[ZOOKEEPER-4836] Inconsist aclIndex leads to MarshallingError

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Duplicate
• Affects Version/s: 3.10.0
• Component/s: server
• Assignee: Unassigned
• Reporter: mutu
• Created: 05/Jun/24 11:17
• Updated: 29/Oct/24 11:44
• Resolved: 29/Oct/24 11:44

Description

There is a three-node cluster.  A client create  /1 with a new acl1 to leader.

When leader sends the snapshot to the follower, the nic becomes fail-slow. Hence, the follower1 get aclcache(aclIndex 2) without acl1. But the follower1 get a correct datatree. When deserializing the datatree, aclIndex is reset to 1. At this time, the aclIndex is inconsistent (leader:3, follower1:1,follower2:3). The follower1 execute Txn that logs the create operation.

In follower1, the aclIndex is set to 2 and node/1 points to 3. When executing getAcl /1, MarshallingError arises.

System logs are attached.

Are there any comments to figure out this issues？ I will very appreciate them.

Issue Links

• is superceded by: ZOOKEEPER-4689 Node may not accessible due the the inconsistent ACL reference map after SNAP sync (again)