def call_external_scheduler(binpath, scheduled_jobs, all_slot_sets, resource_set, job_security_time, queue, initial_time_sec, initial_time_sql): # pragma: no cover cmd_scheduler = binpath + "schedulers/" + queue.scheduler_policy child_launched = True # TODO TO CONFIRM sched_exit_code = 0 sched_signal_num = 0 sched_dumped_core = 0 try: child = Popen([cmd_scheduler, queue.name, str( initial_time_sec), initial_time_sql], stdout=subprocess.PIPE) for line in iter(child.stdout.readline, ''): logger.debug("Read on the scheduler output:" + str(line.rstrip())) # TODO SCHEDULER_LAUNCHER_OPTIMIZATION # if # ((get_conf_with_default_param('SCHEDULER_LAUNCHER_OPTIMIZATION', # 'yes') eq 'yes') and child.wait() rc = child.returncode sched_exit_code, sched_signal_num, sched_dumped_core = rc >> 8, rc & 0x7f, bool( rc & 0x80) except OSError as e: child_launched = False logger.warn(str(e) + " Cannot run: " + cmd_scheduler + " " + queue.name + " " + str(initial_time_sec) + " " + initial_time_sql) if (not child_launched) or (sched_signal_num != 0) or (sched_dumped_core != 0): logger.error("Execution of " + queue.scheduler_policy + " failed, inactivating queue " + queue.name + " (see `oarnotify')") # stop queue db.query(Queue).filter(Queue.name == queue.name).update({'state': 'notActive'}) if sched_exit_code != 0: logger.error("Scheduler " + queue.scheduler_policy + " returned a bad value: " + str(sched_exit_code) + ". Inactivating queue " + queue.scheduler_policy + " (see `oarnotify')") # stop queue db.query(Queue).filter(Queue.name == queue.name).update({'state': 'notActive'}) # retrieve jobs and assignement decision from previous scheduling step scheduled_jobs = get_after_sched_no_AR_jobs(queue.name, resource_set, job_security_time, initial_time_sec) if scheduled_jobs != []: if queue.name == 'besteffort': filter_besteffort = False else: filter_besteffort = True set_slots_with_prev_scheduled_jobs(all_slot_sets, scheduled_jobs, job_security_time, initial_time_sec, filter_besteffort)
def call_internal_scheduler(plt, scheduled_jobs, all_slot_sets, job_security_time, queue, now): # Place running besteffort jobs if their queue is considered if queue.name == 'besteffort': set_slots_with_prev_scheduled_jobs(all_slot_sets, scheduled_jobs, job_security_time, now, False, True) internal_schedule_cycle(plt, now, all_slot_sets, job_security_time, queue.name)
def test_schedule_placeholder_prev_sched(): res = [(1, 32)] ss = SlotSet(Slot(1, 0, 0, res, 0, 1000)) all_ss = {"default": ss} hy = {"node": [[(1, 8)], [(9, 16)], [(17, 24)], [(25, 32)]]} j1 = JobPseudo( id=1, types={}, deps=[], key_cache={}, res_set=[(7, 27)], start_time=200, walltime=150, mld_res_rqts=[(1, 60, [([("node", 2)], res[:])])], ts=False, ph=1, ph_name="yop", ) j2 = JobPseudo( id=2, types={}, deps=[], key_cache={}, mld_res_rqts=[(1, 150, [([("node", 4)], res[:])])], ts=False, ph=0 ) j3 = JobPseudo( id=3, types={}, deps=[], key_cache={}, mld_res_rqts=[(1, 500, [([("node", 2)], res[:])])], ts=False, ph=2, ph_name="yop", ) # pdb.set_trace() set_slots_with_prev_scheduled_jobs(all_ss, [j1], 20) all_ss["default"].show_slots() # pdb.set_trace() schedule_id_jobs_ct(all_ss, {2: j2, 3: j3}, hy, [2, 3], 20) print("j1.start_time:", j1.start_time, "j2.start_time:", j2.start_time, " j3.start_time:", j3.start_time) all_ss["default"].show_slots() # pdb.set_trace() # assert j3.start_time == 150 assert j3.res_set == [(1, 16)]
def test_schedule_w_temporally_fragmented_container(): res = [(1, 32)] ss = SlotSet(Slot(1, 0, 0, res, 0, 5000)) all_ss = {"default": ss} hy = {"node": [[(1, 8)], [(9, 16)], [(17, 24)], [(25, 32)]]} j1 = JobPseudo( id=1, types={"container": "yop"}, deps=[], key_cache={}, res_set=[(7, 32)], start_time=200, walltime=50, ts=False, ph=0, ) j2 = JobPseudo( id=2, types={"container": "yop"}, deps=[], key_cache={}, res_set=[(15, 25)], start_time=1000, walltime=200, ts=False, ph=0, ) j3 = JobPseudo( id=3, types={"inner": "yop"}, deps=[], key_cache={}, mld_res_rqts=[(1, 100, [([("node", 1)], res[:])])], ts=False, ph=0, ) set_slots_with_prev_scheduled_jobs(all_ss, [j1, j2], 20) all_ss["yop"].show_slots() schedule_id_jobs_ct(all_ss, {3: j3}, hy, [3], 20) all_ss["yop"].show_slots() assert j3.start_time == 1000 assert j3.res_set == [(17, 24)]
