def get_arn_role_from_current_aws_user():
sts_client = boto3.client("sts")
identity = sts_client.get_caller_identity()
sts_arn = identity["Arn"]
sts_arn_list = sts_arn.split(":")
type_role = sts_arn_list[-1].split("/")
iam_client = boto3.client("iam")
if type_role[0] in ("user", "root"):
role_list = iam_client.list_roles()
arn = None
for role in role_list["Roles"]:
policy_document = role["AssumeRolePolicyDocument"]
statement = policy_document["Statement"][0]
if (
statement["Effect"] == "Allow"
and statement["Principal"]["Service"] == "sagemaker.amazonaws.com"
):
arn = role["Arn"]
if arn is None:
raise BentoMLDeploymentException(
"Can't find proper Arn role for Sagemaker, please create one and try "
"again"
)
return arn
elif type_role[0] == "role":
role_response = iam_client.get_role(RoleName=type_role[1])
return role_response["Role"]["Arn"]
raise BentoMLDeploymentException(
"Not supported role type {}; sts arn is {}".format(type_role[0], sts_arn)
)
def apply_deployment(deployment_info, yatai_service=None):
if yatai_service is None:
from bentoml.yatai import get_yatai_service
yatai_service = get_yatai_service()
try:
if isinstance(deployment_info, dict):
deployment_pb = deployment_dict_to_pb(deployment_info)
elif isinstance(deployment_info, str):
deployment_pb = deployment_yaml_string_to_pb(deployment_info)
else:
raise BentoMLDeploymentException(
'Unexpected argument type, expect deployment info to be str in yaml '
'format or a dict, instead got: {}'.format(str(type(deployment_info)))
)
validation_errors = validate_deployment_pb_schema(deployment_pb)
if validation_errors:
return ApplyDeploymentResponse(
status=Status.INVALID_ARGUMENT(
'Failed to validate deployment: {errors}'.format(
errors=validation_errors
)
)
)
return yatai_service.ApplyDeployment(
ApplyDeploymentRequest(deployment=deployment_pb)
)
except BentoMLException as error:
return ApplyDeploymentResponse(status=Status.INTERNAL(str(error)))
def apply(self, deployment_pb, yatai_service, prev_deployment=None):
try:
ensure_docker_available_or_raise()
deployment_spec = deployment_pb.spec
sagemaker_config = deployment_spec.sagemaker_operator_config
if sagemaker_config is None:
raise BentoMLDeploymentException('Sagemaker configuration is missing.')
bento_pb = yatai_service.GetBento(
GetBentoRequest(
bento_name=deployment_spec.bento_name,
bento_version=deployment_spec.bento_version,
)
)
if bento_pb.bento.uri.type not in (BentoUri.LOCAL, BentoUri.S3):
raise BentoMLException(
'BentoML currently not support {} repository'.format(
bento_pb.bento.uri.type
)
)
return self._apply(
deployment_pb,
bento_pb,
yatai_service,
bento_pb.bento.uri.uri,
prev_deployment,
)
except BentoMLException as error:
return ApplyDeploymentResponse(status=exception_to_return_status(error))
def describe(self, deployment_pb, repo=None):
deployment_spec = deployment_pb.spec
sagemaker_config = deployment_spec.sagemaker_operator_config
if sagemaker_config is None:
raise BentoMLDeploymentException(
'Sagemaker configuration is missing.')
sagemaker_client = boto3.client('sagemaker', sagemaker_config.region)
endpoint_name = generate_aws_compatible_string(
deployment_pb.namespace + '-' + deployment_spec.bento_name)
try:
endpoint_status_response = sagemaker_client.describe_endpoint(
EndpointName=endpoint_name)
except ClientError as e:
status = _parse_aws_client_exception_or_raise(e)
status.error_message = (
'Failed to describe SageMaker deployment: %s',
status.error_message,
)
return DescribeDeploymentResponse(status=status)
logger.debug("AWS describe endpoint response: %s",
endpoint_status_response)
endpoint_status = endpoint_status_response["EndpointStatus"]
service_state = ENDPOINT_STATUS_TO_STATE[endpoint_status]
deployment_state = DeploymentState(
state=service_state,
info_json=json.dumps(endpoint_status_response, default=str),
)
return DescribeDeploymentResponse(state=deployment_state,
status=Status.OK())
def delete(self, name, namespace):
with create_session(self.sess_maker) as sess:
try:
deployment = (sess.query(Deployment).filter_by(
name=name, namespace=namespace).one())
return sess.delete(deployment)
except NoResultFound:
raise BentoMLDeploymentException(
"Deployment '%s' in namespace: '%s' is not found" % name,
namespace)
def delete(self, deployment_pb, yatai_service=None):
try:
deployment_spec = deployment_pb.spec
sagemaker_config = deployment_spec.sagemaker_operator_config
if sagemaker_config is None:
raise BentoMLDeploymentException('Sagemaker configuration is missing.')
