Manually creating EndpointSlice object doesn't work

Kubernetes
#1

Hello,

I have an nginx pod running in my cluster, and tried to expose it via the combination of an EndpointSlice object and a service without a selector.

First, I created a service without a selector by:
kubectl create svc clusterip nginx --tcp 80:80

Then. I created an EndpointSlice manually based on the snippet from

with a single endpoint address, that of the nginx pod running in my cluster.
I used the label ‘kubernetes.io/service-name: nginx’ as suggested in the docs.

I was expecting for K8s to associate the service with the EndpointSlice object, however that didn’t happen. The service has no endpoints, and after creating the service a new EndpointSlice object with a random suffix in its name, no ports and no endpoints is automatically created.

If someone has tried and succeeded in exposing a set of pods via a manually created EndpointSlice with the pod IP addresses, I would appreciate your feedback and help on how to make this work.

Thanks in advance.

#2

Hi! Can you share your configuration for pod and EndpointSlice?

#3 
apiVersion: v1
kind: Pod
metadata:
  name: nginx
  labels:
    app: test
spec:
  containers:
  - name: nginx
    image: nginx:alpine
---
apiVersion: v1
kind: Service
metadata:
  name: nginx
  labels:
    app: test
spec:
  type: ClusterIP
  ports:
  - port: 80
    targetPort: 80

Then I check the IP assigned to the pod, let’s say it’s 192.168.1.4. I go on and create an endpointslice object as follows.

apiVersion: discovery.k8s.io/v1
kind: EndpointSlice
metadata:
  name: nginx
  labels:
    kubernetes.io/service-name: nginx
    app: test
addressType: IPv4
ports:
  - name: http
    protocol: TCP
    port: 80
endpoints:
  - addresses:
      - "192.168.1.4"

So after having created these 3 objects, I check for endpoints with “kubectl get ep”, but no endpoints object has been created for the nginx service.

What am I missing?

#4

It looks like this issue has recently been acknowledged and fixed.

We’ll have to keep an eye on which upcoming Kubernetes release will include the fix.