Implement deployment update for sagemaker

bentoml/cli/deployment.py

@@ -46,6 +46,7 @@
     get_deployment,
     describe_deployment,
     list_deployments,
+    update_deployment,
 )
 from bentoml.yatai import get_yatai_service
 
@@ -346,6 +347,7 @@ def create(
                 ),
                 CLI_COLOR_ERROR,
             )
+            return
         else:
             if wait:
                 result_state = get_state_after_await_action_complete(
@@ -372,6 +374,133 @@ def create(
             _echo('Successfully created deployment {}'.format(name), CLI_COLOR_SUCCESS)
             _print_deployment_info(result.deployment, output)
 
+    @deployment.command(help='Update existing deployment')
+    @click.argument("name", type=click.STRING, required=True)
+    @click.option(
+        '-n',
+        '--namespace',
+        type=click.STRING,
+        help='Deployment namespace managed by BentoML, default value is "default" which'
+        'can be changed in BentoML configuration file',
+    )
+    @click.option(
+        '-b',
+        '--bento',
+        '--bento-service-bundle',
+        type=click.STRING,
+        callback=parse_bento_tag_callback,
+        help='Target BentoService to be deployed, referenced by its name and version '
+        'in format of name:version. For example: "iris_classifier:v1.2.0"',
+    )
+    @click.option(
+        '--instance-type',
+        help='Type of instance will be used for inference. Option applicable to '
+        'platform: AWS SageMaker. Default to "m1.m4.xlarge"',
+        type=click.STRING,
+    )
+    @click.option(
+        '--instance-count',
+        help='Number of instance will be used. Option applicable to platform: AWS '
+        'SageMaker. Default value is 1',
+        type=click.INT,
+    )
+    @click.option(
+        '--num-of-gunicorn-workers-per-instance',
+        help='Number of gunicorn worker will be used per instance. Option applicable '
+        'to platform: AWS SageMaker. Default value for gunicorn worker is '
+        'based on the instance\' cpu core counts. The formula is num_of_cpu/2 + 1',
+        type=click.INT,
+    )
+    @click.option(
+        '--api-name',
+        help='User defined API function will be used for inference. Required for AWS '
+        'SageMaker',
+    )
+    @click.option('-o', '--output', type=click.Choice(['json', 'yaml']), default='json')
+    @click.option(
+        '--wait/--no-wait',
+        default=True,
+        help='Wait for apply action to complete or encounter an error.'
+        'If set to no-wait, CLI will return immediately. The default value is wait',
+    )
+    def update(
+        name,
+        namespace,
+        bento,
+        instance_type,
+        instance_count,
+        num_of_gunicorn_workers_per_instance,
+        api_name,
+        output,
+        wait,
+    ):
+        yatai_service = get_yatai_service()
+        get_deployment_result = get_deployment(namespace, name, yatai_service)
+        if get_deployment_result.status.status_code != status_pb2.Status.OK:
+            get_deployment_status = get_deployment_result.status
+            _echo(
+                f'Failed to updated deployment {name}.'
+                f'{status_pb2.Status.Code.Name(get_deployment_status.status_code)}'
+                f':{get_deployment_status.error_message}',
+                CLI_COLOR_ERROR,
+            )
+            return
+        current_deployment = get_deployment_result.deployment
+        track_cli(
+            'deploy-update',
+            deploy_platform=DeploymentSpec.DeploymentOperator.Name(
+                current_deployment.spec.operator
+            ),
+        )
+        bento_name, bento_version = bento.split(':')
+        updated_spec = {
+            'bento_name': bento_name,
+            'bento_version': bento_version,
+            'instance_type': instance_type,
+            'instance_count': instance_count,
+            'num_of_gunicorn_workers_per_instance': num_of_gunicorn_workers_per_instance,  # noqa E501
+            'api_name': api_name,
+        }
+        try:
+            result = update_deployment(current_deployment, updated_spec, yatai_service)
+        except BentoMLException as e:
+            _echo(f'Failed to update deployment {name}: {str(e)}', CLI_COLOR_ERROR)
+            return
+        if result.status.status_code != status_pb2.Status.OK:
+            update_deployment_status = result.status
+            _echo(
+                f'Failed to update deployment {name}. '
+                f'{status_pb2.Status.Code.Name(update_deployment_status.status_code)}:'
+                f'{update_deployment_status.error_message}',
+                CLI_COLOR_ERROR,
+            )
+            return
+        else:
+            if wait:
+                result_state = get_state_after_await_action_complete(
+                    yatai_service=yatai_service,
+                    name=name,
+                    namespace=namespace,
+                    message='Updating deployment',
+                )
+                if result_state.status.status_code != status_pb2.Status.OK:
+                    describe_deployment_status = result_state.status
+                    _echo(
+                        f'Updated deployment {name}. Failed to retrieve latest status. '
+                        f'{status_pb2.Status.Code.Name(describe_deployment_status.status_code)}:'  # noqa E501
+                        f'{describe_deployment_status.error_message}'
+                    )
+                    return
+                result.deployment.state.CopyFrom(result_state.state)
+        track_cli(
+            'deploy-update-success',
+            deploy_platform=DeploymentSpec.DeploymentOperator.Name(
+                current_deployment.spec.operator
+            ),
+        )
+        _echo(f'Successfully updated deployment {name}', CLI_COLOR_SUCCESS)
+        _print_deployment_info(result.deployment, output)
+
     @deployment.command(help='Apply model service deployment from yaml file')
     @click.option(
         '-f',

