Currently, if there are not enough resources (RAM or CPU) to start a large VM on an individual node, the VM will not start, but remains unscheduled until enough resources become available on at least one node.

If there are enough resources in the cluster, but enough resources spread out on individual nodes across the cluster, we can live migrate workloads to other nodes, and free up enough CPU and memory to allows the large VM to start.

This VM rebalancing will still need to adhere to the principles of node selectors, affinity and anti-affinity rules, and any CPU pinning.