Kubernetes Networking at Scale
summary
Targets clusters with thousands of nodes, tens of thousands of pods and thousands of services and shows concrete fixes: use native pod routing (avoid overlays and bridges), prefer AWS ENI or CNI IP allocations and right‑sized alias ranges, replace iptables kube‑proxy with IPVS over iptables to avoid giant rule syncs, run a node-local DNS cache to absorb DNS query storms, and use container‑native ingress (ALB/NEG targeting pod IPs) to cut extra hops; most prototypes are deployed and working, with ELB node limits (1000) and graceful‑termination kinks noted.
AI-written summary · about
chapters (30)
- 0:00Introduction
- 0:28Scaling Challenges
- 2:26Classic approaches
- 4:23Overlay data plane
- 5:12Overlay control plane
- 5:54Pod routing - best practices
- 7:40Pod routing (GCP)
- 10:58AWS EKS CNI plugin
- 12:01Lyft AWS CNI plugin
- 13:35Role of kube-proxy
- 14:28kube-proxy modes
- 15:02iptables overview
- 15:24iptables: how it works
- 15:52iptables challenges
- 16:58IPVS overview
- 17:43IPVS implementation
- 18:02IPVS challenges
- 19:46kube-proxy alternatives
- 20:17Features in preparation
- 21:17Load-balancer services
- 22:20External Traffic Policy: Local
- 23:35L7-proxy ingress controller
- 24:04With native pod routing
- 24:21Implementations
- 24:57Remaining challenges
- 25:39DNS challenges
- 26:29DNS node local cache
- 27:09DNS future directions
- 27:44Current DNS setup at Datadog
- 30:37Where should community go?
related talks
32mThe Easy--Don't Drive Yourself Crazy--Way to Kubernetes Networking
22mNavigating the Kubernetes Networking Maze
23mMaking thousands of ingresses available on independent K8s clusters
15mSimplify Kubernetes Connectivity
37mPanel Discussion: Cloud Native Networking State of the Union
40mCloud, Kubernetes, and Service Networking - Taming the Turtles
34mLow Latency Multi-cluster Kubernetes Networking in AWS
32mUnderstand and Troubleshoot the “Magic” of Kubernetes Networking - Minhan Xia & Rohit Ramkumar