Stranded VMs (manual migration)
A host that loses its network cable keeps running — the VMs on it are alive, the CPU is busy, nothing crashed. But nobody can reach them. Any single-tier VM on that host goes dark from the customer's point of view, and it will not move itself somewhere reachable. It just sits there, stranded.
Single-tier VMs have no automatic failover, so recovery is on you. Open the customer row in the ops console and pick a fresh host to migrate them to. The moment they land on a reachable, healthy host, the customer is back. Higher tiers self-heal; single-tier is hand-driven until you sell the customer redundancy.
Detailed explanation
Single-tier customers contract for one VM on one host with no automatic failover. When that host loses reachability, the VM is stranded and stays stranded until the player issues a MigrateVm { vm_id, target_host_id }. There is no engine-side relocation for the Single tier by design — that is exactly what the customer chose not to pay for.
The engine validates the target host is online + reachable + same-AZ + capacity-clear before swapping the allocation, so a migrate cannot strand the VM a second time. Replicated / MultiAz tiers auto-failover onto their surviving replica; for Single tier this manual migrate is the only recovery path.