class TestAIFlowContext(unittest.TestCase):
def setUp(self):
if os.path.exists(_SQLITE_DB_FILE):
os.remove(_SQLITE_DB_FILE)
self.server = AIFlowServer(store_uri=_SQLITE_DB_URI, port=_PORT,
start_default_notification=False,
start_meta_service=True,
start_metric_service=False,
start_model_center_service=False,
start_scheduler_service=False)
self.server.run()
def tearDown(self):
self.server.stop()
if os.path.exists(_SQLITE_DB_FILE):
os.remove(_SQLITE_DB_FILE)
def test_init_ai_flow_context(self):
init_ai_flow_context()
project_config = current_project_config()
self.assertEqual('test_project', project_config.get_project_name())
self.assertEqual('a', project_config.get('a'))
project_context = current_project_context()
self.assertEqual('test_project', project_context.project_name)
workflow_config_ = current_workflow_config()
self.assertEqual('test_ai_flow_context', workflow_config_.workflow_name)
self.assertEqual(5, len(workflow_config_.job_configs))
class TestTensorFlowIrisModel(unittest.TestCase):
def setUp(self) -> None:
if os.path.exists(_SQLITE_DB_FILE):
os.remove(_SQLITE_DB_FILE)
self.server = AIFlowServer(store_uri=_SQLITE_DB_URI,
port=_PORT,
start_scheduler_service=False)
self.server.run()
self.client = AIFlowClient(server_uri='localhost:' + _PORT)
def tearDown(self) -> None:
self.client.stop_listen_event()
self.server.stop()
os.remove(_SQLITE_DB_FILE)
def test_save_and_load_model(self):
iris_model = fit_and_save_model()
tf_graph = tf.Graph()
registered_model = self.client.create_registered_model(
model_name='iris_model', model_desc='iris model')
self.client.create_model_version(
model_name=registered_model.model_name,
model_path=iris_model.path,
model_type=
'{"meta_graph_tags":["serve"],"signature_def_map_key":"predict"}',
version_desc='iris model')
class IrisWatcher(EventWatcher):
def process(self, notifications):
for notification in notifications:
model_path = json.loads(
notification.value).get('_model_path')
model_flavor = json.loads(
notification.value).get('_model_type')
print(json.loads(notification.value).keys())
print(model_path)
signature_def = load_tensorflow_saved_model(
model_uri=model_path,
meta_graph_tags=json.loads(model_flavor).get(
'meta_graph_tags'),
signature_def_map_key=json.loads(model_flavor).get(
'signature_def_map_key'),
tf_session=tf.Session(graph=tf_graph))
for _, input_signature in signature_def.inputs.items():
t_input = tf_graph.get_tensor_by_name(
input_signature.name)
assert t_input is not None
for _, output_signature in signature_def.outputs.items():
t_output = tf_graph.get_tensor_by_name(
output_signature.name)
assert t_output is not None
self.client.start_listen_event(key=registered_model.model_name,
watcher=IrisWatcher())
def start_aiflow_server(host, port):
port = str(port)
server_uri = host + ":" + port
server = AIFlowServer(store_uri=_SQLITE_DB_URI,
port=port,
enabled_ha=True,
start_scheduler_service=False,
ha_server_uri=server_uri,
notification_uri='localhost:30031',
start_default_notification=False)
server.run()
return server
class TestSklearnModel(unittest.TestCase):
def setUp(self) -> None:
if os.path.exists(_SQLITE_DB_FILE):
os.remove(_SQLITE_DB_FILE)
self.server = AIFlowServer(store_uri=_SQLITE_DB_URI,
port=_PORT,
start_scheduler_service=False)
self.server.run()
self.client = AIFlowClient(server_uri='localhost:' + _PORT)
def tearDown(self) -> None:
self.client.stop_listen_event()
self.server.stop()
os.remove(_SQLITE_DB_FILE)
def test_save_and_load_model(self):
knn_model = fit_model()
model_path = tempfile.mkdtemp()
if not os.path.exists(model_path):
os.makedirs(model_path)
model_path = os.path.join(model_path, 'model.pkl')
save_model(sk_model=knn_model.model,
output_path=model_path,
serialization_format=SERIALIZATION_FORMAT_PICKLE)
registered_model = self.client.create_registered_model(
model_name='knn_model', model_desc='knn model')
self.client.create_model_version(
model_name=registered_model.model_name,
model_path=model_path,
model_type='sklearn',
version_desc='knn model')
class KnnWatcher(EventWatcher):
def process(self, notifications):
for notification in notifications:
load_path = json.loads(
notification.value).get('_model_path')
reloaded_knn_model = load_scikit_learn_model(
model_uri=load_path)
numpy.testing.assert_array_equal(
knn_model.model.predict(knn_model.inference_data),
reloaded_knn_model.predict(knn_model.inference_data))
os.remove(load_path)
self.client.start_listen_event(key=registered_model.model_name,
watcher=KnnWatcher())
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
#
import os
import sys
import tempfile
sys.path.append(os.path.abspath(os.path.join(os.getcwd(), "../../../..")))
