In an HA cluster, we rotate one node at a time during an upgrade, so the other two nodes in the cluster will continue to service traffic, while the 3rd node is being upgraded, which prevents downtime.