[HADOOP-9850] RPC kerberos errors don't trigger relogin

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 2.1.0-beta, 3.0.0-alpha1
• Fix Version/s: 2.1.0-beta
• Component/s: ipc
• Target Version/s: 2.1.0-beta
• Hadoop Flags: Reviewed
• Assignee: Daryn Sharp
• Reporter: Daryn Sharp
• Created: 07/Aug/13 20:25
• Updated: 12/May/16 18:26
• Resolved: 08/Aug/13 15:09

Description

Hadoop auto-renews a ticket cache TGT. However, a TGT acquired via keytab is just allowed to expire. To compensate, any exception during a kerberos RPC connection triggers a relogin.

Prior to HADOOP-9698, the RPC client "knew" the SASL client was attempting authMethod kerberos. Now the SASL client negotiates and returns the authMethod to the RPC Client. When an exception occurs, such as TGT expired, the Client doesn't know what the SASL client was attempting so no relogin is attempted. After 24 hours, keytab based services that act as clients (ex. RM for token renewal) go dead.