在 RPI 上运行的 K8 集群中无法访问 NGINX 节点端口服务

Question

I have set-up a small Kubernetes Cluster on Raspberry Pi currently consisting of 1 Master and 1 Worker. I have created a simple deployment of NGINX and created a NodePort service for this. My YAML looks like this:

apiVersion: v1
kind: Service
metadata:
  name: nginx-service
  labels:
    app: nginx
spec:
  selector:
    app: nginx
  type: NodePort
  ports:
    - nodePort: 30333
      port: 80
      targetPort: 80
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx
  labels:
    app: nginx
spec:
  replicas: 2
  selector:
  matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: nginx:1.16.1
        imagePullPolicy: IfNotPresent

       ports:
        - containerPort: 80
          protocol: TCP
      restartPolicy: Always

The PODS are up and running and so is the service

$ kubectl get pods
NAME                     READY   STATUS    RESTARTS   AGE
nginx-5d66cc795f-bgqdp   1/1     Running   0          65m
nginx-5d66cc795f-mb8qw   1/1     Running   0          65m

$ kubectl get svc
NAME            TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes      ClusterIP   10.96.0.1        <none>        443/TCP        13d
nginx-service   NodePort    10.104.133.199   <none>        80:30333/TCP   66m

But I am unable to access the NGINX from the master node.

curl http://192.168.178.101:30333
curl: (7) Failed to connect to 192.168.178.101 port 30333: Connection timed out

If I try from the Worker node it works fine, and NGINX responds. From the worker node I can use the IP address:30333 or the hostname:30333, although localhost:30333 does not work!

Connectivity from my Master to Worker seems fine. I can ping, SSH from there etc. either on IP address or hostname.

Any ideas what I have done wrong?

Output from get nodes -o wide:

 $ kubectl get nodes -o wide
NAME        STATUS   ROLES    AGE   VERSION   INTERNAL-IP       EXTERNAL-IP   OS-IMAGE                         KERNEL-VERSION   CONTAINER-RUNTIME
k8-master   Ready    master   13d   v1.17.4   192.168.178.100   <none>        Raspbian GNU/Linux 10 (buster)   4.19.97-v7+      docker://19.3.8
k8-w1       Ready    worker   13d   v1.17.4   192.168.178.101   <none>        Raspbian GNU/Linux 10 (buster)   4.19.97-v7+      docker://19.3.8

Output from describe service:

$ kubectl describe service nginx-service
Name:                     nginx-service
Namespace:                default
Labels:                   app=nginx
Annotations:              kubectl.kubernetes.io/last-applied-configuration:
                            {"apiVersion":"v1","kind":"Service","metadata":{"annotations":{},"labels":{"app":"nginx"},"name":"nginx-service","namespace":"default"},"s...
Selector:                 app=nginx
Type:                     NodePort
IP:                       10.104.133.199
Port:                     <unset>  80/TCP
TargetPort:               80/TCP
NodePort:                 <unset>  30333/TCP
Endpoints:                10.44.0.1:80,10.44.0.2:80
Session Affinity:         None
External Traffic Policy:  Cluster
Events:                   <none>

Update: I tried a simple Telnet command to the Worker on the NodePort 30333. But I get a connection timed out error.

So then, I removed the NodePort service and tried a simple Port forward command

kubectl port-forward pod/nginx-545b8fdd47-brg7r 8080:80

This worked OK, and I could connect from Master -> Worker via this port.

But NodePort service still doesn't work!

Answer 1

So after many hours and days I think I have found the source of the problem.

I found this blog: https://limpygnome.com/2019/09/21/raspberry-pi-kubernetes-cluster/

Which led me to this bug report: https://github.com/kubernetes-sigs/kubespray/issues/4674

So executing the following would allow the connection:

sudo iptables -P FORWARD ACCEPT

But I could not get this to stick after a reboot (even using iptables-persistent), I assume as the rules are updated by Docker/K8s dynamically during startup and after changes.

This led me to investigate further and I found information in the Weave network documentation about issues regarding K8s network with hosts that use iptables v 1.8 and higher. (This seems to have also applied to other K8s networking providers although some of these might have been resolved). I then saw in my Weave log files that it was indeed rejecting requests.

So by executing the following on my nodes I was able to get this working permanently.

sudo update-alternatives --set iptables /usr/sbin/iptables-legacy

I can now access via my NodePort between all nodes, and externally from the cluster onto all nodes.