This page:

These instructions will help you create a Kubernetes cluster on a set of Nimbus instances. This procedure takes approximately two hours of elapsed time.

Procedure

Before beginning, ensure that you have a running Ubuntu instance on Nimbus, and that you can connect to it.

1. Install Juju on your seed instance

1.1. Run this command to install Juju. It will take approximately 15 minutes to process.

sudo snap install juju --classic

1.2. When processing completes, list the clouds Juju has preconfigured. Note that Nimbus is not listed:

juju clouds --all

1.3a. Create a working directory in your home directory:

mkdir ~/juju

1.3. Add configuration for a Nimbus cloud.

Further details can be found at https://juju.is/docs/openstack-cloud

cat > ~/juju/nimbus-cloud.yaml <<EOF
clouds:
  nimbus:
    type: openstack
    auth-types: [userpass]
    endpoint: https://nimbus.pawsey.org.au:5000/v3
    regions:
      RegionOne:
        endpoint: https://nimbus.pawsey.org.au:5000/v3
EOF

juju add-cloud --client nimbus ~/juju/nimbus-cloud.yaml
juju clouds --all

There should now be a 'nimbus' entry in the listing

1.4. Add your credentials for the Nimbus cloud:

cat > ~/juju/credentials.yaml <<EOF
credentials:
  nimbus:
    CHANGE_THIS_USERNAME-nimbus:
      auth-type: userpass
      password: 'CHANGE_THIS_PASSWORD'
      project-domain-name: pawsey
      tenant-name: CHANGE_THIS_PROJECT
      user-domain-name: pawsey
      username: CHANGE_THIS_USERNAME
EOF

# store your nimbus username, project name, and password in these variables
username=OS_USERNAME
project=OS_PROJECT_NAME
read -s OS_PASSWORD

# then replace them in the file
sed -i'' -e "s/CHANGE_THIS_PROJECT/$OS_PROJECT_NAME/" ~/juju/credentials.yaml
sed -i'' -e "s/CHANGE_THIS_USERNAME/$OS_USERNAME/" ~/juju/credentials.yaml
sed -i'' -e "s/CHANGE_THIS_PASSWORD/$OS_PASSWORD/" ~/juju/credentials.yaml
unset OS_PASSWORD

juju add-credential --client nimbus -f ~/juju/credentials.yaml
rm ~/juju/credentials.yaml

2. Prepare an image for use by Juju

2.1. Run these commands to configure the metadata for an image (click for details). In this case we have given the IMAGE_ID for the Pawsey Ubuntu 18.04 image "Pawsey - Ubuntu 18.04 - 2022-05" (updated as of May 2022).

Further details can be found at https://juju.is/docs/cloud-image-metadata

IMAGE_ID=97c4562f-1087-4f2c-abc6-a7b02fb9f9b9
OS_SERIES=bionic
REGION=RegionOne
OS_AUTH_URL=https://nimbus.pawsey.org.au:5000/v3
mkdir ~/juju/simplestreams
juju metadata generate-image -d ~/juju/simplestreams -i $IMAGE_ID -s $OS_SERIES -r $REGION -u $OS_AUTH_URL

If you want to instead use the Pawsey Ubuntu 20.04 image "Pawsey - Ubuntu 20.04 - 2022-05", replace the first two lines above with:

IMAGE_ID=67bab16e-453b-46a8-a262-c0796fa35d85
OS_SERIES=focal

3. Use Juju to bootstrap the controller instance

During this step, you need to get the network ID of your private network. This should be connected via your router to the Public external network. You can check this at the Network Topology page. To bootstrap the controller instance:

3.1. Go to the Networks page.
3.2. Click on your network, which will be named something like projectname-network
3.3. Select and copy the ID (not Project ID), ready to paste into the NETID=... command below.
3.4. Back on the seed instance, run the following commands. The final bootstrap command will take approximately 20 minutes to process. While it is running, you can check https://nimbus.pawsey.org.au/project/instances/ to see a new controller instance being created.

