Taints Propagation Proposal implementation

[chrischdi]

This issue is for tracking progress of implementation for the taints propagation proposal:

• https://github.com/kubernetes-sigs/cluster-api/blob/main/docs/proposals/20250513-propogate-taints.md

We plan to split the implementation across Cluster API v1.12 and v1.13, which is considered safe because the changes will be feature-gated.

1.12:

• Machines (@chrischdi) :
  • Implement MachineSpec API changes (including validation)
  • Implement Machine controller changes
  • Implement unit-test coverage
  • PR: ✨ taint propagation: machine related API changes, conversion and feature gate #12936
• Implement e2e-test coverage for taints (@chrischdi)
  • Maybe fits into the MD Rollout test, make sure its easily portable to ClusterClass rollout test later on
  • PR: 🌱 taint propagation: e2e coverage via md rollout test #12966
• Check the repository for the handling of the NodeDrainTimeoutSeconds field (prior art in-place propagation to Machine) and if same was applied for the Taints field.

1.13

• We should look into numeric taints and gt/lt operator and figure out what that means for this feature (xref: KEP-5471: Extend tolerations operators kubernetes/kubernetes#134665)

  • Note: This does not effect Machine Taints at all

• CABPK: implement merging behavior

  • discuss desired behavior: add to kubeadm config (which results in adding as --register-with-taints for kubelet / re-registration issue) or not
    • Note: the proposed behavior was documented here https://github.com/kubernetes-sigs/cluster-api/blob/main/docs/proposals/20250513-propogate-taints.md#behavior-with-cluster-api-bootstrap-provider-kubeadms-taints-fields but there were concerns if we should do that

• ControlPlane contract:

  • PR: ✨ KCP: Implement support for machine taints #13181
  • Implement ControlPlane contract changes
  • Implement KCP changes to fulfill the contract
  • Implement test coverage (extend the e2e used for MachineDeployments)
  • Check the repository for the handling of the NodeDrainTimeoutSeconds field (prior art in-place propagation to Machine) and if same was applied for the Taints field.

• topology:

  • PR: [WIP] ✨ topology: implement support for taints #13192

  • Implement Cluster related API changes

  • Implement ClusterClass related API changes

  • Implement topology controller changes / field propagation

  • Check the repository for the handling of the NodeDrainTimeoutSeconds field (prior art in-place propagation to Machine) and if same was applied for the Taints field.

• Migrate e2e test coverage for taints from non-topology to topology

  • PR: [WIP] ✨ topology: implement support for taints #13192
  • probably fits at the existing ClusterClass rollout test
  • we might want to keep the existing coverage for non-topology

• Migrate tracking annotation names to contain a internal prefix (for labels, annotations and taints tracking)

• Add documentation to the book below 1.11 Experimental Features (via a subpage)

Additional work:

• Cluster autoscaler to be aware of the taints field
• MachinePool: implement MachinePool controller changes
  • Note: maybe should be tracked in MachinePool graduation tracking issue (edited)