Karmada
Runs many Kubernetes clusters as though they were one. You submit a perfectly ordinary resource plus a rule for which clusters it should end up in.
The project came out of Huawei and now lives under the CNCF. The architecture is an API server of its own that looks exactly like Kubernetes', where you send your manifests as usual. The difference is two extra object types: a PropagationPolicy deciding which member clusters the resource spreads to and how replicas are divided — evenly, by weight, or by how much spare capacity each cluster has — and an OverridePolicy for the fields that must differ per cluster, such as image registry or ingress hostname. If a cluster goes down, workloads can fail over automatically. Note the distinction from Cluster API, which creates clusters; Karmada puts things onto them.