HACKING.md

Developer setup for working on cluster-api-provider-libvirt

Firstly you should have a Openshift 4.0 cluster up and running. This is already documented elsewhere so I won't get into that. Then you need to build a custom image for cluster-api-provider-libvirt:

# In ${GOPATH}/src/github.com/openshift/
git clone git@github.com:openshift/cluster-api-provider-libvirt.git
cd cluster-api-provider-libvirt
# Make any changes you need to in the source code
podman build -t ${YOUR_DOCKER_NS}/capl:test .
podman login # If not already done
podman push ${YOUR_DOCKER_NS}/capl:test

Let's avoid having to specify namespace all the time to oc:

oc project openshift-machine-api

Deploying the custom image

Before you can deploy your custom image, you need to disable CVO:

oc scale --replicas 0 deployments/cluster-version-operator -n openshift-cluster-version

There are at least two paths to deploying the image, the intrusive one and less-intrusive one.

The intrusive method

This one means you also disable MAO:

oc scale --replicas 0 deployments/machine-api-operator

Now you just edit the machine-api-controllers deployment and tell it to use your build image:

oc edit deployments machine-api-controllers

It's the image under the /machine-controller-manager command that you need to modify to ${YOUR_DOCKER_NS}/capl:test.

Wait a bit and your new image will be deployed. You can monitor the status with oc get pods. You can also see the logs with oc logs deploy/machine-api-controllers -c machine-controller | less.

The less intrusive method:

Instead of disabling MAO, you just update the relevant configmaps:

oc edit configmaps/machine-api-operator-images

and override the clusterAPIControllerLibvirt to your image. Then reset the relevant controller to make it get the new image:

```sh
oc scale --replicas 0 deployments/machine-api-controllers
# If we scale up immediately after, new images will not be used.
sleep 5
oc scale --replicas 1 deployments/machine-api-controllers

Wait a bit and your new image will be deployed. You can monitor in the same way as the other method.

Testing

So far you've only seen your new modified fancy provider coming up successfully. Let's make it work a bit by scaling up the worker nodes:

oc get machinesets
# The output will tell you the name of the worker machineset
# Let's assume it's "test1-wk7xq-worker-0" below
oc scale --replicas 2 machinesets/test1-wk7xq-worker-0

You should be immediately able to see machines coming up with oc get machines but it'll take a bit before machine is ready. You can monitor that using oc get machinesets. Also you can see the logs using the command mentioned previously to see what's happening.