I Replaced kube-proxy with eBPF in Production (And Why My Monitoring Went Blind for 6 Hours)

A Cilium upgrade on a 4-node bare-metal homelab cluster, consisting of Dell OptiPlex and Raspberry Pis running Talos Linux, ArgoCD, and Longhorn, caused a six-hour monitoring outage. After replacing kube-proxy with eBPF, the user discovered that their SIEM could no longer detect any packets. The system had previously operated in kube-proxy-replacement partial mode, but the transition to a full eBPF datapath created an invisible path that blinded the existing monitoring tools.