Kubernetes clusters should be ephemeral
summary
Claim: make Kubernetes clusters disposable and replace them regularly instead of in-place upgrades. Automation uses software-defined manifests, software-defined networking, and orchestration of node and app manifests to stand up 2–4 clusters across regions, shift traffic with blue‑green or canary, then tear down the old. Practice runs monthly for a year, using EKS and seeing problems like Prometheus missing during brings up but fixing automation so replacements reliably work.
AI-written summary · about
related talks
39mRunning Kubernetes on Ephemeral Infrastructure
25mAutomating Kubernetes Upgrades for Zero Downtime
18mAutomating Kubernetes Cluster Updates: Achieving Zero Downtime Effortlessly
35mKubernetes journey
47mPostmortem: A Modern Approach to Managing Kubernetes Clusters
32mDon't Start With 500 Clusters: Patterns in Scaling Multi-Cluster Kubernetes…
26mMulti-Tenant vs Micro-Clusters in Kubernetes-Driven Platform Engineering
32mBuilding Reliability into Kubernetes using Chaos Experiments with Terraform