void *
sge_worker_main(void *arg)
{
bool do_endlessly = true;
cl_thread_settings_t *thread_config = (cl_thread_settings_t*)arg;
sge_gdi_ctx_class_t *ctx = NULL;
monitoring_t monitor;
monitoring_t *monitorp = &monitor;
time_t next_prof_output = 0;
DENTER(TOP_LAYER, "sge_worker_main");
DPRINTF(("started"));
cl_thread_func_startup(thread_config);
sge_monitor_init(&monitor, thread_config->thread_name, GDI_EXT, MT_WARNING, MT_ERROR);
sge_qmaster_thread_init(&ctx, QMASTER, WORKER_THREAD, true);
/* register at profiling module */
set_thread_name(pthread_self(), "Worker Thread");
conf_update_thread_profiling("Worker Thread");
while (do_endlessly) {
sge_gdi_packet_class_t *packet = NULL;
/*
* Wait for packets. As long as packets are available cancelation
* of this thread is ignored. The shutdown procedure in the main
* thread takes care that packet producers will be terminated
* before all worker threads so that this won't be a problem.
*/
MONITOR_IDLE_TIME(
sge_tq_wait_for_task(Master_Task_Queue, 1, SGE_TQ_GDI_PACKET, (void *)&packet),
&monitor, mconf_get_monitor_time(), mconf_is_monitor_message());
MONITOR_SET_QLEN((monitorp), sge_tq_get_task_count(Master_Task_Queue));
if (packet != NULL) {
sge_gdi_task_class_t *task = packet->first_task;
bool is_only_read_request = true;
thread_start_stop_profiling();
#ifdef SEND_ANSWER_IN_LISTENER
#else
/*
* prepare buffer for sending an answer
*/
if (packet->is_intern_request == false && packet->is_gdi_request == true) {
init_packbuffer(&(packet->pb), 0, 0);
}
#endif
MONITOR_MESSAGES((monitorp));
if (packet->is_gdi_request == true) {
/*
* test if a write lock is necessary
*/
task = packet->first_task;
while (task != NULL) {
u_long32 command = SGE_GDI_GET_OPERATION(task->command);
if (command != SGE_GDI_GET) {
is_only_read_request = false;
break;
}
task = task->next;
}
} else {
is_only_read_request = false;
}
/*
* acquire the correct lock
*/
if (is_only_read_request) {
MONITOR_WAIT_TIME(SGE_LOCK(LOCK_GLOBAL, LOCK_READ), monitorp);
} else {
MONITOR_WAIT_TIME(SGE_LOCK(LOCK_GLOBAL, LOCK_WRITE), monitorp);
}
if (packet->is_gdi_request == true) {
/*
* do the GDI request
*/
task = packet->first_task;
while (task != NULL) {
sge_c_gdi(ctx, packet, task, &(task->answer_list), &monitor);
task = task->next;
}
} else {
task = packet->first_task;
sge_c_report(ctx, packet->host, packet->commproc, packet->commproc_id,
task->data_list, &monitor);
}
/*
* do unlock
*/
if (is_only_read_request) {
SGE_UNLOCK(LOCK_GLOBAL, LOCK_READ)
} else {
SGE_UNLOCK(LOCK_GLOBAL, LOCK_WRITE)
}
if (packet->is_gdi_request == true) {
#ifdef SEND_ANSWER_IN_LISTENER
sge_gdi_packet_broadcast_that_handled(packet);
#else
/*
* Send the answer to the client
*/
if (packet->is_intern_request == false) {
MONITOR_MESSAGES_OUT(monitorp);
sge_gdi2_send_any_request(ctx, 0, NULL,
packet->host, packet->commproc, packet->commproc_id,
&(packet->pb), TAG_GDI_REQUEST,
packet->response_id, NULL);
clear_packbuffer(&(packet->pb));
# ifdef BLOCK_LISTENER
sge_gdi_packet_broadcast_that_handled(packet);
# else
sge_gdi_packet_free(&packet);
# endif
/*
* Code only for TS:
*
* Following if-block will only be executed in testsuite if the qmaster
* parameter __TEST_SLEEP_AFTER_REQUEST is defined. This will block the
* worker thread if it handled a request. Only this makes sure that
* other worker threads can handle incoming requests. Otherwise
* it might be possible that one worker threads handles all requests
* on fast qmaster hosts if testsuite is not fast enough to generate
* gdi requests.
