Argo Rollouts Canary Delivery

Argo Rollouts: replaces the Deployment with a Rollout custom resource that adds fine-grained progressive delivery. Beyond the basic rolling update, it orchestrates canary and blue-green releases, integrates with traffic providers to shift percentages of traffic, and can automatically promote or abort based on metric analysis.

Canary steps: strategy.canary.steps is an ordered list like setWeight: 20, then pause: {} or pause: { duration: 5m }, then setWeight: 50, and so on. Each step shifts a defined percentage of traffic to the new (canary) ReplicaSet and optionally pauses for observation or manual approval before continuing.

AnalysisTemplate: declares metrics (such as a Prometheus success-rate query) with success/failure thresholds. During a canary step, a referenced AnalysisRun samples those metrics; if they fall outside the success criteria, the rollout aborts and rolls back automatically — enabling metric-driven, hands-off progressive delivery.

Traffic routing: rather than relying solely on pod ratios, strategy.canary.trafficRouting plugs into providers like Istio, NGINX, SMI, or AWS ALB to direct an exact weight (e.g. 20%) to the canary version. This decouples the traffic split from the number of replicas, giving precise control over exposure during a release.

Automated rollback: if an AnalysisRun reports failure (metrics breach the failure condition or exceed the failure limit), Argo Rollouts aborts the rollout, shifts traffic back to the stable ReplicaSet, and halts promotion. This automatic safety net is the core benefit of pairing canary steps with metric analysis.