It was previously being pulled in via weak dependencies of fwupd,
but we removed fwupd in #47 and now it's not there any more. It
is needed for the first boot resize by cloud-init to work, since
we use a btrfs filesystem.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Use consistent network device names for network devices instead of
forcing the old "ethX" names from pre-2017. This ensures that
specialized network devices, such as SR-IOV devices, are easy to
recognize and configure inside a Fedora instance on a public cloud or
OpenStack cloud.
FESCo ticket: https://pagure.io/fesco/issue/3190
Change proposal: https://fedoraproject.org/wiki/Changes/EnableConsistentDeviceNamingCloud
Signed-off-by: Major Hayden <major@redhat.com>
It's the *Google* image that's required to have a 10 GB root
for performance reasons, not the EC2 image, as the comment says,
but the change was inadvertently applied to the EC2 image not
the Google one. This means our Google image is slow and our EC2
images are failing to be published as AMIs.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Most of the Fedora Cloud-owned profiles are limited to a subset of
architectures, generally x86_64 and aarch64 (with the exception of
the VirtualBox Vagrant image, which is x86_64 only).
Azure provides an optional "Accelerated Networking" feature. Enabling
this feature when creating a VM enables SR-IOV and provides the guest
with a virtual function.
Supporting this requires two changes. Firstly, the kernel-modules
package is required as it contains the Mellanox drivers required for the
VF provided to the guest. Secondly, the interface needs to be ignored by
NetworkManager or it'll try and fail to bring up the device and the VM
will be unreachable except by serial console. If this is observed, it's
likely new hardware has been deployed and additional drivers need to be
added to the NetworkManager config.
See https://learn.microsoft.com/en-us/azure/virtual-network/accelerated-networking-overview
for details.
