def _create_sagemaker_model(sagemaker_client, sagemaker_model_name,
ecr_image_path, spec):
execution_role_arn = get_arn_role_from_current_aws_user()
sagemaker_model_info = {
"ModelName": sagemaker_model_name,
"PrimaryContainer": {
"ContainerHostname": sagemaker_model_name,
"Image": ecr_image_path,
"Environment": {
"API_NAME": spec.api_name,
"BENTOML_GUNICORN_TIMEOUT": str(spec.timeout),
},
},
"ExecutionRoleArn": execution_role_arn,
}
# Will set envvar, if user defined gunicorn workers per instance. EnvVar needs
# to be string instead of the int.
if spec.num_of_gunicorn_workers_per_instance:
sagemaker_model_info["PrimaryContainer"]["Environment"][
"BENTOML_GUNICORN_NUM_OF_WORKERS"] = str(
spec.num_of_gunicorn_workers_per_instance)
try:
create_model_response = sagemaker_client.create_model(
**sagemaker_model_info)
except ClientError as e:
raise generate_bentoml_exception_from_aws_client_error(
e, "Failed to create sagemaker model")
logger.debug("AWS create model response: %s", create_model_response)
def test_generate_bentoml_exception_from_aws_client_error():
error = ClientError(
error_response={
'Error': {'Code': 'ValidationException', 'Message': 'error message'}
},
operation_name='failed_operation',
)
exception = generate_bentoml_exception_from_aws_client_error(error)
assert isinstance(exception, AWSServiceError)
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 generate_bentoml_exception_from_aws_client_error(
e, "Failed to update sagemaker endpoint")
def _create_sagemaker_endpoint_config(sagemaker_client, sagemaker_model_name,
endpoint_config_name, sagemaker_config):
create_endpoint_config_arguments = {
"EndpointConfigName":
endpoint_config_name,
"ProductionVariants": [{
"VariantName": sagemaker_model_name,
"ModelName": sagemaker_model_name,
"InitialInstanceCount": sagemaker_config.instance_count,
"InstanceType": sagemaker_config.instance_type,
}],
}
if sagemaker_config.data_capture_s3_prefix:
logger.debug(
"data_capture_s3_prefix %s found, creating data capture config",
sagemaker_config.data_capture_s3_prefix,
)
if not sagemaker_config.data_capture_sample_percent:
logger.debug(
"data_capture_sample_percent not found, using default 100%")
sagemaker_config.data_capture_sample_percent = 100
create_endpoint_config_arguments["DataCaptureConfig"] = {
"EnableCapture":
True,
"DestinationS3Uri":
sagemaker_config.data_capture_s3_prefix,
"InitialSamplingPercentage":
sagemaker_config.data_capture_sample_percent,
"CaptureOptions": [{
"CaptureMode": "Input"
}, {
"CaptureMode": "Output"
}],
}
logger.debug("Creating Sagemaker endpoint %s configuration",
endpoint_config_name)
try:
create_config_response = sagemaker_client.create_endpoint_config(
**create_endpoint_config_arguments)
except ClientError as e:
raise generate_bentoml_exception_from_aws_client_error(
e, "Failed to create sagemaker endpoint config")
logger.debug("AWS create endpoint config response: %s",
create_config_response)
def _delete_sagemaker_model_if_exist(sagemaker_client, sagemaker_model_name):
try:
delete_model_response = sagemaker_client.delete_model(
ModelName=sagemaker_model_name)
logger.debug("AWS delete model response: %s", delete_model_response)
except ClientError as e:
error_response = e.response.get("Error", {})
error_code = error_response.get("Code", "Unknown")
error_message = error_response.get("Message", "Unknown")
if (error_code == "ValidationException"
and "Could not find model" in error_message):
# sagemaker model does not exist
return
raise generate_bentoml_exception_from_aws_client_error(
e, f"Failed to cleanup sagemaker model '{sagemaker_model_name}'")
return
def _delete_sagemaker_endpoint_if_exist(sagemaker_client,
sagemaker_endpoint_name):
try:
delete_endpoint_response = sagemaker_client.delete_endpoint(
EndpointName=sagemaker_endpoint_name)
logger.debug("AWS delete endpoint response: %s",
delete_endpoint_response)
except ClientError as e:
error_response = e.response.get("Error", {})
error_code = error_response.get("Code", "Unknown")
error_message = error_response.get("Message", "Unknown")
if (error_code == "ValidationException"
and "Could not find endpoint" in error_message):
# sagemaker endpoint does not exist
return
raise generate_bentoml_exception_from_aws_client_error(
e,
f"Failed to delete sagemaker endpoint '{sagemaker_endpoint_name}'")
def _delete_sagemaker_endpoint_config_if_exist(sagemaker_client,
sagemaker_endpoint_config_name):
try:
delete_endpoint_config_response = sagemaker_client.delete_endpoint_config(
EndpointConfigName=sagemaker_endpoint_config_name)
logger.debug("AWS delete endpoint config response: %s",
delete_endpoint_config_response)
except ClientError as e:
error_response = e.response.get("Error", {})
error_code = error_response.get("Code", "Unknown")
error_message = error_response.get("Message", "Unknown")
if (error_code == "ValidationException"
and "Could not find endpoint configuration" in error_message):
# endpoint config does not exist
return
raise generate_bentoml_exception_from_aws_client_error(
e,
f"Failed to cleanup sagemaker endpoint config "
f"'{sagemaker_endpoint_config_name}' after creation failed",
)
return
