Networking Basics

In Docker world, when you use docker-compose, all the networking is done for you automatically behind the scenes. However with Kubernetes, you will need to set up a lot of the networking yourself, by creating service and ingress. Kubernetes does come with some basic networking features out-of-the-box:

  • A pod's internal networking
  • IP based pod-to-pod networking
  • IP based kube-node-to-pod networking
  • dns-service (kube-dns)

A pod's internal networking (eg1-networking-inside-pods)

If you have 2+ containers inside a single pod, then these containers can reach each other via localhost. Let's say we create the following 2-container pod:

apiVersion: v1
kind: Pod
  name: pod-demo      # both container's hostname will be set to this
    component: centos
    - name: cntr-httpd
      image: httpd:latest
        - containerPort: 80
    - name: cntr-centos
      image: centos
      command: ["/bin/bash", "-c"]
        - |
          while true ; do
            curl http://localhost
            sleep 10

This defines a single pod that's housing 2 containers (and in our case kubernetes has auto-assigned an IP of to the pod):

$ kubectl get pods -o wide
pod-demo   2/2     Running   0          28m   minikube   <none>           <none>

These containers should be able to talk to each other without needing any further configurations. Let's start by logging into one of the containers:

$ kubectl exec pod-demo -c cntr-centos -it -- /bin/bash
[root@pod-demo /]#

Notice that I also specified the -c (container) flag followed by the container name. That's only required when logging into a container in a multi-container pod, so that kubectl knows which container you want to access. If you omit this then kubectl will default to logging into the first container that's listed in the pod yaml descriptor. With multi-container pods it's really important to only have one primary container, and all other containers act as secondary containers. Meaning that, if the secondary containers fails, then the pod's main app (primary container) still continues to function. All containers in a pod are actually assigned the same hostname, which is the pod's hostname.:

[root@pod-demo /]# hostname

You'll find is that each container is attached to 2 network interfaces:

[root@pod-centos /]# yum install -q -y net-tools      # we have to install ifconfig before we can use it.

[root@pod-centos /]# ifconfig
eth0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST>  mtu 1500
        inet  netmask  broadcast
        ether 02:42:ac:11:00:07  txqueuelen 0  (Ethernet)
        RX packets 3475  bytes 9754927 (9.3 MiB)
        RX errors 0  dropped 0  overruns 0  frame 0
        TX packets 2819  bytes 161157 (157.3 KiB)
        TX errors 0  dropped 0 overruns 0  carrier 0  collisions 0

lo: flags=73<UP,LOOPBACK,RUNNING>  mtu 65536
        inet  netmask
        loop  txqueuelen 1000  (Local Loopback)
        RX packets 40  bytes 3516 (3.4 KiB)
        RX errors 0  dropped 0  overruns 0  frame 0
        TX packets 40  bytes 3516 (3.4 KiB)
        TX errors 0  dropped 0 overruns 0  carrier 0  collisions 0

These are virtual network interfaces that exists as part of the pod. These same network interfaces are actually attached to all the containers in the pod at the same time. Because of that, it means that, from cntr-centos container you can reach the cntr-httpd via the localhost ip address, or the pod's ip address:

[root@pod-centos /]# curl
<html><body><h1>It works!</h1></body></html>

[root@pod-centos /]# curl
<html><body><h1>It works!</h1></body></html>

Here our pod routes this curl request internally to the cntr-httpd container.

Question: If our pod had several other containers, then how would our pod know which container to route the curl request to?

Answer: Our yaml descriptor specifies which port each container is listening on (if any). Therefore our pod knows that requests destined to port 80 should be routed to cntr-httpd container.

You also find that the following works as well:

[root@pod-centos /]# curl http://localhost
<html><body><h1>It works!</h1></body></html>

[root@pod-centos /]# curl http://pod-centos
<html><body><h1>It works!</h1></body></html>

That's because of the /etc/hosts file:

[root@pod-centos /]# cat /etc/hosts
# Kubernetes-managed hosts file.       localhost
::1     localhost ip6-localhost ip6-loopback
fe00::0 ip6-localnet
fe00::0 ip6-mcastprefix
fe00::1 ip6-allnodes
fe00::2 ip6-allrouters      pod-centos

In both 'localhost' and 'hostname' curl requests, a static internal dns-lookup occured thanks to the /etc/hosts entry.

