[HDFS-6510] WebHdfs clients clears the delegation token on retry (for HA), thus failing retry requests

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.4.0
• Assignee: Yongjun Zhang
• Reporter: Yongjun Zhang
• Created: 10/Jun/14 16:39
• Updated: 20/Jun/14 04:51
• Resolved: 20/Jun/14 04:51

Description

In WebHdfs clients connected to a HA HDFS service, when a failure (that is inferred as a failover) happens and retry is attempted, the delegation token previously initialized is cleared. For token based auth, this causes all the subsequent retries to fail due to auth errors.
See "delegationToken = null" in method WebHdfsFileSystem.resetStateToFailOver.
This issue would not only show up on failover, but happens more commonly when the first NN specified in the config is not reachable.

Issue Links

• duplicates: HDFS-6312 WebHdfs HA failover is broken on secure clusters