When I replicate the application on more than one pod, the web-app will return a Http 504 while using a NGINX load-balancer.

The NGINX load-balancer is outside of the K8 cluster and acts as a reverse-proxy + load-balancer. Hence, NGINX will forward the requests to one node, hosting the web-app container. Important: I don't want the NGINX host to be part of the cluster. (As long as it can be prevented)

upstream website {
             ip_hash;
             server 1.1.1.1:30300;
             #server 2.2.2.2:30300;
}

server {
    listen                          443 ssl http2;
    server_name                     example.com;

    location / {
            proxy_pass http://website;

            proxy_cache off;
            proxy_buffering off;

            proxy_read_timeout 1d;
            proxy_connect_timeout 4;
            proxy_send_timeout 1d;

            proxy_set_header Upgrade $http_upgrade;
            proxy_set_header Connection $http_connection;

            proxy_set_header Host $http_host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    }
}

This config does work, if, and only if, the app has been published to the 1.1.1.1 node only. If I replicate the web-app to 2.2.2.2 as well, the snippet above will already lead to a 504, even thought 2.2.2.2 is still commented out. Commenting the 2.2.2.2 in, won't change anything.

As far as I understood, the NodePort is a public-available port, mapping to an internal port. (Called port) Hence, NodePort 30300 will be forwarded to 2000, which is also my targetport the web-app listens on. Upon replication the second pod will also host the web-app (+ microservices) and expose itself to NodePort 30300. So we do have two NodePorts 30300 within our k8 network and I guess this might lead to confusion and routing issues.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: swiper-web
  namespace: default
spec:
  replicas: 2
  selector:
    matchLabels:
      app: swiper

  template:
    metadata:
      labels:
        app: swiper
    spec:
      containers:
 
      - name: swiper-web-app-example
        image: docker.example.com/swiper.web.app.webapp:$(Build.BuildId)
        ports:
        - containerPort: 2000
        resources:
          limits:
            memory: "2.2G"
            cpu: "0.6"


      - name: swiper-web-api-oauth
        image: docker.example.com/swiper.web.api.oauth:$(Build.BuildId)
        ports:
        - containerPort: 2010
        resources:
          limits:
            memory: "100M"
            cpu: "0.1"

      imagePullSecrets:
      - name: regcred

      dnsPolicy: "None"
      dnsConfig:
        nameservers:
        - 8.8.8.8

---

apiVersion: v1
kind: Service
metadata:
  name: swiper-web-service
  namespace: default
spec:
  type: NodePort
  selector:
    app: swiper
  ports:
  - name: swiper-web-app-example
    port: 2000
    nodePort: 30300

  - name: swiper-web-api-oauth
    port: 2010

Edit:

Adding externalTrafficPolicy: Local to the swiper-web-service solves the issue. Both endpoints are now reachable. But the load-balacing of the other microservices is now disabled.