Status · Alpha
Flagger Demo Kind
Open-source macOS lab: progressive delivery with Flagger, Gateway API, Cilium and Envoy Gateway — participants apply YAML, not only make shortcuts.
Problem faced
Teams still confuse Kubernetes rolling updates with traffic decisions: Ready does not mean a revision deserves 100% of users.
Context. 60-minute teaching workshop, isolated Kind cluster, FR/EN docs, Kind/Cilium/Envoy/Flagger E2E CI.
Role
Author
Structural choices
- Gateway API + Envoy Gateway instead of Istio to match modern platform paths
- Canary metric = Podinfo app SLI (teaching) with rate window aligned to Flagger analysis
- Scenarios via manifests/tp (kubectl patch/apply) so the control loop stays readable
- Teaching lab, not a production template
- Tools and kubeconfig isolated in the repo (.tools)
- FR canonical for facilitators; EN for participant entry and architecture
How it works
- Repository EtienneDeneuve/flagger-demo-kind (MIT)
- Participant / facilitator guides and English mirrors
- Validate CI + real Kind E2E
- Flagger
- Gateway API
- Envoy Gateway
- Cilium
- Kind
- Prometheus
- Kubernetes
Concrete limits
- No GitOps, TLS, NetworkPolicy, or end-to-end Envoy metrics
- Targets macOS + Docker engine; no documented Linux-desktop participant path
Learnings
- Without config to apply, make good/fail/ab stays magical in a workshop
- A PromQL window that outlives fault injection fails the next A/B