uint32_t SessionTime)
{
JCR *jcr;
foreach_jcr(jcr) {
if (jcr->VolSessionId == SessionId &&
jcr->VolSessionTime == SessionTime) {
jcr->inc_use_count();
Dmsg3(dbglvl, "Inc get_jcr jid=%u use_count=%d Job=%s\n",
jcr->JobId, jcr->use_count(), jcr->Job);
break;
}
}
endeach_jcr(jcr);
return jcr;
}
static void job_monitor_watchdog(watchdog_t *self)
{
JCR *control_jcr, *jcr;
control_jcr = (JCR *)self->data;
Dsm_check(100);
Dmsg1(800, "job_monitor_watchdog %p called\n", self);
foreach_jcr(jcr) {
bool cancel = false;
if (jcr->JobId == 0 || job_canceled(jcr) || jcr->no_maxtime) {
Dmsg2(800, "Skipping JCR=%p Job=%s\n", jcr, jcr->Job);
continue;
}
/* check MaxWaitTime */
if (job_check_maxwaittime(jcr)) {
jcr->setJobStatus(JS_Canceled);
Qmsg(jcr, M_FATAL, 0, _("Max wait time exceeded. Job canceled.\n"));
cancel = true;
/* check MaxRunTime */
} else if (job_check_maxruntime(jcr)) {
jcr->setJobStatus(JS_Canceled);
Qmsg(jcr, M_FATAL, 0, _("Max run time exceeded. Job canceled.\n"));
cancel = true;
/* check MaxRunSchedTime */
} else if (job_check_maxrunschedtime(jcr)) {
jcr->setJobStatus(JS_Canceled);
Qmsg(jcr, M_FATAL, 0, _("Max run sched time exceeded. Job canceled.\n"));
cancel = true;
}
if (cancel) {
Dmsg3(800, "Cancelling JCR %p jobid %d (%s)\n", jcr, jcr->JobId, jcr->Job);
UAContext *ua = new_ua_context(jcr);
ua->jcr = control_jcr;
cancel_job(ua, jcr);
free_ua_context(ua);
Dmsg2(800, "Have cancelled JCR %p Job=%d\n", jcr, jcr->JobId);
}
}
/* Keep reference counts correct */
endeach_jcr(jcr);
}
/*
* Given a thread id, find the JobId
*
* Returns: JobId on success
* 0 on failure
*/
uint32_t get_jobid_from_tid(pthread_t tid)
{
JCR *jcr = NULL;
bool found = false;
foreach_jcr(jcr) {
if (pthread_equal(jcr->my_thread_id, tid)) {
found = true;
break;
}
}
endeach_jcr(jcr);
if (found) {
return jcr->JobId;
}
return 0;
}
/*
* Given a Job, find the JCR requires an exact match of names.
*
* Returns: jcr on success
* NULL on failure
*/
JCR *get_jcr_by_full_name(char *Job)
{
JCR *jcr;
if (!Job) {
return NULL;
}
foreach_jcr(jcr) {
if (bstrcmp(jcr->Job, Job)) {
jcr->inc_use_count();
Dmsg3(dbglvl, "Inc get_jcr jid=%u use_count=%d Job=%s\n",
jcr->JobId, jcr->use_count(), jcr->Job);
break;
}
}
endeach_jcr(jcr);
return jcr;
}
/*
* Given a Job, find the JCR compares on the number of
* characters in Job thus allowing partial matches.
*
* Returns: jcr on success
* NULL on failure
*/
JCR *get_jcr_by_partial_name(char *Job)
{
JCR *jcr;
int len;
if (!Job) {
return NULL;
}
len = strlen(Job);
foreach_jcr(jcr) {
if (bstrncmp(Job, jcr->Job, len)) {
jcr->inc_use_count();
Dmsg3(dbglvl, "Inc get_jcr jid=%u use_count=%d Job=%s\n",
jcr->JobId, jcr->use_count(), jcr->Job);
break;
}
}
endeach_jcr(jcr);
return jcr;
}
void reload_config(int sig)
{
static bool already_here = false;
#if !defined(HAVE_WIN32)
sigset_t set;
#endif
JCR *jcr;
int njobs = 0; /* number of running jobs */
int table, rtable;
bool ok;
if (already_here) {
abort(); /* Oops, recursion -> die */
}
already_here = true;
#if !defined(HAVE_WIN32)
sigemptyset(&set);
sigaddset(&set, SIGHUP);
sigprocmask(SIG_BLOCK, &set, NULL);
#endif
lock_jobs();
LockRes();
table = find_free_reload_table_entry();
if (table < 0) {
Jmsg(NULL, M_ERROR, 0, _("Too many open reload requests. Request ignored.\n"));
goto bail_out;
}
/**
* Flush the sql connection pools.