from ai_flow.endpoint.server.server import AIFlowServer
if __name__ == '__main__':
fd, temp_dbfile = tempfile.mkstemp()
os.close(fd)
db_uri = '%s%s' % ('sqlite:///', temp_dbfile)
server = AIFlowServer(store_uri=db_uri)
server.run(is_block=True)
class TestHighAvailableAIFlowServer(unittest.TestCase):
@staticmethod
def start_aiflow_server(host, port):
port = str(port)
server_uri = host + ":" + port
server = AIFlowServer(store_uri=_SQLITE_DB_URI,
port=port,
enabled_ha=True,
start_scheduler_service=False,
ha_server_uri=server_uri,
notification_uri='localhost:30031',
start_default_notification=False)
server.run()
return server
def wait_for_new_members_detected(self, new_member_uri):
while True:
living_member = self.client.living_aiflow_members
if new_member_uri in living_member:
break
else:
time.sleep(1)
def setUp(self) -> None:
SqlAlchemyStore(_SQLITE_DB_URI)
self.notification = NotificationMaster(
service=NotificationService(storage=MemoryEventStorage()),
port=30031)
self.notification.run()
self.server1 = AIFlowServer(store_uri=_SQLITE_DB_URI,
port=50051,
enabled_ha=True,
start_scheduler_service=False,
ha_server_uri='localhost:50051',
notification_uri='localhost:30031',
start_default_notification=False)
self.server1.run()
self.server2 = None
self.server3 = None
self.config = ProjectConfig()
self.config.set_enable_ha(True)
self.config.set_notification_service_uri('localhost:30031')
self.client = AIFlowClient(
server_uri='localhost:50052,localhost:50051',
project_config=self.config)
def tearDown(self) -> None:
self.client.stop_listen_event()
self.client.disable_high_availability()
if self.server1 is not None:
self.server1.stop()
if self.server2 is not None:
self.server2.stop()
if self.server3 is not None:
self.server3.stop()
if self.notification is not None:
self.notification.stop()
store = SqlAlchemyStore(_SQLITE_DB_URI)
base.metadata.drop_all(store.db_engine)
def test_server_change(self) -> None:
self.client.register_project("test_project")
projects = self.client.list_project(10, 0)
self.assertEqual(self.client.current_aiflow_uri, "localhost:50051")
self.assertEqual(projects[0].name, "test_project")
self.server2 = self.start_aiflow_server("localhost", 50052)
self.wait_for_new_members_detected("localhost:50052")
self.server1.stop()
projects = self.client.list_project(10, 0)
self.assertEqual(self.client.current_aiflow_uri, "localhost:50052")
self.assertEqual(projects[0].name, "test_project")
self.server3 = self.start_aiflow_server("localhost", 50053)
self.wait_for_new_members_detected("localhost:50053")
self.server2.stop()
projects = self.client.list_project(10, 0)
self.assertEqual(self.client.current_aiflow_uri, "localhost:50053")
self.assertEqual(projects[0].name, "test_project")
class TestSchedulerService(unittest.TestCase):
def setUp(self):
config = SchedulerServiceConfig()
config.set_scheduler_class_name(SCHEDULER_CLASS)
config.set_scheduler_config({})
if os.path.exists(_SQLITE_DB_FILE):
os.remove(_SQLITE_DB_FILE)
self.server = AIFlowServer(store_uri=_SQLITE_DB_URI,
port=_PORT,
start_default_notification=False,
start_meta_service=False,
start_metric_service=False,
start_model_center_service=False,
start_scheduler_service=True,
scheduler_service_config=config)
self.server.run()
def tearDown(self):
self.server.stop()
if os.path.exists(_SQLITE_DB_FILE):
os.remove(_SQLITE_DB_FILE)
def test_submit_workflow(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
instance.submit_workflow.return_value = WorkflowInfo(
workflow_name='test_workflow')
client = SchedulerClient("localhost:{}".format(_PORT))
with self.assertRaises(Exception) as context:
workflow = client.submit_workflow_to_scheduler(
namespace='namespace',
workflow_name='test_workflow',
workflow_json='')
self.assertTrue('workflow json is empty' in str(context.exception))
def test_pause_workflow(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.pause_workflow_scheduling.return_value = WorkflowInfo(
workflow_name='test_workflow')
client = SchedulerClient("localhost:{}".format(_PORT))
workflow = client.pause_workflow_scheduling(
