On Wed, Jul 20, 2022 at 11:42 AM Martin Pitt <mpitt@redhat.com> wrote:

Hello Mark,

Mark O'Brien [2022-07-20 15:48 +0100]:
> Apologies for the delay in response but access to /dev/kvm is not something
> we want to support in the new setup.
> Sorry we can't provide this for you but if your workflow can change to fit
> in our infrastructure we would welcome you.

Ack, thanks for confirming! We can't do without /dev/kvm, so no need to migrate
our project then.

Out of interest, why is it so bad/hard to support? It's not like you need it to
burn CPU cycles or allocate memory or so :-)

FWIW for OpenShift's own upstream CI (based on Prow) we deployed https://github.com/cgwalters/kvm-device-plugin

which is a new fork of the minimized infrastructure from KubeVirt needed for this.

It runs on a GCP cluster ("build02") and we use it for e.g. testing RHEL CoreOS and other upstream projects (some parts of Fedora CoreOS too).

There's some use of the OCP Prow instance for projects that live outside of the github.com/openshift namespace, in e.g. github.com/containers and github.com/coreos,

although it's not a focus for the CI team.

But I will support use of the plugin elsewhere and happy to share tips/advice!