NO_READ_ACCESS_TO_S3 while calling StartTextTranslationJob on AWS Translate

Question

I am trying to run AWS Translate job on Batch data stored in S3 with the following python code

role_arn='arn:aws:iam::<account-id>:role/service-role/AmazonTranslateServiceRoleS3FullAccess-ExploreML'
inp_data_path='s3://exploring-ml/aws-translate/assets/input/'
opt_data_path='s3://exploring-ml/aws-translate/assets/output/'

response = translate_client.start_text_translation_job(
    JobName='string',
    InputDataConfig={
        'S3Uri': inp_data_path,
        'ContentType': 'text/plain'
    },
    OutputDataConfig={
        'S3Uri': opt_data_path
    },
    DataAccessRoleArn=role_arn,
    SourceLanguageCode='zh',
    TargetLanguageCodes=[
        'en'
    ]
)

I can confirm that role has full S3 access, However, when executing above code, I get following error

---------------------------------------------------------------------------
InvalidRequestException                   Traceback (most recent call last)
<ipython-input-11-2bf8de09e0fe> in <module>
     11     SourceLanguageCode='zh',
     12     TargetLanguageCodes=[
---> 13         'en'
     14     ]
     15 )

~/.local/share/virtualenvs/exploring-ml-tools-zug9J9gH/lib/python3.6/site-packages/botocore/client.py in _api_call(self, *args, **kwargs)
    314                     "%s() only accepts keyword arguments." % py_operation_name)
    315             # The "self" in this scope is referring to the BaseClient.
--> 316             return self._make_api_call(operation_name, kwargs)
    317 
    318         _api_call.__name__ = str(py_operation_name)

~/.local/share/virtualenvs/exploring-ml-tools-zug9J9gH/lib/python3.6/site-packages/botocore/client.py in _make_api_call(self, operation_name, api_params)
    633             error_code = parsed_response.get("Error", {}).get("Code")
    634             error_class = self.exceptions.from_code(error_code)
--> 635             raise error_class(parsed_response, operation_name)
    636         else:
    637             return parsed_response

InvalidRequestException: An error occurred (InvalidRequestException) when calling the StartTextTranslationJob operation: NO_READ_ACCESS_TO_S3: The provided data access role does not have proper access to the input/output S3 URI.

Any pointer if I am missing anything. Also for clarification, this is running in Ireland where the service is available.

Answer 1

I tried to reproduce this situation and received:

botocore.errorfactory.InvalidRequestException: An error occurred (InvalidRequestException) when calling the StartTextTranslationJob operation: OUTPUT_S3_URI_INVALID: The provided output S3 URI is invalid or does not exist.

I then changed my output to point purely to the bucket (with no sub-directory) and it worked!

I then copied a file to a sub-directory (eg aws cp foo.txt s3://my-bucket/aws-translate/assets/output/) and re-ran the program and it worked!

Bottom line: Make sure that the output directory already exists (either by putting a file in that path, or using "Create Folder" in the S3 management console to create the output directory).

For the record, my IAM Role had the AmazonS3FullAccess policy and a trust policy of:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "",
      "Effect": "Allow",
      "Principal": {
        "Service": "translate.amazonaws.com"
      },
      "Action": "sts:AssumeRole"
    }
  ]
}