sagemaker_client = boto3.client('sagemaker', sagemaker_config.region)
endpoint_name = generate_aws_compatible_string(
deployment_pb.namespace + '-' + deployment_spec.bento_name
)
try:
delete_endpoint_response = sagemaker_client.delete_endpoint(
EndpointName=endpoint_name
)
logger.debug(
"AWS delete endpoint response: %s", delete_endpoint_response
)
except ClientError as e:
status = _parse_aws_client_exception_or_raise(e)
status.error_message = 'Failed to delete SageMaker endpoint: {}'.format(
status.error_message
)
return DeleteDeploymentResponse(status=status)
delete_config_error = _cleanup_sagemaker_endpoint_config(
client=sagemaker_client,
name=deployment_spec.bento_name,
version=deployment_spec.bento_version,
)
if delete_config_error:
delete_config_error.error_message = (
'Failed to delete SageMaker endpoint config: %s',
delete_config_error.error_message,
)
return DeleteDeploymentResponse(status=delete_config_error)
delete_model_error = _cleanup_sagemaker_model(
client=sagemaker_client,
name=deployment_spec.bento_name,
version=deployment_spec.bento_version,
)
if delete_model_error:
delete_model_error.error_message = (
'Failed to delete SageMaker model: %s',
delete_model_error.error_message,
)
return DeleteDeploymentResponse(status=delete_model_error)
return DeleteDeploymentResponse(status=Status.OK())
except BentoMLException as error:
return DeleteDeploymentResponse(status=exception_to_return_status(error))
def get_deployment_operator(deployment_pb):
operator = deployment_pb.spec.operator
if operator == DeploymentOperator.AWS_SAGEMAKER:
from bentoml.deployment.sagemaker import SageMakerDeploymentOperator
return SageMakerDeploymentOperator()
elif operator == DeploymentOperator.AWS_LAMBDA:
pass
elif operator == DeploymentOperator.GCP_FUNCTION:
pass
elif operator == DeploymentOperator.KUBERNETES:
pass
elif operator == DeploymentOperator.CUSTOM:
pass
else:
raise BentoMLDeploymentException("DeployOperator must be set")
def delete(self, deployment_pb, repo=None):
deployment_spec = deployment_pb.spec
sagemaker_config = deployment_spec.sagemaker_operator_config
if sagemaker_config is None:
raise BentoMLDeploymentException('Sagemaker configuration is missing.')
sagemaker_client = boto3.client('sagemaker', sagemaker_config.region)
endpoint_name = generate_aws_compatible_string(
deployment_pb.namespace + '-' + deployment_spec.bento_name
)
delete_endpoint_response = sagemaker_client.delete_endpoint(
EndpointName=endpoint_name
)
logger.debug("AWS delete endpoint response: %s", delete_endpoint_response)
if delete_endpoint_response["ResponseMetadata"]["HTTPStatusCode"] == 200:
