Managing costs for staging or development Kubernetes environments that aren't always active is a common challenge. While scaling services to zero helps, implementing this with dedicated proxy layers for each service can be operationally complex. These proxies often need careful configuration, can become another point of failure, and make applying consistent network policies more difficult.

This session explores a different approach: achieving scale-to-zero for TCP services directly at the CNI level using eBPF and Cilium. We'll discuss how eBPF can intercept initial TCP packets in the kernel, then trigger a Kubernetes operator to scale up the necessary backend pods only when there's actual traffic. This method bypasses the need for individual service proxies, simplifying the operational overhead. We'll look at the architecture, how it interacts with the Kubernetes control plane, and the benefits of a CNI-native solution that "just works" for many TCP-based applications.