Freedom of Kubernetes requires Chaos Engineering to shine in production
summary
Kubernetes freedom without controls leads to hundreds or thousands of clusters with unpredictable behavior; missing requests/limits and no resource quotas cause node pressure and pod evictions. A simple CPU limit test shows response time jumping from 200 ms → 3.6 s due to CPU throttling. Practical experiments include removing nodes, injecting CPU/memory hogs to force evictions, running load tests with k6 + Prometheus, using litmus chaos for failures and Captain SLO scoring to automate pass/fail decisions.
AI-written summary · about
chapters (10)
- 0:00Introduction and agenda
- 1:44Kubernetes basics and challenges
- 5:40Resource requests and limits
- 10:13CPU throttling and OOM kills
- 12:01What is Chaos Engineering
- 15:13Observability for Kubernetes
- 17:16Chaos experiments for evictions
- 19:45Automating chaos and load tests
- 21:49Quality gates with SLI and SLO
- 23:17Key takeaways and conclusion
related talks
39mChaos: Breaking your Systems to Make them Unbreakable
34mChaos Engineering For Hybrid Targets With LitmusChaos
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
39mWhy SREs can't afford to NOT do Chaos Engineering
32mImplementing Chaos Engineering in Kubernetes
27mRunning Kubernetes in Production A Million Ways to Crash Your Cluster
28mContaining Chaos with Kubernetes
26mAchieving Cloud-Native Resiliency: The Evolution of Chaos Engineering