*/
if (mconf_get_enable_test_sleep_after_request() == true) {
sleep(5);
}
} else {
sge_gdi_packet_broadcast_that_handled(packet);
/* this is an internal request, packet will get destroyed later,
* where the caller waits for the answer
* make sure it is no longer accessed here
*/
packet = NULL;
}
#endif
} else {
sge_gdi_packet_free(&packet);
}
thread_output_profiling("worker thread profiling summary:\n",
&next_prof_output);
sge_monitor_output(&monitor);
} else {
void *
sge_event_master_main(void *arg)
{
bool do_endlessly = true;
cl_thread_settings_t *thread_config = (cl_thread_settings_t*)arg;
sge_gdi_ctx_class_t *ctx = NULL;
monitoring_t monitor;
monitoring_t *monitorp = &monitor;
lListElem *report = NULL;
lList *report_list = NULL;
time_t next_prof_output = 0;
DENTER(TOP_LAYER, "sge_event_master_main");
DPRINTF(("started"));
cl_thread_func_startup(thread_config);
sge_monitor_init(&monitor, thread_config->thread_name, EDT_EXT, EMT_WARNING, EMT_ERROR);
sge_qmaster_thread_init(&ctx, QMASTER, DELIVERER_THREAD, true);
/* register at profiling module */
set_thread_name(pthread_self(), "Deliver Thread");
conf_update_thread_profiling("Deliver Thread");
report_list = lCreateListHash("report list", REP_Type, false);
report = lCreateElem(REP_Type);
lSetUlong(report, REP_type, NUM_REP_REPORT_EVENTS);
lSetHost(report, REP_host, ctx->get_qualified_hostname(ctx));
lAppendElem(report_list, report);
while (do_endlessly) {
int execute = 0;
thread_start_stop_profiling();
/*
* did a new event arrive which has a flush time of 0 seconds?
*/
MONITOR_IDLE_TIME(sge_event_master_wait_next(), (&monitor), mconf_get_monitor_time(),
mconf_is_monitor_message());
MONITOR_MESSAGES((monitorp));
MONITOR_EDT_COUNT((&monitor));
MONITOR_CLIENT_COUNT((&monitor), lGetNumberOfElem(Event_Master_Control.clients));
sge_event_master_process_requests(&monitor);
sge_event_master_send_events(ctx, report, report_list, &monitor);
sge_monitor_output(&monitor);
thread_output_profiling("event master thread profiling summary:\n",
&next_prof_output);
/* pthread cancelation point */
pthread_cleanup_push((void (*)(void *))sge_event_master_cleanup_monitor,
(void *)&monitor);
pthread_cleanup_push((void (*)(void *))sge_event_master_cleanup_report_list,
(void *)&report_list);
cl_thread_func_testcancel(thread_config);
pthread_cleanup_pop(execute);
pthread_cleanup_pop(execute);
if (sge_thread_has_shutdown_started()) {
DPRINTF(("waiting for termination\n"));
sleep(1);
}
}
/*
* Don't add cleanup code here. It will never be executed. Instead register
* a cleanup function with pthread_cleanup_push()/pthread_cleanup_pop() before
* and after the call of cl_thread_func_testcancel()
*/
DRETURN(NULL);
}
/****** qmaster/threads/sge_scheduler_main() **********************************
* NAME
* sge_scheduler_main() -- main function of the scheduler thread
*
* SYNOPSIS
* void * sge_scheduler_main(void *arg)
*
* FUNCTION
* Main function of the scheduler thread,
*
* INPUTS
* void *arg - pointer to the thread function (type cl_thread_settings_t*)
*
* RESULT
* void * - always NULL
*
* NOTES
* MT-NOTE: sge_scheduler_main() is MT safe
*
* MT-NOTE: this is a thread function. Do NOT use this function
* MT-NOTE: in any other way!
