Global Stock Price Viewer

What is being asked

Design a live quote service. Exchanges emit a flood of price updates; each user watches a handful of symbols and should see changes almost as they happen.

Requirements

Functional

• Subscribe to a list of symbols and receive live prices.
• Show the latest price immediately on opening the page.
• Historical charts.

Non-functional

• Hundreds of thousands of updates per second at the source.
• Millions of concurrent connections.
• Low latency worldwide; a missed intermediate tick is acceptable, a stale final price is not.

Design

1. Feed handlers convert each exchange's format into a common tick and publish to a stream partitioned by symbol.
2. The latest price per symbol is kept in a cache, so a new viewer gets a value at once.
3. A fan-out tier conflates: it sends at most a few updates per symbol per second, always the most recent, since nobody can read faster.
4. Edge servers near users hold the WebSocket connections and keep a map from symbol to subscribed connections. They subscribe upstream only to symbols someone there is watching.

Key decisions and trade-offs

• Conflation cuts volume by orders of magnitude and protects slow clients. Professional traders who need every tick use a different, costlier product.
• Push at the edge. Terminating connections regionally keeps the last hop short; the backbone carries one copy per symbol per region, not one per user.
• Slow consumers. If a client cannot keep up, drop intermediate updates for it rather than queueing without bound.
• Snapshot then stream. Send the cached price first, then live updates with sequence numbers so the client can detect a gap and ask for a fresh snapshot.

Follow-up questions

• How do you handle a symbol that half of all users watch during a crash?
• How would you reconnect a million clients after an edge server restarts?
• How are historical candles built from the tick stream?
• What changes for delayed quotes for unpaid users?