So far we did a cntr-centos-->cntr-httpd, where both containers are in the same pod. But we didn't do it in the other direction. That's because cntr-centos is in theory reachable, but we haven't setup anytihng (e.g. web server) on cntr-centos to listen on any ports.

IP based kube-node-to-pod networking

The container's eth0 interface is on the same network as the the kube masters/worker nodes. This means that you can also curl from inside the master/worker nodes:

$ minikube ssh
                         _             _
            _         _ ( )           ( )
  ___ ___  (_)  ___  (_)| |/')  _   _ | |_      __
/' _  _ \| |/' _ \| || , <  ( ) ( )| '_\  /'__`\
| ( ) ( ) || || ( ) || || |\`\ | (_) || |_) )(  ___/
(_) (_) (_)(_)(_) (_)(_)(_) (_)\___/'(_,__/'\____)

$ curl
<html><body><h1>It works!</h1></body></html>

IP based Pod-2-Pod networking (eg2-pod2pod-networking)

Pod-2-pod is possible thanks to the fact that Kubernetes auto-assigns a private ip address to all pods during the pod's launch time. The pods can reach each other with these ip addresses.

apiVersion: v1
kind: Pod
  name: pod-httpd
    - name: cntr-httpd
      image: httpd:latest
        - containerPort: 80
apiVersion: v1
kind: Pod
  name: pod-nginx
    - name: cntr-nginx
      image: nginx:latest
        - containerPort: 80

The pod's ip are shown below.

$ kubectl get pods -o wide
pod-httpd   1/1     Running   0          54s   minikube   <none>           <none>
pod-nginx   1/1     Running   0          54s   minikube   <none>           <none>

To test this, we need to log into one of the pods, and run curl to the other pod. Let's log into pod-httpd:

kubectl exec pod-httpd -it -- /bin/bash
root@pod-httpd:/usr/local/apache2# apt-get update
root@pod-httpd:/usr/local/apache2# apt-get install -y curl

We also had to install the curl command itself. Before we can perform the test:

root@pod-httpd:/usr/local/apache2# curl
<!DOCTYPE html>
<title>Welcome to nginx!</title>
    body {
        width: 35em;
        margin: 0 auto;
        font-family: Tahoma, Verdana, Arial, sans-serif;
<h1>Welcome to nginx!</h1>
<p>If you see this page, the nginx web server is successfully installed and
working. Further configuration is required.</p>

<p>For online documentation and support please refer to
<a href=""></a>.<br/>
Commercial support is available at
<a href=""></a>.</p>

<p><em>Thank you for using nginx.</em></p>

However we can't rely on IP addresses because they are prone to changing, e.g. when a pod is rebuilt. So hard coding ip addresses in various places is not an option. The conventional way address this problem is giving each pod a human-readable dns name, which is kept up to date dynamically, that's where kube-dns comes to the rescue

DNS Service

Kubernetes comes with a builtin internal dns service, kube-dns (soon to be coredn).

$ kubectl get svc kube-dns --namespace=kube-system -o wide
kube-dns   ClusterIP   <none>        53/UDP,53/TCP   3d3h   k8s-app=kube-dns

This ip address is automatically added to all other pods /etc/resolv.conf file:

root@pod-nginx:/# cat /etc/resolv.conf
search default.svc.cluster.local svc.cluster.local cluster.local
options ndots:5

So all we need to do to use this dns service, is create dns entries in our internal dns server (kube-dns). That's done by creating service objects. There are different types of service objects, such as nodePort, and ClusterIP service. We'll cover them later.

The 'kubectl expose' command

In this course we'll focus on creating service objects declaratively using yaml files. But you can create them implicitly from the command line using the 'expose' subcommand:

kubectl expose pod podname --type=NodePort --name servicename