def scheduler(args):
db = whiplash.db(args.host,args.port,token=args.token)
logging.info('%s batcher scheduler connected to db', args.user)
while True:
time_window = get_times(args,db)
if time_window > 0:
make_batches(db,time_window)
time.sleep(1)
if args.test:
break
def __init__(self, server, port, username, password, use_pymongo=False):
self.log_file = "benchmark_" + str(int(time.time())) + '.dat'
self.log_handle = open(self.log_file, 'w')
self.use_pymongo = use_pymongo
if self.use_pymongo:
client = pymongo.MongoClient(server, 27017)
client.wdb.authenticate(username,
password,
mechanism='SCRAM-SHA-1')
self.wdb = client.wdb
else:
self.wdb = whiplash.db(server,
port,
username=username,
password=password,
save_token=True)
def scheduler(args):
end_time = time.time() + args.time_limit
logging.info('local scheduler started')
db = whiplash.db(args.host,args.port,token=args.token)
logging.info('local scheduler connected to db')
num_cpus = mp.cpu_count()
if args.num_cpus != None:
num_cpus = min(args.num_cpus,num_cpus)
assert num_cpus > 0
is_work = check_for_work(db, end_time)
logging.info('starting workers')
context = mp.get_context('fork')
procs = {}
for pid in range(num_cpus):
procs[pid] = context.Process(target=worker, args=(pid,db,args,end_time,is_work,))
procs[pid].start()
while True:
time.sleep(1)
n_alive = 0
for pid in procs:
is_work = check_for_work(db, end_time)
if is_work and (not procs[pid].is_alive()):
logging.info('worker %i restarting', pid)
procs[pid].join()
procs[pid] = context.Process(target=worker, args=(pid,db,args,end_time,is_work,))
procs[pid].start()
n_alive += 1
elif procs[pid].is_alive():
n_alive += 1
if n_alive == 0:
logging.info('stopping workers')
for pid in procs:
procs[pid].join()
break
logging.info('local scheduler shutting down')
def scheduler(args):
db = whiplash.db(args.host, args.port, token=args.token)
logging.info('slurm scheduler connected to db')
while True:
try:
num_pending = int(
sp.check_output(
"ssh " + args.user + "@" + args.cluster + " \'squeue -u " +
args.user +
" | grep \" PD \" | grep \"whiplash\" | wc -l\'",
shell=True))
except:
num_pending = 1
if (db.collection('work_batches').count({}) > 0) and (num_pending
== 0):
submit_job(args)
time.sleep(5)
logging.info('slurm scheduler shutting down')
#!/usr/bin/env python3
import sys, random
import whiplash
print("Login as test user")
db = whiplash.db("localhost", 1337, username="******", password="******")
print("Check the status of a query for the spin glass solver and instances")
filters = {
'input_model': {
"set": "test_set"
},
'executable': {
"name": "an_ss_ge_fi_vdeg"
},
'params': {
"n_sweeps": 100,
"n_reps": 10,
"seed": 0
},
'output_model': {}
}
print(db.status(filters))
#!/usr/bin/env python3
import sys, os, random
import whiplash
import whipbench as bench
print("Login")
db = whiplash.db(sys.argv[1], int(sys.argv[2]), username=sys.argv[3], password=sys.argv[4])
print("Reset database")
bench.reset_db(db)
print("Benchmarking collections")
sizes = [2]
numbers = [10, 100, 1000]
print("sizes:", sizes)
print("numbers:", numbers)
collections = [
['models', []],
['executables', ['name', 'description', 'algorithm', 'version', 'build', 'path', 'params']]
]
for collection, required_fields in collections:
bench.commit(db, collection, sizes, numbers, required_fields=required_fields)
bench.commit(db, collection, sizes, numbers, required_fields=required_fields)
bench.count(db, collection, sizes, numbers)
bench.query_collection(db, collection, sizes, numbers)
bench.update(db, collection, sizes, numbers)
if collection == 'models':
bench.stats(db, collection, sizes, numbers)
print("Benchmarking submission")
bench.submit(db, sizes, numbers)
#!/usr/bin/env python3
import sys, os, json, random, copy
import whiplash
print("Read inputs")
host = sys.argv[1]
port = int(sys.argv[2])
print("Login as test")
db = whiplash.db(host, port, username="******", password="******")
print("Commit models")
n_models = 10
models = []
random.seed(0)
for k in range(n_models):
N = 5
hamiltonian = []
for i in range(N):
for j in range(i + 1, N):
value = 2.0 * random.random() - 1.0
hamiltonian.append([[i, j], value])
tags = {
"n_spins": N,
"name": "test_set_" + str(k),
"description": "This is a test model",
"set": "test_set",
"format": "json"
}
model = {"content": {"edges": hamiltonian}}
model.update(tags)
def scheduler(args):
db = whiplash.db(args.host,
args.port,
username="******",
password=os.environ['WHIPLASH_ADMIN_PASSWORD'],
save_token=True)
logging.info('admin connected to db')
time_limit = 24 * 3600
schedulers = {}
while True:
all_users = get_users(args, db)
if len(all_users) > 0:
for db_user in all_users:
username = db_user['username']
if username not in schedulers:
schedulers[username] = {'batcher': th.Thread()}
if args.cluster:
schedulers[username]['slurm'] = th.Thread()
else:
schedulers[username]['local'] = th.Thread()
flags = " --user " + username + " --token " + db_user[
'token'] + " --host " + args.host + " --port " + str(
args.port) + " --log_dir " + args.log_dir
if (args.verbose):
flags += " --verbose"
if not schedulers[username]['batcher'].is_alive():
logging.info('starting batcher for user %s', username)
schedulers[username]['batcher'] = th.Thread(
target=start_batcher, args=(
args,
flags,
))
schedulers[username]['batcher'].start()
flags += " --num_cpus " + str(
args.num_cpus) + " --work_dir " + db_user[
'work_dir'] + " --time_limit " + str(time_limit)
if args.docker:
flags += " --docker"
if args.dind:
flags += " --dind"
if args.cluster:
if not schedulers[username]['slurm'].is_alive():
if check_access(db_user):
flags += " --cluster " + db_user['cluster']
logging.info(
'starting slurm scheduler for user %s',
username)
schedulers[username]['slurm'] = th.Thread(
target=start_slurm_scheduler,
args=(
args,
flags,
))
schedulers[username]['slurm'].start()
else:
logging.info('access denied for user %s', username)
else:
if not schedulers[username]['local'].is_alive():
logging.info('starting local scheduler for user %s',
username)
schedulers[username]['local'] = th.Thread(
target=start_local_scheduler, args=(
args,
flags,
))
schedulers[username]['local'].start()
else:
logging.info('local scheduler still alive')
if args.test:
break
else:
logging.info('no users found')
time.sleep(5)
logging.info('terminating user schedulers')
for username in schedulers:
schedulers[username]['batcher'].join()
if args.cluster:
schedulers[username]['slurm'].join()
else:
schedulers[username]['local'].join()
logging.info('user scheduler exiting')
sys.exit(0)