Client Could not Connect to server if more than 1 node is down in the cluster simultaniously


#1

Server Version : 3.1.1.2 Java Client Version : 3.2.4

3 node cluster with single namespace with replication factor 2

Observation

  1. When read/writes are happening on a cluster and 1 node goes down then read/writes failed due to connection failure but that happens for couple of seconds and then read/writes continue without any issue

  2. Same behavior if 2nd node also goes down after some gap of time

  3. However if two nodes go down simultaneously, then the clients 60-70% read/write keep failing forever until the java code re initiates the connection to server . This behavior is not as expected and expected behavior was as in 1 and 2.

Issue is easily reproducible. Seems like some bug in the client library


#2

There is a bug when multiple nodes fail simultaneously leaving exactly one node. This will be fixed in the next client release.