How to solve the issue where a problem with any single node in a cluster causes the entire cluster to stop working?