[IOTDB-1404] Write blocking problem for three node three replica cluster

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Sprint: 2021-09
• Assignee: Xinyu Tan
• Reporter: Xinyu Tan
• Created: 29/May/21 12:12
• Updated: 08/Oct/21 02:59
• Resolved: 22/Sep/21 01:45

Description

Set up a cluster with three nodes and three replicas. Firstly, manually registered all metadata, and then wrote the data several times by using multiple sessionpools which are encapsulated in the upper layer and can automatically fault tolerant. The results show that shutting down a node service has little effect on performance, but pulling out a node network cable has a great effect on performance. For the latter, use a SessionPool directly to connect to one of the normal two nodes, and it still takes a long time to write data.

This is unusual because the quorum of a raft group is 2, and the slowness of a node should not block writes

Issue Links

• is a child of: IOTDB-68 IoTDB supports distributed version