Even if there are stateful workloads, one has to be prepared to migrate at some point. Nodes can crash, become unreachable, etc. - workloads have to be able to survive such events.
Edit: But I see your point - if their stateful workloads are not "migration/K8S-incident" ready, it's hard to suddenly build that overnight.
Edit: But I see your point - if their stateful workloads are not "migration/K8S-incident" ready, it's hard to suddenly build that overnight.