Leveraging Class E address space to mitigate IPv4 exhaustion issues in GKE

While I get that large enterprises battle problems that are hard to imagine for most people, I feel that if my Kubernetes cluster wants to use above 16 million addresses ... I might be battling wrong problems.

I'd love to read how they (snap) split their ip pools across clusters and regions, as the article sounds like they span single clusters with one subnet globally.