def test_schedule_container_prev_sched(): res = [(1, 32)] ss = SlotSet(Slot(1, 0, 0, res, 0, 1000)) all_ss = {"default": ss} hy = {"node": [[(1, 8)], [(9, 16)], [(17, 24)], [(25, 32)]]} j1 = JobPseudo( id=1, types={"container": ""}, deps=[], key_cache={}, res_set=[(7, 27)], start_time=200, walltime=150, mld_res_rqts=[(1, 60, [([("node", 2)], res[:])])], ts=False, ph=0, ) j2 = JobPseudo( id=2, types={"inner": "1"}, deps=[], key_cache={}, res_set=[(9, 16)], start_time=210, walltime=70, mld_res_rqts=[(1, 30, [([("node", 1)], res[:])])], ts=False, ph=0, ) j3 = JobPseudo( id=3, types={"inner": "1"}, deps=[], key_cache={}, mld_res_rqts=[(1, 30, [([("node", 1)], res[:])])], ts=False, ph=0, ) set_slots_with_prev_scheduled_jobs(all_ss, [j1, j2], 20) schedule_id_jobs_ct(all_ss, {3: j3}, hy, [3], 20) assert j3.start_time == 200 assert j3.res_set == [(17, 24)]
def test_set_slots_with_prev_scheduled_jobs_1(): v = [ (1, 4, [(1, 32)]), (5, 14, [(1, 9), (21, 32)]), (15, 29, [(1, 32)]), (30, 49, [(1, 4), (16, 19), (29, 32)]), (50, 100, [(1, 32)]), ] j1 = JobPseudo(id=1, start_time=5, walltime=10, res_set=[(10, 20)], types={}, ts=False, ph=0) j2 = JobPseudo(id=2, start_time=30, walltime=20, res_set=[(5, 15), (20, 28)], types={}, ts=False, ph=0) ss = SlotSet(Slot(1, 0, 0, [(1, 32)], 1, 100)) all_ss = {"default": ss} set_slots_with_prev_scheduled_jobs(all_ss, [j1, j2], 10) assert compare_slots_val_ref(ss.slots, v) is True
def test_schedule_error_1(): # Be careful you need a deepcopy for resources constraint when declare res = [(1, 32)] ss = SlotSet(Slot(1, 0, 0, deepcopy(res), 0, 10000)) all_ss = {"default": ss} hy = {"node": [[(1, 8)], [(9, 16)], [(17, 24)], [(25, 32)]]} j2 = JobPseudo(id=2, start_time=0, walltime=30, res_set=[(1, 8)], types={}, ts=False, ph=0) set_slots_with_prev_scheduled_jobs(all_ss, [j2], 10) j4 = JobPseudo( id=4, types={}, deps=[], key_cache={}, mld_res_rqts=[(1, 60, [([("node", 2)], deepcopy(res))])], ts=False, ph=0 ) schedule_id_jobs_ct(all_ss, {4: j4}, hy, [4], 5) assert j4.res_set == [(9, 24)]
def test_quotas_four_jobs_rule_1(): config['QUOTAS'] = 'yes' # quotas.set_quotas_rules({('*', '*', '*', '/'): [1, -1, -1]}) # global quotas_rules qts.quotas_rules = {('*', '*', '*', '/'): [16, -1, -1], ('*', 'yop', '*', '*'): [-1, 1, -1]} res = [(1, 32)] rs.default_resource_itvs = deepcopy(res) ss = SlotSet(Slot(1, 0, 0, deepcopy(res), 0, 10000)) all_ss = {"default": ss} hy = {'node': [[(1, 8)], [(9, 16)], [(17, 24)], [(25, 32)]]} j1 = JobPseudo(id=1, start_time=0, walltime=20, queue='default', user='******', project='', res_set=[(9, 24)], types={}, ts=False, ph=0) j2 = JobPseudo(id=2, start_time=0, walltime=50, queue='default', user='******', project='yop', res_set=[(1, 8)]) j3 = JobPseudo(id=3, queue='default', user='******', project='') j3.simple_req(('node', 1), 10, res) j4 = JobPseudo(id=4, queue='default', user='******', project='yop') j4.simple_req(('node', 1), 60, res) set_slots_with_prev_scheduled_jobs(all_ss, [j1, j2], 5) ss.show_slots() # pdb.set_trace() schedule_id_jobs_ct(all_ss, {3: j3, 4: j4}, hy, [3, 4], 5) print(j3.start_time, j4.start_time) assert j3.start_time == 20 assert j3.res_set == [(9, 16)] assert j4.start_time == 50 assert j4.res_set == [(1, 8)]
