def deploy(env, model_name, model_type, model_dir, project, url, min_replica,
max_replica, cpu_request, memory_request):
merlin.set_url(url)
target_env = merlin.get_environment(env)
resource_request = target_env.default_resource_request
if min_replica is not None:
resource_request.min_replica = int(min_replica)
if max_replica is not None:
resource_request.max_replica = int(max_replica)
if cpu_request is not None:
resource_request.cpu_request = cpu_request
if memory_request is not None:
resource_request.memory_request = memory_request
merlin.set_project(project)
merlin.set_model(model_name, ModelType(model_type))
with merlin.new_model_version() as v:
merlin.log_model(model_dir=model_dir)
try:
endpoint = merlin.deploy(v, env, resource_request)
if endpoint:
print('Model deployed to {}'.format(endpoint))
except Exception as e:
print(e)
def test_sklearn(integration_test_url, project_name, use_google_oauth):
merlin.set_url(integration_test_url, use_google_oauth=use_google_oauth)
merlin.set_project(project_name)
merlin.set_model("sklearn-sample", ModelType.SKLEARN)
model_dir = "test/sklearn-model"
MODEL_FILE = "model.joblib"
undeploy_all_version()
with merlin.new_model_version() as v:
clf = svm.SVC(gamma='scale')
iris = load_iris()
X, y = iris.data, iris.target
clf.fit(X, y)
dump(clf, os.path.join(model_dir, MODEL_FILE))
# Upload the serialized model to MLP
merlin.log_model(model_dir=model_dir)
endpoint = merlin.deploy(v)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()['predictions']) == len(request_json['instances'])
merlin.undeploy(v)
sleep(5)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 404
def test_model_version_with_labels(integration_test_url, project_name,
use_google_oauth):
merlin.set_url(integration_test_url, use_google_oauth=use_google_oauth)
merlin.set_project(project_name)
merlin.set_model("sklearn-labels", ModelType.SKLEARN)
model_dir = "test/sklearn-model"
MODEL_FILE = "model.joblib"
undeploy_all_version()
with merlin.new_model_version(labels={"model": "T-800"}) as v:
clf = svm.SVC(gamma="scale")
iris = load_iris()
X, y = iris.data, iris.target
clf.fit(X, y)
dump(clf, os.path.join(model_dir, MODEL_FILE))
# Upload the serialized model to MLP
merlin.log_model(model_dir=model_dir)
assert len(v.labels) == 1
assert v.labels["model"] == "T-800"
merlin_active_model = merlin.active_model()
all_versions = merlin_active_model.list_version(
labels={"model": ["T-800"]})
for version in all_versions:
assert version.labels["model"] == "T-800"
should_not_exist_versions = merlin_active_model.list_version(
labels={"model": ["T-1000"]})
assert len(should_not_exist_versions) == 0
def test_tensorflow(integration_test_url, project_name, use_google_oauth):
merlin.set_url(integration_test_url, use_google_oauth=use_google_oauth)
merlin.set_project(project_name)
merlin.set_model("tensorflow-sample", ModelType.TENSORFLOW)
model_dir = "test/tensorflow-model"
undeploy_all_version()
with merlin.new_model_version() as v:
merlin.log_model(model_dir=model_dir)
endpoint = merlin.deploy(v)
request_json = {
"signature_name": "predict",
"instances": [
{"sepal_length": 2.8, "sepal_width": 1.0, "petal_length": 6.8,
"petal_width": 0.4},
{"sepal_length": 0.1, "sepal_width": 0.5, "petal_length": 1.8,
"petal_width": 2.4}
]
}
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()['predictions']) == len(request_json['instances'])
merlin.undeploy(v)
sleep(5)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 404
def test_resource_request(integration_test_url, project_name,
use_google_oauth):
merlin.set_url(integration_test_url, use_google_oauth=use_google_oauth)
merlin.set_project(project_name)
merlin.set_model("resource-request", ModelType.XGBOOST)
model_dir = "test/xgboost-model"
BST_FILE = "model.bst"
envs = merlin.list_environment()
assert len(envs) >= 1
default_env = merlin.get_default_environment()
assert default_env is not None
undeploy_all_version()
with merlin.new_model_version() as v:
iris = load_iris()
y = iris['target']
X = iris['data']
dtrain = xgb.DMatrix(X, label=y)
param = {
'max_depth': 6,
'eta': 0.1,
'silent': 1,
'nthread': 4,
