Description of problem:

We suspect that https://github.com/openshift/oc/pull/1521 has broken all Metal jobs, an example of a failure: https://prow.ci.openshift.org/view/gs/origin-ci-test/pr-logs/pull/openshift_cluster-baremetal-operator/355/pull-ci-openshift-cluster-baremetal-operator-master-e2e-metal-ipi-ovn-ipv6/1691359315740332032.

Details:

The testing scripts we use set KUBECONFIG in advance to the location where we'll create it. At the time "oc adm extract" is called, the file does not exist yet. While you could argue that we should not do it, it has worked for years, and it's quite possible that customers have similar automation (e.g. setting KUBECONFIG as a global variable in their playbooks). In any case, I don't think "oc adm extract" should try to read the configuration if it does not explicitly need it.

Updated details:

After the change, "oc adm extract" expects KUBECONFIG to be present, but at the point when we call it, there is no cluster. I initially assumed that unsetting KUBECONFIG will help but it does not.