Minikube is a project allowing you to run a local Kubernetes node simulation for development and testing purposes.

It’s possible to run QHub on Minikube, and this can allow quicker feedback loops for development, as well as being less expensive than running cloud Kubernetes clusters.

Local testing is a great way to test the components of QHub. It’s important to highlight that while it’s possible to test most of QHub with this version, components that are Cloud provisioned such as VPCs, managed Kubernetes cluster and managed container registries can’t be locally tested, due to their Cloud dependencies.


Currently, QHub local deployment is primarily compatible with Linux-based Operating Systems. The main limitation for the installation on MacOS relates to Docker Desktop for Mac being unable to route traffic to containers. Theoretically, the installation of HyperKit Driver could solve the issue, although the proposed solution isn’t tested. There some workarounds for running Minikube on Mac below.

This guide assumes that you have the QHub repository downloaded, and you are at the root of the repository.


NOTE: The following instructions apply only to Linux OS.

To deploy QHub locally requires the installation of the following dependencies:

The installation of a hypervisor isn’t necessary.

NOTE: Minikube requires kubectl OR you can use the embedded kubectl appropriate for your Minikube cluster version using minikube kubectl. To install kubectl follow the instructions according to your operating system.

Initialize Kubernetes cluster

Before proceeding with the initialization, make sure to add yourself to the Docker group by executing the command sudo usermod -aG docker <username> && newgrp docker.

Testing is done with Minikube. To confirm successful installation of both Docker and Minikube, you can run the following command to start up a local Kubernetes cluster:

minikube start --cpus 2 --memory 4096 --driver=docker

The command downloads a Docker image of around 500Mb in size and initialise a cluster with 2 CPUs and 4 GB of RAM, with Docker as the chosen driver.

Once minikube start finishes, run the command below to select the status of the cluster:

minikube status

If your cluster is running, the output from minikube status should be similar to:

type: Control Plane
host: Running
kubelet: Running
apiserver: Running
kubeconfig: Configured
timeToStop: Nonexistent

After you have confirmed that Minikube is working, you can either continue to use, or you can stop your cluster. To stop your cluster, run:

minikube stop

Next, the user installs nfs-common drivers. This is required by the JupyterLab instances, which require NFS mount for PersistentVolumeClaims (PVCs). To install it, run:

minikube ssh "sudo apt update; sudo apt install nfs-common -y"

For more details on PVs and PVCs, read the JupyterHub documentation.

Optional pre-pulling and caching of Docker images

Click to expand note

Why pre-pull Docker images

As part of deployment, Minikube downloads Docker images that have a combined size of several Gigabytes. Each time minikube is destroyed and created it re-pulls these images. Also, Terraform times out on slower internet connections if it takes longer than 10 minutes to pull the images.

Images can be pre-pulled and added to the Minikube cache. This greatly reduce the time required for future deployments and reduces the data requiring download during deployment.

Pre-pulling and caching

The following assumes that docker is currently installed.

The first step is to configure the minikube home directory environment variable. To set this to the home directory of the current user, run:

export MINIKUBE_HOME=$HOME/.minikube

The current list of Docker images is visible in the qhub-config.yaml file under the default_images key. Each image is pulled.

docker pull quansight/qhub-jupyterhub:v0.x.x
docker pull quansight/qhub-jupyterlab:v0.x.x
docker pull quansight/qhub-dask-worker:v0.x.x
docker pull quansight/qhub-dask-gateway:v0.x.x

Replacing v0.x.x with the current version that’s listed. Note this may take several minutes. Once the process is complete, the user can copy them to the Minikube cache.

minikube image load quansight/qhub-jupyterhub:v0.x.x
minikube image load quansight/qhub-jupyterlab:v0.x.x
minikube image load quansight/qhub-dask-worker:v0.x.x
minikube image load quansight/qhub-dask-gateway:v0.x.x

Again, adding the correct version. With this completed local Minikube deployment no longer requires pulling the preceding docker images.

