Question

A machine learning engineer deploys a real-time SageMaker endpoint using a model whose artifacts are stored in an S3 bucket encrypted with a customer managed KMS key. The endpoint deployment fails during creation, and CloudWatch logs show the container could not download the model.tar.gz file. The SageMaker execution role has s3:GetObject permission on the bucket. What is the MOST likely cause and the correct remediation?

Accepted Answer

The SageMaker execution role lacks kms:Decrypt permission (and a corresponding grant in the KMS key policy) for the customer managed key; add the decrypt permission.

Answer

The endpoint must be deployed inside a VPC with an S3 gateway endpoint; add the VPC configuration to the endpoint.

Answer

S3 default encryption with a customer managed KMS key is not supported for SageMaker model artifacts; re-upload the artifacts with SSE-S3 encryption.

Answer

The model.tar.gz must be encrypted in transit using a separate TLS certificate configured on the endpoint; enable HTTPS on the model package.