# We will also try to delete both model and endpoint configuration for user.
# Since they are not critical, even they failed, we will still count delete
# deployment a success action
model_name = create_sagemaker_model_name(
deployment_spec.bento_name, deployment_spec.bento_version
)
delete_model_response = sagemaker_client.delete_model(ModelName=model_name)
logger.debug("AWS delete model response: %s", delete_model_response)
if delete_model_response["ResponseMetadata"]["HTTPStatusCode"] != 200:
logger.error(
"Encounter error when deleting model: %s", delete_model_response
)
endpoint_config_name = create_sagemaker_endpoint_config_name(
deployment_spec.bento_name, deployment_spec.bento_version
)
delete_endpoint_config_response = sagemaker_client.delete_endpoint_config( # noqa: E501
EndpointConfigName=endpoint_config_name
)
logger.debug(
"AWS delete endpoint config response: %s",
delete_endpoint_config_response,
)
return DeleteDeploymentResponse(status=Status.OK())
else:
return DeleteDeploymentResponse(
status=Status.INTERNAL(str(delete_endpoint_response))
)
def get_deployment_operator(deployment_pb):
operator = deployment_pb.spec.operator
if operator == deployment_pb2.AWS_SAGEMAKER:
from bentoml.deployment.sagemaker import SageMakerDeploymentOperator
return SageMakerDeploymentOperator()
elif operator == deployment_pb2.AWS_LAMBDA:
pass
elif operator == deployment_pb2.GCP_FUNCTION:
raise NotImplementedError(
"GCP function deployment operator is not implemented")
elif operator == deployment_pb2.KUBERNETES:
raise NotImplementedError(
"Kubernetes deployment operator is not implemented")
elif operator == deployment_pb2.CUSTOM:
raise NotImplementedError(
"Custom deployment operator is not implemented")
else:
raise BentoMLDeploymentException("DeployOperator must be set")
def get_deployment_operator(deployment_pb):
operator = deployment_pb.spec.operator
if operator == DeploymentSpec.AWS_SAGEMAKER:
from bentoml.deployment.sagemaker import SageMakerDeploymentOperator
return SageMakerDeploymentOperator()
elif operator == DeploymentSpec.AWS_LAMBDA:
from bentoml.deployment.serverless.aws_lambda import AwsLambdaDeploymentOperator
return AwsLambdaDeploymentOperator()
elif operator == DeploymentSpec.GCP_FUNCTION:
from bentoml.deployment.serverless.gcp_function import (
GcpFunctionDeploymentOperator, )
return GcpFunctionDeploymentOperator()
elif operator == DeploymentSpec.KUBERNETES:
raise NotImplementedError(
"Kubernetes deployment operator is not implemented")
elif operator == DeploymentSpec.CUSTOM:
raise NotImplementedError(
"Custom deployment operator is not implemented")
else:
raise BentoMLDeploymentException("DeployOperator must be set")
def apply(self, deployment_pb, repo, prev_deployment=None):
deployment_spec = deployment_pb.spec
sagemaker_config = deployment_spec.sagemaker_operator_config
if sagemaker_config is None:
raise BentoMLDeploymentException('Sagemaker configuration is missing.')
archive_path = repo.get(
deployment_spec.bento_name, deployment_spec.bento_version
)
# config = load_bentoml_config(bento_path)...
sagemaker_client = boto3.client('sagemaker', sagemaker_config.region)
with TemporarySageMakerContent(
archive_path, deployment_spec.bento_name, deployment_spec.bento_version
) as temp_path:
ecr_image_path = create_push_image_to_ecr(
deployment_spec.bento_name, deployment_spec.bento_version, temp_path
)
execution_role_arn = get_arn_role_from_current_user()
model_name = create_sagemaker_model_name(
deployment_spec.bento_name, deployment_spec.bento_version
)
sagemaker_model_info = {
"ModelName": model_name,
"PrimaryContainer": {
"ContainerHostname": model_name,
"Image": ecr_image_path,
"Environment": {
"API_NAME": sagemaker_config.api_name,
"BENTO_SERVER_TIMEOUT": config().get(
'apiserver', 'default_timeout'
),
"BENTO_SERVER_WORKERS": config().get(
'apiserver', 'default_gunicorn_workers_count'
),
},