namespace='namespace', workflow_name='test_workflow')
self.assertTrue('test_workflow', workflow.name)
def test_resume_workflow(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.resume_workflow_scheduling.return_value = WorkflowInfo(
workflow_name='test_workflow')
client = SchedulerClient("localhost:{}".format(_PORT))
workflow = client.resume_workflow_scheduling(
namespace='namespace', workflow_name='test_workflow')
self.assertTrue('test_workflow', workflow.name)
def test_start_new_workflow_execution(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.start_new_workflow_execution.return_value \
= WorkflowExecutionInfo(workflow_execution_id='id', status=Status.INIT)
client = SchedulerClient("localhost:{}".format(_PORT))
workflow_execution = client.start_new_workflow_execution(
namespace='namespace', workflow_name='test_workflow')
self.assertEqual('id', workflow_execution.execution_id)
self.assertEqual(StateProto.INIT,
workflow_execution.execution_state)
def test_kill_all_workflow_execution(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.stop_all_workflow_execution.return_value \
= [WorkflowExecutionInfo(workflow_execution_id='id_1', status=Status.INIT),
WorkflowExecutionInfo(workflow_execution_id='id_2', status=Status.INIT)]
client = SchedulerClient("localhost:{}".format(_PORT))
workflow_execution_list = client.kill_all_workflow_executions(
namespace='namespace', workflow_name='test_workflow')
self.assertEqual(2, len(workflow_execution_list))
def test_kill_workflow_execution(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.stop_workflow_execution.return_value \
= WorkflowExecutionInfo(workflow_execution_id='id', status=Status.RUNNING)
client = SchedulerClient("localhost:{}".format(_PORT))
workflow_execution = client.kill_workflow_execution(
execution_id='id')
self.assertEqual('id', workflow_execution.execution_id)
self.assertEqual(StateProto.RUNNING,
workflow_execution.execution_state)
def test_get_workflow_execution(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.get_workflow_execution.return_value \
= WorkflowExecutionInfo(workflow_execution_id='id', status=Status.INIT)
client = SchedulerClient("localhost:{}".format(_PORT))
workflow_execution = client.get_workflow_execution(
execution_id='id')
self.assertEqual('id', workflow_execution.execution_id)
self.assertEqual(StateProto.INIT,
workflow_execution.execution_state)
def test_list_workflow_executions(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.list_workflow_executions.return_value \
= [WorkflowExecutionInfo(workflow_execution_id='id_1', status=Status.INIT),
WorkflowExecutionInfo(workflow_execution_id='id_2', status=Status.INIT)]
client = SchedulerClient("localhost:{}".format(_PORT))
workflow_execution_list = client.list_workflow_executions(
namespace='namespace', workflow_name='test_workflow')
self.assertEqual(2, len(workflow_execution_list))
def test_start_job(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.start_job_execution.return_value \
= JobExecutionInfo(job_name='job_name',
status=Status.RUNNING,
workflow_execution=WorkflowExecutionInfo(workflow_execution_id='id',
status=Status.INIT))
client = SchedulerClient("localhost:{}".format(_PORT))
job = client.start_job(job_name='job_name', execution_id='id')
self.assertEqual('job_name', job.name)
self.assertEqual(StateProto.RUNNING, job.job_state)
self.assertEqual('id', job.workflow_execution.execution_id)
self.assertEqual(StateProto.INIT,
job.workflow_execution.execution_state)
def test_stop_job(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.stop_job_execution.return_value \
= JobExecutionInfo(job_name='job_name',
status=Status.RUNNING,
workflow_execution=WorkflowExecutionInfo(workflow_execution_id='id',
status=Status.INIT))
client = SchedulerClient("localhost:{}".format(_PORT))
job = client.stop_job(job_name='job_name', execution_id='id')
self.assertEqual('job_name', job.name)
self.assertEqual(StateProto.RUNNING, job.job_state)
self.assertEqual('id', job.workflow_execution.execution_id)
self.assertEqual(StateProto.INIT,