bentoml/deployment/operator.py

@@ -65,7 +65,7 @@ def add(self, deployment_pb):
         """
 
     @abstractmethod
-    def update(self, deployment_pb):
+    def update(self, deployment_pb, previous_deployment):
         """
         Update existing deployment based on deployment_pb spec
         """

bentoml/deployment/sagemaker/__init__.py

@@ -437,6 +437,19 @@ def _create_sagemaker_endpoint(sagemaker_client, endpoint_name, endpoint_config_
         )
 
 
+def _update_sagemaker_endpoint(sagemaker_client, endpoint_name, endpoint_config_name):
+    try:
+        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", str(update_endpoint_response))
+    except ClientError as e:
+        raise _aws_client_error_to_bentoml_exception(
+            e, "Failed to update sagemaker endpoint"
+        )
+
+
 class SageMakerDeploymentOperator(DeploymentOperatorBase):
     def add(self, deployment_pb):
         try:
@@ -526,11 +539,160 @@ def _add(self, deployment_pb, bento_pb, bento_path):
 
         return ApplyDeploymentResponse(status=Status.OK(), deployment=deployment_pb)
 
-    def update(self, deployment_pb):
-        raise NotImplementedError(
-            "Updating AWS SageMaker deployment is not supported in current version of "
-            "BentoML"
-        )
+    def update(self, deployment_pb, previous_deployment):
+        try:
+            ensure_docker_available_or_raise()
+            deployment_spec = deployment_pb.spec
+            bento_pb = self.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(
+                        BentoUri.StorageType.Name(bento_pb.bento.uri.type)
+                    )
+                )
+            return self._update(
+                deployment_pb, previous_deployment, bento_pb, bento_pb.bento.uri.uri
+            )
+        except BentoMLException as error:
+            deployment_pb.state.state = DeploymentState.ERROR
+            deployment_pb.state.error_message = (
+                f'Error updating SageMaker deployment: {str(error)}'
+            )
+            return ApplyDeploymentResponse(
+                status=error.status_proto, deployment=deployment_pb
+            )
+
+    def _update(self, deployment_pb, current_deployment, bento_pb, bento_path):
+        if loader._is_remote_path(bento_path):
+            with loader._resolve_remote_bundle_path(bento_path) as local_path:
+                return self._update(
+                    deployment_pb, current_deployment, bento_pb, local_path
+                )
+        updated_deployment_spec = deployment_pb.spec
+        updated_sagemaker_config = updated_deployment_spec.sagemaker_operator_config
+        sagemaker_client = boto3.client('sagemaker', updated_sagemaker_config.region)
+
+        try:
+            raise_if_api_names_not_found_in_bento_service_metadata(
+                bento_pb.bento.bento_service_metadata,
+                [updated_sagemaker_config.api_name],
+            )
+            describe_latest_deployment_state = self.describe(deployment_pb)
+            current_deployment_spec = current_deployment.spec
+            current_sagemaker_config = current_deployment_spec.sagemaker_operator_config
+            latest_deployment_state = json.loads(
+                describe_latest_deployment_state.state.info_json
+            )
+
+            current_ecr_image_tag = latest_deployment_state['ProductionVariants'][0][
+                'DeployedImages'
+            ][0]['SpecifiedImage']
+            if (
+                updated_deployment_spec.bento_name != current_deployment_spec.bento_name
+                or updated_deployment_spec.bento_version
+                != current_deployment_spec.bento_version
+            ):
+                logger.debug(
+                    'BentoService tag is different from current deployment, '
+                    'creating new docker image and push to ECR'
+                )
+                with TempDirectory() as temp_dir:
+                    sagemaker_project_dir = os.path.join(
+                        temp_dir, updated_deployment_spec.bento_name
+                    )
+                    _init_sagemaker_project(sagemaker_project_dir, bento_path)
+                    ecr_image_path = create_and_push_docker_image_to_ecr(
+                        updated_sagemaker_config.region,
+                        updated_deployment_spec.bento_name,
+                        updated_deployment_spec.bento_version,
+                        sagemaker_project_dir,
+                    )
+            else:
+                logger.debug('Using existing ECR image for Sagemaker model')
