Analytical stores reward batches
Tiny synchronous inserts create excessive parts and coordination overhead. Producers or Kafka consumers should batch within bounded latency, row and byte budgets. Backpressure must propagate when ClickHouse maintenance or merges cannot keep up.