The Machine objects should not be deployed in the shoot namespace.

[krgostev]

Currently the machine objects are deployed in the shoot namespace in the seed. This is not compatible with the control plane migration, because at the end of the migration that namespace is deleted and with it all machine objects. This is not desired in a scenario with control plane migration, so the the machine objects should be created in a different namespace, most probably dedicated only for them.

cc @plkokanov @rfranzke

[rfranzke]

Does this matter much for the CPM? Can't this be treated as implementation detail of how machines work for the local provider? In the end, all that matters (from g/g's PoV) is that the Machine object is migrated and that the cluster is healthy again, right?

[plkokanov]

I see how this can be implementation dependant, but one of the requirements for the control plane migration was that the nodes do not get recreated during the procedure. Otherwise we could miss issues like this one: gardener/gardener#5423

[rfranzke]

Let's start simple for now and then make increments as needed. Let's get the local e2e test done for CPM first.