Progressive delivery with Argo Rollouts

Argo Rollouts replaces a Deployment with a Rollout that shifts traffic step by step (canary) or switches a preview stack (blue-green), and can roll back on bad metrics.

Argo CD decides what runs; a Deployment's rolling update decides how Pods are replaced — and it cannot pause at 10% traffic or check error rates. Argo Rollouts (a sister project) adds that. It is a controller plus a CRD, Rollout, that you keep in Git like any other manifest.

Canary

apiVersion: argoproj.io/v1alpha1
kind: Rollout
metadata:
  name: shop-api
spec:
  replicas: 5
  selector: { matchLabels: { app: shop-api } }
  template: { ... same Pod template as the Deployment ... }
  strategy:
    canary:
      canaryService: shop-api-canary
      stableService: shop-api-stable
      trafficRouting:
        nginx: { stableIngress: shop-api }
      steps:
        - setWeight: 10
        - pause: { duration: 5m }
        - analysis: { templates: [{ templateName: error-rate }] }
        - setWeight: 50
        - pause: {}                 # wait for a human: kubectl argo rollouts promote
        - setWeight: 100

Analysis with Prometheus

apiVersion: argoproj.io/v1alpha1
kind: AnalysisTemplate
metadata:
  name: error-rate
spec:
  metrics:
    - name: error-rate
      interval: 1m
      count: 5
      failureLimit: 1
      successCondition: result[0] < 0.01
      provider:
        prometheus:
          address: http://prometheus.monitoring:9090
          query: |
            sum(rate(http_requests_total{app="shop-api",rollout="canary",code=~"5.."}[2m]))
            / sum(rate(http_requests_total{app="shop-api",rollout="canary"}[2m]))

If the analysis fails, the Rollout aborts: traffic goes back to the stable version automatically, without waiting for a human.

Blue-green

strategy:
  blueGreen:
    activeService: shop-api
    previewService: shop-api-preview
    autoPromotionEnabled: false
    prePromotionAnalysis: { templates: [{ templateName: smoke }] }

The new version runs next to the old one behind a preview Service; promotion switches the active Service in one step.

	Canary	Blue-green
Traffic shift	Gradual (10% → 50% → 100%)	All at once
Extra capacity	Small	A full second stack while both run
Best for	High-traffic APIs with good metrics	Apps that need a full check before any user sees them

Working with Argo CD

Argo CD knows the Rollout's health: Paused shows as Suspended, an aborted rollout as Degraded.
A Git commit with a new image starts the rollout; promotion and abort happen through the Rollouts plugin or UI, not through Git.
With selfHeal, avoid changing spec by hand during a rollout; use the plugin commands.

kubectl argo rollouts get rollout shop-api -n shop-prod --watch
kubectl argo rollouts promote shop-api -n shop-prod
kubectl argo rollouts abort shop-api -n shop-prod