*
* SEE ALSO
* qmaster/threads/sge_scheduler_initialize()
* qmaster/threads/sge_scheduler_cleanup_thread()
* qmaster/threads/sge_scheduler_terminate()
* qmaster/threads/sge_scheduler_main()
*******************************************************************************/
void *
sge_scheduler_main(void *arg)
{
time_t next_prof_output = 0;
monitoring_t monitor;
sge_gdi_ctx_class_t *ctx = NULL;
sge_evc_class_t *evc = NULL;
lList *alp = NULL;
sge_where_what_t where_what;
cl_thread_settings_t *thread_config = (cl_thread_settings_t*)arg;
bool do_shutdown = false;
bool do_endlessly = true;
bool local_ret = true;
DENTER(TOP_LAYER, "sge_scheduler_main");
memset(&where_what, 0, sizeof(where_what));
/*
* startup
*/
if (local_ret) {
/* initialize commlib thread */
cl_thread_func_startup(thread_config);
/* initialize monitoring */
sge_monitor_init(&monitor, thread_config->thread_name, SCH_EXT, SCT_WARNING, SCT_ERROR);
sge_qmaster_thread_init(&ctx, SCHEDD, SCHEDD_THREAD, true);
/* register at profiling module */
set_thread_name(pthread_self(), "Scheduler Thread");
conf_update_thread_profiling("Scheduler Thread");
DPRINTF((SFN" started\n", thread_config->thread_name));
/* initialize schedd_runnlog logging */
schedd_set_schedd_log_file(ctx);
}
/* set profiling parameters */
prof_set_level_name(SGE_PROF_EVENTMASTER, NULL, NULL);
prof_set_level_name(SGE_PROF_SPOOLING, NULL, NULL);
prof_set_level_name(SGE_PROF_CUSTOM0, "scheduler", NULL);
prof_set_level_name(SGE_PROF_CUSTOM1, "pending ticket calculation", NULL);
prof_set_level_name(SGE_PROF_CUSTOM3, "job sorting", NULL);
prof_set_level_name(SGE_PROF_CUSTOM4, "job dispatching", NULL);
prof_set_level_name(SGE_PROF_CUSTOM5, "send orders", NULL);
prof_set_level_name(SGE_PROF_CUSTOM6, "scheduler event loop", NULL);
prof_set_level_name(SGE_PROF_CUSTOM7, "copy lists", NULL);
prof_set_level_name(SGE_PROF_SCHEDLIB4, NULL, NULL);
/* set-up needed for 'schedule' file */
serf_init(schedd_serf_record_func, schedd_serf_newline);
schedd_set_serf_log_file(ctx);
/*
* prepare event client/mirror mechanism
*/
if (local_ret) {
local_ret = sge_gdi2_evc_setup(&evc, ctx, EV_ID_SCHEDD, &alp, "scheduler");
DPRINTF(("prepared event client/mirror mechanism\n"));
}
/*
* register as event mirror
*/
if (local_ret) {
sge_mirror_initialize(evc, EV_ID_SCHEDD, "scheduler",
false, &event_update_func, &sge_mod_event_client,
&sge_add_event_client, &sge_remove_event_client,
&sge_handle_event_ack);
evc->ec_register(evc, false, NULL, &monitor);
evc->ec_set_busy_handling(evc, EV_BUSY_UNTIL_RELEASED);
DPRINTF(("registered at event mirror\n"));
}
/*
* subscribe necessary data
*/
if (local_ret) {
ensure_valid_what_and_where(&where_what);
subscribe_scheduler(evc, &where_what);
DPRINTF(("subscribed necessary data from event master\n"));
}
/*
* schedulers main loop
*/
if (local_ret) {
while (do_endlessly) {
bool handled_events = false;
lList *event_list = NULL;
int execute = 0;
double prof_copy = 0.0;
double prof_total = 0.0;
double prof_init = 0.0;
double prof_free = 0.0;
double prof_run = 0.0;
lList *orders = NULL;
if (sconf_get_profiling()) {
prof_start(SGE_PROF_OTHER, NULL);
prof_start(SGE_PROF_PACKING, NULL);
prof_start(SGE_PROF_EVENTCLIENT, NULL);
prof_start(SGE_PROF_MIRROR, NULL);
prof_start(SGE_PROF_GDI, NULL);
prof_start(SGE_PROF_HT_RESIZE, NULL);
prof_start(SGE_PROF_CUSTOM0, NULL);
prof_start(SGE_PROF_CUSTOM1, NULL);
prof_start(SGE_PROF_CUSTOM3, NULL);
