Making changes to OKD¶

The source code for OKD is available on github. OKD is made up of many components bundled into a release. You can find the exact commit for each component included in a release using the oc adm release info command with the --commit-urls option, as outlined in the overview section.

To make a change to OKD you need to:

1. Identify the component(s) that needs to be changed
2. Clone/fork the git repository (you can choose to fork the exact commit used to create the image referenced by the OKD release or a newer version of the source)
3. Make the change
4. Build the image and push to a container registry that the OKD cluster will be able to access
5. Run the modified container on a cluster

Building images¶

Most component repositories contain a Dockerfile, so building the image is as simple as podman build or docker build depending on your container tool of choice.

Some component repositories contain a Makefile, so building the image can be done using the Makefile, typically with make build

First thing to do is to replace the FROM images in Dockerfile.rhel7. You may want to just copy it to Dockerfile and then make the changes.

    FROM registry.ci.openshift.org/openshift/release:golang-1.17 AS builder

    FROM registry.ci.openshift.org/origin/4.10:base

The original images are unavailable to the public. There is an effort to update the Dockerfiles with publically available images.

Example Scenario¶

• Modify console-operator to have a link to the community site okd.io instead of docs.okd.io
• add to pre-existing cluster
• build a custom release to include the modified console-operator, then install a new cluster will custom release

To complete the scenario the following steps need to be performed:

1. Fork the console-operator repository
2. Clone the new fork locally: git clone https://github.com/<username>/console-operator.git
3. create new branch from master (or main): git switch -c <branch name>
4. Make needed modifications. Commit/squash as needed. Maintainers like to see 1 commit rather than several.
5. Create the image: podman build -f <Dockerfile file> -t <target repo>/<username>/console-operator:4.11-<some additional identifier>
6. Push image to external repository: podman push <target repo>/<username>/console-operator:4.11-<some additional identifier>

7. Create new release to test with. This requires the oc command to be available. I use the following script (make_payload.sh). It can be modified as needed, such as adding the correct container registry and username:

   server=https://api.ci.openshift.org
   
   from_release=registry.ci.openshift.org/origin/release:4.11.0-0.okd-2022-04-12-000907
   release_name=4.11.0-0.jef-2022-04-12-0
   to_image=quay.io/fortinj66/origin-release:v4.11-console-operator
   
   oc adm release new --from-release ${from_release} \
                   --name ${release_name} \
                   --to-image ${to_image} \
                   console-operator=<target repo>/<username>/console-operator:4.11-<some additional identifier>
   

   from_release, release_name, to_image will need to be updated as needed

8. Pull installer for cluster release: oc adm release extract --tools <to_image from above> (Make sure image is publically available)

If it is not clear how to build a component you can look in the release repository at https://github.com/openshift/release/tree/master/ci-operator/config/openshift/<operator repo name>, this is used by the Red Hat build system to build components so can be used to determine how to build a component.

You should also check the repo README.md file or any documentation, typically in a doc folder, as there may be some repo specific details

Running the modified image on a cluster¶

An OKD release contain a specific set of images and there are operators that ensure that only the correct set of images are running a cluster, so you need to do some specific actions to be able to run your modified image on a cluster. You can do this by:

1. configuring an existing cluster to run a modified image
2. create a new installer containing your image then creating a new cluster with the modified installer

Running on an existing cluster¶

The Cluster Version Operator watches the deployments and images related to the core OKD services to ensure that only valid images are running in the core. This prevents you from changing any of the core images. If you want to replace an image you need to scale the Cluster Version Operator down to 0 replicas:

oc scale --replicas=0 deployment/cluster-version-operator -n openshift-cluster-version

Some images, such as the Cluster Cloud Controller Manager Operator and the Machine API Operator need additional steps to be able to make changes, but these typically have a docs folder containing additional information about how to make changes to these images.

Create custom release¶