Design Item Favorites and Like Counts

Design a system that lets millions of Roblox users “favorite” catalog items and instantly see accurate favorite counts on every item page. The system must support two high-traffic access patterns: (1) a user toggling the favorite star on/off and immediately seeing the correct state (favorite or not) for that item, and (2) every user viewing an item page and seeing the global favorite count within ~200 ms. Expect 1 M read QPS for count/status checks and 100 k write QPS for fav/unfav operations at steady state, with 5× peaks during item drops. Counts may be eventually consistent for other users, but a user must always read their own write. Storage cost is not a concern; correctness, low latency, and horizontal scalability are. Design the data model, API, and distributed backend that keeps both flows fast and available even when a single item suddenly goes viral.