Latest update: November 2020
This repository contains material I've gathered to pass the CKA exam + some clusters which can be created using vagrant
Contributions and fixes are welcome :)
- Kelsey Hightower "Kubernetes the Hard Way": how to build a k8s cluster from scratch with no automation tools.
- walidshaari's Kubernetes Certified Administration repo: it links curriculum parts with documentation pages. Also contains interesting tips for the exam, and link to paid courses (killer.sh is a very interesting one, it provides a CKA exam simulator with 25 scenarios)
- dgkanatsios' CKAD exercises: quite good to level up in "normal" k8s usage, and to train parts of the exam which matches the CKAD exam (workloads, services, etc...)
- CKA example exam questions: from the same creator than
killer.sh, 6 really good scenarios to practice locally - backup/restore ETCD: quick guide to check how to backup and restore ETCD
These tips assume you are already familiar with k8s and you want to tune up your skills for the exam. If you are not familiar a good starting point are the tutorials from kubernetes.io, don't hesitate to practice as much as you want with minikube, even trying questions from the CKAD exam.
I loosely followed this order to study:
- Reading the docs. I started by following the curriculum sections proposed in the walidshaari's Kubernetes Certified Administration repo. Then I read both Concepts and Tasks sections while I was dealing with the next points, while I was facing problems or challenges during my progress.
- Do Kelsey Hightower "Kubernetes the Hard Way", I did it several times until I got familiar with which certificates are needed, how to create ETCD, etc... the
multi-node-clustercan be used to follow this guide - Try creating a cluster with a controller node and a worker node with kubeadm following the official documentation
- Rinse and repeat creating a multi node cluster with kubeadm. There are scripts which will do it by you in the
multi-node-cluster, you can check them if you get stuck. - Create a k8s cluster version 1.18, and upgrade it.
- Go through all the dgkanatsios' CKAD exercises at least two times. In the first run trying to gather all information from official documentation only and/or using
kubectl explain, without minding the time, learing how to locate the information. On the second and subsequent runs, trying to be fast. - Trying the CKA example exam questions (a cluster is also provided there).
- Practice some ETCD backup and restore (not needed in a multi node environment, but it doesn't hurt doing it)
This is not a strict order, it helps to follow it to acquire the installation/upgrade knowledge, but you can mix them up. For example, you can start digging how Deployments works, running examples in one of the clusters, then try to create a cluster, then go back to other Documentation section... It's worth to note half of the curriculum has to do with installation, upgrade and troubleshooting, so you should be really thorough with those parts.
For the exam, you should get used to easily locate what you need in the documentation (learn how to quickly navigate and to find sections you may need in your travels).
I have seen many places which recommend to create several aliases for kubectl in order to type quicker in the exam, but my take on this is to just create the alias alias k=kubectl as in the exam you will be context-switching (specially if you go back or don't follow the questions order) and having to type several alias each time you do a context switch may backfire. Of course this is a personal opinion and if you prefer to use several aliases you should do it, just keep in mind that you may lose them with the context changing.
Another really good command you should get used is kubectl explain (and kubectl explain --recursive). It's very handful when you know what you are doing but you need to check exact names for the different fields of the object you are creating.
Finally, the kubectl --dry-run=client -o yaml trick to populate "skeleton" YAMLs which you can edit afterwards with anything you need. Learning how to create pods, deployments... with this helps a lot for the complex cases which are not supported by kubectl
This a WIP section. Also, the intention of these clusters is to have a testbed where to practice for the CKA, with no other applications or purposes.
In the kubernetes_clusters folder you can spin up two kinds of Kubernetes clusters based on Vagrant machines. The Vagrantfile and the different provisioners are heavily based on this fork from "kubernetes the hard way". I started using that repository to play with k8s installations and I ended up building some scripts to have different practice environments. The clusters available are:
- "two-nodes-cluster": it contains a controller node and a worker node
- "multi-node-cluster": it spins up a lb, 3 controller nodes and 3 workers nodes (same setup than Kelsey Hightower "Kubernetes the Hard Way" )
I have set a 2GB memory size for each worker and controller node, but the controller probably would work with 1GB or less. You can also downsize the workers if you have memory constraints to run this cluster, it will only limit the size of the containers you can run inside.
When the clusters are created with the start-full-* scripts, they also provide:
- Weaveworks network addon
- Creating an StorageClass and several local volumes you can use for testing PVCs, StatefulSets...
You can spin up a simple cluster with a controller node and a worker node with the script start-full-two-node-cluster.sh. It should take about 5 minutes to be ready, which is handy to practice with a running cluster. As it provision only some instances from the Vagrantfile, you can manually start more instances and try to join them in the cluster by yourself, as part of practice.
If you are interested only in creating the virtual machines but you want to test Kubernetes installation by yourself, you can use the script start-two-nodes.sh.
Contains a LB for the controller nodes, 3 controller nodes and 3 worker nodes.
If you want to provision the "bare metal" cluster you only need to execute vagrant up. This will only create the VMs but you will have to install everything by yourself. This cluster can be used to follow the Kelsey Hightower "Kubernetes the Hard Way" (ignoring the parts where it sets up load balancers, as they are already in place) or if you want to provision a cluster following the official guide using kubeadm. In this case, I had to install ETCD separately as kubeadm was conflicting with the internal vagrant IP address, once ETCD was running you can install the rest of the nodes following the guide.
There are scripts to do the different installation steps if you want to speed up the process or if you bump into problems and need some inspiration to continue. There is also a start-cluster.sh script which do the vagrant up and runs the scripts until the cluster is fully provisioned. There is plenty of room for improvement on this script (and PRs are welcome!) as it's solely purpose is to mimic what someone should do manually to learn about installing k8s clusters. As there are not steps executed in parallel the provision is a bit slow (about 15 minutes in a modern laptop). Once provisioned kubectl will be available on controller-0 node.
- allow selecting kubernetes version in the full provisioning script
- configure
kubectlin all nodes