Change to network interface naming on new installs

TLDR: Due to a bug, all current FCOS machines are using legacy network interface naming (e.g. eth0). Fixing and changing the default naming scheme will be targeted at freshly provisioned instances ONLY. Existing instances will be able to switch to the new scheme via manual intervention.

Hi all,

Due to an oversight, network interfaces (NICs) on Fedora CoreOS are currently named using the kernel’s default scheme (e.g. eth0). Depending on the order that devices are discovered, these names can change from boot to boot.

Unfortunately we did not realize this until recently, as described in #484. This means that currently all FCOS nodes on all streams (thus including stable) are not using predictable interface names as used in other Fedora Editions and Linux distributions.

Starting with the next release cycle we plan to fix this bug, thus changing the default behavior for new Fedora CoreOS releases.

As this is a quite invasive behavioral change, it will only affect newly-provisioned machines. NIC names on existing systems will not change, as upgrades there will automatically pin the current behavior via the kernel command line.

The ticket above has further details on the plan of action, but in short we are going to:

  • (in the upcoming release cycle) pin existing nodes into their current naming scheme
  • (in the release cycle after that) switch to persistent naming scheme for new nodes

If everything goes as planned, the end result is that:

  • freshly provisioned instances will use the persistent naming scheme only
  • existing instances will stick to their existing naming scheme
  • in those cases where it is safe to change existing behavior, manual intervention can be applied to opt-in into the new default behavior

If you deploy Fedora CoreOS using an Ignition config that references NICs by name, you’ll need to update your config to reflect the new naming. Also, if you have a mixed fleet of older and newer Fedora CoreOS installs, note that NIC names may not be consistent across machines (unless manually switched).

We will keep posting progress and updates to the bug tracker, please keep watching #484 for details.