The preceding process is repeated with the updated tags when a new version of QHub is deployed.


MetalLB is the load balancer for bare-metal Kubernetes clusters. The user needs to configure MetalLB to match the QHub configuration.

Automation of MetalLB with Python script

Skip to next section for configuration without Python

Minikube doesn’t provide a simple interactive way to configure addons, (as shown in this repository issue). It’s recommended to set load balancer start/stop IP address using a Python script with pre-established values. This recommendation is due to an existing DNS name that uses some addresses.

To do so, paste this Python script in a text file named and then run:


Manually configure MetalLB

Skip this section if the preceding Python script is used

First the user needs to obtain the Docker image ID:

$ docker ps --format "{{.Names}} {{.ID}}"
minikube <image-id>

Copy the output image id and use it in the following command to obtain the Docker interface subnet CIDR range:

$ docker inspect --format='{{range .NetworkSettings.Networks}}{{.IPAddress}}/{{.IPPrefixLen}}{{end}}' <image-id>

A example subnet range looks like This CIDR range has a starting IP of and ending address of The metallb load balancer is provided a range of IP addresses contained in the Docker CIDR range. If your CIDR is different, you can determine your range IP addresses from a CIDR address at this website.

For this example case, the user assigns metallb a start IP address of and an end of

The user can enable metallb as shown below. The command-line tool interface prompts the user for the start and stop IP range:

minikube addons configure metallb
-- Enter Load Balancer Start IP:
-- Enter Load Balancer End IP:

If successful, the output should be   metallb was successfully configured.

Enable MetalLB

After configuration enable MetalLB by running

minikube addons enable metallb

The output should be The 'metallb' addon is enabled.

Note for development on Windows Subsystem for Linux 2

Click to expand note

The browser can have trouble reaching the load balancer running on WSL2. A workaround is to port forward the proxy-pod to the host IP Get the ip address of the WSL2 machine via ip a, which should be a 127.x.x.x address. To change the port forwarding after opening k9s you can type :pods <enter>, hover over the proxy-… pod and type <shift-s>, and enter the IP addresses.

Deploy QHub

To deploy QHub handle setup dependencies and create a new sub-directory by running:

pip install -e .
mkdir -p data
cd data

Initialize configuration

Then, initialize the configuration file qhub-config.yaml with:

python -m qhub init local --project=thisisatest  --domain --auth-provider=password --terraform-state=local

Deploy and render the infrastructure

Next, the user renders the infrastructure files from qhub-config.yaml running.

python -m qhub deploy --config qhub-config.yaml --disable-prompt

To ease development, the DNS record is pointed to so the next step is optional unless you end up with the load-balancer giving you a different IP address.

Make sure to point the DNS domain to from the previous commands. This is done in many ways, the easiest one is by modifying /etc/hosts and adding the line below. The command overrides any DNS server.

Verify the local deployment

Finally, if everything is set properly you should be able to cURL the JupyterHub Server. Run

curl -k

It’s also possible to visit in your web browser to select the deployment.

Since this is a local deployment, hence it’s not visible to the internet; https certificates isn’t signed by Let’s Encrypt. Thus, the certificates is self-signed by Traefik.

Several browsers makes it difficult to view a self-signed certificate that’s not added to the certificate registry.

Each web browser handles this differently. A workaround for Firefox:

  • Visit about:config and change the network.stricttransportsecurity.preloadlist to false

And a workaround for Chrome:


To delete all the QHub resources run the destroy command. Note that this won’t delete your qhub-config.yaml and related rendered files thus a re-deployment via deploy is possible afterwards.

python -m qhub destroy --config qhub-config.yaml

To delete the Minikube Kubernetes cluster run the following command:

minikube delete

The command deletes all instances of QHub, cleaning up the deployment environment.

