16 May 2019, 18:07

Kubernetes/kubectl crash course - engineering POV


Brain dump of various kubernetes/kubectl things before I forget them all. I’m coming from the deploying analytics side, not the person who sets up kubernetes side, so this will be a list of useful day 1 type stuff for anyone who is learning kubernetes practically via deployments and kubectl.


  • service - something with an IP/hostname you want to hit, e.g. a RESTful front end.
  • deployment - powers a service by specifying things like number of instances, and how ports are shuffled.
  • pods - basic atomic-ish unit that make up a deployment. An instance.

Suppose you want to stand up your RESTful analytic with 8 instances, available at foo.internal.com. Start with a service, then a deployment. Pods come via the deployment.


pacman -S kubectl

Get config file from ops-type person. Put or link in ~/.kube/config, then

kubectl config show

to make sure you’re running with the correct one.

Change contexts, if needed, for various projects.

Basic commands


kubectl get pods
kubectl delete pod ...

You can delete a pod to have K8s immediately kick another one off, useful for getting info about why things aren’t working


kubectl logs pod-name
kubectl logs pod-name --previous

Get logs, or see why app crashed before via --previous. Anything else, you probably need some log aggregation service (Helm/fluentd?)


kubectl deploy -f ./service.yaml
kubectl get services

Deploy services, which are sort of the interface between pods/containers and the rest of the cluster (and outside world).


kubectl deploy -f ./deployment.yaml
kubectl delete deployment your-deployment

Deployments power services by specifying instances and memory, ports, docker images, etc.

Interaction outside cluster

kubectl port-forward services/your-service 8080:8080

Test whether or not your API/service/deployments actually work.

Next up: how Terraform does all this for you.