In past we discussed kcp distributing ClusterDeployments amongst various Hive shards. Investigate instead a deployment model where Hive literally ran against the kcp API, and when it needs to run a pod, sends this out to a shard cluster to run provision/deprovision.

See if we can get Hive running against kcp reasonably easily.

See if we can get provision/deprovision pods out to clusters and mirror status back to Hive.

Explore scale. If pods move to spoke clusters, is the remaining controller workload for hive day 2 controllers able to hold up against a kcp for a reasonable scale gain?