Minikube on Mac

At one point developing with Minikube on a Mac worked, unfortunately this appears to no longer be the case. A few differences exists between how Minikube is deployed on Mac versus Linux and if you’re interested to try the instructions as they appeared when this worked, open the collapsed tab below:

Previous instructions

The earlier instructions for Minikube on Linux nearly works on Mac except things that break without clever use of port forwarding at the right times.

1 - When working out the IP addresses to configure metallb try this:

docker ps --format "{{.Names}} {{.ID}}"
docker inspect --format='{{range .NetworkSettings.Networks}}{{.IPAddress}}/{{.IPPrefixLen}}{{end}}' <ID of minikube from previous cmd>

This displays something like, in which case a suitable IP range would be on the same subnet, for example start IP, end IP

2 - This load balancer won’t actually work, so you need to port-forward directly to the JupyterHub service:

minikube kubectl -- --namespace=dev port-forward svc/proxy-public 8000:80

Then you could access QHub on

3 - However, the qhub deploy step needs to communicate with the Keycloak server, but this isn’t possible without the correct hostname.

It might be possible to set /etc/hosts to include as they are done for the AWS minikube, below. And meanwhile use kubectl port-forward to actually forward the traffic (from port 443 to something similar?). But you’d have to start that forwarding at the right point in the deployment. (When Kubernetes is ready, but before terraform runs the Keycloak operator…)


Minikube on AWS

It’s possible to run Minikube on AWS (and probably the other clouds). This is useful where you don’t have enough memory to run QHub in a local Minikube cluster on your laptop, or if you are using Mac or Windows and struggling to get Minikube to work.

Please understand that running Minikube on an AWS EC2 instance isn’t the same as ‘proper’ deployment of QHub to AWS EKS (Kubernetes service). You might prefer Minikube on AWS over full AWS EKS deployment for testing purposes if you find Minikube easier to debug, cheaper to run, or if you want to replicate the Minikube setup directly - for example, if trying to fix the automated Cypress tests which use Minikube within a GitHub actions workflow.

There are some tricks that can make allow Minikube on AWS to feel much like local Minikube for development.

The instructions below should work for Mac.

Set up AWS credentials

Set your environment variables for your AWS account:

export AWS_PROFILE=quansight
export AWS_DEFAULT_REGION="eu-west-2"

This assumes you have a ‘quansight’ profile in your ~/.aws/config and credentials files, but instead you can set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY directly.

Create a Key Pair

You can use a Key Pair that’s already registered with AWS, or create one as follows:

export MYKEYNAME=aws-quansight-<mykey>

aws ec2 create-key-pair \
    --key-name ${MYKEYNAME} \
    --query "KeyMaterial" \
    --output text > ~/.ssh/${MYKEYNAME}.pem

chmod 400 ~/.ssh/${MYKEYNAME}.pem

Run the EC2 instance

The recommended image is an Ubuntu 20.04 with Docker installed. It’s recommended to be to run on a 16 GB/4 Core image, and increase the EBS disk space to 48 GB or so, up from the standard 8 GB.

aws ec2 run-instances --image-id ami-0cd5fb602c264fbd6 --instance-type t3a.xlarge --count 1 --key-name ${MYKEYNAME} --block-device-mappings 'DeviceName=/dev/sda1,Ebs={VolumeSize=48}'

Once running, get the instance ID and public DNS:

aws ec2 describe-instances --query "Reservations[*].Instances[*].{InstanceId:InstanceId,Host:PublicDnsName}"

This should show all instances, so work out which one you need if there are multiple.

Open SSH port access

Using the instance ID you obtained just preceding (for example i-01bd8a4ee6016e1fe), use that to first query for the ‘security GroupSet ID’ (for example sg-96f73feb).

