Support for Spot Instances eviction notice on Windows workers

1. Proposed title of this feature request

Support for Spot Instances eviction notice on Windows workers

2. What is the nature and description of the request?

As defined in the OpenShift documentation, when AWS/Azure sends the notification for preemption/eviction of the node (2 minutes in advance for AWS and 30 seconds for Azure), the workloads start to move away from the affected instance. However, this case only applies for the Linux workers, Windows workers do not start the draining process for the node, leaving the workloads around until machine-api recognized that the machine is not available anymore, moment in which the workloads are moved and a new machine is provisioned to comply with the replicas. However, this situation causes workload outage for the deployments running on the Windows nodes, in opposition to the Linux workers case.

3. Why does the customer need this? (List the business requirements here)

Taking into account the importance of managing costs, some organizations which use OpenShift for the CI/testing infrastructure might considering moving their Worker nodes into Spot Instances and the lack of this behavior in the case of Windows workers might block them from migrating into Spot Instances. Causing first a disappointment in the customer (this behaviour is stated in the documentation and there is no explicit notice that it isn't supported by Windows workers) and second it will incur into higher expenditure and unsatisfaction, if only if the customer use Windows workloads.

4. List any affected packages or components.

WMCO

Machine-api