Techniques and Insights To Test Kubernetes Limits With Kind
Talkat CNCF Maintainer Summit Europe 2025 · on Monday, 2025-03-31· at 16:40· Level 3 | ICC Capital Suite 14-16·
summary
Shows how Kind creates compact clusters that emulate large-scale failures: fakes 50,000 endpoints via Endpointslices to trigger 50,000 endpoints, finds a Go 1.22 DNS resolver bug that disappears with Go 1.23, and demonstrates kube-proxy in iptables mode generating ~100k iptables rules that stall iptables-restore while nftables programs the datapath orders of magnitude faster. Covers practical commands (kind create, kind load docker-image with full docker.io path), metrics to measure programming latency, and when Kind is sufficient vs when real clusters are required.
AI-written summary · about
chapters (12)
- 0:00Kubernetes scalability testing
- 3:51Why Kubernetes APIs matter
- 5:59Debugging DNS issues
- 7:45Creating a kind cluster
- 12:27Testing kube-proxy
- 13:30kube-proxy performance testing
- 20:22Scalability testing takeaways
- 21:25kind limitations for production
- 24:14Q&A scalability limits
- 26:43Q&A real-world clusters
- 27:46Q&A optimizing metrics
- 29:32Closing remarks
related talks
29mKubernetes Advanced Networking Testing with KIND
31mDeep Dive: Kind
36mTesting your K8s apps with KIND
01h 31mTutorial: A Kind Workflow for Contri...
48mScale and Performance Testing of Kubernetes: Answers for Specific Applications
35mKubernetes, Lessons Learnt
21mDeep Dive: Testing SIG
34mA Huge Cluster or Multi-Clusters? Identifying the Bottleneck