Canary Ingress only update the first backend per rule

[rafaelgaspar]

Flagger version: 1.0.0 revision 682a1bf
Flagger CRD version: v1beta1
Kubernetes version: v1.16.8-eks-e16311
Mesh Provider: nginx

Hi,

I have an ingress resource that has 2 rules with the second rule having 3 paths, all of them pointing to the apex service.

The related canary ingress is properly created, but only the backend of the first rule, and the backend of first path of the second rule is properly updated to point to the canary service.

Follows the spec for the apex ingress:

spec:
  rules:
    - host: ui.my.service.com
      http:
        paths:
          - path: /path0/
            backend:
              serviceName: my-service
              servicePort: http
    - http:
        paths:
          - path: /path1/
            backend:
              serviceName: my-service
              servicePort: http
          - path: /path2/
            backend:
              serviceName: my-service
              servicePort: http
          - path: /path3/
            backend:
              serviceName: my-service
              servicePort: http

And the canary ingress:

spec:
  rules:
    - host: ui.my.service.com
      http:
        paths:
          - path: /path0/
            backend:
              serviceName: my-service-canary
              servicePort: http
    - http:
        paths:
          - path: /path1/
            backend:
              serviceName: my-service-canary
              servicePort: http
          - path: /path2/
            backend:
              serviceName: my-service
              servicePort: http
          - path: /path3/
            backend:
              serviceName: my-service
              servicePort: http

[rafaelgaspar]

I'm not very fluent in go, but it looks like the break here is the cause of the bug.

https://github.com/weaveworks/flagger/blob/1a6593727871f1b5af19158c9eb41fe98d005316/pkg/router/ingress.go#L43-L52

Was there a reason for the break other than performance implications?

[stefanprodan]

@rafaelgaspar my assumption was that there is a single backend. Would you mind opening a PR and remove the break? It should fix the bug.