[HADOOP-11038] Support AWS roles to talk to AWS-S3 largely for cross-AWS-account integration

• Type: New Feature
• Status: Resolved
• Priority: Major
• Resolution: Won't Fix
• Component/s: fs, fs/s3
• Assignee: Unassigned
• Reporter: Vishal Gupta
• Created: 31/Aug/14 23:07
• Updated: 06/Feb/15 21:45
• Resolved: 06/Feb/15 21:45

Description

Currently "hdfs dfs -lsr s3://..." supports acess-keys/secret-keys only as the way to authenticate to s3. This should support AWS-roles also because of the following reasons :

1) AWS-roles is a AWS best-practice and is highly recommended by AWS themselves.
2) This helps in cross-AWS-account integration also. An AWS-account-holder can provide another AWS-account-holder a cross-account-AWS-role to perform operations over his S3-buckets.

The current syntax is "hdfs dfs" is :
hdfs dfs -Dfs.s3n.awsAccessKeyId=XXXX -Dfs.s3n.awsSecretAccessKey=XXXX -ls s3n://.../

This should change to :
hdfs dfs -Dfs.s3n.awsAccessKeyId=XXXX -Dfs.s3n.awsSecretAccessKey=XXXX -Dfs.s3n.awsRoleToBeAssumed=arn:aws:iam::XXXX:role/XXXX -Dfs.s3n.awsExternalId=XXXX -ls s3n://.../

Extending the use-case a little further, for a client AWS-account to integrate with multiple different AWS-accounts, configuration for s3-bucket to role-to-be-assumed mapping ( which will override the master-role ) should be provided :
hdfs dfs -Dfs.s3.awsAccessKeyId=XXXX -Dfs.s3.awsSecretAccessKey=XXXX -Dfs.s3.awsRoleToBeAssumed=arn:aws:iam::XXXX:role/XXXX -Dfs.s3.awsBucketToRoleMapping="{\"bucket1\": { \"roleName\":\"arn:aws:iam::XXXX:role/role1\", \"externalId\":\"....\"}}" -ls s3://.../

Since, AWS treats a cross-account-AWS-role the same as an AWS-role within a AWS-account, the above flows remain same for a role within a AWS-account.

Issue Links

• duplicates: HADOOP-9384 Update S3 native fs implementation to use AWS SDK to support authorization through roles