},
"ExecutionRoleArn": execution_role_arn,
}
logger.info("Creating sagemaker model %s", model_name)
create_model_response = sagemaker_client.create_model(**sagemaker_model_info)
logger.debug("AWS create model response: %s", create_model_response)
production_variants = [
{
"VariantName": generate_aws_compatible_string(
deployment_spec.bento_name
),
"ModelName": model_name,
"InitialInstanceCount": sagemaker_config.instance_count,
"InstanceType": sagemaker_config.instance_type,
}
]
endpoint_config_name = create_sagemaker_endpoint_config_name(
deployment_spec.bento_name, deployment_spec.bento_version
)
logger.info(
"Creating Sagemaker endpoint %s configuration", endpoint_config_name
)
create_endpoint_config_response = sagemaker_client.create_endpoint_config(
EndpointConfigName=endpoint_config_name,
ProductionVariants=production_variants,
)
logger.debug(
"AWS create endpoint config response: %s", create_endpoint_config_response
)
endpoint_name = generate_aws_compatible_string(
deployment_pb.namespace + '-' + deployment_spec.bento_name
)
if prev_deployment:
logger.info("Updating sagemaker endpoint %s", endpoint_name)
update_endpoint_response = sagemaker_client.update_endpoint(
EndpointName=endpoint_name, EndpointConfigName=endpoint_config_name
)
logger.debug("AWS update endpoint response: %s", update_endpoint_response)
else:
logger.info("Creating sagemaker endpoint %s", endpoint_name)
create_endpoint_response = sagemaker_client.create_endpoint(
EndpointName=endpoint_name, EndpointConfigName=endpoint_config_name
)
logger.debug("AWS create endpoint response: %s", create_endpoint_response)
res_deployment_pb = Deployment(state=DeploymentState())
res_deployment_pb.CopyFrom(deployment_pb)
return ApplyDeploymentResponse(status=Status.OK(), deployment=res_deployment_pb)
def apply(self, deployment_pb, yatai_service, prev_deployment=None):
try:
ensure_docker_available_or_raise()
deployment_spec = deployment_pb.spec
sagemaker_config = deployment_spec.sagemaker_operator_config
if sagemaker_config is None:
raise BentoMLDeploymentException('Sagemaker configuration is missing.')
bento_pb = yatai_service.GetBento(
GetBentoRequest(
bento_name=deployment_spec.bento_name,
bento_version=deployment_spec.bento_version,
)
)
if bento_pb.bento.uri.type != BentoUri.LOCAL:
raise BentoMLException(
'BentoML currently only support local repository'
)
else:
bento_path = bento_pb.bento.uri.uri
ensure_deploy_api_name_exists_in_bento(
[api.name for api in bento_pb.bento.bento_service_metadata.apis],
[sagemaker_config.api_name],
)
sagemaker_client = boto3.client('sagemaker', sagemaker_config.region)
with TempDirectory() as temp_dir:
sagemaker_project_dir = os.path.jon(
temp_dir, deployment_spec.bento_name
)
init_sagemaker_project(sagemaker_project_dir, bento_path)
ecr_image_path = create_push_docker_image_to_ecr(
deployment_spec.bento_name,
deployment_spec.bento_version,
sagemaker_project_dir,
)
execution_role_arn = get_arn_role_from_current_aws_user()
model_name = create_sagemaker_model_name(
deployment_spec.bento_name, deployment_spec.bento_version
)
sagemaker_model_info = {
"ModelName": model_name,
"PrimaryContainer": {
"ContainerHostname": model_name,
"Image": ecr_image_path,
"Environment": {
"API_NAME": sagemaker_config.api_name,
"BENTO_SERVER_TIMEOUT": config().get(
'apiserver', 'default_timeout'
),
"BENTO_SERVER_WORKERS": config().get(
'apiserver', 'default_gunicorn_workers_count'
),
},
},
"ExecutionRoleArn": execution_role_arn,
}
logger.info("Creating sagemaker model %s", model_name)
try:
create_model_response = sagemaker_client.create_model(
**sagemaker_model_info
)
logger.debug("AWS create model response: %s", create_model_response)
except ClientError as e:
status = _parse_aws_client_exception_or_raise(e)
status.error_message = (
'Failed to create model for SageMaker Deployment: %s',
status.error_message,
)
