[HADOOP-14808] Hadoop keychain

• Type: New Feature
• Status: Patch Available
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.7.0
• Component/s: security
• Target Version/s: 3.5.0
• Assignee: John Zhuge
• Reporter: John Zhuge
• Created: 25/Aug/17 16:49
• Updated: 04/Jan/24 12:16

Description

Extend the idea from HADOOP-6520 "UGI should load tokens from the environment" to a generic lightweight "keychain" design. Load keys (secrets) into a keychain in UGI (secret map) at startup. YARN will distribute them securely into each container. The Hadoop code running in the container can then retrieve the credentials from UGI.

The use case is Bring Your Own Key (BYOK) credentials for cloud connectors (adl, wasb, s3a, etc.), while Hadoop authentication is still Kerberos. No configuration change, no admin involved. It will support YARN applications initially, e.g., DistCp, Tera Suite, Spark-on-Yarn, etc.

Implementation is surprisingly simple because almost all pieces are in place:

• Retrieve secrets from UGI using conf.getPassword backed by the existing Credential Provider class UserProvider
• Reuse Credential Provider classes and interface to define local permanent or transient credential store, e.g., LocalJavaKeyStoreProvider
• New: create a new transient Credential Provider that logs into AAD with username/password or device code, and then put the Client ID and Refresh Token into the keychain
• New: create a new permanent Credential Provider based on Hadoop configuration XML, for dev/testing purpose.

Links

• HADOOP-11766 Generic token authentication support for Hadoop
• HADOOP-11744 Support OAuth2 in Hadoop
• HADOOP-10959 A Kerberos based token authentication approach
• HADOOP-9392 Token based authentication and Single Sign On

Issue Links

• is related to: HADOOP-6520 UGI should load tokens from the environment
• is related to: HADOOP-9392 Token based authentication and Single Sign On