flagger

Create a Canary resource targeting your Deployment with an analysis block defining interval, threshold, stepWeight, and maxWeight. Flagger creates primary and canary Deployments and services, then shifts traffic gradually while checking metrics before promoting.

Canary uses stepWeight and maxWeight for gradual traffic shifting, A/B testing uses iterations plus header or cookie match conditions to route specific users, and blue/green uses iterations for an instant switch, optionally with traffic mirroring.

No. With the kubernetes provider (plain CNI), Flagger only supports blue/green deployments using iterations. Weighted canary requires a service mesh like Istio or Linkerd, or an ingress controller such as NGINX or Contour.

The most common cause is an unreachable Prometheus metricsServer, which makes the analysis loop fail at the first check. Also verify CRDs are installed, the targetRef points to a Deployment, and webhook URLs are reachable from the Flagger pod.

Define a MetricTemplate CRD with your Prometheus address and query using the template variables namespace, target, and interval. Reference it in the Canary analysis metrics list via templateRef with thresholdRange values.