*/
db_sql_pool_flush();
Dmsg1(100, "Reload_config njobs=%d\n", njobs);
reload_table[table].res_table = my_config->save_resources();
Dmsg1(100, "Saved old config in table %d\n", table);
ok = parse_dir_config(my_config, configfile, M_ERROR);
Dmsg0(100, "Reloaded config file\n");
if (!ok || !check_resources() || !check_catalog(UPDATE_CATALOG) || !initialize_sql_pooling()) {
rtable = find_free_reload_table_entry(); /* save new, bad table */
if (rtable < 0) {
Jmsg(NULL, M_ERROR, 0, _("Please correct configuration file: %s\n"), configfile);
Jmsg(NULL, M_ERROR_TERM, 0, _("Out of reload table entries. Giving up.\n"));
goto bail_out;
} else {
Jmsg(NULL, M_ERROR, 0, _("Please correct configuration file: %s\n"), configfile);
Jmsg(NULL, M_ERROR, 0, _("Resetting previous configuration.\n"));
}
reload_table[rtable].res_table = my_config->save_resources();
/* Now restore old resource values */
int num = my_config->m_r_last - my_config->m_r_first + 1;
RES **res_tab = reload_table[table].res_table;
for (int i=0; i<num; i++) {
my_config->m_res_head[i] = res_tab[i];
}
table = rtable; /* release new, bad, saved table below */
} else {
invalidate_schedules();
/*
* Hook all active jobs so that they release this table
*/
foreach_jcr(jcr) {
if (jcr->getJobType() != JT_SYSTEM) {
reload_table[table].job_count++;
job_end_push(jcr, reload_job_end_cb, (void *)((long int)table));
njobs++;
}
}
endeach_jcr(jcr);
}
/* Reset globals */
set_working_directory(me->working_directory);
Dmsg0(10, "Director's configuration file reread.\n");
/* Now release saved resources, if no jobs using the resources */
if (njobs == 0) {
free_saved_resources(table);
}
bail_out:
UnlockRes();
unlock_jobs();
#if !defined(HAVE_WIN32)
sigprocmask(SIG_UNBLOCK, &set, NULL);
signal(SIGHUP, reload_config);
#endif
already_here = false;
}
/*
* Check for duplicate jobs.
* Returns: true if current job should continue
* false if current job should terminate
*/
bool allow_duplicate_job(JCR *jcr)
{
JCR *djcr; /* possible duplicate job */
JOBRES *job = jcr->res.job;
bool cancel_dup = false;
bool cancel_me = false;
/*
* See if AllowDuplicateJobs is set or
* if duplicate checking is disabled for this job.
*/
if (job->AllowDuplicateJobs || jcr->IgnoreDuplicateJobChecking) {
return true;
}
Dmsg0(800, "Enter allow_duplicate_job\n");
/*
* After this point, we do not want to allow any duplicate
* job to run.
*/
foreach_jcr(djcr) {
if (jcr == djcr || djcr->JobId == 0) {
continue; /* do not cancel this job or consoles */
}
/*
* See if this Job has the IgnoreDuplicateJobChecking flag set, ignore it
* for any checking against other jobs.
