Resolution: Works for Me
Affects Version/s: 2.7.3
Fix Version/s: None
hadoop fs s3a:// operations fail authentication for s3 buckets hosted in regions other than default us-east-1
Steps to reproduce:
- create s3 bucket in eu-west-1
- Using IAM instance profile or fs.s3a.access.key/fs.s3a.secret.key run following command:
You will see listing of the bucket
You will get 403 Authentication Denied response for AWS S3.
Reason is mismatch in string to sign as defined in http://docs.aws.amazon.com/AmazonS3/latest/dev/RESTAuthentication.html provided by hadoop and expected by AWS.
If you use https://aws.amazon.com/code/199 to analyse StringToSignBytes returned by AWS, you will see that AWS expects CanonicalizedResource to be in form /your-eu-west-1-hosted-bucket.s3.eu-west-1.amazonaws.com/.
Hadoop provides it as /your-eu-west-1-hosted-bucket/
Note that AWS documentation doesn't explicitly state that endpoint or full dns address should be appended to CanonicalizedResource however practice shows it is actually required.
I've also submitted this to AWS for them to correct behaviour or documentation.