The most fundamental step to ensure high availability is eliminating (SPOF). A SPOF is a component within a system which, if fails, causes the entire system to fail.
For example, if we deploy only one instance of our backend API, the single Node process running the instance becomes a SPOF. If that Node process exits for whatever reason, then our whole application goes down.
Fortunately, eliminating a SPOF is relatively simple—replication; you simply have to deploy multiple instances of that component. However, that comes with challenges of its own—when a new request is received, which instance should handle it?
