Not able to access statefulset pod via headless service using fqdn

user_1099142 0
Viewed 1621

I have a k8 setup that looks like this

ingress -> headless service (k8 service with clusterIp: none) -> statefulsets ( 2pods)

Fqdn looks like this:

nslookup my-service
Server:         100.4.0.10
Address:        100.4.0.10#53

Name:   my-service.my-namespace.svc.cluster.local
Address: 100.2.2.8
Name:   my-service.my-namespace.svc.cluster.local
Address: 100.1.4.2

I am trying to reach one of the pod directly via the service using the following fqdn but not able to do so.

curl -I my-pod-0.my-service.my-namespace.svc.cluster.local:8222
curl: (6) Could not resolve host: my-pod-0.my-service.my-namespace.svc.cluster.local

If I try to hit the service directly then it works correctly (as a loadbalancer)

curl -I my-service.my-namespace.svc.cluster.local:8222
HTTP/1.1 200 OK
Date: Sat, 31 Jul 2021 21:24:42 GMT
Content-Length: 656

If I try to hit the pod directly using it's cluster ip, it also works fine

curl -I 100.2.2.8:8222
HTTP/1.1 200 OK
Date: Sat, 31 Jul 2021 21:29:22 GMT
Content-Length: 656
Content-Type: text/html; charset=utf-8

But my use case requires me to be able to hit the statefulset pod using fqdn i.e my-pod-0.my-service.my-namespace.svc.cluster.local . What am I missing here?

kubernetes-pod kubernetes-ingress kubernetes dns
2 Answers

example statefulset called foo with image nginx:

k get statefulsets.apps
NAME   READY   AGE
foo    3/3     8m55s

This stateful set created following pods(foo-0,foo-1,foo-2):

k get pod -o wide
NAME      READY   STATUS    RESTARTS   AGE     IP             NODE        NOMINATED NODE   READINESS GATES
busybox   1/1     Running   1          3h47m   10.1.198.71    ps-master   <none>           <none>
foo-0     1/1     Running   0          12m     10.1.198.121   ps-master   <none>           <none>
foo-1     1/1     Running   0          12m     10.1.198.77    ps-master   <none>           <none>
foo-2     1/1     Running   0          12m     10.1.198.111   ps-master   <none>           <none>

Now create a headless service(clusterIP is none) as follow:(make sure to use correct selector same as your statefulset)

apiVersion: v1
kind: Service
metadata:
  name: nginx
  labels:
    app: foo
spec:
  type: ClusterIP
  clusterIP: None
  ports:
  - port: 80
    name: web
  selector:
    app: foo

Now, do nslookup to see the dns resolution working for the service.(Optional step)

k exec -it busybox -- nslookup nginx.default.svc.cluster.local
Server:    10.152.183.10
Address 1: 10.152.183.10 kube-dns.kube-system.svc.cluster.local

Name:      nginx.default.svc.cluster.local
Address 1: 10.1.198.77 foo-1.nginx.default.svc.cluster.local
Address 2: 10.1.198.111 foo-2.nginx.default.svc.cluster.local
Address 3: 10.1.198.121 foo-0.nginx.default.svc.cluster.local

Now validate that, individual resolution per-pod is working:

k exec -it busybox -- nslookup foo-1.nginx.default.svc.cluster.local
Server:    10.152.183.10
Address 1: 10.152.183.10 kube-dns.kube-system.svc.cluster.local

Name:      foo-1.nginx.default.svc.cluster.local
Address 1: 10.1.198.77 foo-1.nginx.default.svc.cluster.local

More info: Here

Note: In this case OP had incorrect mapping of headless service and the statefulset, this can be verified with below command:

k get statefulsets.apps foo -o jsonpath="{.spec.serviceName}{'\n'}"
nignx

Ensure that, the mapping.

user_63096010

Original answer didn't clarify how OP fixed the issue, the problem was in serviceName property under statefulset implementation.

user_23966740
Related