K
Q

UDP response issues

1/24/2016

I'm having an issue with port forwarding UDP traffic in kubernetes. I am running a coreos baremetal setup and in the past just used fleet to run my containers, so these containers to work and my network and port forwarding is setup correctly. I can manually run the container and port forward to it and things work as expected, so it seems something is going on with kubernetes and UDP. I have multiple services that are not working properly, but the easiest one is this mumble server. Here is the setup for it starting with replication controller.

apiVersion: v1
kind: ReplicationController
metadata:
  name: mumble-v0
  labels:
    app: mumble
    version: v0
spec:
  replicas: 1
  selector:
    app: mumble
    version: v0
  template:
    metadata:
      labels:
        app: mumble
        version: v0
    spec:
      containers:
      - name: mumble
        image: coppit/mumble-server
        imagePullPolicy: Always
        resources:
          limits:
            cpu: 0.5
            memory: 500Mi
        ports:
        - containerPort: 64738
          name: mumble
        - containerPort: 64738
          name: mudp
          protocol: UDP

And the service itself:

apiVersion: v1
kind: Service
metadata:
  name: mumble
  labels:
    app: mumble
    kubernetes.io/name: "mumble"
spec:
  selector:
    app: mumble
  ports:
  - name: mumble
    port: 64738
  - name: mumble-udp
    port: 64738
    protocol: UDP
  externalIPs: ["10.0.1.19"]

Clients are able to connect to the service and see the server, but no voice traffic flows properly. This voice traffic is sent and received via UDP. Looking around, I see other reports of UDP issues that seem to be related Responses from kubernetes containers getting lost and Problems on running a SIP application (UDP) on Kubernetes.

Does anyone know what could be going wrong here or the fix?

-- Mike Wilson
coreos
docker
flannel
kubernetes

Similar Questions

Using Kubernetes to manage deployment on specific and targeted nodes
Google Container Engine Private Kubernetes LoadBalancer
Kubernetes service in java does not resolve restarted service/replicationcontroller
Why the pods returns Error or ExitCode:0 even they run successfully?

1 Answer

8/17/2016

If you kubectl get svc you should see that it either opened the UDP or the TCP port. Kubernetes supporting both UDP and TCP on the same port seems to be an ongoing issue.

If you changed port number, you'd get an error "cannot create an external load balancer with mix protocols" (see related issue).

What you can do is create two services with the same static IP (should be reserved):

apiVersion: v1
kind: Service
metadata:
  name: mumble-tcp
  labels:
    project: mumble
spec:
  type: LoadBalancer
  loadBalancerIP: 10.10.10.10
  ports:
    - port: 64738
  selector:
    name: mumble
    project: mumble
---
apiVersion: v1
kind: Service
metadata:
  name: mumble-udp
  labels:
    project: mumble
spec:
  type: LoadBalancer
  loadBalancerIP: 10.10.10.10
  ports:
    - port: 64738
      protocol: UDP
  selector:
    name: mumble
    project: mumble
-- Wernight
Source: StackOverflow