'num_class': 10,
'objective': 'multi:softmax'
}
xgb_model = xgb.train(params=param, dtrain=dtrain)
model_file = os.path.join(model_dir, BST_FILE)
xgb_model.save_model(model_file)
# Upload the serialized model to MLP
merlin.log_model(model_dir=model_dir)
resource_request = ResourceRequest(1, 1, "100m", "200Mi")
endpoint = merlin.deploy(v,
environment_name=default_env.name,
resource_request=resource_request)
sleep(5)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()['predictions']) == len(request_json['instances'])
merlin.undeploy(v)
sleep(5)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 404
def test_set_traffic(integration_test_url, project_name, use_google_oauth):
merlin.set_url(integration_test_url, use_google_oauth=use_google_oauth)
merlin.set_project(project_name)
merlin.set_model("set-traffic-sample", ModelType.SKLEARN)
model_dir = "test/sklearn-model"
MODEL_FILE = "model.joblib"
undeploy_all_version()
with merlin.new_model_version() as v:
clf = svm.SVC(gamma='scale')
iris = load_iris()
X, y = iris.data, iris.target
clf.fit(X, y)
dump(clf, os.path.join(model_dir, MODEL_FILE))
# Upload the serialized model to MLP
merlin.log_model(model_dir=model_dir)
endpoint = merlin.deploy(v)
sleep(5)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()['predictions']) == len(request_json['instances'])
# Undeploy deployed model version
merlin.undeploy(v)
sleep(5)
# Redeploy and set traffic
merlin.deploy(v)
endpoint = merlin.set_traffic({v: 100})
sleep(5)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()['predictions']) == len(request_json['instances'])
# Try to undeploy serving model version. It must be fail
with pytest.raises(Exception):
assert merlin.undeploy(v)
# Undeploy other running model version endpoints
undeploy_all_version()
def test_stop_serving_traffic(integration_test_url, project_name,
use_google_oauth):
merlin.set_url(integration_test_url, use_google_oauth=use_google_oauth)
merlin.set_project(project_name)
merlin.set_model("stop-serving-traffic", ModelType.SKLEARN)
model_dir = "test/sklearn-model"
MODEL_FILE = "model.joblib"
undeploy_all_version()
with merlin.new_model_version() as v:
clf = svm.SVC(gamma='scale')
iris = load_iris()
X, y = iris.data, iris.target
clf.fit(X, y)
dump(clf, os.path.join(model_dir, MODEL_FILE))
# Upload the serialized model to MLP
merlin.log_model(model_dir=model_dir)
endpoint = merlin.deploy(v)
sleep(5)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()['predictions']) == len(request_json['instances'])
model_endpoint = merlin.serve_traffic({endpoint: 100})
sleep(5)
resp = requests.post(f"{model_endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()['predictions']) == len(request_json['instances'])
merlin.stop_serving_traffic(model_endpoint.environment_name)
endpoints = merlin.list_model_endpoints()
for endpoint in endpoints:
if endpoint.environment_name == model_endpoint.environment_name:
assert endpoint.status == Status.TERMINATED
# Undeploy other running model version endpoints
undeploy_all_version()
def test_xgboost(integration_test_url, project_name, use_google_oauth):
merlin.set_url(integration_test_url, use_google_oauth=use_google_oauth)
merlin.set_project(project_name)
merlin.set_model("xgboost-sample", ModelType.XGBOOST)
model_dir = "test/xgboost-model"
BST_FILE = "model.bst"
undeploy_all_version()
with merlin.new_model_version() as v:
iris = load_iris()
y = iris["target"]
X = iris["data"]
dtrain = xgb.DMatrix(X, label=y)
param = {
"max_depth": 6,
"eta": 0.1,
"silent": 1,
"nthread": 4,
"num_class": 10,
"objective": "multi:softmax",
}
xgb_model = xgb.train(params=param, dtrain=dtrain)
model_file = os.path.join(model_dir, BST_FILE)
xgb_model.save_model(model_file)
# Upload the serialized model to MLP
merlin.log_model(model_dir=model_dir)
resource_request = ResourceRequest(1, 1, "100m", "200Mi")
endpoint = merlin.deploy(v, resource_request=resource_request)
resp = requests.post(f"{endpoint.url}", json=request_json)
assert resp.status_code == 200
assert resp.json() is not None
assert len(resp.json()["predictions"]) == len(request_json["instances"])
merlin.undeploy(v)