+                ecr_image_path = current_ecr_image_tag
+
+            (
+                updated_sagemaker_model_name,
+                updated_sagemaker_endpoint_config_name,
+                sagemaker_endpoint_name,
+            ) = _get_sagemaker_resource_names(deployment_pb)
+            (
+                current_sagemaker_model_name,
+                current_sagemaker_endpoint_config_name,
+                _,
+            ) = _get_sagemaker_resource_names(current_deployment)
+
+            if (
+                updated_sagemaker_config.api_name != current_sagemaker_config.api_name
+                or updated_sagemaker_config.num_of_gunicorn_workers_per_instance
+                != current_sagemaker_config.num_of_gunicorn_workers_per_instance
+                or ecr_image_path != current_ecr_image_tag
+            ):
+                logger.debug(
+                    'Sagemaker model requires update. Delete current sagemaker model %s'
+                    'and creating new model %s',
+                    current_sagemaker_model_name,
+                    updated_sagemaker_model_name,
+                )
+                _delete_sagemaker_model_if_exist(
+                    sagemaker_client, current_sagemaker_model_name
+                )
+                _create_sagemaker_model(
+                    sagemaker_client,
+                    updated_sagemaker_model_name,
+                    ecr_image_path,
+                    updated_sagemaker_config,
+                )
+            # When bento service tag is not changed, we need to delete the current
+            # endpoint configuration in order to create new one to avoid name collation
+            if (
+                current_sagemaker_endpoint_config_name
+                == updated_sagemaker_endpoint_config_name
+            ):
+                logger.debug(
+                    'Current sagemaker config name %s is same as updated one, '
+                    'delete it before create new endpoint config',
+                    current_sagemaker_endpoint_config_name,
+                )
+                _delete_sagemaker_endpoint_config_if_exist(
+                    sagemaker_client, current_sagemaker_endpoint_config_name
+                )
+            logger.debug(
+                'Create new endpoint configuration %s',
+                updated_sagemaker_endpoint_config_name,
+            )
+            _create_sagemaker_endpoint_config(
+                sagemaker_client,
+                updated_sagemaker_model_name,
+                updated_sagemaker_endpoint_config_name,
+                updated_sagemaker_config,
+            )
+            logger.debug(
+                'Updating endpoint to new endpoint configuration %s',
+                updated_sagemaker_endpoint_config_name,
+            )
+            _update_sagemaker_endpoint(
+                sagemaker_client,
+                sagemaker_endpoint_name,
+                updated_sagemaker_endpoint_config_name,
+            )
+            logger.debug(
+                'Delete old sagemaker endpoint config %s',
+                current_sagemaker_endpoint_config_name,
+            )
+            _delete_sagemaker_endpoint_config_if_exist(
+                sagemaker_client, current_sagemaker_endpoint_config_name
+            )
+        except AWSServiceError as e:
+            _try_clean_up_sagemaker_deployment_resource(deployment_pb)
+            raise e
+
+        return ApplyDeploymentResponse(status=Status.OK(), deployment=deployment_pb)
 
     def delete(self, deployment_pb):
         try:

bentoml/utils/validator/__init__.py

@@ -19,13 +19,15 @@
 from cerberus import Validator
 
 from bentoml.utils import ProtoMessageToDict
-from bentoml.proto.deployment_pb2 import DeploymentSpec
+from bentoml.proto.deployment_pb2 import DeploymentSpec, DeploymentState
 
 deployment_schema = {
     'name': {'type': 'string', 'required': True, 'minlength': 4},
     'namespace': {'type': 'string', 'required': True, 'minlength': 3},
     'labels': {'type': 'dict', 'allow_unknown': True},
     'annotations': {'type': 'dict', 'allow_unknown': True},
+    'created_at': {'type': 'string'},
+    'last_updated_at': {'type': 'string'},
     'spec': {
         'type': 'dict',
         'required': True,
@@ -80,9 +82,10 @@
     'state': {
         'type': 'dict',
         'schema': {
-            'state': {'type': 'integer'},
+            'state': {'type': 'string', 'allowed': DeploymentState.State.keys()},
             'error_message': {'type': 'string'},
             'info_json': {'type': 'string'},
+            'timestamp': {'type': 'string'},
         },
     },
 }