Rewriting the Rules: Why Kubernetes Is Moving To NFTables
summary
Claims that nftables replaces iptables/IPVS for kube-proxy by switching from linear rule processing to hash maps, delivering constant-time lookups and 30k services scale tested on ~100-node clusters. Migration uses kube-proxy in nftables mode (configMap or flag), requires kernel >5.13 (kernel 6.1 fixes an invalid-packet bug) and Kubernetes 1.29+ (GA in 1.33). Measured restore/programmability goes from 6 minutes vs 10 seconds, but NodePort-to-127.0.0.1 no longer works without risky route_localnet changes.
AI-written summary · about
chapters (5)
- 0:00Kubernetes and NF tables intro
- 2:31IP tables limitations
- 7:23NF tables advantages
- 10:53Recommendations with NF tables
- 12:15Q&A
related talks
31mBuilding a Hypervisor Firewall with nftables and Rust
33mNavigating the Container Jungle
31mBuilding a Hypervisor Firewall with nftables and Rust
29mSimplifying the Networking and Secur...
22mFirewalld Changes Coming to Your OS
40mAttacking Kubernetes Clusters Through Your Network Plumbing
21mFrom Squire To Network Knight: Defending the Kubernetes Castle the Cloud Native...
45mTo the tables! Routing efficiently with iproute and Netfilter
