def test_send_file(self):
port = 1239
master = Master(port=port)
th = threading.Thread(target=master.run)
th.start()
worker = Worker('localhost:{}'.format(port), 1)
time.sleep(2)
tmp_dir = 'rom_files'
tmp_file = os.path.join(tmp_dir, 'pong.bin')
os.system('mkdir {}'.format(tmp_dir))
if _IS_WINDOWS:
os.system('type NUL >> {}'.format(tmp_file))
else:
os.system('touch {}'.format(tmp_file))
assert os.path.exists(tmp_file)
parl.connect('localhost:{}'.format(port), distributed_files=[tmp_file])
time.sleep(5)
actor = Actor()
for _ in range(10):
if actor.check_local_file():
break
time.sleep(10)
self.assertEqual(True, actor.check_local_file())
del actor
time.sleep(10)
worker.exit()
master.exit()
def test_sync_config_file(self):
master = Master(port=1335)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker = Worker('localhost:1335', 1)
random_file = 'random.npy'
random_array = np.random.randn(3, 5)
np.save(random_file, random_array)
random_sum = random_array.sum()
with open('config.json', 'w') as f:
config_file = {'test': 1000}
json.dump(config_file, f)
parl.connect('localhost:1335', ['random.npy', 'config.json'])
actor = Actor('random.npy', 'config.json')
time.sleep(5)
os.remove('./random.npy')
os.remove('./config.json')
remote_sum = actor.random_sum()
self.assertEqual(remote_sum, random_sum)
time.sleep(10)
remote_config = actor.read_config()
self.assertEqual(config_file['test'], remote_config)
del actor
worker.exit()
master.exit()
def test_connect_and_create_actor_in_multiprocessing_with_connected_in_main_process(
self):
# start the master
master = Master(port=8238)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker1 = Worker('localhost:8238', 4)
parl.connect('localhost:8238')
proc1 = multiprocessing.Process(target=self._connect_and_create_actor,
args=('localhost:8238', ))
proc2 = multiprocessing.Process(target=self._connect_and_create_actor,
args=('localhost:8238', ))
proc1.start()
proc2.start()
proc1.join()
proc2.join()
# make sure that the client of the main process still works
self._create_actor()
worker1.exit()
master.exit()
def test_create_actor_in_multiprocessing(self):
# start the master
master = Master(port=8240)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker1 = Worker('localhost:8240', 4)
parl.connect('localhost:8240')
if not _IS_WINDOWS: # In windows, fork process cannot access client created in main process.
proc1 = multiprocessing.Process(target=self._create_actor)
proc2 = multiprocessing.Process(target=self._create_actor)
proc1.start()
proc2.start()
proc1.join()
proc2.join()
print("[test_create_actor_in_multiprocessing] Join")
# make sure that the client of the main process still works
self._create_actor()
worker1.exit()
master.exit()
def test_max_memory(self):
port = 3001
master = Master(port=port)
th = threading.Thread(target=master.run)
th.start()
time.sleep(5)
cluster_addr = 'localhost:{}'.format(port)
worker = Worker(cluster_addr, 1)
cluster_monitor = ClusterMonitor(cluster_addr)
time.sleep(5)
parl.connect(cluster_addr)
actor = Actor()
time.sleep(20)
self.assertEqual(1, cluster_monitor.data['clients'][0]['actor_num'])
del actor
time.sleep(10)
p = Process(target=self.actor, args=(cluster_addr, ))
p.start()
for _ in range(6):
x = cluster_monitor.data['clients'][0]['actor_num']
if x == 0:
break
else:
time.sleep(10)
if x == 1:
raise ValueError("Actor max memory test failed.")
self.assertEqual(0, cluster_monitor.data['clients'][0]['actor_num'])
p.terminate()
worker.exit()
master.exit()
def test_job_exit_exceptionally(self):
master = Master(port=1334)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker1 = Worker('localhost:1334', 4)
time.sleep(10)
self.assertEqual(worker1.job_buffer.full(), True)
time.sleep(1)
self.assertEqual(master.cpu_num, 4)
print("We are going to kill all the jobs.")