return ApplyDeploymentResponse(status=status, deployment=deployment_pb)
production_variants = [
{
"VariantName": generate_aws_compatible_string(
deployment_spec.bento_name
),
"ModelName": model_name,
"InitialInstanceCount": sagemaker_config.instance_count,
"InstanceType": sagemaker_config.instance_type,
}
]
endpoint_config_name = create_sagemaker_endpoint_config_name(
deployment_spec.bento_name, deployment_spec.bento_version
)
logger.info(
"Creating Sagemaker endpoint %s configuration", endpoint_config_name
)
try:
create_config_response = sagemaker_client.create_endpoint_config(
EndpointConfigName=endpoint_config_name,
ProductionVariants=production_variants,
)
logger.debug(
"AWS create endpoint config response: %s", create_config_response
)
except ClientError as e:
# create endpoint failed, will remove previously created model
cleanup_model_error = _cleanup_sagemaker_model(
sagemaker_client,
deployment_spec.bento_name,
deployment_spec.bento_version,
)
if cleanup_model_error:
cleanup_model_error.error_message = (
'Failed to clean up model after unsuccessfully '
'create endpoint config: %s',
cleanup_model_error.error_message,
)
return ApplyDeploymentResponse(
status=cleanup_model_error, deployment=deployment_pb
)
status = _parse_aws_client_exception_or_raise(e)
status.error_message = (
'Failed to create endpoint config for SageMaker deployment: %s',
status.error_message,
)
return ApplyDeploymentResponse(status=status, deployment=deployment_pb)
endpoint_name = generate_aws_compatible_string(
deployment_pb.namespace + '-' + deployment_spec.bento_name
)
try:
if prev_deployment:
logger.debug("Updating sagemaker endpoint %s", endpoint_name)
update_endpoint_response = sagemaker_client.update_endpoint(
EndpointName=endpoint_name,
EndpointConfigName=endpoint_config_name,
)
logger.debug(
"AWS update endpoint response: %s", update_endpoint_response
)
else:
logger.debug("Creating sagemaker endpoint %s", endpoint_name)
create_endpoint_response = sagemaker_client.create_endpoint(
EndpointName=endpoint_name,
EndpointConfigName=endpoint_config_name,
)
logger.debug(
"AWS create endpoint response: %s", create_endpoint_response
)
except ClientError as e:
# create/update endpoint failed, will remove previously created config
# and then remove the model
cleanup_endpoint_config_error = _cleanup_sagemaker_endpoint_config(
client=sagemaker_client,
name=deployment_spec.bento_name,
version=deployment_spec.bento_version,
)
if cleanup_endpoint_config_error:
cleanup_endpoint_config_error.error_message = (
'Failed to clean up endpoint config after unsuccessfully '
'apply SageMaker deployment: %s',
cleanup_endpoint_config_error.error_message,
)
return ApplyDeploymentResponse(
status=cleanup_endpoint_config_error, deployment=deployment_pb
)
cleanup_model_error = _cleanup_sagemaker_model(
client=sagemaker_client,
name=deployment_spec.bento_name,
version=deployment_spec.bento_version,
)
if cleanup_model_error:
cleanup_model_error.error_message = (
'Failed to clean up model after unsuccessfully '
'apply SageMaker deployment: %s',
cleanup_model_error.error_message,
)
return ApplyDeploymentResponse(
status=cleanup_model_error, deployment=deployment_pb
)
status = _parse_aws_client_exception_or_raise(e)
status.error_message = (
'Failed to apply SageMaker deployment: %s',
status.error_message,
)
return ApplyDeploymentResponse(status=status, deployment=deployment_pb)
res_deployment_pb = Deployment(state=DeploymentState())
res_deployment_pb.CopyFrom(deployment_pb)
return ApplyDeploymentResponse(
status=Status.OK(), deployment=res_deployment_pb
)
except BentoMLException as error:
return ApplyDeploymentResponse(status=exception_to_return_status(error))
def create_deployment(
deployment_name,
namespace,
bento_name,
bento_version,
platform,
operator_spec,
labels=None,
annotations=None,
yatai_service=None,
):
if yatai_service is None:
from bentoml.yatai import get_yatai_service
