Have you lost your node...again?
summary
Explains why changing host networking on Kubernetes often kills nodes and shows a concrete solution: Kubernetes nmstate runs an agent that turns policies into per-node node network configuration enactment objects, applies changes one node at a time as a canary test, runs connectivity checks to gateway/DNS/API and automatically rolls back failed nodes. Demonstrates a real failure (bonding left without IP/DHCP), the fix (declare IPv4 DHCP on the bond), and gaps remaining: no pod draining before reconfigure and no group-based rollout API yet.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:40What is Kubernetes
- 6:38Network configuration
- 11:01nmc
- 17:03Network configuration enactment
- 19:01Missing features
related talks
28mHow to Break your Kubernetes Cluster with Networking
32mUnderstand and Troubleshoot the “Magic” of Kubernetes Networking - Minhan Xia & Rohit Ramkumar
35mKubernetes, Lessons Learnt
11mSIG-NETWORK: Updates and Directions
34mIt's Not Always the Network (But Here's How to Prove It): Kubernetes Packet...
20mKubernetes Networking 101
35mDeclarative Networking in Declarative World, ver. 2025
51mKubernetes SIG-Network: Intro and Deep-Dive