We run a significant amount of stuff on spot-instances (AKS nodes) and use the service detect, monitor and gracefully handle the imminent shutdown on the Kubernetes side.

https://learn.microsoft.com/en-us/azure/virtual-machines/lin...