[ZOOKEEPER-4388] Recover from network partition, follower/observer ephemerals nodes is inconsistent with leader

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.6.0, 3.6.3, 3.6.1, 3.6.2
• Component/s: server
• Labels: inconsistency partitoned zookeeper

Environment

zk version 3.6.0 3.6.1 3.6.2

the follower/observer network disconnection time exceeds session timeout

• Assignee: Unassigned
• Reporter: shixiaoxiao
• Created: 26/Sep/21 14:33
• Updated: 24/Apr/22 20:00

Description

The follower/observer enable read only. When the node returns to normal from the partitioned, the ephemerals nodes will be  inconsistent with the leader node. The reason is that the request to close the timeout sessions is processed by the ReadOnly follower or observer  when they are partitioned and the ephemerals nodes created by these sessions also are delete. When the leader node uses diff to synchronize data with the follower/observer node, the transaction that needs to be synchronized does not include the creation of temporary nodes which created by sessions closed by followers.So the follower/observer  ephemerals nodes is inconsistent with leader.