Local mode fails to read serialized bytes from S3 alternative storage

[cuongvd23]

Describe the bug
I encountered ServiceError: Failed to read serialized bytes from S3 alternative storage (MinIO).

To reproduce

• Run MinIO image with docker, expose the server to port 9000, the endpoint address is http://localhost:9000.
• Utilize LocalPipelineSession with s3_endpoint_url = 'http://localhost:9000'.
• Start SageMaker local mode.

Expected behavior
Can read objects from MinIO without any problems.

Screenshots or logs

INFO [sagemaker.remote_function] Uploading serialized function code to s3://sagemaker-data/pipelines/step-name/function
INFO [sagemaker.remote_function] Uploading serialized function arguments to s3://sagemaker-data/pipelines/step-name/arguments
...
sagemaker.remote_function.errors.ServiceError: Failed to read serialized bytes from s3://sagemaker-data/pipelines/pipeline-name/step-name-xxx/xxx/function/metadata.json: ClientError('An error occurred (403) when calling the HeadObject operation: Forbidden')

System information
A description of your system. Please provide:

• SageMaker Python SDK version: 2.250.0
• Framework name (eg. PyTorch) or algorithm (eg. KMeans): Not used
• Framework version: Not used
• Python version: 3.13.0
• CPU or GPU: CPU
• Custom Docker image (Y/N):

Additional context

• Successfully upload the serialized resources (function code, function arguments) to MinIO.
• Double-check on MinIO console to make sure the objects exist.

[cuongvd23]

The step invoke-function is currently using boto3.session.Session with the default s3_endpoint_url pointing to AWS S3.

• sagemaker-python-sdk/src/sagemaker/remote_function/invoke_function.py

  Line 63 in 73bdd08

  boto_session = boto3.session.Session(region_name=region)

Setting AWS_ENDPOINT_URL_S3 environment variable would override the default value.

• https://github.com/boto/botocore/blob/25cbbe2251d746b3f5ef0a4fcdb644def86937dc/botocore/configprovider.py#L1011

However, sagemaker.local.image.S3_ENDPOINT_URL_ENV_NAME is set to "S3_ENDPOINT_URL" instead.

• sagemaker-python-sdk/src/sagemaker/local/image.py

  Line 53 in 73bdd08

  S3_ENDPOINT_URL_ENV_NAME = "S3_ENDPOINT_URL"

[nmadan]

Thanks for catching and fixing this issue. The fix has been merged in #5264