Details
-
Bug
-
Status: In Progress
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
Description
daryn reported that the ZK delegation token secret manager uses a SharedCounter to synchronize increments of a monotonically increasing sequence number for new tokens. Yet the KMS logs occasionally, depending on load, contains an odd error indicating collisions:
org.apache.zookeeper.KeeperException$NodeExistsException: KeeperErrorCode = NodeExists for /zkdtsm/ZKDTSMRoot/ZKDTSMTokensRoot/DT_137547444
ZKDTSM does a CAS get and set of the sequence number. Rather than return the value it set, it returns the current value which may have already been incremented by another KMS.