Creating a Gossip-Based Kubernetes Cluster on AWS

Creating a Kubernetes cluster using Kops requires a top-level domain or a subdomain and setting up Route 53-hosted zones. This domain allows the worker nodes to discover the master and the master to discover all the etcd servers. This is also needed for kubectl to be able to talk directly with the master. This worked well, but poses an additional hassle for the developers.

Kops 1.6.2 adds an experimental support for gossip-based discovery of nodes. This makes the process of setting up Kubernetes cluster using Kops DNS-free, and much more simplified.

Let’s take a look!

Install or upgrade kops:

brew upgrade kops

Check the version:

~ $ kops version
Version 1.6.2

Create an S3 bucket as “state store”:

~ $ kops version
Version 1.6.2

Create a Kubernetes cluster:

kops create cluster cluster.k8s.local --zones us-east-1a --yes

It shows the output as:

I0622 16:52:07.494558   83656 create_cluster.go:655] Inferred --cloud=aws from zone "us-east-1a"
I0622 16:52:07.495012   83656 create_cluster.go:841] Using SSH public key: /Users/argu/.ssh/id_rsa.pub
I0622 16:52:08.540445   83656 subnets.go:183] Assigned CIDR 172.20.32.0/19 to subnet us-east-1a
I0622 16:52:16.327523   83656 apply_cluster.go:396] Gossip DNS: skipping DNS validation
I0622 16:52:25.539755   83656 executor.go:91] Tasks: 0 done / 67 total; 32 can run
I0622 16:52:29.843320   83656 vfs_castore.go:422] Issuing new certificate: "kubecfg"
I0622 16:52:30.108046   83656 vfs_castore.go:422] Issuing new certificate: "kubelet"
I0622 16:52:30.139629   83656 vfs_castore.go:422] Issuing new certificate: "kube-scheduler"
I0622 16:52:31.072229   83656 vfs_castore.go:422] Issuing new certificate: "kube-proxy"
I0622 16:52:31.082560   83656 vfs_castore.go:422] Issuing new certificate: "kube-controller-manager"
I0622 16:52:31.579158   83656 vfs_castore.go:422] Issuing new certificate: "kops"
I0622 16:52:32.742807   83656 executor.go:91] Tasks: 32 done / 67 total; 13 can run
I0622 16:52:43.057189   83656 executor.go:91] Tasks: 45 done / 67 total; 18 can run
I0622 16:52:50.047375   83656 executor.go:91] Tasks: 63 done / 67 total; 3 can run
I0622 16:53:02.047610   83656 vfs_castore.go:422] Issuing new certificate: "master"
I0622 16:53:03.027007   83656 executor.go:91] Tasks: 66 done / 67 total; 1 can run
I0622 16:53:04.197637   83656 executor.go:91] Tasks: 67 done / 67 total; 0 can run
I0622 16:53:04.884362   83656 update_cluster.go:229] Exporting kubecfg for cluster
Kops has set your kubectl context to cluster.k8s.local

Cluster is starting.  It should be ready in a few minutes.

Suggestions:
 * validate cluster: kops validate cluster
 * list nodes: kubectl get nodes --show-labels
 * ssh to the master: ssh -i ~/.ssh/id_rsa admin@api.cluster.k8s.local
The admin user is specific to Debian. If not using Debian please use the appropriate user based on your OS.
 * read about installing addons: https://github.com/kubernetes/kops/blob/master/docs/addons.md

Wait for a few minutes for the cluster to be created.

Then, validate the cluster:

~ $ kops validate cluster
Using cluster from kubectl context: cluster.k8s.local

Validating cluster cluster.k8s.local

INSTANCE GROUPS
NAME			ROLE	MACHINETYPE	MIN	MAX	SUBNETS
master-us-east-1a	Master	m3.medium	1	1	us-east-1a
nodes			Node	t2.medium	2	2	us-east-1a

NODE STATUS
NAME				ROLE	READY
ip-172-20-36-52.ec2.internal	node	True
ip-172-20-38-117.ec2.internal	master	True
ip-172-20-49-179.ec2.internal	node	True

Your cluster cluster.k8s.local is ready

Get the list of nodes using kubectl:

~ $ kubectl get nodes
NAME                            STATUS         AGE       VERSION
ip-172-20-36-52.ec2.internal    Ready,node     4h        v1.6.2
ip-172-20-38-117.ec2.internal   Ready,master   4h        v1.6.2
ip-172-20-49-179.ec2.internal   Ready,node     4h        v1.6.2

Deleting a cluster is pretty straight forward as well:

kops delete cluster cluster.k8s.local --yes

That’s it!

Here, you can find several examples of getting started with Kubernetes.

File issues at github.com/kubernetes/kops/issues.