prof_start(SGE_PROF_CUSTOM4, NULL);
prof_start(SGE_PROF_CUSTOM5, NULL);
prof_start(SGE_PROF_CUSTOM6, NULL);
prof_start(SGE_PROF_CUSTOM7, NULL);
prof_start(SGE_PROF_SCHEDLIB4, NULL);
} else {
prof_stop(SGE_PROF_OTHER, NULL);
prof_stop(SGE_PROF_PACKING, NULL);
prof_stop(SGE_PROF_EVENTCLIENT, NULL);
prof_stop(SGE_PROF_MIRROR, NULL);
prof_stop(SGE_PROF_GDI, NULL);
prof_stop(SGE_PROF_HT_RESIZE, NULL);
prof_stop(SGE_PROF_CUSTOM0, NULL);
prof_stop(SGE_PROF_CUSTOM1, NULL);
prof_stop(SGE_PROF_CUSTOM3, NULL);
prof_stop(SGE_PROF_CUSTOM4, NULL);
prof_stop(SGE_PROF_CUSTOM5, NULL);
prof_stop(SGE_PROF_CUSTOM6, NULL);
prof_stop(SGE_PROF_CUSTOM7, NULL);
prof_stop(SGE_PROF_SCHEDLIB4, NULL);
}
/*
* Wait for new events
*/
MONITOR_IDLE_TIME(sge_scheduler_wait_for_event(evc, &event_list), (&monitor), mconf_get_monitor_time(),
mconf_is_monitor_message());
/* If we lost connection we have to register again */
if (evc->ec_need_new_registration(evc)) {
lFreeList(&event_list);
if (evc->ec_register(evc, false, NULL, &monitor) == true) {
DPRINTF(("re-registered at event master!\n"));
}
}
if (event_list != NULL) {
/* check for shutdown */
do_shutdown = (lGetElemUlong(event_list, ET_type, sgeE_SHUTDOWN) != NULL) ? true : false;
/* update mirror and free data */
if (do_shutdown == false && sge_mirror_process_event_list(evc, event_list) == SGE_EM_OK) {
handled_events = true;
DPRINTF(("events handled\n"));
} else {
DPRINTF(("events contain shutdown event - ignoring events\n"));
}
lFreeList(&event_list);
}
/* if we actually got events, start the scheduling run and further event processing */
if (handled_events == true) {
lList *answer_list = NULL;
scheduler_all_data_t copy;
lList *master_cqueue_list = *(object_type_get_master_list(SGE_TYPE_CQUEUE));
lList *master_job_list = *object_type_get_master_list(SGE_TYPE_JOB);
lList *master_userset_list = *object_type_get_master_list(SGE_TYPE_USERSET);
lList *master_project_list = *object_type_get_master_list(SGE_TYPE_PROJECT);
lList *master_exechost_list= *object_type_get_master_list(SGE_TYPE_EXECHOST);
lList *master_rqs_list= *object_type_get_master_list(SGE_TYPE_RQS);
lList *master_centry_list = *object_type_get_master_list(SGE_TYPE_CENTRY);
lList *master_ckpt_list = *object_type_get_master_list(SGE_TYPE_CKPT);
lList *master_user_list = *object_type_get_master_list(SGE_TYPE_USER);
lList *master_ar_list = *object_type_get_master_list(SGE_TYPE_AR);
lList *master_pe_list = *object_type_get_master_list(SGE_TYPE_PE);
lList *master_hgrp_list = *object_type_get_master_list(SGE_TYPE_HGROUP);
lList *master_sharetree_list = *object_type_get_master_list(SGE_TYPE_SHARETREE);
/* delay scheduling for test purposes, see issue GE-3306 */
if (SGE_TEST_DELAY_SCHEDULING > 0) {
sleep(SGE_TEST_DELAY_SCHEDULING);
}
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM6);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
if (__CONDITION(INFOPRINT)) {
dstring ds;
char buffer[128];
sge_dstring_init(&ds, buffer, sizeof(buffer));
DPRINTF(("================[SCHEDULING-EPOCH %s]==================\n",
sge_at_time(0, &ds)));
sge_dstring_free(&ds);
}
/*
* If there were new events then
* copy/filter data necessary for the scheduler run
* and run the scheduler method
*/
memset(©, 0, sizeof(copy));
copy.dept_list = lSelect("", master_userset_list, where_what.where_dept, where_what.what_acldept);
copy.acl_list = lSelect("", master_userset_list, where_what.where_acl, where_what.what_acldept);
