Design rolling-window top-K click tracker

You receive a high-volume stream of click events (timestamp in ms, url). Implement a data structure with two operations: 1) record(event) inserts one event; 2) queryTopK(T, K, now) returns the K most-clicked URLs in the half-open interval (now − T, now], and queryUnique(T, now) returns the number of distinct URLs in that interval. Requirements: maintain near real-time results with efficient insert and eviction; target O(log K) or better per event for top-K maintenance; ensure memory stays bounded via time-based eviction. Handle out-of-order events with maximum lateness L and deduplicate by an event id for idempotency. Discuss the data structures you would use (e.g., time-bucketed sliding windows, hash maps for counts, heaps for top-K, deques for expirations), analyze time/space complexity, and explain how you would adapt the design for concurrency (multiple producer threads, lock contention minimization, atomicity of updates) and horizontal sharding.