KUBERNETES, A GUIDE FOR DOCKER USERS

Jon “The Nice Guy” Spriggs

https://jon.sprig.gs | jon@sprig.gs | @jontheniceguy

Why might you be here?

How is it SIMILAR?

Good News!
Containers are still containers!

Bad News!
You don’t work directly with containers!

Pods are like
docker-compose.yml

A deployment adds high availability

Getting to your service

Getting exposure

How is it DIFFERENT?

More nodes are needed

Storage is different

ConfigMap
Secret
PersistentVolumeClaim
EmptyDir

API Access & Granular RBAC

Service Accounts for Pods, User Accounts for users.

Roles and RoleBindings for namespaces.
ClusterRoles and ClusterRoleBindings for the cluster.

Webhooks for every action.

A red and white label on a wooden surface. The label reads "Hello my name is Dave"

Namespaces?!

Addressing

Inside every cluster is a DNS service. This service serves for everything inside the cluster, and resolves external resources, like the DNS name of your NFS service (if you’re using one), but also internal services, like my-api.my-namespace.srv.cluster.local. On top of that, everything inside the cluster can be protected by TLS, because there’s a certificate authority that all the pods trust.

If you want to use IP addresses, there’s a few of them to think about! Your worker has at least one “Internal” IP address (on an RFC1918 address range, hopefully), but usually more, because each pod also has it’s own IP address. Inside the pod, the containers have their own IP addresses, but these are on a separate address range which isn’t routable outside of the worker. Each service will also have an IP address to expose content to your load balancers, and the load balancer will often have a public and an internal IP address. You can also find all of these listed as “Endpoints” on your cluster.

Is it all worth it?

harry potter shrug GIF from Giphy

But really… is it all worth it?

But really [CLICK] Is it worth it? [CLICK] If you’ve got more than 10 users who all want to run services at the same time, then yes, I believe it’s worth it. If it’s not just users, but it’s also how many workers you’re using? It is possible to start working with Kubernetes at 1 node which self-manages, using something like k3s or minikube, but you lose out on a lot of the nuance at that level. When you get above 3 nodes it’s probably worth it, but you might just be better off with one or two servers and an HA “thing” in front of it. But where it really gets going is 7 nodes… or 3 if you consume a managed service like EKS from Amazon or AKS from Azure. But you should be aware that you’ll be paying more for your service, so make sure the cost is offset by the benefits it will bring you.