Developing for Kubernetes with Minikube

This guide is meant to serve as a cross-plaform resource for setting up a local Kubernetes development environment. In this guide, we’ll be using Minikube as it is the defacto standard.

Getting Started with Minikube

We’ll extract and expound on the official documentation from the Kubernetes project, Running Kubernetes Locally with Minikube.

Installing kubectl

The official documentation provides several options, but the result is that you can do one of three things:

  • Download as a part of the Google Cloud SDK from Google Cloud Platform’s Cloud SDK page. Once you have gcloud installed, you can install kubectl:

    sudo gcloud components install kubectl

    If you’ve already installed kubectl via this method, ensure it is updated:

    sudo gcloud components update
  • Install directly from the Google Storage APIs.
  • Install with the appropriate package management system:
    • Linux: your package manager of choice, or Snap.
    • macOS
    • Windows

Installing Minikube

See the Kubernetes documentation where they suggest directly installing from the releases on GitHub.

Choosing a VM driver

For the purposes of cross-platform compatibility in this guide, we’ll stick with VirtualBox, however there are drivers for VMware Fusion, HyperV, KVM, and Xhyve.

Starting / Stopping Minikube

Minikube resource requests must be set higher than the default for developing the GitLab chart. The key configuration items can be found with minikube start --help. A selection is provided below, for what we may want to change according to the pieces being tested, and the requirements as listed:

  • --cpus int: Number of CPUs allocated to the Minikube VM (default 2). The absolute minimum necessary CPU is 2. Deploying the complete chart requires 3.
  • --memory int: Amount of RAM allocated to the Minikube VM (default 2048). The absolute same minimum is 5120 (5 GB). Recommendation is 8192 (8 GB).
  • --disk-size string: Disk size allocated to the Minikube VM (format: <number>[<unit>], where unit = b, k, m or g) (default 20g). See GitLab’s storage and database requirements.

    Note: This is created in your home directory under ~/.minikube/machines/minikube/.
  • --kubernetes-version string: The Kubernetes version that the Minikube VM will use (e.g., v1.2.3).
  • --registry-mirror stringSlice: Registry mirrors to pass to the Docker daemon.
Note: Changing these values in a second start command, requires to first delete the existing instance with minikube delete, or manually you can alter the properties with VirtualBox Manager.

Once you have all the tools installed and configured, starting at stopping Minikube can be done with:

minikube start --cpus 3 --memory 8192

This command should output something similar to:

Starting local Kubernetes v1.7.0 cluster...
Starting VM...
Downloading Minikube ISO
 97.80 MB / 97.80 MB [==============================================] 100.00% 0s
Getting VM IP address...
Moving files into cluster...
Setting up certs...
Starting cluster components...
Connecting to cluster...
Setting up kubeconfig...
Kubectl is now configured to use the cluster.
[]$ minikube ip
[]$ minikube stop
Stopping local Kubernetes cluster...
Machine stopped.

Take note of the result from running the minikube ip command. If the output is not, the output IP will be needed later.

Using Minikube

Minikube can be used directly as a Kubernetes installation, and treated as a single node cluster. There are some behaviors that are slightly different between Minikube and full-fledged Kubernetes clusters, such as Google Container Engine (GKE).


  • Persistent Volumes: hostPath only.


  • Load Balancers (requires cloud provider).
  • Advanced Scheduling Policies (requires multiple nodes).

Gotcha: Persistent Volumes

Minikube supports PersistenVolumes of the hostPath type, which are mapped to directories inside the VM. As Minikube boots into a tmpfs, most directories will not persist across reboots via minikube stop.

Further details and listings of directories that do persist, can be found in the Minikube getting started guide.

Enable Addons

Minikube handles some features apart from the base configuration. For the development of this project, we’ll need access to Ingress:

minikube addons enable ingress

Connecting to the dashboard

You can find the URL for the dashboard by calling:

minikube dashboard --url

Hooking Helm to Minikube

If you are using Helm v2, then once your Minikube is up and running, you can initialize Helm with the command helm init.

Using Helm v3 does not require any initialization commands and will work out of the box.

For further details on Helm, see Developing for Helm.

Deploying the chart

When deploying this chart into Minikube, some chart resources need to be reduced or disabled. It is not possible to use the nginx-ingress chart to provide ports 22, 80, 443. It’s best to disable it and set the Ingress class by setting nginx-ingress.enabled=false,global.ingress.class="nginx".

The certmanager chart can not be used with Minikube. You must disable this by setting certmanager.install=false,global.ingress.configureCertmanager=false. As a result, if you don’t provide your own SSL certificates, self-signed certificates will be generated. The gitlab-runner chart will accept the self-signed certificates via gitlab-runner.certsSecretName. Assuming your release name is gitlab, the certificate name will be gitlab-wildcard-tls-chain.

The gitlab-shell chart can be used with Minikube, but requires mapping to a port other than 22 as it used by Minikube already. You can configure gitlab.gitlab-shell.service.type=NodePort and gitlab.gitlab-shell.service.nodePort=<high-numbered port>, which will allow cloning a repository via the specified port. To ensure this port is reflected in the clone link in the UI, configure<high-numbered port>.

In the following sections, we’ll show how to install these charts from your local Git clone. Be sure that you have checked out the desired branch or tag, and are at the base folder of that checkout.

When using the recommended 3 CPU and 8 GB of RAM, use values-minikube.yaml as a base.

helm repo add gitlab
helm repo update
helm upgrade --install gitlab . \
  --timeout 600s \
Note: If using Helm v2, please see notes about the --timeout option in the Deployment documentation.

Deploying GitLab with minimal settings

If using absolute minimum resources, 2 CPU and 4GB of RAM, you must reduce all replicas and disable unneeded services. See values-minikube-minimum.yaml as a reasonable base.

helm repo add gitlab
helm repo update
helm upgrade --install gitlab . \
  --timeout 600s \
Note: If using Helm v2, please see notes about the --timeout option in the Deployment documentation.

If the output of minikube ip was not, add these arguments to override the IP endpoints in the example configuration files:

  --set global.hosts.domain=$(minikube ip) \
  --set global.hosts.externalIP=$(minikube ip)

Handling DNS

The example configurations provided, configure the domain as in an attempt to reduce the overhead of handling alterations to host files, or other domain name resolution services. However, this relies on the network reachability of

If this is not available to you, then you may need to make alterations to your /etc/hosts file, or provide another means of DNS resolution.

Example /etc/hosts file addition: gitlab.some.domain registry.some.domain minio.some.domain

Incorporating Self-Signed CA

Once the chart is deployed, if using self-signed certificates, the user will be given the notice on how to fetch the CA certificate that was generated. This certificate can be added to the system store, so that all browsers, Docker daemon, and git command recognize the deployed certificates as trusted. The method depends on your operating system.

BounCA has a good tutorial, covering most operating systems.

Logging in

You can access the GitLab instance by visiting the domain specified, is used in these examples. If you manually created the secret for initial root password, you can use that to sign in as root user. If not, GitLab automatically created a random password for the root user. This can be extracted by the following command (replace <name> by name of the release - which is gitlab if you used the command above).

kubectl get secret <name>-gitlab-initial-root-password -ojsonpath='{.data.password}' | base64 --decode ; echo