if _IS_WINDOWS:
command = r'''for /F "skip=2 tokens=2 delims=," %a in ('wmic process where "commandline like '%remote\\job.py%'" get processid^,status /format:csv') do taskkill /F /T /pid %a'''
print(os.popen(command).read())
else:
command = (
"ps aux | grep remote/job.py | awk '{print $2}' | xargs kill -9"
)
subprocess.call([command], shell=True)
parl.connect('localhost:1334')
actor = Actor()
self.assertEqual(actor.add_one(1), 2)
time.sleep(20)
master.exit()
worker1.exit()
def test_actor_exception_2(self):
logger.info("running: test_actor_exception_2")
master = Master(port=8236)
th = threading.Thread(target=master.run)
th.start()
time.sleep(3)
worker1 = Worker('localhost:8236', 1)
self.assertEqual(1, master.cpu_num)
parl.connect('localhost:8236')
actor = Actor()
try:
actor.will_raise_exception_func()
except:
pass
actor2 = Actor()
for _ in range(5):
if master.cpu_num == 0:
break
time.sleep(10)
self.assertEqual(actor2.add_one(1), 2)
self.assertEqual(0, master.cpu_num)
del actor
del actor2
worker1.exit()
master.exit()
def test_actor_exception(self):
logger.info("running:test_actor_exception")
master = Master(port=8235)
th = threading.Thread(target=master.run)
th.start()
time.sleep(3)
worker1 = Worker('localhost:8235', 1)
for _ in range(3):
if master.cpu_num == 1:
break
time.sleep(10)
self.assertEqual(1, master.cpu_num)
logger.info("running:test_actor_exception: 0")
parl.connect('localhost:8235')
logger.info("running:test_actor_exception: 1")
with self.assertRaises(exceptions.RemoteError):
actor = Actor(abcd='a bug')
logger.info("running:test_actor_exception: 2")
actor2 = Actor()
for _ in range(3):
if master.cpu_num == 0:
break
time.sleep(10)
self.assertEqual(actor2.add_one(1), 2)
self.assertEqual(0, master.cpu_num)
master.exit()
worker1.exit()
def test_send_file2(self):
port = 1240
master = Master(port=port)
th = threading.Thread(target=master.run)
th.start()
worker = Worker('localhost:{}'.format(port), 1)
time.sleep(2)
tmp_file = os.path.join('rom_files', 'no_pong.bin')
self.assertRaises(Exception, parl.connect, 'localhost:{}'.format(port),
[tmp_file])
worker.exit()
master.exit()
def test_add_worker(self):
logger.info("running: test_add_worker")
master = Master(port=8234)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker1 = Worker('localhost:8234', 4)
for _ in range(3):
if master.cpu_num == 4:
break
time.sleep(10)
self.assertEqual(master.cpu_num, 4)
worker2 = Worker('localhost:8234', 4)
for _ in range(3):
if master.cpu_num == 8:
break
time.sleep(10)
self.assertEqual(master.cpu_num, 8)
worker2.exit()
for _ in range(10):
if master.cpu_num == 4:
break
time.sleep(10)
self.assertEqual(master.cpu_num, 4)
master.exit()
worker1.exit()
def test_one_worker(self):
port = 1439
master = Master(port=port)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker = Worker('localhost:{}'.format(port), 1)
cluster_monitor = ClusterMonitor('localhost:{}'.format(port))
time.sleep(1)
self.assertEqual(1, len(cluster_monitor.data['workers']))
worker.exit()
time.sleep(40)
self.assertEqual(0, len(cluster_monitor.data['workers']))
master.exit()
def test_get_attribute(self):
logger.info("running:test_get_attirbute")
master = Master(port=8507)
th = threading.Thread(target=master.run)
th.start()
time.sleep(3)
worker1 = Worker('localhost:8507', 1)
arg1 = np.random.randint(100)
arg2 = np.random.randn()
arg3 = np.random.randn(3, 3)
parl.connect('localhost:8507')
actor = Actor(arg1, arg2, arg3)
self.assertTrue(arg1 == actor.arg1)
self.assertTrue(arg2 == actor.arg2)
self.assertTrue((arg3 == actor.arg3).all())
master.exit()
worker1.exit()
def test_cluster_status(self):
port = 4321
master = Master(port=port)
th = threading.Thread(target=master.run)
th.start()
time.sleep(5)
worker = Worker('localhost:{}'.format(port), 1)
time.sleep(5)
status_info = master.cluster_monitor.get_status_info()
self.assertEqual(status_info, 'has 0 used cpus, 1 vacant cpus.')
