Replace Round-Robin Routing with Least-Latency Routing

Problem: Replace Round-Robin Routing with Least-Latency Routing An existing load balancer chooses backend servers using round robin. Replace it with a least request latency policy: for each new request, select the healthy, non-circuit-broken backend with the lowest predicted request latency. Provide a design and implementation explanation covering: The latency statistics maintained for each backend. A latency-estimation method, such as the latest latency, sliding-window average, or EWMA (exponentially weighted moving average). How the metric is updated when a backend request completes. How to handle cold-start backends with no latency history. How to avoid continually selecting a low-latency backend that already has many queued or in-flight requests. How health-check and circuit-breaker state filter candidates. Time complexity per routing decision and optimizations for a large backend pool. Assume the actual processing or end-to-end latency is available when each request completes.