def gantt_init_with_running_jobs(plt, initial_time_sec, job_security_time): # # Determine Global Resource Intervals and Initial Slot # resource_set = plt.resource_set() initial_slot_set = SlotSet((resource_set.roid_itvs, initial_time_sec)) logger.debug("Processing of processing of already handled reservations") accepted_ar_jids, accepted_ar_jobs = \ get_waiting_reservations_already_scheduled(resource_set, job_security_time) gantt_flush_tables(accepted_ar_jids) # TODO Can we remove this step, below ??? # why don't use: assigned_resources and job start_time ??? in get_scheduled_jobs ??? logger.debug("Processing of current jobs") current_jobs = get_jobs_in_multiple_states(['Running', 'toLaunch', 'Launching', 'Finishing', 'Suspended', 'Resuming'], resource_set) plt.save_assigns(current_jobs, resource_set) # # Resource availabilty (Available_upto field) is integrated through pseudo job # pseudo_jobs = [] for t_avail_upto in sorted(resource_set.available_upto.keys()): itvs = resource_set.available_upto[t_avail_upto] j = JobPseudo() j.start_time = t_avail_upto j.walltime = MAX_TIME - t_avail_upto j.res_set = itvs j.ts = False j.ph = NO_PLACEHOLDER pseudo_jobs.append(j) if pseudo_jobs != []: initial_slot_set.split_slots_jobs(pseudo_jobs) # # Get already scheduled jobs advanced reservations and jobs from more higher priority queues # # TODO?: Remove resources of the type specified in # SCHEDULER_AVAILABLE_SUSPENDED_RESOURCE_TYPE scheduled_jobs = plt.get_scheduled_jobs( resource_set, job_security_time, initial_time_sec) # retrieve ressources used by besteffort jobs besteffort_rid2job = {} for job in scheduled_jobs: # print("job.id:", job.id, job.queue_name, job.types, job.res_set, job.start_time) if 'besteffort' in job.types: for r_id in itvs2ids(job.res_set): besteffort_rid2job[r_id] = job # Create and fill gantt all_slot_sets = {'default': initial_slot_set} if scheduled_jobs != []: filter_besteffort = True set_slots_with_prev_scheduled_jobs(all_slot_sets, scheduled_jobs, job_security_time, initial_time_sec, filter_besteffort) return (all_slot_sets, scheduled_jobs, besteffort_rid2job)
def schedule_cycle(plt, now, queue="default"): logger.info("Begin scheduling....now: " + str(now) + ", queue: " + queue) # # Retrieve waiting jobs # waiting_jobs, waiting_jids, nb_waiting_jobs = plt.get_waiting_jobs(queue) if nb_waiting_jobs > 0: logger.info("nb_waiting_jobs:" + str(nb_waiting_jobs)) for jid in waiting_jids: logger.debug("waiting_jid: " + str(jid)) job_security_time = int(config["SCHEDULER_JOB_SECURITY_TIME"]) # # Determine Global Resource Intervals and Initial Slot # resource_set = plt.resource_set() initial_slot_set = SlotSet((resource_set.roid_itvs, now)) # # Resource availabilty (Available_upto field) is integrated through pseudo job # pseudo_jobs = [] for t_avail_upto in sorted(resource_set.available_upto.keys()): itvs = resource_set.available_upto[t_avail_upto] j = JobPseudo() # print t_avail_upto, max_time - t_avail_upto, itvs j.start_time = t_avail_upto j.walltime = MAX_TIME - t_avail_upto j.res_set = itvs j.ts = False j.ph = NO_PLACEHOLDER pseudo_jobs.append(j) if pseudo_jobs != []: initial_slot_set.split_slots_jobs(pseudo_jobs) # # Get additional waiting jobs' data # plt.get_data_jobs( waiting_jobs, waiting_jids, resource_set, job_security_time) # Job sorting (karma and advanced) waiting_ordered_jids = karma_job_sorting(queue, now, waiting_jids, waiting_jobs, plt) # # Get already scheduled jobs advanced reservations and jobs from more higher priority queues # scheduled_jobs = plt.get_scheduled_jobs(resource_set, job_security_time, now) all_slot_sets = {'default': initial_slot_set} if scheduled_jobs != []: if queue == 'besteffort': filter_besteffort = False else: filter_besteffort = True set_slots_with_prev_scheduled_jobs(all_slot_sets, scheduled_jobs, job_security_time, now, filter_besteffort) # # Scheduled # schedule_id_jobs_ct(all_slot_sets, waiting_jobs, resource_set.hierarchy, waiting_ordered_jids, job_security_time) # # Save assignement # logger.info("save assignement") plt.save_assigns(waiting_jobs, resource_set) else: logger.info("no waiting jobs")