NETID=YOUR_NET_ID # paste your network ID here instead of the YOUR_NET_ID
PUBEXT=dfb2cfd9-b746-410d-ab4b-f2e7d5bafacf
juju bootstrap nimbus nimbus-k8s-controller \
 --constraints "arch=amd64" \
 --metadata-source ~/juju/simplestreams \
 --model-default network=$NETID \
 --model-default external-network=$PUBEXT \
 --model-default use-floating-ip=true \
 --bootstrap-constraints "instance-type=n3.1c4r"

If you are using an Ubuntu 20.04 image, you must also include the flag "--bootstrap-series=focal" in the above "juju bootstrap" command.

4. Create a new model for the Kubernetes deployment

Models are Juju's concept of a workspace, so it is a good idea to create one for each deployment, to encapsulate them. This is true even with only one deployment. Refer to https://juju.is/docs/models for more information.

4.1. Create the model, then switch to ensure you are using it.

juju add-model nimbus-k8s-model
juju switch nimbus-k8s-model

From here onward, you can revert to this point in the instructions by deleting the model and recreating it:

juju destroy-model -y nimbus-k8s-model

Then go back and start again from step 4.

5. Use Juju to deploy Kubernetes, install kubectl

5.1. Install Kubernetes.

Installing Kubernetes takes approximately one hour.
- If you are familiar with the screen program, this would be a good time to use it.
View the status command to see installation progress.
Once everything is active and started the installation is ready to use.
Note: This approach installs Kubernetes with default machine constraints. For further information, including increasing those constraints, refer to https://jaas.ai/canonical-kubernetes

juju status
wget https://api.jujucharms.com/charmstore/v5/bundle/charmed-kubernetes-541/archive/bundle.yaml
sed -i'' -e "s/focal/bionic/" bundle.yaml
juju deploy ./bundle.yaml
watch -c 'juju status --color'
# Ctrl-C to stop the 'watch' command

5.2. While waiting, use another window to install the Kubernetes controller client kubectl. This will take 10 minutes.

sudo snap install kubectl --classic

6. Configure the kubectl Kubernetes client

6.1. Use the following commands to install and configure the client:

mkdir -p ~/.kube
juju scp kubernetes-master/0:config ~/.kube/config
kubectl get all
kubectl cluster-info

7. (Optional) Run an autoscaling stress test on Kubernetes

7.1. Use the commands below to run a stress test.

These steps are taken from https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale-walkthrough/

7.2. Launch a web server deployment and create an autoscaler:

kubectl apply -f https://k8s.io/examples/application/php-apache.yaml
kubectl autoscale deployment php-apache --cpu-percent=50 --min=1 --max=10
kubectl get hpa

7.3. In a separate window, generate load on the deployment:

kubectl run -it --rm load-generator --image=busybox /bin/sh
If you don't see a command prompt, try pressing enter.
/ # while true; do wget -q -O- http://php-apache; done
OK!OK!OK!OK!OK!OK!OK!OK!OK!OK!OK!...

7.4. In the first window again, oberve the scaling as it stabilises:

while true; do kubectl get hpa; sleep 30; done
NAME         REFERENCE               TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
php-apache   Deployment/php-apache   0%/50%    1         10        1          62s
NAME         REFERENCE               TARGETS    MINPODS   MAXPODS   REPLICAS   AGE
php-apache   Deployment/php-apache   250%/50%   1         10        4          92s
NAME         REFERENCE               TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
php-apache   Deployment/php-apache   69%/50%   1         10        5          2m3s
# Ctrl-C to stop the while loop

7.5. Stop and delete the pod and autoscaler:

kubectl delete hpa php-apache
kubectl delete -f https://k8s.io/examples/application/php-apache.yaml

User Support Documentation

Create a Kubernetes Cluster