SageMaker endpoint production variants should have an initial instance count greater than 1
Confirms that SageMaker endpoint production variants are configured with more than one initial instance to avoid single points of failure. Serverless variants are outside the control, which applies only to instance-based endpoint configurations.
- service
- SageMaker
- severity
- Medium
- scope
- evaluated in every region
How to fix it
Set at creation: fixing it means replacing the resource.
Give every instance-based production variant in the endpoint configuration an initial instance count of two or more, so SageMaker spreads them across Availability Zones. Read the finding as being about the configuration rather than the running endpoint: resizing a live endpoint, by hand or through Application Auto Scaling, changes the desired instance count and leaves the configuration as it was, so an endpoint serving on four instances can still fail this. There is no update operation for an endpoint configuration, so the fix is a new one carrying the higher count and an endpoint update pointing at it. Serverless variants are not evaluated and have no instance count to set.
In the console
SageMaker Studio → Models → Deploy → Initial instance count, which writes a new endpoint configuration
With the AWS CLI
aws sagemaker create-endpoint-config --endpoint-config-name <endpoint-config-name> --production-variants '[{"VariantName":"<variant-name>","ModelName":"<model-name>","InstanceType":"<instance-type>","InitialInstanceCount":2}]' --region <region>With Terraform
resource "aws_sagemaker_endpoint_configuration" "this" {
production_variants {
variant_name = var.variant_name
model_name = var.model_name
instance_type = var.instance_type
initial_instance_count = 2
}
}
See which of your accounts this check is failing on.
Book a demo