GKE Dataplane v2 uses eBPF and Cilium to provide a programmable path for Kubernetes Service traffic and NetworkPolicy enforcement. eBPF hooks let the Linux kernel use workload context when handling packets, addressing observability, scalability, and functional requirements. section 10.2.1
Two Cilium components take over responsibilities previously handled by Calico and kube-proxy:
- The cilium-agent DaemonSet programs the eBPF data path, resolves Services, and enforces NetworkPolicies. section 10.2.1
- The cilium-operator Deployment manages Cilium-internal CRDs and reduces the need for agents to watch every Pod. section 10.2.1
Compare the implementations before and after the switch: ClusterIP resolution moves from kube-proxy iptables rules to eBPF socket hooks; NodePort, load-balancer resolution, and policy enforcement use the hooks shown in the table. section 10.2.1
