Why all requests are going to the same pod when replica is set to 2?

General Discussions
1

Here is a Kubernetes yaml file:


apiVersion: apps/v1
kind: Deployment
metadata:
  name: opp-deployment
spec:
  replicas: 2
  selector:
    matchLabels:
      app: opp
  template:
    metadata:
      labels:
        app: opp
    spec:
      containers:
      - name: opp-container
        image: opp_svc:1.0.2
        ports:
        - containerPort: 51061
  ---
apiVersion: v1
kind: Service
metadata:
  name: opp-service
spec:
  selector:
    app: opp
  ports:
    - protocol: TCP
      port: 8080
      targetPort: 51061
  type: LoadBalancer

So, “kubectl apply -f ./kube-start-opp-svc.yaml” run succesfully and able to see 2 pods running.

Now, to be able to make a http call to the “opp-service”, run in dedicated terminal this command:
kubectl port-forward service/opp-service 8080:8080
to forward request from my local machine to the service and I am expecting the service to use “round-robin” algorithm to send request to different pods.

So, I have 2 pods:
opp-deployment-85dc5b8496-b7qnn and opp-deployment-85dc5b8496-xsnvd

For some reason all request are going only to the pod: opp-deployment-85dc5b8496-xsnvd

same-pod-q1
same-pod-q11920×1080 173 KB

I’ve used “curl http://localhost:8080/health” command to run serveral times from one terminal, from another terminal and the I coded a go program that makes 10 parallel request every 1 seconds.
All calls goes to the same pod. Why?

2

With port-forward, the service is resolved to a specific endpoint one time, not every new connection. To do it on every connection would be effectively implementing kube-proxy.

See ResourceLocation in kubernetes/pkg/registry/core/service/storage/storage.go at master · kubernetes/kubernetes · GitHub

Port-forward is not really a production-use thing.

3

make sense now!
Thank you