[ZOOKEEPER-4039] accpetedEpoch过大导致对应的节点无法加入集群

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 3.5.5
• Component/s: server
• Assignee: Damien Diederen
• Reporter: pengfei
• Created: 28/Dec/20 10:02
• Updated: 30/Dec/20 20:06
• Resolved: 30/Dec/20 20:06

Description

leader会在收到过半的节点的accpetedEpoch后会将本身的accpetedEpoch设置为这些节点的最大值加1，但是此时leader宕机会导致leader节点的accpetedEpoch比其他节点大1，然后此节点再重启，再次被选为leader，再次宕机，然后剩下的节点再重新选举一个leader，这个leader的epoch会比原来的leader的accpetedEpoch要小，从而导致原来的节点一直在looking和follower状态切换

 

复现步骤:

3个节点，server1,server2,server3

• 启动server1,server2,然后在下面红点位置停止server1和server2此时server2的对应的accpetedEpoch=1
• 再启动server1,server2,然后再在下面红点位置停止server1和server2此时server2的对应的accpetedEpoch=2
• 再启动server1，server3，等server1和server3选举出对应的leader为server3,然后再启动server2，就会一直重复下面的异常

 

 

Issue Links

• duplicates: ZOOKEEPER-4040 java.io.IOException: Leaders epoch, 1 is less than accepted epoch, 2