Ability to run containerd without k3s

[retr0h]

Is it possible to configure k3s such that only containerd starts up?

I'd like to pre-seed the local registry with container images, which would be pulled down from our remote registry, and added to the local containrd image registry. However, I do not want to start k3s yet, as I need to ensure the images were successfully pulled first, prior to pivoting the system to a new k3s architecture.

[brandond]

No, there is not currently any way to do that.

There is however some undocumented functionality to pre-pull images during containerd startup. Similar to how you can place image tarballs in /var/lib/rancher/k3s/agent/images for preload on airgapped nodes, any .txt files found in that directory will be treated as an image list, and the images pulled before the kubelet is started.

[retr0h]

Yeah, understood. For us, it is about accounting for unreliable networks and small links, prior to pivoting a system over to k3s. We do not want to deploy k3s for the first time, if we do not first have the images downloaded.

[brandond]

That sounds like a good case for preloading airgap images and/or running a pull-through registry mirror. Alternately, you can install your Linux distribution's containerd package, configure and prepare it however you like, and point K3s at it with the --container-runtime-endpoint option.