Design CoinSwitch's Real-Time Price Aggregation Service Across 10 Exchanges

Problem

Design a price aggregation service that consumes live price feeds from 10 crypto exchanges and serves a unified best-price view to CoinSwitch users.

Requirements

Functional:

• Aggregate BTC/INR, ETH/INR, etc. prices from 10 exchange WebSocket feeds
• Compute best bid (highest buy price) and best ask (lowest sell price) across all exchanges
• Serve aggregated price to 5M users with < 500ms latency
• Price history for charting (1m, 5m, 1h candles)

Non-functional:

• 10 exchanges × 100 price updates/sec = 1000 inbound events/sec
• Fanout to 5M user sessions
• 99.9% uptime — exchange feed drops should not cause outage

Discussion points

1. Exchange feed ingestion: WebSocket reconnect with exponential backoff
2. Aggregation logic: per-pair sorted map of (exchange → price)
3. Fanout: Redis Pub/Sub vs Kafka → WebSocket gateways
4. Stale price handling: if exchange feed drops, mark price as stale after TTL
5. OHLCV candle computation: streaming aggregation with time windows