I am developing a microservice locally and I have built a Docker image on my machine. However, when I try to use this image in a Kubernetes deployment on Minikube, it fails with an "ImagePullBackOff" error because it tries to find the image on Docker Hub. How do I configure my environment so that Minikube's internal Docker daemon can see and use the images I just built on my local system?
3 answers
The most efficient way to use local images in Minikube is to reuse the Docker daemon inside the Minikube virtual machine. By default, your local Docker CLI communicates with your host's daemon, but Minikube runs its own. You can point your terminal to Minikube's engine by running eval $(minikube docker-env). Once you run this command, any docker build you perform will happen directly inside the Minikube environment. This makes the image immediately available to your Kubernetes pods. Crucially, in your Deployment YAML, you must set imagePullPolicy: IfNotPresent or Never, otherwise Kubernetes will still attempt to reach out to an external registry and fail.
Have you tried the minikube image load command as an alternative, and are you aware of how it handles image tagging? This command allows you to keep your host Docker daemon separate while manually "sideloading" specific images into the cluster nodes, which can sometimes be safer if you don't want to accidentally clutter the internal Minikube registry with every intermediate build layer from your local development.
The "eval" trick is the best. Just remember that it only lasts for the current terminal session. If you open a new tab, you have to run the command again!
I agree with Sarah. I actually added that eval line to my .bashrc file so it happens automatically. It saved me so much time during my last Agile sprint when I was constantly rebuilding my machine learning containers for testing.
Marcus, that minikube image load command is actually my preferred method lately. To answer your question, it works by exporting the image from your host and importing it into the node. It is slightly slower than the eval method because of the transfer time, but it's much more explicit. You just run minikube image load your-image-name:tag and it ensures the cluster has it. It’s perfect for one-off tests where you don't want to mess with your shell environment variables.