Design a Real-Time Commenting System

To design a real-time commenting system for Meta-scale traffic, I would prioritize read performance since users consume far more content than they create. First, I'd define the API contract to support pagination and sorting by recency. For the write path, incoming requests hit an API Gateway that validates input and pushes the comment payload into a Kafka topic partitioned by Article_ID. This decouples ingestion from storage, allowing us to absorb traffic spikes. A consumer group processes these messages, enriches them with user metadata, and persists them to a NoSQL database like Cassandra or DynamoDB, which offers horizontal scalability. For reads, we implement a strict separation. When a user loads an article, the system checks a Redis cache first. If the comments are present, we return them immediately; otherwise, we fetch from the database and populate the cache. To achieve real-time updates, we utilize a Pub/Sub pattern where the Kafka consumer broadcasts a notification to a WebSocket service whenever a new comment arrives. Connected clients subscribe to specific Article_ID channels and receive instant push notifications. We must also handle edge cases like spam filtering, which runs asynchronously before persistence. Finally, we ensure global availability by deploying this across multiple regions with active-active replication for the cache layer, ensuring sub-second latency for users worldwide regardless of their physical location.