Should be solved by just adding kubeletExtraArgs to the initConfiguration (and joinConfiguration?) below?

Ack, removed.

This works (at least for me) if you just patch the quick-start-control-plane KubeadmControlPlaneTemplate. So these comments could be dropped?

Yes, seems to work now. Removed.

I think this should be dropped (so that the autoscaler knows that it can control the field)

I am removing the entire "machinePools" section here because that is for a pool with an experimental API. What we want is just the "machineDeployments". This is what is left after patching:

    workers:
      machineDeployments:
      - class: default-worker
        name: md-0
        replicas: 1

@marquiz: agreed?

I pushed an update that addresses the other points.

Sorry for the delay, I had to put this aside for a while to focus on 1.30. It's still useful to have for 1.31, because that is where we hopefully will have cluster autoscaler support.

/cc @towca

@pohly sorry my comment was off-by-a-few-lines. I think the replicas field should be removed for the machineDeployments

As it stands, the cluster comes up with one worker node. Then autoscaler can scale up or down. I prefer that over not bringing up any worker node initially because then a problem with that only surfaces later.

Your comment was "so that the autoscaler knows that it can control the field" - I don't think that setting the initial value prevents that.

IIRC in my testing it did. When the autoscaler sees that it's set it determines that it's controlled by some other entity and refuses to act on it. It default to 1 (if not set), IIRC

You are right, the instructions are not enough to actually make the autoscaler do anything. It finds no node groups. But did you really get it to work as you suggested above?

What is missing is something else, the annotations on the MachineDeployment:
https://cluster-api.sigs.k8s.io/tasks/automated-machine-management/autoscaling#enabling-autoscaling

@sbueringer: any suggestion how to get those annotations added automatically to the MachineDeployment?

Do we need --node-group-auto-discovery=clusterapi:clusterName=capi-quickstart or is it enabled by default?

https://cluster-api.sigs.k8s.io/tasks/automated-machine-management/autoscaling#configuring-node-group-auto-discovery says "you must configure node group auto discovery" but https://cluster-api.sigs.k8s.io/tasks/automated-machine-management/autoscaling#enabling-autoscaling says "The autoscaler will monitor any MachineSet, MachineDeployment, or MachinePool containing both of these annotations."

When the cluster comes up after following the instructions in this README.md, it has a generated machine deployment with a variable name. I could come up with a kubectl invocation that adds the annotations, but it would be nicer to include that in the cluster configuration.

Also, does the cloud provider in use here (Docker, from clusterctl init --infrastructure docker) support scale from zero automatically? https://cluster-api.sigs.k8s.io/tasks/automated-machine-management/autoscaling#scale-from-zero-support documents some additional annotations, but it is not clear if they are needed.

@sbueringer: any suggestion how to get those annotations added automatically to the MachineDeployment?

You can patch:

        - class: default-worker
          name: md-0
          replicas: 1

to be this instead:

        - class: default-worker
          name: md-0
          metadata:
            annotations:
              cluster.x-k8s.io/cluster-api-autoscaler-node-group-min-size: "1"
              cluster.x-k8s.io/cluster-api-autoscaler-node-group-max-size: "3"

The result is:

• the annotations will be set on the MachineDeployment
• the MachineDeployment webhook will pick the value of min-size as initial value for MD.spec.replicas

If you are setting replicas and the autoscaler annotations here, the CAPI controller and the autoscaler would both try to write the MD.spec.replicas field continuously.

Do we need --node-group-auto-discovery=clusterapi:clusterName=capi-quickstart or is it enabled by default?

I'm not sure if you have to set --node-group-auto-discovery. In our e2e test we do it, but we also have other tests running in other namespaces. You can give it a try without it.

Also, does the cloud provider in use here (Docker, from clusterctl init --infrastructure docker) support scale from zero automatically?

CAPD falls under:

If your Cluster API provider does not have support for scaling from zero, you may still use this feature through the capacity annotations.

So you'll have to set the annotations. Also in CAPD you can't really define via DockerMachine spec which size your "Machine" has like e.g. in AWS. But you can check on the created Nodes via Nodes.status.capacity what the capacity of a Node is. Not really sure where that is coming from though.

Reminder to self: this still needs to be included in the instructions.

you should use kindnetd is much more stable kubernetes-sigs/cluster-api@d0c495a and consume less resources

+1 Good catch!

what can I say, it is my baby :)

While we're here. Thx for maintaining kind & kindnet. It's a very nice and stable foundation to build our own testing in CAPI on. Safes us so much time & effort.

kind is Ben baby ,

appreciate the compliments ❤️