DPRINTF(("RAW CQ:%d, J:%d, H:%d, C:%d, A:%d, D:%d, P:%d, CKPT:%d,"
" US:%d, PR:%d, RQS:%d, AR:%d, S:nd:%d/lf:%d\n",
lGetNumberOfElem(master_cqueue_list),
lGetNumberOfElem(master_job_list),
lGetNumberOfElem(master_exechost_list),
lGetNumberOfElem(master_centry_list),
lGetNumberOfElem(copy.acl_list),
lGetNumberOfElem(copy.dept_list),
lGetNumberOfElem(master_project_list),
lGetNumberOfElem(master_ckpt_list),
lGetNumberOfElem(master_user_list),
lGetNumberOfElem(master_project_list),
lGetNumberOfElem(master_rqs_list),
lGetNumberOfElem(master_ar_list),
lGetNumberOfNodes(NULL, master_sharetree_list, STN_children),
lGetNumberOfLeafs(NULL, master_sharetree_list, STN_children)
));
sge_rebuild_job_category(master_job_list, master_userset_list,
master_project_list, master_rqs_list);
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_init = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
sge_before_dispatch(evc);
/* prepare data for the scheduler itself */
copy.host_list = lCopyList("", master_exechost_list);
/*
* Within the scheduler we do only need QIs
*/
{
lListElem *cqueue = NULL;
lEnumeration *what_queue3 = NULL;
for_each(cqueue, master_cqueue_list) {
lList *qinstance_list = lGetList(cqueue, CQ_qinstances);
lList *t;
if (!qinstance_list) {
continue;
}
/* all_queue_list contains all queue instances with state and full queue name only */
if (!what_queue3) {
what_queue3 = lWhat("%T(%I%I)", lGetListDescr(qinstance_list), QU_full_name, QU_state);
}
t = lSelect("t", qinstance_list, NULL, what_queue3);
if (t) {
if (copy.all_queue_list == NULL) {
copy.all_queue_list = lCreateList("all", lGetListDescr(t));
}
lAppendList(copy.all_queue_list, t);
lFreeList (&t);
}
t = lSelect("t", qinstance_list, where_what.where_queue, where_what.what_queue2);
if (t) {
if (copy.queue_list == NULL) {
copy.queue_list = lCreateList("enabled", lGetListDescr(t));
}
lAppendList(copy.queue_list, t);
lFreeList (&t);
}
t = lSelect("t", qinstance_list, where_what.where_queue2, where_what.what_queue2);
if (t) {
if (copy.dis_queue_list == NULL) {
copy.dis_queue_list = lCreateList("disabled", lGetListDescr(t));
}
lAppendList(copy.dis_queue_list, t);
lFreeList (&t);
}
}
if (what_queue3) {
lFreeWhat(&what_queue3);
}
}
if (sconf_is_job_category_filtering()) {
copy.job_list = sge_category_job_copy(copy.queue_list, &orders, evc->monitor_next_run);
} else {
copy.job_list = lCopyList("", master_job_list);
}
/* no need to copy these lists, they are read only used */
copy.centry_list = master_centry_list;
copy.ckpt_list = master_ckpt_list;
copy.hgrp_list = master_hgrp_list;
/* these lists need to be copied because they are modified during scheduling run */
copy.share_tree = lCopyList("", master_sharetree_list);
copy.pe_list = lCopyList("", master_pe_list);
copy.user_list = lCopyList("", master_user_list);
copy.project_list = lCopyList("", master_project_list);
copy.rqs_list = lCopyList("", master_rqs_list);
copy.ar_list = lCopyList("", master_ar_list);
/* report number of reduced and raw (in brackets) lists */
DPRINTF(("Q:%d, AQ:%d J:%d(%d), H:%d(%d), C:%d, A:%d, D:%d, P:%d, CKPT:%d,"
" US:%d, PR:%d, RQS:%d, AR:%d, S:nd:%d/lf:%d \n",
lGetNumberOfElem(copy.queue_list),
lGetNumberOfElem(copy.all_queue_list),
lGetNumberOfElem(copy.job_list),
lGetNumberOfElem(master_job_list),
lGetNumberOfElem(copy.host_list),
lGetNumberOfElem(master_exechost_list),
lGetNumberOfElem(copy.centry_list),
