deadlock when trying to delete a machine about machine-controller HOT 4 CLOSED

From a user perspective orphaned instances at cloud provider mean losing some money. Thus setting finalizers as we do at the moment seems to be good and prevents exactly that scenario.

In situation like the one described we will make use of terminal errors (defined in the spec) to indicate errors that have to be resolved manually.

from machine-controller.

Would it solve the problem when we add the finalizer directly after we successfully created the instance at the cloud provider?

Or we do a getInstance in the sync loop and add the finalizer as soon as we get an instance?
That way we would solve the problem when the Patch/Updated fails in the case above.

from machine-controller.

Would it solve the problem when we add the finalizer directly after we successfully created the instance at the cloud provider?

I think so, but then we introduce another issue - a user decided to delete Machine while we were making a call to "create an instance"

from machine-controller.

Noticed that issue some time ago as well. IMHO the approach of adding the finalizer after the first getInstance succeeded makes most sense.

If for whatever reason getInstance stops working later on (e.G. credentials got invalidated) then the user has to delete the finalizer manually to be able to delete the machine, but I think that is way better than risking to leave orphaned instances at the cloud provider.

from machine-controller.