In this case Argo uses the Kubernetes distributed state store (etcd or equivalent) and is a stateless service so can deal with failures quite well.