Flagger

Flagger is an open-source progressive delivery tool for Kubernetes that automates canary releases and A/B testing. It aims to reduce deployment risk by gradually shifting traffic from a baseline version to a new candidate while evaluating real-time metrics before promoting or rolling back.

Flagger operates as a Kubernetes controller and uses a CustomResourceDefinition called Canary to describe a rollout. A Canary resource specifies the baseline and the canary versions, the traffic routing method, and the analysis configuration. It can route traffic through service meshes such as Istio or Linkerd, or through compatible ingress controllers like NGINX. Flagger provisions and manages the canary deployment, gradually increases traffic to the new version, and monitors metrics from a backend (commonly Prometheus).

During the rollout, Flagger performs automated analysis based on defined criteria for metrics such as error rate, latency, or custom signals. If the metrics meet the success conditions, Flagger promotes the canary to the primary deployment; if not, it rolls back to the previous stable version. The traffic-shifting process is configurable in steps, allowing observable rollout progress and easier rollback if issues arise.

Origins and scope: Flagger was originally developed by Weaveworks and later expanded by an active community of contributors. It supports multiple traffic routing options across Kubernetes deployments, enabling progressive delivery workflows in diverse environments, from simple deployments behind NGINX to advanced service-mesh configurations with Istio or Linkerd.

Limitations and considerations: Effective use requires compatible routing and monitoring infrastructure, thoughtful analysis configurations, and coordination with CI/CD processes. It is best suited for Kubernetes-based microservices architectures seeking automated, metric-driven release promotion and rollback.