It was recently the time change in Chile. My cluster works with an embedded zookeper and with three nodes.
When the time change was generated, the cluster had the processes stopped but it was running. After the time change I restarted it and a node did not return, it did not communicate with the other two that were started and forming the cluster.
The solution to this problem was to raise the nodes one by one in an odd order. Once all the nodes are disconnected, the nodes are restarted again forming the cluster mode.
Do you have an idea why that behavior?