lGetNumberOfElem(copy.acl_list),
lGetNumberOfElem(copy.dept_list),
lGetNumberOfElem(copy.pe_list),
lGetNumberOfElem(copy.ckpt_list),
lGetNumberOfElem(copy.user_list),
lGetNumberOfElem(copy.project_list),
lGetNumberOfElem(copy.rqs_list),
lGetNumberOfElem(copy.ar_list),
lGetNumberOfNodes(NULL, copy.share_tree, STN_children),
lGetNumberOfLeafs(NULL, copy.share_tree, STN_children)
));
if (getenv("SGE_ND")) {
printf("Q:%d, AQ:%d J:%d(%d), H:%d(%d), C:%d, A:%d, D:%d, "
"P:%d, CKPT:%d, US:%d, PR:%d, RQS:%d, AR:%d, S:nd:%d/lf:%d \n",
lGetNumberOfElem(copy.queue_list),
lGetNumberOfElem(copy.all_queue_list),
lGetNumberOfElem(copy.job_list),
lGetNumberOfElem(master_job_list),
lGetNumberOfElem(copy.host_list),
lGetNumberOfElem(master_exechost_list),
lGetNumberOfElem(copy.centry_list),
lGetNumberOfElem(copy.acl_list),
lGetNumberOfElem(copy.dept_list),
lGetNumberOfElem(copy.pe_list),
lGetNumberOfElem(copy.ckpt_list),
lGetNumberOfElem(copy.user_list),
lGetNumberOfElem(copy.project_list),
lGetNumberOfElem(copy.rqs_list),
lGetNumberOfElem(copy.ar_list),
lGetNumberOfNodes(NULL, copy.share_tree, STN_children),
lGetNumberOfLeafs(NULL, copy.share_tree, STN_children)
);
} else {
schedd_log("-------------START-SCHEDULER-RUN-------------", NULL, evc->monitor_next_run);
}
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_copy = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
scheduler_method(evc, &answer_list, ©, &orders);
answer_list_output(&answer_list);
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_run = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
/* .. which gets deleted after using */
lFreeList(&(copy.host_list));
lFreeList(&(copy.queue_list));
lFreeList(&(copy.dis_queue_list));
lFreeList(&(copy.all_queue_list));
lFreeList(&(copy.job_list));
lFreeList(&(copy.acl_list));
lFreeList(&(copy.dept_list));
lFreeList(&(copy.pe_list));
lFreeList(&(copy.share_tree));
lFreeList(&(copy.user_list));
lFreeList(&(copy.project_list));
lFreeList(&(copy.rqs_list));
lFreeList(&(copy.ar_list));
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_free = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
/*
* need to sync with event master thread
* if schedd configuration changed then settings in evm can be adjusted
*/
if (sconf_is_new_config()) {
/* set scheduler interval / event delivery interval */
u_long32 interval = sconf_get_schedule_interval();
if (evc->ec_get_edtime(evc) != interval) {
evc->ec_set_edtime(evc, interval);
}
/* set job / ja_task event flushing */
set_job_flushing(evc);
/* no need to ec_commit here - we do it when resetting the busy state */
/* now we handled the new schedd config - no need to do it twice */
sconf_reset_new_config();
}
/* block till master handled all GDI orders */
sge_schedd_block_until_orders_processed(evc->get_gdi_ctx(evc), NULL);
schedd_order_destroy();
/*
* Stop profiling for "schedd run total" and the subcategories
*/
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM6);
prof_total = prof_get_measurement_wallclock(SGE_PROF_CUSTOM6, true, NULL);
if (prof_is_active(SGE_PROF_CUSTOM6)) {
PROFILING((SGE_EVENT, "PROF: schedd run took: %.3f s (init: %.3f s, copy: %.3f s, "
"run:%.3f, free: %.3f s, jobs: %d, categories: %d/%d)",
prof_total, prof_init, prof_copy, prof_run, prof_free,
lGetNumberOfElem(*object_type_get_master_list(SGE_TYPE_JOB)), sge_category_count(),
sge_cs_category_count() ));
}
if (getenv("SGE_ND") != NULL) {