parl.connect('localhost:{}'.format(port))
actor = Actor()
time.sleep(50)
status_info = master.cluster_monitor.get_status_info()
self.assertEqual(status_info, 'has 1 used cpus, 0 vacant cpus.')
worker.exit()
master.exit()
def test_log_server(self):
master_port = 8401
# start the master
master = Master(port=master_port)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
cluster_addr = 'localhost:{}'.format(master_port)
log_server_port = 8402
worker = Worker(cluster_addr, 4, log_server_port=log_server_port)
outputs = self._connect_and_create_actor(cluster_addr)
# Get status
status = master._get_status()
client_jobs = pickle.loads(status).get('client_jobs')
self.assertIsNotNone(client_jobs)
# Get job id
client = get_global_client()
jobs = client_jobs.get(client.client_id)
self.assertIsNotNone(jobs)
for job_id, log_server_addr in jobs.items():
log_url = "http://{}/get-log".format(log_server_addr)
# Test response without job_id
r = requests.get(log_url)
self.assertEqual(r.status_code, 400)
# Test normal response
r = requests.get(log_url, params={'job_id': job_id})
self.assertEqual(r.status_code, 200)
log_content = json.loads(r.text).get('log')
self.assertIsNotNone(log_content)
log_content = log_content.replace('\r\n', '\n')
self.assertIn(log_content, outputs)
# Test download
download_url = "http://{}/download-log".format(log_server_addr)
r = requests.get(download_url, params={'job_id': job_id})
self.assertEqual(r.status_code, 200)
log_content = r.text.replace('\r\n', '\n')
self.assertIn(log_content, outputs)
disconnect()
worker.exit()
master.exit()
def test_connect_and_create_actor_in_multiprocessing_without_connected_in_main_process(
self):
# start the master
master = Master(port=8239)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker1 = Worker('localhost:8239', 4)
proc1 = multiprocessing.Process(target=self._connect_and_create_actor,
args=('localhost:8239', ))
proc2 = multiprocessing.Process(target=self._connect_and_create_actor,
args=('localhost:8239', ))
proc1.start()
proc2.start()
proc1.join()
proc2.join()
self.assertRaises(AssertionError, self._create_actor)
worker1.exit()
master.exit()
def test_acor_exit_exceptionally(self):
port = 1337
master = Master(port)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
worker1 = Worker('localhost:{}'.format(port), 1)
file_path = __file__.replace('reset_job_test', 'simulate_client')
command = [sys.executable, file_path]
proc = subprocess.Popen(command)
for _ in range(6):
if master.cpu_num == 0:
break
else:
time.sleep(10)
self.assertEqual(master.cpu_num, 0)
proc.kill()
parl.connect('localhost:{}'.format(port))
actor = Actor()
master.exit()
worker1.exit()
disconnect()
def test_reset_actor(self):
logger.info("running: test_reset_actor")
# start the master
master = Master(port=8237)
th = threading.Thread(target=master.run)
th.start()
time.sleep(3)
worker1 = Worker('localhost:8237', 4)
parl.connect('localhost:8237')
for _ in range(10):
actor = Actor()
ret = actor.add_one(1)
self.assertEqual(ret, 2)
del actor
for _ in range(10):
if master.cpu_num == 4:
break
time.sleep(10)
self.assertEqual(master.cpu_num, 4)
worker1.exit()
master.exit()
def test_twenty_worker(self):
port = 1440
