def job_sorting_simple_priority(queue, now, jids, jobs, str_config, plt):
priority_config = json.loads(str_config)
# import pdb; pdb.set_trace()
if 'WAITING_TIME_WEIGHT' in config:
waiting_time_weight = float(priority_config['WAITING_TIME_WEIGHT'])
else:
waiting_time_weight = 0.0
#
# establish job priori
#
for job in itervalues(jobs):
if 'priority' in job.types:
try:
priority = float(job.types['priority'])
except ValueError:
logger.warn("job priority failed to convert to float: " % job.types['priority'])
priority = 0.0
job.karma = priority + waiting_time_weight * float(now-job.submission_time) / float(now)
# sort jids according to jobs' karma value
# print jids
karma_ordered_jids = sorted(jids, key=lambda jid: jobs[jid].karma, reverse=True)
# print karma_ordered_jids
return karma_ordered_jids
def save_assigns_redis_0(jobs, resource_set):
if len(jobs) > 0:
r = redis.Redis()
mld_id_start_time_s = []
for j in itervalues(jobs):
mld_id_start_time_s.append({"moldable_job_id": j.moldable_id, "start_time": j.start_time})
riods = itvs2ids(j.res_set)
str_mld_id_rids = ",".join(map(lambda x: str(resource_set.rid_o2i[x]), riods))
r.set(str(j.moldable_id), str_mld_id_rids)
db.session.execute(GanttJobsPrediction.__table__.insert(), mld_id_start_time_s)
def save_assigns_bulk_0(jobs, resource_set):
if len(jobs) > 0:
mld_id_start_time_s = []
mld_id_rid_s = []
for j in itervalues(jobs):
mld_id_start_time_s.append((j.moldable_id, j.start_time))
riods = itvs2ids(j.res_set)
mld_id_rid_s.extend(
[(j.moldable_id, resource_set.rid_o2i[rid]) for rid in riods])
with db.engine.connect() as to_conn:
cursor = to_conn.connection.cursor()
pg_bulk_insert(cursor, db['gantt_jobs_predictions'], mld_id_start_time_s,
('moldable_job_id', 'start_time'), binary=True)
pg_bulk_insert(cursor, db['queues'], mld_id_rid_s,
('moldable_job_id', 'resource_id'), binary=True)
def get_jobs_types(jids, jobs):
import oar.kao.advanced_scheduling
jobs_types = {}
for j_type in db.query(JobType).filter(JobType.job_id.in_(tuple(jids))):
jid = j_type.job_id
job = jobs[jid]
t_v = j_type.type.split("=")
t = t_v[0]
if t == "timesharing":
job.ts = True
job.ts_user, job.ts_name = t_v[1].split(',')
elif t == "placeholder":
job.ph = PLACEHOLDER
job.ph_name = t_v[1]
elif t == "allow":
job.ph = ALLOW
job.ph_name = t_v[1]
elif t == "assign":
job.assign = True
raw_args = '='.join(t_v[1:])
funcname, job.assign_args, job.assign_kwargs = extract_find_assign_args(raw_args)
job.assign_func = getattr(oar.kao.advanced_scheduling, 'assign_%s' % funcname)
elif t == "find":
job.find = True
raw_args = '='.join(t_v[1:])
funcname, job.find_args, job.find_kwargs = extract_find_assign_args(raw_args)
job.find_func = getattr(oar.kao.advanced_scheduling, 'find_%s' % funcname)
else:
if len(t_v) == 2:
v = t_v[1]
else:
v = ""
if jid not in jobs_types:
jobs_types[jid] = dict()
(jobs_types[jid])[t] = v
for job in itervalues(jobs):
if job.id in jobs_types:
job.types = jobs_types[job.id]
else:
job.types = {}
def save_assigns_redis_pipeline_0(jobs, resource_set):
print('# save_assigns_redis_pipeline_0')
t = 0
if len(jobs) > 0:
r = redis.Redis()
pipe = r.pipeline()
mld_id_start_time_s = []
for j in itervalues(jobs):
t0 = time.time()
mld_id_start_time_s.append(
{'moldable_job_id': j.moldable_id, 'start_time': j.start_time})
riods = itvs2ids(j.res_set)
str_mld_id_rids = ','.join(map(lambda x: str(resource_set.rid_o2i[x]), riods))
t += (time.time() - t0)
pipe.set(str(j.moldable_id), str_mld_id_rids)
db.session.execute(
GanttJobsPrediction.__table__.insert(), mld_id_start_time_s)
pipe.execute()
print("Cumlated mld_id_start_time_s.append time:", t)
def save_assigns_bulk(jobs, resource_set):
if len(jobs) > 0:
logger.debug("nb job to save: " + str(len(jobs)))
mld_id_start_time_s = []
mld_id_rid_s = []
for j in itervalues(jobs):
logger.debug("first job_id to save: " + str(j.id))
mld_id_start_time_s.append((j.moldable_id, j.start_time))
riods = itvs2ids(j.res_set)
mld_id_rid_s.extend(
[(j.moldable_id, resource_set.rid_o2i[rid]) for rid in riods])
logger.info("save assignements")
with db.engine.connect() as to_conn:
cursor = to_conn.connection.cursor()
