Skip to content

Fast local Development using Ko, Golang, and Minikube

Posted on:January 1, 2021 at 03:22 PM

ko is a simple, fast container image builder for Go applications. In combination with minikube, it enables an easy local development cycle.

In this post, I will explain how to set up ko to directly build and deploy Golang applications to minikube.

First, we need to start a local minikube cluster. Use the following command to deploy a local cluster and set it as the current kubectl context.

minikube start

Ko can build and push containers to all kinds of registries. Ko depends on an environment variable, KO_DOCKER_REPO, to identify where it should push images that it builds. For local development, we don’t want to push the images to a remote registry, we will publish the images to the local Docker daemon by exporting the following environment variable:

export KO_DOCKER_REPO=ko.local

Next, we need to point the shell to minikube’s docker-daemon by running the following command

eval $(minikube -p minikube docker-env)

Now we are ready to deploy Go applications directly to minikube. We will deploy a basic application that utilizes the Go prometheus-client using ko:

import (


func main() {
    http.Handle("/metrics", promhttp.Handler())
    http.ListenAndServe(":8080", nil)
    fmt.Println("Starting the server on port 8080")

The Go application is named testGoExport. Next, we will define the Kubernetes Deployment. However, instead of using a path to a remote image, we will reference the Go binary by its importpath, prefixed with ko:// For this to work, make sure that your GOPATH is set correctly.

# deployment.yaml
apiVersion: apps/v1
kind: Deployment
  creationTimestamp: null
    app: prom-test
  name: prom-test

  replicas: 1
      app: prom-test
  strategy: {}
      creationTimestamp: null
        app: prom-test
        - image: ko://
          name: testgoexporter
            - containerPort: 8080
          imagePullPolicy: Never
              path: /metrics
              port: 8080
          resources: {}

To deploy the Deployment we will now run ko apply -f deployment.yaml instead of kubectl apply -f deployment.yaml. This will:

  1. Scan the YAML file for the ko:// prefix
  2. Build the Go application
  3. Publish the application to the local docker daemon
  4. Replace the image name with the fully-specified image reference of the build image
  5. Deploy the Deployment After executing:
ko apply -f deployment.yaml

2022/01/04 17:22:21 Using base for
2022/01/04 17:22:22 Building for linux/amd64
2022/01/04 17:22:24 Loading ko.local/testgoexporter-18931265053c2910578ee5844178b709:beb82aa376e6822d87b72e69a48356427674630e844d720357afb532b73b6896
2022/01/04 17:22:24 Loaded ko.local/testgoexporter-18931265053c2910578ee5844178b709:beb82aa376e6822d87b72e69a48356427674630e844d720357afb532b73b6896
2022/01/04 17:22:24 Adding tag latest
2022/01/04 17:22:24 Added tag latest
deployment.apps/prom-test created

If we now check the current deployments we see our deployed applications:

kubectl get deployments

NAME                            READY   UP-TO-DATE   AVAILABLE   AGE
prom-test                       1/1     1            1           68s

The Deployment uses the image called ko.local/testgoexporter-18931265053c2910578ee5844178b709:beb82aa376e6822d87b72e69a4

To learn more, make sure to check out ko documentation at the GitHub repository!