One thing that can be a little surprising when you use minikube with the docker driver, is that it doesn’t actually use the images from your local docker when it runs. It actually runs with containerd.

Your kubernetes node is “running in your docker” (KIC / Kubernetes-in-Container), but that node uses containerd and its own registry when running pods itself.

Why are large images a problem?

When you go to use an image, you have to do this to move it from your docker registry to your minikube registry.

minikube image load appname:tagname 2>&1 

This operation basically does a docker-image-save to a large tar file, copies that file into the container running your kubernetes, and unpacks it into containerd. This is very expensive on memory; it can freeze WSL2 (ubuntu) on my high end laptop on a 8GB image. If you are finding the image load command hangs your WSL, try opening two terminals and use htop to watch the memory usage as the command runs; you’ll see it soak up far more than you would expect, and then it likely starts swapping/etc and slowing everything down until it is unusable.

So… How do we use big images?

Building big images isn’t much of a problem as it happens in a more efficient way. So, the solution is to build directly into minkube instead of building into docker and importing to minikube.

minikube image build -t appname:tagname -f Dockerfile .

With this change, the image builds into minikube quite easily and is readily useable.