Description
Hello, did a quick test and found that the new redis master cannot be elected correctly when the old redis master is down:
Here is what I did:
1: Deployed a 3 nodes redis-cluster with the compose example here
2: docker stop the current redis-master node (192.168.128.171)
3: Then found other redis-slaves are still point to the old master
4: Checked logs, it seems other redis-slaves still elect the old master as the new master:
77988:X 24 Aug 18:15:29.786 # +sdown master mymaster 192.168.128.171 6379
77988:X 24 Aug 18:15:29.844 # +odown master mymaster 192.168.128.171 6379 #quorum 2/2
77988:X 24 Aug 18:15:29.844 # +new-epoch 1
77988:X 24 Aug 18:15:29.844 # +try-failover master mymaster 192.168.128.171 6379
77988:X 24 Aug 18:15:29.845 # +vote-for-leader f5c9d950d522a745146040ddad51f8d6b5179972 1
77988:X 24 Aug 18:15:29.847 # e5f4a970e493af668f61b1cf63ba5ac608fc11fd voted for f5c9d950d522a745146040ddad51f8d6b5179972 1
77988:X 24 Aug 18:15:29.904 # +elected-leader master mymaster 192.168.128.171 6379
77988:X 24 Aug 18:15:29.904 # +failover-state-select-slave master mymaster 192.168.128.171 6379
77988:X 24 Aug 18:15:29.975 # -failover-abort-no-good-slave master mymaster 192.168.128.171 6379
77988:X 24 Aug 18:15:30.052 # Next failover delay: I will not start a failover before Thu Aug 24 18:16:30 2017
77633:S 24 Aug 18:16:01.339 # Timeout connecting to the MASTER...
So from my testing, it seems the sentinel HA doesn't work at all. Not sure what I miss? Thanks!