Labels and annotations for MachineDeployments and KubeadmControlPlane created by topology controller

[MaxFedotov]

User Story

As a user I would like to specify labels and annotations for MachineDeployment and KubeadmControlPlane created by topology controller to be used used by cluster-autoscaler or for filtering.

Detailed Description

It is possible to specify labels and annotations in cluster topology spec, but they are propagated only to machines created by MachineDeployment and KubeadmControlPlane.

As documented in cluster-autoscaler readme, in order to use it with Managed Topology, users had to skip spec.topology.workers.machineDeployments[].replicas field, which cause topology controller to always create MachineDeployment with only one replica. Users can later add specific cluster-autoscaler annotations in order to increase number of replicas.
And as this annotations cannot be added to MachineDeployment during provisioning, it is impossible to create them with any other number of replicas other then one.

It will be nice to have a way to pass labels and annotation from Cluster topology spec to MachineDeployment and KubeadmControlPlane and also be able to update them without triggering rollout for corresponding Machines

/kind feature

[fabriziopandini]

/area topology
@vincepri another possible improvement for cluster.spec.topolgy and/or ClusterClass

[enxebre]

Is there any other use cases we can think of for labels/annotations? e.g Syncing from class to Nodes.
For autoscaling in particular I'd expect this to be baked in MachineDeploymentClass API eventually e.g .autoscaling {min: , max: }

[fabriziopandini]

Syncing from classes to a node can happen for free if we use existing metadata fields, otherwise, if we add a new field for node labels it should be surfaced in CC as well

[fabriziopandini]

/triage accepted

As discussed in the Aug 5th issue triage meeting we should evaluate options:

• Propagate existing metadata both to MD/KCP and Machines (to be verified if this as undesirable side effects)
• Add new metadata fields for MD/KCP objects, TBD UX

[MaxFedotov]

Propagate existing metadata both to MD/KCP and Machines (to be verified if this as undesirable side effects)

One side effect would be that any update of labels\annotations (e.g. for autoscaler) will cause Machine rollout

[jackfrancis]

Related discussion about node annotations not propagating via Machine update: #6255

[jackfrancis]

Also related: #5880

[MaxFedotov]

After the discussion during Cluster API meeting it was proposed to separate this into two different issues:

1. Propagate labels and annotations from ClusterClass and Cluster.spec.topology not only to Machines, but also to MD and KCP
2. Prevent triggering of Machine rollout when labels or annotations in MachineDeployment.spec.template.metadata or KubeadmControlPlane.spec.machineTemplate.metadata are updated (can be done as a part of Support to propagate properties in-place from MachineDeployments to Machines #5880 or 📖 Label Sync Between MachineDeployment and underlying Kubernetes Nodes #6255)