yatai_service = get_yatai_service()
try:
# Make sure there is no active deployment with the same deployment name
get_deployment_pb = yatai_service.GetDeployment(
GetDeploymentRequest(deployment_name=deployment_name, namespace=namespace)
)
if get_deployment_pb.status.status_code == status_pb2.Status.OK:
raise BentoMLDeploymentException(
'Deployment "{name}" already existed, use Update or Apply for updating'
'existing deployment, or create the deployment with a different name or'
'under a different deployment namespace'.format(name=deployment_name)
)
if get_deployment_pb.status.status_code != status_pb2.Status.NOT_FOUND:
raise BentoMLDeploymentException(
'Failed accesing YataiService deployment store. {error_code}:'
'{error_message}'.format(
error_code=Status.Name(get_deployment_pb.status.status_code),
error_message=get_deployment_pb.status.error_message,
)
)
deployment_dict = {
"name": deployment_name,
"namespace": namespace or config().get('deployment', 'default_namespace'),
"labels": labels,
"annotations": annotations,
"spec": {
"bento_name": bento_name,
"bento_version": bento_version,
"operator": platform,
},
}
operator = platform.replace('-', '_').upper()
try:
operator_value = DeploymentSpec.DeploymentOperator.Value(operator)
except ValueError:
return ApplyDeploymentResponse(
status=Status.INVALID_ARGUMENT('Invalid platform "{}"'.format(platform))
)
if operator_value == DeploymentSpec.AWS_SAGEMAKER:
deployment_dict['spec']['sagemaker_operator_config'] = {
'region': operator_spec.get('region')
or config().get('aws', 'default_region'),
'instance_count': operator_spec.get('instance_count')
or config().getint('sagemaker', 'default_instance_count'),
'instance_type': operator_spec.get('instance_type')
or config().get('sagemaker', 'default_instance_type'),
'api_name': operator_spec.get('api_name', ''),
}
elif operator_value == DeploymentSpec.AWS_LAMBDA:
deployment_dict['spec']['aws_lambda_operator_config'] = {
'region': operator_spec.get('region')
or config().get('aws', 'default_region')
}
if operator_spec.get('api_name'):
deployment_dict['spec']['aws_lambda_operator_config'][
'api_name'
] = operator_spec['api_name']
elif operator_value == DeploymentSpec.GCP_FCUNTION:
deployment_dict['spec']['gcp_function_operatorConfig'] = {
'region': operator_spec.get('region')
or config().get('google-cloud', 'default_region')
}
if operator_spec.get('api_name'):
deployment_dict['spec']['gcp_function_operator_config'][
'api_name'
] = operator_spec['api_name']
elif operator_value == DeploymentSpec.KUBERNETES:
deployment_dict['spec']['kubernetes_operator_config'] = {
'kube_namespace': operator_spec.get('kube_namespace', ''),
'replicas': operator_spec.get('replicas', 0),
'service_name': operator_spec.get('service_name', ''),
'service_type': operator_spec.get('service_type', ''),
}
else:
raise BentoMLDeploymentException(
'Platform "{}" is not supported in the current version of '
'BentoML'.format(platform)
)
return apply_deployment(deployment_dict, yatai_service)
except BentoMLException as error:
return ApplyDeploymentResponse(status=Status.INTERNAL(str(error)))
def create(
name,
bento,
platform,
output,
namespace,
labels,
annotations,
region,
instance_type,
instance_count,
api_name,
kube_namespace,
replicas,
service_name,
service_type,
wait,
):
# converting platform parameter to DeploymentOperator name in proto
# e.g. 'aws-lambda' to 'AWS_LAMBDA'
platform = platform.replace('-', '_').upper()
operator = DeploymentSpec.DeploymentOperator.Value(platform)
track_cli('deploy-create', platform)
yatai_service = get_yatai_service()
# Make sure there is no active deployment with the same deployment name
get_deployment = yatai_service.GetDeployment(
GetDeploymentRequest(deployment_name=name, namespace=namespace))
if get_deployment.status.status_code != Status.NOT_FOUND:
raise BentoMLDeploymentException(
