Consider adding Deleting phase for machine deployments

[fabriziopandini]

Before 1.11, machine deployment wasn't updating replica counters of phase during deletion

With v1.11 this was fixed, but when deleting a machine deployment phase is scalingDown which is correct except for when replicas = 0 and deletion is blocked by a finalizer.

We can consider adding a Deleting phase for machine deployments to use at the end of the deletion process (or during the entire deletion process).

However, doing this now has the downside to figure out what to do for up/down conversions, so I think we should defer to when v1beta1 is removed (or to when all the priority work for v1beta2 is completed)

[sbueringer]

Sounds good

/triage accepted

[arshadd-b]

/assign

[sbueringer]

so I think we should defer to when v1beta1 is removed (or to when all the priority work for v1beta2 is completed)

I agree. I don't see how we can solve the conversion problem if we introduce the phase today when we still support conversion to v1beta1

[arshadd-b]

So we will wait for the v1beta2 items to be completed, then we can start on this.

[sbueringer]

when we still support conversion to v1beta1

We will have to support conversion to v1beta1 until including Cluster API v1.17 (August 2027) (or maybe only as long as we serve v1beta1, which would be until including v1.13 | April 2026).

It's not clear to me how we can do it before. But maybe someone else has an idea