Skip to content
/ os Public
forked from openshift/os

Latest commit

 

History

History
130 lines (98 loc) · 3.52 KB

development.md

File metadata and controls

130 lines (98 loc) · 3.52 KB
Raw

Background

RHCOS is a derivative of both RHEL and Fedora CoreOS. The tool to build both RHCOS and FCOS is coreos-assembler. See the upstream documentation there first, and pull down the container.

coreos-assembler (or "cosa") accepts a configuration git repository as input. This repository is that configuration for RHCOS, just like fedora-coreos-config is for FCOS.

For example, you would use

$ cosa init https://github.com/openshift/os

to start. However, currently you need to configure the rpm-md repositories manually:

Repositories

The RHEL repositories are only available to Red Hat customers, and it's likely that you want to use a mirror. And further, a common scenario is to inject specific versioned rpm-md repositories to test different snapshots or "composes". For these reasons, you need to provide the repos of input RPMs.

To do so, create a rhcos.repo file that looks like this:

# RHEL repos
[rhel-8-baseos]
baseurl=<url>

[rhel-8-appstream]
baseurl=<url>

[rhel-8-fast-datapath]
baseurl=<url>

# These are the OpenShift RPMs, see https://mirror.openshift.com/pub/openshift-v4/dependencies/rpms/
# except there's things like afterburn that are only internal right now unfortunately.
[rhel-8-server-ose]
baseurl=<url>

The names of the repos must match those in manifest.yaml.

Updating FCOS

As noted above, RHCOS uses FCOS as an upstream. This is implemented by inheriting from the fedora-coreos-config repository as a git submodule; the RHCOS manifests and overlays include parts (though not all) of FCOS.

After a change is landed in FCOS, it's often OK to also immediately update RHCOS' main branch to the latest FCOS. We try to keep things compatible.

$ (cd fedora-coreos-config && git fetch origin && git reset --hard origin/testing-devel)
$ git commit -a -m 'Update FCOS'

Debugging failures in app.ci

Occasionally it is necessary to debug test failures on the OpenShift CI clusters by scheduling a cosa pod on the cluster. There's a few additional steps after getting the pod scheduled, in order to be able to build and test RHCOS.

  1. Create the pod

Using the YAML below, create a new pod with oc create -f pod.yaml

apiVersion: v1
kind: Pod
metadata:
  # Replace this with e.g. $USER-cosa
  name: test-cosa
spec:
  containers:
  - args:
    - shell
    - sleep
    - infinity
    image: registry.ci.openshift.org/coreos/coreos-assembler:latest
    name: cosa
    resources:
      requests:
        memory: "6Gi"
        devices.kubevirt.io/kvm: '1'
      limits:
        memory: "6Gi"
        devices.kubevirt.io/kvm: '1'
    volumeMounts:
    - mountPath: /srv
      name: workdir
    securityContext:
      privileged: false
  volumes:
  - name: workdir
    emptyDir: {}
  restartPolicy: Never
  1. Set the umask
$ pushd /srv
$ umask 0022
  1. Init the repo

$ cosa init https://github.com/openshift/os

  1. Populate the repo files

Determine which repo the failing CI job is using and replicate it.

This can be done by inspecting the results of the failing job and looking for the portion of the test that uses curl to populate the ocp.repo (see https://github.com/openshift/os/blob/master/ci/prow-build-test-qemu.sh#L35).

For example:

$ curl -Ls http://base-4-11-rhel8.ocp.svc.cluster.local > src/config/ocp.repo

  1. Build and test normally