K
Q

How to resolve pod hostnames from other pods?

12/9/2019
  • I have 2 pods running on 2 nodes, each pod runs in different node.
  • These nodes are on the same subnet and can TCP/UDP/ICMP themselves.

These pods got some hostnames, ie:

  • drill-staging-75cddd789-kbzsq
  • drill-staging-75cddd789-amsrj

From pod drill-staging-75cddd789-kbzsq I cannot resolve host name for drill-staging-75cddd789-amsrj and vice versa. Resolving self pod's name works.

I tried setting various dnsPolicies:

  • ClusterFirst: no luck
  • Default: no luck
  • ClusterFirstWithHostNet: no luck and it event couldn't resolve hostname of it's own node
  • None: not tried (I don't think it's a good way)
apiVersion: apps/v1
kind: Deployment
metadata:
  name: {{ include "app.name" . }}
  namespace: {{ .Values.global.namespace }}
spec:
  replicas: 2
  selector:
    matchLabels:
      app: {{ include "app.name" . }}
  template:
    metadata:
      labels:
        app: {{ include "app.name" . }}
    spec:
      containers:
      - name: {{ include "app.name" . }}
        image: ...
        resources:
          ...
        ports:
          ...
        imagePullPolicy: Always
      restartPolicy: Always
-- rudolfdobias
kube-dns
kubernetes
kubernetes-networking
networking

Similar Questions

zalando skipper east-to-west communication (internal service to service) cross namespace
Helm v3 cannot find my charts in private repo
non ASCII filenames on GCP persistantvolume not displaying correctly
Minikube Retriable failure: create: Error creating machine: Error in driver during machine creation: hyperkit crashed! command line:
Why doesn't "helm install --dry-run" give me an error about a conflict but an actual "helm install" does?
Skaffold config dependencies with profiles

1 Answer

12/10/2019

Normally, only Services get DNS names, not Pods. So, by default, you can't refer to another Pod directly by a domain name, only by its IP address.

Pods get DNS names only under certain condidtions that include a headless Service, as explained in the documentation. In particular, the conditions are:

  • The Pods have a hostname field
  • The Pods have a subdomain field
  • There is a headless Service (in the same namespace) that selects the Pods
  • The name of the headless Service equals the subdomain field of the Pods

In this case, each Pod gets a fully-qualified domain name of the following form:

my-hostname.my-subdomain.default.svc.cluster.local

Where my-hostname is the hostname field of the Pod and my-subdomain is the subdomain field of the Pod.

Note: the DNS name is created for the "hostname" of the Pod and not the "name" of the Pod.

You can test this with the following setup:

apiVersion: v1
kind: Service
metadata:
  name: my-subdomain
spec:
  selector:
    name: my-test
  clusterIP: None
---
apiVersion: v1
kind: Pod
metadata:
  name: my-pod-1
  labels:
    name: my-test
spec:
  hostname: my-hostname-1
  subdomain: my-subdomain
  containers:
  - image: weibeld/ubuntu-networking
    command: [sleep, "3600"]
    name: ubuntu-networking
---
apiVersion: v1
kind: Pod
metadata:
  name: my-pod-2
  labels:
    name: my-test
spec:
  hostname: my-hostname-2
  subdomain: my-subdomain
  containers:
  - image: weibeld/ubuntu-networking
    command: [sleep, "3600"]
    name: ubuntu-networking

After applying this, you can exec into one of the Pods:

kubectl exec -ti my-pod-1 bash

And you should be able to resolve the fully-qualifed domain names of the two Pods:

host my-hostname-1.my-subdomain.default.svc.cluster.local
host my-hostname-2.my-subdomain.default.svc.cluster.local

Since you're making the requests from the same namespace as the target Pods, you can abbreviate the domain name to:

host my-hostname-1.my-subdomain
host my-hostname-2.my-subdomain
-- weibeld
Source: StackOverflow