Adaptive Replica Selection (ARS) replaced Simple Round-Robin for routing shard requests. What specific metrics does ARS use to improve P99 tail latency?

1. A It routes based only on the physical proximity of the node to the client's IP address.
2. B Dynamic Health: It tracks real-time response times (EMA), query queue size on each node, and service-time metrics, routing traffic away from 'struggling' nodes (e.g., those in long GC pauses).
3. C It uses the total number of open files on the node as the sole routing heuristic.
4. D It always routes to the Primary Shard unless the Primary is at 100% CPU.