Progressive Delivery Canary Blue Green and Safe Rollbacks

• Compare standard, blue-green, and canary strategies.
• Explain when to use each strategy.
• Show how Argo Rollouts enables metric-driven progressive delivery on Kubernetes.

​

Why progressive delivery matters — a real-world example

• 100% blast radius if the deployment is faulty.
• No native traffic control for staged exposure.
• Rollbacks can be slow because pods must terminate, be rescheduled, and become ready.
• No built-in automated checks tied to production metrics.

​

Blue-green deployments

• Instant rollback by switching routing to the previous environment.
• Zero-downtime deployments for stateless services when readiness and routing are handled correctly.
• Full testing of the new version before exposing users.

• Requires roughly double the resources during deployment.
• The cutover is all-or-nothing; traffic moves 100% at the switch.
• Database migrations can be complicated because both versions may need compatibility.

​

Canary deployments

1. Route a small percentage (e.g., 5%) to the canary.
2. Run automated analysis on latency, error rates, and business metrics.
3. If checks pass, promote to larger percentages (e.g., 25%, 50%).
4. If any analysis step fails, roll the canary back to 0%.

• Gradual exposure reduces blast radius.
• Promotions can be automated and metric-driven.
• Real user traffic provides meaningful validation.

• Requires traffic splitting support (ingress, service mesh, or load balancer).
• More operational complexity to set up and maintain.
• Requires an observability stack (Prometheus, tracing, logs) to drive decisions.

​

Argo Rollouts — Kubernetes-native progressive delivery

• Sequential steps: adjust weight, pause, or run analysis.
• setWeight: change the percentage of traffic sent to the canary.
• pause: wait for a duration or for manual approval.
• analysis: run automated checks (Prometheus queries, webhooks) to decide promotion or rollback.

apiVersion: argoproj.io/v1alpha1
kind: Rollout
metadata:
  name: my-app
spec:
  replicas: 4
  selector:
    matchLabels:
      app: my-app
  template:
    metadata:
      labels:
        app: my-app
    spec:
      containers:
        - name: my-app
          image: my-app:v2
  strategy:
    canary:
      steps:
        - setWeight: 5
        - pause:
            duration: 10s
        - setWeight: 25
        - analysis:
            templates:
              - templateName: success-rate
        - setWeight: 100

apiVersion: argoproj.io/v1alpha1
kind: AnalysisTemplate
metadata:
  name: success-rate
spec:
  metrics:
    - name: request-success-rate
      interval: 30s
      count: 3
      successCondition: result > 0.99
      provider:
        prometheus:
          address: http://prometheus-operated.monitoring.svc:9090
          query: |
            sum(rate(http_requests_total{job="my-app",status!~"5.."}[1m]))
            /
            sum(rate(http_requests_total{job="my-app"}[1m]))

• Native support for canary and blue-green strategies.
• Integrations with Istio, Gateway API, and cloud load balancers for traffic split.
• AnalysisTemplates to automate checks using Prometheus, webhooks, or custom providers.
• Integration with Argo CD for visibility and GitOps workflows.

​

Strategy comparison

​

Summary

• Standard Kubernetes rollouts expose all users at once and have a 100% blast radius.
• Blue-green provides an instant switch and rollback at the cost of running duplicate environments and complex migrations.
• Canary provides gradual exposure with metric-driven promotions and safe automatic rollbacks, but requires traffic splitting and observability.
• Argo Rollouts is a Kubernetes-native controller that implements both patterns and adds analysis templates for automated promotion and rollback.

​

Links and references

• Argo Rollouts: https://argoproj.github.io/argo-rollouts/
• Argo CD: https://learn.kodekloud.com/user/courses/gitops-with-argocd
• Flagger: https://flagger.app/
• Istio: https://istio.io
• Gateway API: https://gateway-api.sigs.k8s.io/
• Prometheus: https://prometheus.io