*/
if (djcr->IgnoreDuplicateJobChecking) {
continue;
}
if (bstrcmp(job->name(), djcr->res.job->name())) {
if (job->DuplicateJobProximity > 0) {
utime_t now = (utime_t)time(NULL);
if ((now - djcr->start_time) > job->DuplicateJobProximity) {
continue; /* not really a duplicate */
}
}
if (job->CancelLowerLevelDuplicates &&
djcr->getJobType() == 'B' && jcr->getJobType() == 'B') {
switch (jcr->getJobLevel()) {
case L_FULL:
if (djcr->getJobLevel() == L_DIFFERENTIAL ||
djcr->getJobLevel() == L_INCREMENTAL) {
cancel_dup = true;
}
break;
case L_DIFFERENTIAL:
if (djcr->getJobLevel() == L_INCREMENTAL) {
cancel_dup = true;
}
if (djcr->getJobLevel() == L_FULL) {
cancel_me = true;
}
break;
case L_INCREMENTAL:
if (djcr->getJobLevel() == L_FULL ||
djcr->getJobLevel() == L_DIFFERENTIAL) {
cancel_me = true;
}
}
/*
* cancel_dup will be done below
*/
if (cancel_me) {
/* Zap current job */
jcr->setJobStatus(JS_Canceled);
Jmsg(jcr, M_FATAL, 0, _("JobId %d already running. Duplicate job not allowed.\n"),
djcr->JobId);
break; /* get out of foreach_jcr */
}
}
/*
* Cancel one of the two jobs (me or dup)
* If CancelQueuedDuplicates is set do so only if job is queued.
*/
if (job->CancelQueuedDuplicates) {
switch (djcr->JobStatus) {
case JS_Created:
case JS_WaitJobRes:
case JS_WaitClientRes:
case JS_WaitStoreRes:
case JS_WaitPriority:
case JS_WaitMaxJobs:
case JS_WaitStartTime:
cancel_dup = true; /* cancel queued duplicate */
break;
default:
break;
}
}
if (cancel_dup || job->CancelRunningDuplicates) {
/*
* Zap the duplicated job djcr
*/
UAContext *ua = new_ua_context(jcr);
Jmsg(jcr, M_INFO, 0, _("Cancelling duplicate JobId=%d.\n"), djcr->JobId);
cancel_job(ua, djcr);
bmicrosleep(0, 500000);
djcr->setJobStatus(JS_Canceled);
cancel_job(ua, djcr);
free_ua_context(ua);
Dmsg2(800, "Cancel dup %p JobId=%d\n", djcr, djcr->JobId);
} else {
/*
* Zap current job
*/
jcr->setJobStatus(JS_Canceled);
Jmsg(jcr, M_FATAL, 0, _("JobId %d already running. Duplicate job not allowed.\n"),
djcr->JobId);
Dmsg2(800, "Cancel me %p JobId=%d\n", jcr, jcr->JobId);
}
Dmsg4(800, "curJobId=%d use_cnt=%d dupJobId=%d use_cnt=%d\n",
jcr->JobId, jcr->use_count(), djcr->JobId, djcr->use_count());
break; /* did our work, get out of foreach loop */
}
}
endeach_jcr(djcr);
return true;
}
static void list_running_jobs_api(STATUS_PKT *sp)
{
JCR *njcr;
int len, sec, bps;
POOL_MEM msg(PM_MESSAGE);
char dt[MAX_TIME_LENGTH], b1[32], b2[32], b3[32], b4[32];
/*
* List running jobs for Bat/Bweb (simple to parse)
*/
foreach_jcr(njcr) {
bstrutime(dt, sizeof(dt), njcr->start_time);
if (njcr->JobId == 0) {
len = Mmsg(msg, "DirectorConnected=%s\n", dt);
} else {
len = Mmsg(msg, "JobId=%d\n Job=%s\n",
njcr->JobId, njcr->Job);
sendit(msg, len, sp);
#ifdef WIN32_VSS
len = Mmsg(msg," VSS=%d\n Level=%c\n JobType=%c\n JobStarted=%s\n",
(njcr->pVSSClient && njcr->pVSSClient->IsInitialized()) ? 1 : 0,
njcr->getJobLevel(), njcr->getJobType(), dt);
#else
len = Mmsg(msg," VSS=%d\n Level=%c\n JobType=%c\n JobStarted=%s\n",
