I understand and use standard package layering and see the references in the documentation to work that will potentially make this even easier which is great.
And, I know that it is possible to install custom rpms via say a systemd unit driven privileged container, mounting impacted directories (e.g. /usr/local/bin, /usr/local/lib64). We currently use this method to successfully install nvidia-container-runtime pre-requisites for GPU nodes presently.
But am I right in understanding that if one wanted to layer said custom rpms via rpm-ostree this would mean building an entirely new derivative rpm-ostree, like this? And once down this path it would seem you are in the realm of publishing your own rpm-ostree for updates…and therefore tying some custom rpm-ostree rebuild/update to follow FCOS central?
The option of building your own derivative rpm-ostree is incredible functionality frankly, but the overhead seems high when compared to the downsides of a one-shot systemd invocation of rpm install?
Perhaps we should include some documentation on this point in the FAQ as the ‘next level’ to the customizing OS after the layering topic?