printf("--------------STOP-SCHEDULER-RUN-------------\n");
} else {
schedd_log("--------------STOP-SCHEDULER-RUN-------------", NULL, evc->monitor_next_run);
}
thread_output_profiling("scheduler thread profiling summary:\n", &next_prof_output);
sge_monitor_output(&monitor);
}
/* reset the busy state */
evc->ec_set_busy(evc, 0);
evc->ec_commit(evc, NULL);
/* stop logging into schedd_runlog (enabled via -tsm) */
evc->monitor_next_run = false;
/*
* pthread cancelation point
*
* sge_scheduler_cleanup_thread() is the last function which should
* be called so it is pushed first
*/
pthread_cleanup_push(sge_scheduler_cleanup_thread, (void *) &ctx);
pthread_cleanup_push((void (*)(void *))sge_scheduler_cleanup_monitor,
(void *)&monitor);
pthread_cleanup_push((void (*)(void *))sge_scheduler_cleanup_event_client,
(void *)evc);
cl_thread_func_testcancel(thread_config);
pthread_cleanup_pop(execute);
pthread_cleanup_pop(execute);
pthread_cleanup_pop(execute);
DPRINTF(("passed cancelation point\n"));
}
/****** qmaster/sge_qmaster_main/signal_thread() *******************************
* NAME
* signal_thread() -- signal thread function
*
* SYNOPSIS
* void* signal_thread(void* anArg)
*
* FUNCTION
* Signal handling thread function. Establish recognized signal set. Enter
* signal wait loop. Wait for signal. Handle signal.
*
* If signal is 'SIGINT' or 'SIGTERM', kick-off shutdown and invalidate
* signal thread.
*
* NOTE: The signal thread will terminate on return of this function.
*
* INPUTS
* void* anArg - not used
*
* RESULT
* void* - none
*
* NOTES
* MT-NOTE: signal_thread() is a thread function. Do NOT use this function
* MT-NOTE: in any other way!
*
*******************************************************************************/
void* sge_signaler_main(void* arg)
{
cl_thread_settings_t *thread_config = (cl_thread_settings_t*)arg;
bool is_continue = true;
sigset_t sig_set;
int sig_num;
time_t next_prof_output = 0;
monitoring_t monitor;
sge_gdi_ctx_class_t *ctx = NULL;
DENTER(TOP_LAYER, "sge_signaler_main");
cl_thread_func_startup(thread_config);
sge_monitor_init(&monitor, thread_config->thread_name, NONE_EXT, ST_WARNING, ST_ERROR);
sge_qmaster_thread_init(&ctx, QMASTER, SIGNALER_THREAD, true);
sigemptyset(&sig_set);
sigaddset(&sig_set, SIGINT);
sigaddset(&sig_set, SIGTERM);
/* register at profiling module */
set_thread_name(pthread_self(), "Signal Thread");
conf_update_thread_profiling("Signal Thread");
while (is_continue) {
/*
* Wait for signals
*/
sigwait(&sig_set, &sig_num);
thread_start_stop_profiling();
DPRINTF(("got signal %d\n", sig_num));
switch (sig_num) {
case SIGINT:
case SIGTERM:
/*
* Notify the main thread. It is waiting and will shutdown all other
* threads if it gets signalled
*/
sge_thread_notify_all_waiting();
#if defined(SOLARIS)
if (sge_smf_used() == 1) {
/* We don't do disable on svcadm restart */
if (sge_strnullcmp(sge_smf_get_instance_state(), SCF_STATE_STRING_ONLINE) == 0 &&
sge_strnullcmp(sge_smf_get_instance_next_state(), SCF_STATE_STRING_NONE) == 0) {
sge_smf_temporary_disable_instance();
}
}
#endif
/*
* Now this thread can exit. The main thread does the remaining
* shutdown activities
*/
is_continue = false;
break;
default:
ERROR((SGE_EVENT, MSG_QMASTER_UNEXPECTED_SIGNAL_I, sig_num));
}
sge_monitor_output(&monitor);
thread_output_profiling("signal thread profiling summary:\n",
&next_prof_output);
}
sge_monitor_free(&monitor);
DRETURN(NULL);
}