Design real-time stock price viewer
Overview
This question evaluates competency in designing low-latency, real-time streaming systems including market-data ingestion, ephemeral in-memory state management, pub/sub fan-out and topic design, client delivery protocols, backpressure and slow-consumer strategies, scaling and load-shedding, consistency and ordering guarantees, and operational monitoring. Commonly asked in the System Design domain to probe an engineer's ability to make production-grade trade-offs around availability, ordering, throttling, partitioning, and quotas, it requires both high-level conceptual architectural reasoning and practical application-level considerations such as protocol choices, sharding, and observability.
System Design: Real-Time Stock Price Streaming (No Historical Storage)
You are designing a real-time market data push service that streams live stock prices to client applications. The system must not persist historical price data; it may keep only ephemeral in-memory state needed to serve current subscriptions and short reconnects.
Design and describe the system end-to-end, covering:
-
Market Data Ingestion
- How you connect to one or more upstream market/exchange feeds.
- Normalization, timestamping, sequencing, and deduplication.
-
Symbol/Instrument Management
- Managing the list of tradable symbols and their attributes (e.g., tick size, status, currency).
- Handling symbol adds/deletes, halts, market hours.
-
In-Memory Caching and State
- What per-symbol state is kept (e.g., last trade, top of book, L2 depth).
- Any short in-memory buffers for snapshots/reconnects without violating the "no historical storage" constraint.
-
Publish/Subscribe and Fan-Out
- Internal pub/sub between ingestion and edge fan-out nodes (e.g., NATS, Redis Pub/Sub, Aeron, or similar) without durable storage.
- Client delivery mechanism (WebSockets, SSE), topic design, and subscription semantics.
-
APIs and Contracts
- Authentication and session model.
- REST endpoints (e.g., list symbols) and WebSocket/SSE message formats for subscribe/unsubscribe, heartbeats, and price updates.
-
Update Frequency and Conflation
- Tick-by-tick vs. throttled/coalesced modes.
- Client-selectable throttles and server-side defaults.
-
Backpressure and Slow Consumers
- Per-connection queues, conflation, dropping strategy, and disconnect thresholds.
-
Scaling and Load Shedding
- Horizontal scaling, sharding/partitioning by symbol, multi-AZ/region.
- Anycast/LB strategy and stickiness.
-
Fault Tolerance and Recovery
- Redundant ingest paths, failover, heartbeats.
- What ordering/consistency guarantees you provide and their limits.
-
Consistency and Ordering Guarantees
- Sequencing per symbol, ordering across services, snapshot vs. incremental.
-
Rate Limiting and Quotas
- Limits per API key/connection/symbol and burst behavior.
-
Monitoring, Alerting, and SLOs
- Metrics, logs, traces; synthetic canaries; key alerts and dashboards.
Assume internet-scale clients (mobile/web/servers), heterogeneous network quality, and that only current prices are needed (no historical queries). Keep the design production-grade and specific about contracts, limits, and trade-offs.
Solution
Login required