There is still one more situation we might encounter. A whole availability zone (data center) might fail. Kubernetes will continue operating correctly. It’ll have two instead of three master nodes, and the failed worker nodes will be recreated in healthy zones. However, we’d run into trouble with our stateful services. Kubernetes would not be able to reschedule those that were mounted to EBS volumes from the failed zone. We’d need to wait for the availability zone to come back online, or we’d need to move the EBS volume to a healthy zone manually. The chances are that, in such a case, the EBS would not be available and, therefore, could not be moved.