job.workflow_execution.execution_state)
def test_restart_job(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.restart_job_execution.return_value \
= JobExecutionInfo(job_name='job_name',
status=Status.RUNNING,
workflow_execution=WorkflowExecutionInfo(workflow_execution_id='id',
status=Status.INIT))
client = SchedulerClient("localhost:{}".format(_PORT))
job = client.restart_job(job_name='job_name', execution_id='id')
self.assertEqual('job_name', job.name)
self.assertEqual(StateProto.RUNNING, job.job_state)
self.assertEqual('id', job.workflow_execution.execution_id)
self.assertEqual(StateProto.INIT,
job.workflow_execution.execution_state)
def test_get_job(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.get_job_executions.return_value \
= [JobExecutionInfo(job_name='job_name',
status=Status.RUNNING,
workflow_execution=WorkflowExecutionInfo(workflow_execution_id='id',
status=Status.INIT))]
client = SchedulerClient("localhost:{}".format(_PORT))
job = client.get_job(job_name='job_name', execution_id='id')
self.assertEqual('job_name', job.name)
self.assertEqual(StateProto.RUNNING, job.job_state)
self.assertEqual('id', job.workflow_execution.execution_id)
self.assertEqual(StateProto.INIT,
job.workflow_execution.execution_state)
def test_list_jobs(self):
with mock.patch(SCHEDULER_CLASS) as mockScheduler:
instance = mockScheduler.return_value
self.server.scheduler_service._scheduler = instance
instance.list_job_executions.return_value \
= [JobExecutionInfo(job_name='job_name_1',
status=Status.RUNNING,
workflow_execution=WorkflowExecutionInfo(workflow_execution_id='id',
status=Status.INIT)),
JobExecutionInfo(job_name='job_name_2',
status=Status.RUNNING,
workflow_execution=WorkflowExecutionInfo(workflow_execution_id='id',
status=Status.INIT))]
client = SchedulerClient("localhost:{}".format(_PORT))
job_list = client.list_jobs(execution_id='id')
self.assertEqual(2, len(job_list))
class AIFlowServerRunner(object):
"""
AI flow server runner. This class is the runner class for the AIFlowServer. It parse the server configuration and
manage the live cycle of the AIFlowServer.
"""
def __init__(self,
config_file: Text = None,
enable_ha=False,
server_uri: str = None,
ttl_ms=10000) -> None:
"""
Set the server attribute according to the server config file.
:param config_file: server configuration file.
"""
super().__init__()
self.config_file = config_file
self.server = None
self.server_config = AIFlowServerConfig()
self.enable_ha = enable_ha
self.server_uri = server_uri
self.ttl_ms = ttl_ms
def start(self, is_block=False) -> None:
"""
Start the AI flow runner.
:param is_block: AI flow runner will run non-stop if True.
"""
if self.config_file is not None:
self.server_config.load_from_file(self.config_file)
else:
self.server_config.set_server_port(str(_PORT))
global GLOBAL_MASTER_CONFIG
GLOBAL_MASTER_CONFIG = self.server_config
logging.info("AI Flow Master Config {}".format(GLOBAL_MASTER_CONFIG))
self.server = AIFlowServer(
store_uri=self.server_config.get_db_uri(),
port=str(self.server_config.get_server_port()),
start_default_notification=self.server_config.
start_default_notification(),
notification_uri=self.server_config.get_notification_uri(),
start_meta_service=self.server_config.start_meta_service(),
start_model_center_service=self.server_config.
start_model_center_service(),
start_metric_service=self.server_config.start_metric_service(),
start_scheduler_service=self.server_config.start_scheduler_service(
),
scheduler_service_config=self.server_config.get_scheduler_config(),
enabled_ha=self.server_config.get_enable_ha(),
ha_server_uri=self.server_config.get_server_ip() + ":" +
str(self.server_config.get_server_port()),
ttl_ms=self.server_config.get_ha_ttl_ms())
self.server.run(is_block=is_block)
def stop(self, clear_sql_lite_db_file=True) -> None:
"""
Stop the AI flow runner.
:param clear_sql_lite_db_file: If True, the sqlite database files will be deleted When the server stops working.
"""
self.server.stop(clear_sql_lite_db_file)
def _clear_db(self):
self.server._clear_db()