0, njcr->getJobLevel(), njcr->getJobType(), dt);
#endif
}
sendit(msg, len, sp);
if (njcr->JobId == 0) {
continue;
}
sec = time(NULL) - njcr->start_time;
if (sec <= 0) {
sec = 1;
}
bps = (int)(njcr->JobBytes / sec);
len = Mmsg(msg, " Files=%s\n Bytes=%s\n Bytes/sec=%s\n Errors=%d\n"
" Bwlimit=%s\n",
edit_uint64(njcr->JobFiles, b1),
edit_uint64(njcr->JobBytes, b2),
edit_uint64(bps, b3),
njcr->JobErrors, edit_int64(njcr->max_bandwidth, b4));
sendit(msg, len, sp);
len = Mmsg(msg, " Files Examined=%s\n",
edit_uint64(njcr->num_files_examined, b1));
sendit(msg, len, sp);
if (njcr->JobFiles > 0) {
njcr->lock();
len = Mmsg(msg, " Processing file=%s\n", njcr->last_fname);
njcr->unlock();
sendit(msg, len, sp);
}
if (njcr->store_bsock) {
len = Mmsg(msg, " SDReadSeqNo=%" lld "\n fd=%d\n",
njcr->store_bsock->read_seqno, njcr->store_bsock->m_fd);
sendit(msg, len, sp);
} else {
len = Mmsg(msg, _(" SDSocket=closed\n"));
sendit(msg, len, sp);
}
}
endeach_jcr(njcr);
}
static void list_running_jobs_plain(STATUS_PKT *sp)
{
JCR *njcr;
int len, sec, bps;
bool found = false;
POOL_MEM msg(PM_MESSAGE);
char dt[MAX_TIME_LENGTH], b1[32], b2[32], b3[32], b4[32];
/*
* List running jobs
*/
Dmsg0(1000, "Begin status jcr loop.\n");
len = Mmsg(msg, _("\nRunning Jobs:\n"));
sendit(msg, len, sp);
foreach_jcr(njcr) {
bstrftime_nc(dt, sizeof(dt), njcr->start_time);
if (njcr->JobId == 0) {
len = Mmsg(msg, _("Director connected at: %s\n"), dt);
} else {
len = Mmsg(msg, _("JobId %d Job %s is running.\n"),
njcr->JobId, njcr->Job);
sendit(msg, len, sp);
#ifdef WIN32_VSS
len = Mmsg(msg, _(" %s%s %s Job started: %s\n"),
(njcr->pVSSClient && njcr->pVSSClient->IsInitialized()) ? "VSS " : "",
level_to_str(njcr->getJobLevel()),
job_type_to_str(njcr->getJobType()), dt);
#else
len = Mmsg(msg, _(" %s %s Job started: %s\n"),
level_to_str(njcr->getJobLevel()),
job_type_to_str(njcr->getJobType()), dt);
#endif
}
sendit(msg, len, sp);
if (njcr->JobId == 0) {
continue;
}
sec = time(NULL) - njcr->start_time;
if (sec <= 0) {
sec = 1;
}
bps = (int)(njcr->JobBytes / sec);
len = Mmsg(msg, _(" Files=%s Bytes=%s Bytes/sec=%s Errors=%d\n"
" Bwlimit=%s\n"),
edit_uint64_with_commas(njcr->JobFiles, b1),
edit_uint64_with_commas(njcr->JobBytes, b2),
edit_uint64_with_commas(bps, b3),
njcr->JobErrors,
edit_uint64_with_commas(njcr->max_bandwidth, b4));
sendit(msg, len, sp);
len = Mmsg(msg, _(" Files Examined=%s\n"),
edit_uint64_with_commas(njcr->num_files_examined, b1));
sendit(msg, len, sp);
if (njcr->JobFiles > 0) {
njcr->lock();
len = Mmsg(msg, _(" Processing file: %s\n"), njcr->last_fname);
njcr->unlock();
sendit(msg, len, sp);
}
found = true;
if (njcr->store_bsock) {
len = Mmsg(msg, " SDReadSeqNo=%" lld " fd=%d\n",
njcr->store_bsock->read_seqno, njcr->store_bsock->m_fd);
sendit(msg, len, sp);
} else {
len = Mmsg(msg, _(" SDSocket closed.\n"));
sendit(msg, len, sp);
}
}
endeach_jcr(njcr);
if (!found) {
len = Mmsg(msg, _("No Jobs running.\n"));
sendit(msg, len, sp);
}
len = pm_strcpy(msg, _("====\n"));
sendit(msg, len, sp);
}
