Haproxy controller mis-sets backends weight to 0 using Gateway resulting in 503 error

[rdavyd]

Description of the problem

I'm testing argo-rollouts 1.7.2 with argoproj-labs/gatewayAPI 0.3.0 version plugin. Installed v1.0.0 Gateway API via
kubectl apply -f https://github.com/kubernetes-sigs/gateway-api/releases/download/v1.0.0/standard-install.yaml
Updated haproxy to 0.15.0 alpha3 to make use of v1.0.0 Gateway API standard.

Rollout manifest exempt:

  strategy:
    canary:
      canaryMetadata:
        annotations:
          role: canary
        labels:
          role: canary
      stableMetadata:
        annotations:
          role: stable
        labels:
          role: stable
      canaryService: testapp-canary
      stableService: testapp
      steps:
      - setWeight: 20
      - setWeight: 40
      - pause: {}
      trafficRouting:
        plugins:
          argoproj-labs/gatewayAPI:
            httpRoute: testapp
            namespace: myns

HTTProute and Gateway are configured like this:

---
apiVersion: gateway.networking.k8s.io/v1
kind: Gateway
metadata:
  name: testapp
  namespace: myns
spec:
  gatewayClassName: haproxy
  listeners:
  - name: http-listener
    port: 80
    protocol: HTTP

---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: testapp
  namespace: myns
spec:
  hostnames:
  - testapp.mycluster.dev
  parentRefs:
  - group: gateway.networking.k8s.io
    kind: Gateway
    name: testapp
  rules:
  - backendRefs:
    - group: ""
      kind: Service
      name: testapp
      port: 8080
      weight: 100
    - group: ""
      kind: Service
      name: testapp-canary
      port: 8080
      weight: 0

Both testapp and testapp-canary Services refer to the same PODs when deployment is stable and argo-rollouts controller sets weight to 0 for the canary service when rollout successfully completes.
When this happens haproxy-controller sets weight of backends to 0:

    server srv001 10.10.180.29:8080 weight 0 check inter 2s
    server srv002 10.10.180.90:8080 weight 0 check inter 2s
    server srv003 10.10.180.91:8080 weight 0 check inter 2s
    server srv004 10.10.180.92:8080 weight 0 check inter 2s
    server srv005 10.10.180.93:8080 weight 0 check inter 2s

This results in returning 503 error when curling testapp.mycluster.dev
If I manually edit HTTPRoute and set canary services weight to any non-zero setting everything starts working.

Expected behavior

Haproxy controller shouldn't set backend weight to zero when the one of multiple services in the HTTPRoute has 0 weight set.

Steps to reproduce the problem

• Create Deployment and two Services with different names both referring to the same Deployment PODs
• Create Gateway and HTTPRoute as pointed out above
• Curling the Hostname from HTTPRoute produces 503 error
• Manually set second Service weight to non-zero value
• Curling the Hostname from HTTPRoute produces expected result

Environment information

HAProxy Ingress version: v0.15.0 alpha3
argo-rollouts 1.7.2 with argoproj-labs/gatewayAPI version 0.3.0
Gateway API v1.0.0