feat: development flags for talosctl cluster create docker

[battamatt]

Feature Request

In Talos v1.12+, development flags like --skip-k8s-node-readiness-check were moved under the talosctl cluster create dev command.

However, talosctl cluster create dev appears strictly coupled to the QEMU provisioner. There is currently no apparent way to use these development flags when targeting the Docker provisioner.

Use Case

We use cluster.network.cni.name: none in our machine configuration because we install Cilium via kubectl immediately after bootstrap. Because the nodes never reach a Ready state without a CNI, talosctl cluster create docker hangs indefinitely and eventually times out. In v1.11, we bypassed this using --skip-k8s-node-readiness-check, but this flag is no longer available for the Docker-specific subcommand.

To Reproduce

Try to skip readiness check on Docker:
talosctl cluster create docker --skip-k8s-node-readiness-check
Result: Error: unknown flag: --skip-k8s-node-readiness-check

Try to use the dev command with Docker:
talosctl cluster create dev --provisioner docker
Result: no unknown flag error but defaults to QEMU

Expected Behavior / Proposed Solution

talosctl cluster create docker should support the same development flags as the dev subcommand, or dev should allow a --provisioner docker flag.

[smira]

Thanks for reporting it, we will look into bringing back --provisioner=docker in talosctl cluster dev.

Please note that you can go directly to docker, you don't need talosctl: https://docs.siderolabs.com/talos/v1.12/platform-specific-installations/local-platforms/docker#running-talos-in-docker-manually

This way you can build your own flow as you like it.

talosctl cluster create is not a production deployment tool in general.

[smira]

@Orzelius I wonder if we can solve this more easily though by skipping the checks on node readiness if the CNI is set to none...

[battamatt]

@Orzelius I wonder if we can solve this more easily though by skipping the checks on node readiness if the CNI is set to none...

That would be enough for our use case at least.

[Orzelius]

@Orzelius I wonder if we can solve this more easily though by skipping the checks on node readiness if the CNI is set to none...

I'd like to avoid bringing the dev flags to the docker provider if possible so this solution is perfect!

[anthosz]

Hello,

I'm using talosctl cluster create docker (v1.12.6) with a patch that set cluster.network.cni.name = none but seems the probe is not skipped:

◲ ◱ waiting for all k8s nodes to report ready: some nodes are not ready: [talos-1-controlplane-1 talos-1-worker-1 talos-1-worker-2 talos-1-worker-3]

Do you know how to bypass it?

[smira]

@anthosz this is only part of Talos v1.13.0, which you can try out today at rc.0 or wait for the GA release next Mondy.

See #12331

[anthosz]

@anthosz this is only part of Talos v1.13.0, which you can try out today at rc.0 or wait for the GA release next Mondy.

See #12331

Thank you!