Design a Top-K trending service

## Scenario You implemented a function to compute “top K” (e.g., most frequent items). Now design a backend service that provides similar functionality at scale. ## Requirements Design a service that continuously computes and serves the **top K most frequent items** (e.g., hashtags, products, search queries). ### Functional requirements - Ingest a high-volume stream of events: each event includes at least `(item_id, timestamp)`. - Support an API to query `GET /topk?k=K&window=...` returning the top-K items by frequency. - Support a time window, at minimum: - Either **all-time** counts, or - A **sliding window** (e.g., last 5 minutes / last 1 hour). State your choice/assumptions. ### Non-functional requirements - Low latency queries (e.g., p95 < 100 ms) at high QPS. - High write throughput ingestion. - Horizontal scalability (add machines to scale). - Fault tolerance and reasonable correctness guarantees (clearly state consistency model). ### Clarifications to address - What are the expected event rate, number of distinct items, and query QPS? - Do you need exact results or are approximate results acceptable?