Sorry, I started trying to download them from dockerhub but they didn't exist there (rancher/k3s-upgrade), then I realized your deployment script deploys to
ghcr.io and I found the current image there, so I tested that too in case there was an issue with dockerhub.
The dockerhub images seems to work now, but I think there was a day or so where the channel server was updated but the dockerhub images did not exist. Anyway, the above yaml snippet from the server plan will block until the image works, instead of starting to cordon/drain nodes.