master = Master(port=port)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
workers = []
for _ in range(20):
worker = Worker('localhost:{}'.format(port), 1)
time.sleep(1)
workers.append(worker)
cluster_monitor = ClusterMonitor('localhost:{}'.format(port))
time.sleep(1)
self.assertEqual(20, len(cluster_monitor.data['workers']))
# check if the number of workers drops by 10
for i in range(10):
workers[i].exit()
check_flag = False
for _ in range(10):
if 10 == len(cluster_monitor.data['workers']):
check_flag = True
break
time.sleep(10)
self.assertTrue(check_flag)
for i in range(10, 20):
workers[i].exit()
# check if the number of workers drops to 0
check_flag = False
for _ in range(10):
if 0 == len(cluster_monitor.data['workers']):
check_flag = True
break
time.sleep(10)
self.assertTrue(check_flag)
master.exit()
from parl.remote.master import Master
from parl.remote.worker import Worker
import time
import threading
c = 10
port = 3002
if __name__ == '__main__':
master = Master(port=port)
th = threading.Thread(target=master.run)
th.setDaemon(True)
th.start()
time.sleep(5)
cluster_addr = 'localhost:{}'.format(port)
parl.connect(cluster_addr)
worker = Worker(cluster_addr, 1)
@parl.remote_class
class Actor(object):
def add(self, a, b):
return a + b + c
actor = Actor()
class TestRecursive_actor(unittest.TestCase):
def tearDown(self):
disconnect()
def test_monitor_query_log_server(self):
master_port = 8403
monitor_port = 8404
# start the master
master = Master(port=master_port, monitor_port=monitor_port)
th = threading.Thread(target=master.run)
th.start()
time.sleep(1)
# start the cluster monitor
monitor_file = __file__.replace(
os.path.join('tests', 'log_server_test.pyc'), 'monitor.py')
monitor_file = monitor_file.replace(
os.path.join('tests', 'log_server_test.py'), 'monitor.py')
command = [
sys.executable, monitor_file, "--monitor_port",
str(monitor_port), "--address", "localhost:" + str(master_port)
]
if _IS_WINDOWS:
FNULL = tempfile.TemporaryFile()
else:
FNULL = open(os.devnull, 'w')
monitor_proc = subprocess.Popen(
command,
stdout=FNULL,
stderr=subprocess.STDOUT,
)
# Start worker
cluster_addr = 'localhost:{}'.format(master_port)
log_server_port = 8405
worker = Worker(cluster_addr, 4, log_server_port=log_server_port)
# Test monitor API
outputs = self._connect_and_create_actor(cluster_addr)
time.sleep(5) # Wait for the status update
client = get_global_client()
jobs_url = "{}/get-jobs?client_id={}".format(master.monitor_url,
client.client_id)
r = requests.get(jobs_url)
self.assertEqual(r.status_code, 200)
data = json.loads(r.text)
for job in data:
log_url = job.get('log_url')
self.assertIsNotNone(log_url)
r = requests.get(log_url)
self.assertEqual(r.status_code, 200)
log_content = json.loads(r.text).get('log')
self.assertIsNotNone(log_content)
log_content = log_content.replace('\r\n', '\n')
self.assertIn(log_content, outputs)
# Test download
download_url = job.get('download_url')
r = requests.get(download_url)
self.assertEqual(r.status_code, 200)
log_content = r.text.replace('\r\n', '\n')
self.assertIn(log_content, outputs)
# Clean context
monitor_proc.kill()
monitor_proc.wait()
disconnect()
worker.exit()
master.exit()
