Design a Geo-Distributed Leaderboard with Strong Consistency

To design a geo-distributed leaderboard with strong consistency, we must prioritize data integrity over global write availability. First, I would clarify that strong consistency implies linearizability, meaning every read returns the most recent write. Given Google's focus on reliability, I propose a Single-Region Write Dominance pattern. We designate a specific region, say US-East, as the primary master for all write operations. When a user submits a new score, the request is routed to this primary node. This ensures there is only one authoritative source of truth, eliminating the complexity of conflict resolution found in multi-master setups. For global reads, we can replicate data asynchronously to secondary regions. However, to strictly enforce strong consistency, we must implement a read-your-writes guarantee where clients are directed to the primary or a synchronous replica until the write propagates. If the primary region fails, we trigger a leader election protocol like Raft to promote a follower to primary, ensuring no data is lost. While this introduces higher write latency for users far from the primary, it guarantees that the leaderboard never displays stale scores, which is critical for competitive gaming integrity. We can mitigate latency by caching read-heavy endpoints at the edge but invalidating them immediately upon any write event to prevent serving outdated ranks.