Any issues with NFS root?

I have successfully gotten my diskless machine to boot and run from the CoreOS live PXE image. But it is a rather small machine and I would prefer not to take the ramdisk memory hit. I also don’t want to download everything from the Internet on every boot, so I have to manage caching images, container files, etc. In general I’d be happier with an installed OS, but I also want it to be diskless. Prior to CoreOS, I was booting Ubuntu from an NFS root filesystem (root=/dev/nfs initrd=initrd.img-4.15.0-96-generic nfsroot=…).

Ideally, I would like to boot the installer, have it mount an empty NFS filesystem, stream the installation into it, and then reboot and run from there. Just like installing on bare metal, but with my NFS root instead of /dev/sda.

I feel like I’m entering uncharted territory here (I have found no mentions of it anywhere) and I’m just curious whether this is so wildly off track from what it supported that I shouldn’t waste my time. Or on the other extreme, maybe it’s completely doable. I’d appreciate any advice.

1 Like

Yeah, it’s uncharted territory. NFS root has some potential advantages for clusters, but also 1) poses some challenges around updates and garbage-collecting old releases from the NFS server, 2) has some potential interactions with how we run Ignition in the initramfs, and 3) requires the network interface to remain up during the transition from the initramfs to the real root. Also, the installer currently works exclusively with disk images, not filesystem images or directory trees. Those issues aren’t on our radar right now. You might be able to hack something together, but it’d probably be fragile.

If you’d like, you can file an enhancement request in the Fedora CoreOS issue tracker for consideration.

1 Like