WebSocket not works using GCE Ingress. Ingress set UNHEALTHY status always

[masalinas]

I have 1.14 kubernetes version on GCE so the ingress-gce version is v1.6.0. with many services deployed. One of then is named trazelab-broker that expose the 9001 port of MQTT container throw the 80 ingress port service. I configured ingress but ingress always create a Healh check for this backen service to use HTTP port and not TCP, so the status is UNHEALTHY always and the port is closed for any communication from internet to this websocket.

If I set the correct port using the GCE UI and set the port to TCP the backen service status will convert to HEALTH after some seconds and the port is active correctly. Nevertheless some seconds later 60 seconds aprox, the status of the backend service come back again to UNHEALTHY and the port again is set to HTTP.

My questions is:

• How could I set the TCP connection for this health check asocaited to the backend service created by ingrees and not the default HTTP?
• Why if I set this port to TCP from UI later ingress or GCE set again the default type to HTTP?

These are the manifests for this service:

BackendConfig;

apiVersion: cloud.google.com/v1beta1
kind: BackendConfig
metadata:
  name: config-broker-ws
  namespace: trazelab-mefasa
spec:
  timeoutSec: 300
  connectionDraining:
    drainingTimeoutSec: 300
  healthCheck:
    type: TCP
    port: 80

Deployment and Service

apiVersion: v1
kind: Service
metadata:
  name: trazelab-broker
  namespace: trazelab-mefasa
  annotations:
    beta.cloud.google.com/backend-config: '{"ports": {"80": "config-broker-ws"}}'
  labels:
    app: trazelab-mefasa
spec:
  selector:
    app: trazelab-mefasa
    tier: broker
  type: NodePort
  ports:
    - name: tcp-port
      port: 1883
      targetPort: 1883
      protocol: TCP
    - name: tcp-ws-port
      port: 80
      targetPort: 9001
      protocol: TCP
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: trazelab-broker
  namespace: trazelab-mefasa
  labels:
    app: trazelab-mefasa
spec:
  replicas: 1
  strategy:
    type: Recreate
  selector:
    matchLabels:
      app: trazelab-mefasa
      tier: broker
  template:
    metadata:
      labels:
        app: trazelab-mefasa
        tier: broker
    spec:
      containers:
      - name: mosquitto
        image: eclipse-mosquitto:latest
        imagePullPolicy: Always
        ports:
          - containerPort: 9001
          - containerPort: 1883
        livenessProbe:
          tcpSocket:
            port: 9001
        readinessProbe:
          tcpSocket:
            port: 9001
        volumeMounts:
          - name: trazelab-broker-config
            mountPath: /mosquitto/config
      volumes:
        - name: trazelab-broker-config
          configMap:
            name: trazelab-broker-config
            items:
              - key: broker-config
                path: mosquitto.conf

Ingress manifest:

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: trazelab-ingress
  namespace: trazelab-mefasa
  labels:
    app: trazelab-mefasa
spec:
  rules:
    - host: mefasa.trazelab.io
      http:
        paths:
          - path: /*
            backend:
              serviceName: trazelab-ui
              servicePort: 80
    - host: api.mefasa.trazelab.io
      http:
        paths:
          - path: /*
            backend:
              serviceName: trazelab-api
              servicePort: 3000
    - host: gateway.mefasa.trazelab.io
      http:
        paths:
          - path: /*
            backend:
              serviceName: trazelab-gateway
              servicePort: 1880
    - host: broker-ws.mefasa.trazelab.io
      http:
        paths:
          - path: /*
            backend:
              serviceName: trazelab-broker
              servicePort: 80

[rramkumar1]

@masalinas Sorry for the delay in response. The reason that TCP health checks do not work is that they are not yet supported in the BackendConfig resource. You specified a healthCheck field in your BackendConfig resource in the original comment but that is not yet supported in Ingress-GCE 1.6

We are currently in the process of adding initial support for all health check protocols in the BackendConfig. We can take a feature request to add support for TCP as well.

[masalinas]

I resolved the problem installing nginx ingress controller behind a LoadBalancer. nginxs is prepared with the path /healthz to be checked correctly.

Regards