Design a Top-K trending items service

Design a system that returns the **Top K** items (e.g., videos/posts/products) by popularity. ### Requirements - API to fetch: `GET /topk?window=10m&k=100` returning the top 100 items in the last 10 minutes. - Support multiple windows (e.g., 1m, 10m, 1h, 24h). - Near real-time updates (freshness target e.g. < 5 seconds). - Read QPS is high (e.g., 10k–100k QPS); write events can be much higher (e.g., millions/minute). - Popularity is driven by events (views/clicks/likes). Event stream may include duplicates and out-of-order delivery. ### Non-functional - Low latency for reads (e.g., p99 < 100 ms). - Fault tolerant, scalable, and cost-aware. - Discuss exact vs approximate Top-K and trade-offs. Produce an end-to-end design: ingestion, storage/compute, APIs, data model, and operations/monitoring.