How does a cloud provider keep track of failed server instances?