class EndPointCreator:
def __init__(self):
self.sagemaker_client = Session(profile_name="your_profile").\
client("sagemaker", region_name="ap-northeast-1")
def execute(self):
model_name = self.get_model_name()
self.create_end_point_config(model_name)
self.create_end_point()
self.wait_end_point(max_iter=120)
def get_model_name(self):
model_name = self.sagemaker_client.list_models(
NameContains=Config.base_model_name(),
SortOrder='Descending',
SortBy='CreationTime'
)["Models"][0]["ModelName"]
return model_name
def create_end_point_config(self, model_name):
self.sagemaker_client.create_endpoint_config(
EndpointConfigName=Config.endpoint_config_name(),
ProductionVariants=[
{
'VariantName': 'hoge',
'ModelName': model_name,
'InitialInstanceCount': 1,
'InstanceType': 'ml.m4.xlarge'
}
]
)
def create_end_point(self):
self.sagemaker_client.create_endpoint(
EndpointName=Config.endpoint_name(),
EndpointConfigName=Config.endpoint_config_name()
)
def wait_end_point(self, max_iter=120):
waiter = self.sagemaker_client.get_waiter("endpoint_in_service")
logging.info("polling start")
waiter.wait(
EndpointName=Config.endpoint_name(),
WaiterConfig={"MaxAttempts": max_iter}
)
logging.info("polling end")
res = self.sagemaker_client.describe_endpoint(EndpointName=Config.endpoint_name())
status = res['EndpointStatus']
if status != 'InService':
message = self.sagemaker_client.describe_endpoint(EndpointName=Config.endpoint_name())['FailureReason']
print('Training failed with the following error: {}'.format(message))
raise Exception('Endpoint creation did not succeed')
return status
