Design a Stock Ticker/Price Feed Service

To design a stock ticker service for millions of users, I would prioritize low latency and efficient bandwidth usage. First, we must define the SLA: updates should reach clients within 50 milliseconds of exchange execution. For transport, WebSockets are essential because they maintain a persistent connection, eliminating the overhead of repeated HTTP handshakes and allowing true bidirectional communication. This aligns with Stripe's focus on developer experience and reliability. On the backend, incoming trade data from exchanges enters a high-throughput ingestion service, likely backed by Apache Kafka to handle bursts. We then need a fan-out mechanism. A naive approach of pushing to each socket individually is inefficient. Instead, I propose using a Pub/Sub layer like Redis Streams or NATS JetStream. The service groups active subscriptions by ticker symbol. When a new price arrives, it publishes to the specific ticker channel, and the broker handles the multiplexed delivery to all subscribed sockets simultaneously. To handle scale, we shard the WebSocket servers horizontally. A consistent hashing algorithm routes client connections to specific nodes, while the Pub/Sub layer remains global. Crucially, we implement backpressure; if a client lags, we drop older messages but keep the stream alive, perhaps using a circular buffer. Finally, we add a resilience layer with exponential backoff for reconnections to prevent thundering herd problems during outages.