'Deployment {name} already existed, please use update or apply command'
' instead'.format(name=name))
if operator == DeploymentSpec.AWS_SAGEMAKER:
if not api_name:
raise click.BadParameter(
'api-name is required for Sagemaker deployment')
sagemaker_operator_config = DeploymentSpec.SageMakerOperatorConfig(
region=region or config().get('aws', 'default_region'),
instance_count=instance_count
or config().getint('sagemaker', 'instance_count'),
instance_type=instance_type
or config().get('sagemaker', 'instance_type'),
api_name=api_name,
)
spec = DeploymentSpec(
sagemaker_operator_config=sagemaker_operator_config)
elif operator == DeploymentSpec.AWS_LAMBDA:
aws_lambda_operator_config = DeploymentSpec.AwsLambdaOperatorConfig(
region=region or config().get('aws', 'default_region'))
if api_name:
aws_lambda_operator_config.api_name = api_name
spec = DeploymentSpec(
aws_lambda_operator_config=aws_lambda_operator_config)
elif operator == DeploymentSpec.GCP_FUNCTION:
gcp_function_operator_config = DeploymentSpec.GcpFunctionOperatorConfig(
region=region
or config().get('google-cloud', 'default_region'))
if api_name:
gcp_function_operator_config.api_name = api_name
spec = DeploymentSpec(
gcp_function_operator_config=gcp_function_operator_config)
elif operator == DeploymentSpec.KUBERNETES:
kubernetes_operator_config = DeploymentSpec.KubernetesOperatorConfig(
kube_namespace=kube_namespace,
replicas=replicas,
service_name=service_name,
service_type=service_type,
)
spec = DeploymentSpec(
kubernetes_operator_config=kubernetes_operator_config)
else:
raise BentoMLDeploymentException(
'Custom deployment is not supported in the current version of BentoML'
)
bento_name, bento_version = bento.split(':')
spec.bento_name = bento_name
spec.bento_version = bento_version
spec.operator = operator
result = yatai_service.ApplyDeployment(
ApplyDeploymentRequest(deployment=Deployment(
namespace=namespace,
name=name,
annotations=parse_key_value_pairs(annotations),
labels=parse_key_value_pairs(labels),
spec=spec,
)))
if result.status.status_code != Status.OK:
_echo(
'Failed to create deployment {name}. {error_code}: '
'{error_message}'.format(
name=name,
error_code=Status.Code.Name(result.status.status_code),
error_message=result.status.error_message,
),
CLI_COLOR_ERROR,
)
else:
if wait:
result_state = get_state_after_await_action_complete(
yatai_service=yatai_service,
name=name,
namespace=namespace,
message='Creating deployment ',
)
result.deployment.state.CopyFrom(result_state.state)
_echo('Successfully created deployment {}'.format(name),
CLI_COLOR_SUCCESS)
display_deployment_info(result.deployment, output)
def apply(
bento,
deployment_name,
platform,
output,
namespace,
labels,
annotations,
region,
stage,
instance_type,
instance_count,
api_name,
kube_namespace,
replicas,
service_name,
service_type,
):
track_cli('deploy-apply', platform)
bento_name, bento_verison = bento.split(':')
spec = DeploymentSpec(
bento_name=bento_name,
bento_verison=bento_verison,
operator=get_deployment_operator_type(platform),
)
if platform == 'aws_sagemaker':
spec.sagemaker_operator_config = DeploymentSpec.SageMakerOperatorConfig(
region=region,
instance_count=instance_count,
instance_type=instance_type,
api_name=api_name,
)
elif platform == 'aws_lambda':
spec.aws_lambda_operator_config = DeploymentSpec.AwsLambdaOperatorConfig(
region=region, stage=stage)
elif platform == 'gcp_function':
spec.gcp_function_operator_config = \
DeploymentSpec.GcpFunctionOperatorConfig(
region=region, stage=stage
)
elif platform == 'kubernetes':
spec.kubernetes_operator_config = DeploymentSpec.KubernetesOperatorConfig(
kube_namespace=kube_namespace,
replicas=replicas,
service_name=service_name,
service_type=service_type,
)
else:
raise BentoMLDeploymentException(
'Custom deployment is not supported in current version of BentoML'
)
result = get_yatai_service().ApplyDeployment(
