Connect Studio JupyterLab notebooks to Amazon S3 Access Grants with Training and Processing jobs
Use the following information to grant Amazon S3 Access Grants to access data in Amazon SageMaker Training and Processing jobs.
When a user with trusted identity propagation enabled launches a SageMaker Training or Processing job that needs to access Amazon S3 data:
-
SageMaker AI calls Amazon S3 Access Grants to get temporary credentials based on the user's identity
-
If successful, these temporary credentials access the Amazon S3 data
-
If unsuccessful, SageMaker AI falls back to using the IAM role credentials
Note
To enforce that all of the permission are granted through Amazon S3 Access Grants, you will need to remove related Amazon S3 access permission your execution role and attach them to your corresponding Amazon S3 Access Grant.
Considerations
Amazon S3 Access Grants cannot be used with Pipe mode for both SageMaker Training and Processing for Amazon S3 input.
When trusted identity propagation is enabled, you cannot launch a SageMaker Training Job with the following feature
-
Remote Debug
-
Debugger
-
Profiler
When trusted identity propagation is enabled, you cannot launch a Processing job with the following feature
-
DatasetDefinition
Set up Amazon S3 Access Grants with Training and Processing jobs
After Amazon S3 Access Grants is set up, add the following permissions to your domain or user execution role.
-
is your Amazon Web Services Regionus-east-1
-
is your Amazon Web Services account ID111122223333
-
is your Amazon S3 Access Grant roleS3-ACCESS-GRANT-ROLE
{ "Version": "2012-10-17", "Statement": [ { "Sid": "AllowDataAccessAPI", "Effect": "Allow", "Action": [ "s3:GetDataAccess", "s3:GetAccessGrantsInstanceForPrefix" ], "Resource": [ "arn:aws:s3:
us-east-1
:111122223333
:access-grants/default" ] }, { "Sid": "RequiredForIdentificationPropagation", "Effect": "Allow", "Action": "sts:SetContext", "Resource": "arn:aws:iam::111122223333
:role/S3-ACCESS-GRANT-ROLE
" } ] }