[ZOOKEEPER-4283] Fix of ZOOKEEPER-1653 was not merged in master and 3.6 branch

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.6.3
• Assignee: Unassigned
• Reporter: May
• Created: 15/Apr/21 08:20
• Updated: 09/Dec/21 06:28

Description

I am using the latest stable release 3.6.3. Here is the minimal workload to reproduce the bug:

1. There is a zookeeper cluster with three nodes: node1, node2 and node3;
2. Start the cluster from scratch;
3. node2 becomes the leader;
4. Stop node2;
5. node3 becomes the new leader;
6. Client requests to create a znode;
7. Restart node2;
8. node2 receives SNAP from node3;
9. node2 receives the NEWLEADER message from node3;
10. node2 takes current snapshot;
11. node2 crashes before writing to the current epoch file;
12. Restart node2 but it fails since "Unable to load database on disk".