Then use that to open up port 22 for the security group (and hence for the instance). Multiple instances running in this security group, all of which are now exposed on Port 22.

aws ec2 describe-instance-attribute --instance-id i-01bd8a4ee6016e1fe --attribute groupSet

aws ec2 authorize-security-group-ingress --group-id sg-96f73feb --protocol tcp --port 22 --cidr

SSH into the instance

Using the Public DNS obtained a couple of steps back, you can now SSH into the instance:

ssh -i ~/.ssh/${MYKEYNAME}.pem

Install Minikube etc

Install Minikube and Kubectl:

curl -LO
sudo install minikube-linux-amd64 /usr/local/bin/minikube

curl -LO
chmod +x kubectl

sudo usermod -aG docker ubuntu && newgrp docker

Obtain QHub repo

git clone

Set up Minikube - very similar to setup for local Minikube:

minikube start --cpus 4 --memory 12288 --driver=docker
minikube ssh "sudo apt update; sudo apt install nfs-common -y"

python3 ./qhub/tests/scripts/

minikube addons enable metallb

Initialize and deploy QHub

Install Virtualenv and Pip:

sudo apt update
sudo apt install python3-virtualenv -y
sudo apt install python3-pip -y

Create and activate a virtualenv, install QHub dev:

cd qhub
virtualenv ./data-venv
. ./data-venv/bin/activate

pip3 install -e .[dev]

Create and modify qhub-config.yaml:

mkdir data-test
cd data-test

export QHUB_GH_BRANCH=main
qhub init local --project=thisisatest  --domain --auth-provider=password

sed -i -E 's/(cpu_guarantee):\s+[0-9\.]+/\1: 1/g' "qhub-config.yaml"
sed -i -E 's/(mem_guarantee):\s+[A-Za-z0-9\.]+/\1: 1G/g' "qhub-config.yaml"

The preceding two commands reduce slightly the memory and CPU requirements of JupyterLab sessions etc. Make any other changes needed to the qhub-config.yaml file.

Then deploy:

qhub deploy --config qhub-config.yaml --disable-prompt

Enable Kubernetes access from Mac

This step is optional, but allows you to use kubectl and k9s directly from your Mac. It’s not needed if you are satisfied to use kubectl within an SSH session on AWS instead.

On your Mac laptop:

cd ~
mkdir .minikube_remote

Copy these files from the remote instance (home folder):

  • .minikube/ca.crt to .minikube_remote/ca.crt

  • .minikube/profiles/minikube/client.crt to .minikube_remote/client.crt

  • .minikube/profiles/minikube/client.key to .minikube_remote/client.key

For example:

cd .minikube_remote
scp -i ~/.ssh/${MYKEYNAME}.pem .
scp -i ~/.ssh/${MYKEYNAME}.pem .
scp -i ~/.ssh/${MYKEYNAME}.pem .

Merge the following into your ~/.kube/config file, or just run the command in full to overwrite it:

cat <<EOF > ~/.kube/config
apiVersion: v1
- cluster:
    certificate-authority: ../.minikube_remote/ca.crt
  name: minikube
- context:
    cluster: minikube
    user: minikube
  name: minikube
current-context: minikube
kind: Config
preferences: {}
- name: minikube
    client-certificate: ../.minikube_remote/client.crt
    client-key: ../.minikube_remote/client.key

Now SSH into the AWS instance, enabling port forwarding so you can access the Minikube cluster as though It’s running on your Mac:

ssh -i ~/.ssh/${MYKEYNAME}.pem -L

You should now find that kubectl and k9 work for the Minikube cluster if you run them on your Mac. This can include kubectl port-forward to access Kubernetes services individually.

Access the full QHub website

However, the way Traefik works won’t allow you to port forward to that since it won’t see the right domain name and port.

The users can trick it by setting up a hostname alias. Run sudo vi /etc/hosts on the Mac and add:

And then the users can add an extra port forward when they SSH into their AWS instance.

sudo ssh -i ~/.ssh/${MYKEYNAME}.pem -L -L

This is executed with the sudo privileges because forwarding a low-numbered port, like 443, is not allowed otherwise.

Now you can access in a browser and you should be able to use your QHub. You have to bypass the self-signed cert warnings though - see verify the local deployment for instructions.