ApplyDeploymentRequest(deployment=Deployment(
namespace=namespace,
name=deployment_name,
annotations=parse_key_value_pairs(annotations),
labels=parse_key_value_pairs(labels),
spec=spec,
)))
if result.status.status_code != Status.OK:
_echo(
'Failed to apply deployment {name}. code: {error_code}, message: '
'{error_message}'.format(
name=deployment_name,
error_code=Status.Code.Name(result.status.status_code),
error_message=result.status.error_message,
),
CLI_COLOR_ERROR,
)
else:
_echo(
'Successfully apply deployment {}'.format(deployment_name),
CLI_COLOR_SUCCESS,
)
display_deployment_info(result.deployment, output)
def deployment_dict_to_pb(deployment_dict):
deployment_pb = Deployment()
if deployment_dict.get('name') is not None:
deployment_pb.name = deployment_dict.get('name')
if deployment_dict.get('namespace') is not None:
deployment_pb.namespace = deployment_dict.get('namespace')
if deployment_dict.get('labels') is not None:
deployment_pb.labels.update(deployment_dict.get('labels'))
if deployment_dict.get('annotations') is not None:
deployment_pb.annotations.update(deployment_dict.get('annotations'))
if deployment_dict.get('spec'):
spec_dict = deployment_dict.get('spec')
else:
raise BentoMLDeploymentException(
'"spec" is required field for deployment')
platform = spec_dict.get('operator')
if platform is not None:
# converting platform parameter to DeploymentOperator name in proto
# e.g. 'aws-lambda' to 'AWS_LAMBDA'
deployment_pb.spec.operator = DeploymentSpec.DeploymentOperator.Value(
platform.replace('-', '_').upper())
if spec_dict.get('bento_name'):
deployment_pb.spec.bento_name = spec_dict.get('bento_name')
if spec_dict.get('bento_version'):
deployment_pb.spec.bento_version = spec_dict.get('bento_version')
if deployment_pb.spec.operator == DeploymentSpec.AWS_SAGEMAKER:
sagemaker_config = spec_dict.get('sagemaker_operator_config', {})
sagemaker_operator_config_pb = deployment_pb.spec.sagemaker_operator_config
if sagemaker_config.get('api_name'):
sagemaker_operator_config_pb.api_name = sagemaker_config.get(
'api_name')
if sagemaker_config.get('region'):
sagemaker_operator_config_pb.region = sagemaker_config.get(
'region')
if sagemaker_config.get('instance_count'):
sagemaker_operator_config_pb.instance_count = int(
sagemaker_config.get('instance_count'))
if sagemaker_config.get('instance_type'):
sagemaker_operator_config_pb.instance_type = sagemaker_config.get(
'instance_type')
elif deployment_pb.spec.operator == DeploymentSpec.AWS_LAMBDA:
lambda_config = spec_dict.get('aws_lambda_operator_config', {})
if lambda_config.get('region'):
deployment_pb.spec.aws_lambda_operator_config.region = lambda_config.get(
'region')
if lambda_config.get('api_name'):
deployment_pb.spec.aws_lambda_operator_config.api_name = lambda_config.get(
'api_name')
elif deployment_pb.spec.operator == DeploymentSpec.GCP_FUNCTION:
gcp_config = spec_dict.get('gcp_function_operator_config', {})
if gcp_config.get('region'):
deployment_pb.spec.gcp_function_operator_config.region = gcp_config.get(
'region')
if gcp_config.get('api_name'):
deployment_pb.spec.aws_lambda_operator_config.api_name = gcp_config.get(
'api_name')
elif deployment_pb.spec.operator == DeploymentSpec.KUBERNETES:
k8s_config = spec_dict.get('kubernetes_operator_config', {})
k8s_operator_config_pb = deployment_pb.spec.kubernetes_operator_config
if k8s_config.get('kube_namespace'):
k8s_operator_config_pb.kube_namespace = k8s_config.get(
'kube_namespace')
if k8s_config.get('replicas'):
k8s_operator_config_pb.replicas = k8s_config.get('replicas')
if k8s_config.get('service_name'):
k8s_operator_config_pb.service_name = k8s_config.get(
'service_name')
if k8s_config.get('service_type'):
k8s_operator_config_pb.service_type = k8s_config.get(
'service_type')
else:
raise BentoMLException(
'Platform "{}" is not supported in the current version of '
'BentoML'.format(platform))
return deployment_pb