Based on our conversation, the stream generator will produce the descriptor/namespace keys (either on its own partition or on a chosen partition), then we can emit the generated keys as we do today. Additionally, we could create a partition for each table created in the workload.