Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
Found this problem on beasting LeaderVoteWaitTimeout,
- a replica publish itself as DOWN on RecoveryStrategy.pingLeader()
- It won the election (by vote wait timeout) then skipping publish itself as ACTIVE (by looking into clusterstate, its state still ACTIVE).
- ends up with a leader with DOWN state!
Therefore, replica should look into both clusterstate and its local last published state on skipping publish itself as ACTIVE