Recall that Kubernetes inherently provides a great deal of resiliency across the nodes of its cluster. A properly designed cluster can lose one or more nodes without impacting the operation of the software running on it. However, achieving this same level of resiliency to address the possibility of an entire data center going offline is notably more complicated. We refer to this level of resiliency as 'Geographic Redundancy.'

A georedundant system must be designed to handle a loss of all nodes in either data center, meaning Kubernetes control plane elements must have sufficient replication and distribution between the sites, and each site should be sized to accommodate the entire load of users in the event that the other site should fail. If the size of deployment requires, say, four nodes to serve all users, then a georedundant system would require eight total nodes: four at each site.

As previously described, there are two approaches for creating geographic redundancy in Kubernetes. The following sections describe each approach.