pg_bulk_insert(cursor, db['gantt_jobs_predictions'], mld_id_start_time_s,
('moldable_job_id', 'start_time'), binary=True)
pg_bulk_insert(cursor, db['queues'], mld_id_rid_s,
('moldable_job_id', 'resource_id'), binary=True)
def save_assigns(jobs, resource_set):
# http://docs.sqlalchemy.org/en/rel_0_9/core/dml.html#sqlalchemy.sql.expression.Insert.values
if len(jobs) > 0:
logger.debug("nb job to save: " + str(len(jobs)))
mld_id_start_time_s = []
mld_id_rid_s = []
for j in itervalues(jobs):
logger.debug("first job_id to save: " + str(j.id))
mld_id_start_time_s.append(
{'moldable_job_id': j.moldable_id, 'start_time': j.start_time})
riods = itvs2ids(j.res_set)
mld_id_rid_s.extend(
[{'moldable_job_id': j.moldable_id,
'resource_id': resource_set.rid_o2i[rid]} for rid in riods])
logger.info("save assignements")
db.session.execute(
GanttJobsPrediction.__table__.insert(), mld_id_start_time_s)
db.session.execute(GanttJobsResource.__table__.insert(), mld_id_rid_s)
db.commit()
def karma_jobs_sorting(queue, now, jids, jobs, plt):
# if "SCHEDULER_FAIRSHARING_MAX_JOB_PER_USER" in config:
# fairsharing_nb_job_limit = config["SCHEDULER_FAIRSHARING_MAX_JOB_PER_USER"]
# TODO NOT UDSED
# fairsharing_nb_job_limit = 100000
karma_window_size = 3600 * 30 * 24 # TODO in conf ???
# Set undefined config value to default one
default_config = {
"SCHEDULER_FAIRSHARING_PROJECT_TARGETS": "{default => 21.0}",
"SCHEDULER_FAIRSHARING_USER_TARGETS": "{default => 22.0}",
"SCHEDULER_FAIRSHARING_COEF_PROJECT": "0",
"SCHEDULER_FAIRSHARING_COEF_USER": "******",
"SCHEDULER_FAIRSHARING_COEF_USER_ASK": "1"
}
config.setdefault_config(default_config)
# get fairsharing config if any
karma_proj_targets = perl_hash_2_dict(
config["SCHEDULER_FAIRSHARING_PROJECT_TARGETS"])
karma_user_targets = perl_hash_2_dict(
config["SCHEDULER_FAIRSHARING_USER_TARGETS"])
karma_coeff_proj_consumption = float(
config["SCHEDULER_FAIRSHARING_COEF_PROJECT"])
karma_coeff_user_consumption = float(
config["SCHEDULER_FAIRSHARING_COEF_USER"])
karma_coeff_user_asked_consumption = float(
config["SCHEDULER_FAIRSHARING_COEF_USER_ASK"])
#
# Sort jobs accordingly to karma value (fairsharing) *)
# *)
window_start = now - karma_window_size
window_stop = now
karma_sum_time_asked, karma_sum_time_used = plt.get_sum_accounting_window(
queue, window_start, window_stop)
karma_projects_asked, karma_projects_used = plt.get_sum_accounting_by_project(
queue, window_start, window_stop)
karma_users_asked, karma_users_used = plt.get_sum_accounting_by_user(
queue, window_start, window_stop)
#
# compute karma for each job
#
for job in itervalues(jobs):
if job.project in karma_projects_used:
karma_proj_used_j = karma_projects_used[job.project]
else:
karma_proj_used_j = 0.0
if job.user in karma_users_used:
karma_user_used_j = karma_users_used[job.user]
else:
karma_user_used_j = 0.0
if job.user in karma_users_asked:
karma_user_asked_j = karma_users_asked[job.user]
else:
karma_user_asked_j = 0.0
if job.project in karma_proj_targets:
karma_proj_target = karma_proj_targets[job.project]
else:
karma_proj_target = 0.0
if job.user in karma_user_targets:
karma_user_target = karma_user_targets[job.user] / 100.0
else:
karma_user_target = 0.0
# x1 = karma_coeff_proj_consumption * ((karma_proj_used_j / karma_sum_time_used) - (karma_proj_target / 100.0))
# x2 = karma_coeff_user_consumption * ((karma_user_used_j / karma_sum_time_used) - (karma_user_target / 100.0))
# x3 = karma_coeff_user_asked_consumption * ((karma_user_asked_j / karma_sum_time_asked) - (karma_user_target / 100.0))
# print "yopypop", x1, x2, x3
job.karma = (karma_coeff_proj_consumption * ((karma_proj_used_j / karma_sum_time_used) - (karma_proj_target / 100.0)) +
karma_coeff_user_consumption * ((karma_user_used_j / karma_sum_time_used) - (karma_user_target / 100.0)) +
karma_coeff_user_asked_consumption *
((karma_user_asked_j / karma_sum_time_asked) -
(karma_user_target / 100.0))
)
# print "job.karma", job.karma
# sort jids according to jobs' karma value
# print jids
karma_ordered_jids = sorted(jids, key=lambda jid: jobs[jid].karma)
# print karma_ordered_jids
return karma_ordered_jids
