Gong Zhang

It seems this one is more about testing efforts. Just raised a PR to test. - [ ] P1: Add clusterctl upgrade tests n-2, n-1 (supervisor)

@sbueringer ~~I could take this one once #3024 is merged. Thanks!~~ ~~- [ ] Add clusterctl upgrade tests n-3 (supervisor)~~

@sbueringer Makes sense to me. This will provide n-3 tests with much less effort. Thanks!

@sbueringer Just go ahead, no worries :)

@sbueringer I would like to work on it :-)

Good finding! For records, the current workaround to recover the whole cluster: - Manually create a `VirtualMachineService` and get the new accessible LoadBalancer IP address. - Update the `VSphereCluster.Spec.ControlPlaneEndpoint` with...

Sorry please ignore this... it works in local env as the LoadBalance IP doesn't change...

Considering the cluster has a dependency on VirtualMachineService, VSphereCluster created VirtualMachineService but it's vm-operator who manages VirtualMachineService, I'm thinking in CAPV side: 1. Add a finalizer when VSphereCluster to create...

Yes, I agree, that should indeed be the intention of the finalizer. However, I think it could be beneficial in this scenario to prevent the deletion of the VirtualMachineService until...

Now I get it, the key point is that adding a finalizer can not prevent the deletion of the underneath resources from vm-operator, so it is not a viable option...