Design a Real-Time Analytics Dashboard

Stripe's dashboard shows payment volume updating every second. Uber's operations center displays ride requests flowing across a map in real-time. Cloudflare shows attack traffic hitting their network as it happens. These aren't dashboards that refresh every 5 minutes, they're living, breathing visualizations of data that's seconds old.

Traditional analytics run batch jobs overnight and show you yesterday's numbers by morning. That's fine for quarterly reports. But for operational decisions, is the system healthy right now? Is the campaign performing? Is the launch going well?, you need analytics that are measured in seconds, not hours.

Building real-time analytics is fundamentally different from batch analytics. You can't scan the entire dataset for every query because the dataset is infinite, events never stop flowing. You can't wait for all the data to arrive because more data is always coming. You need to compute results incrementally, as each event arrives, and push updates to dashboards the instant numbers change.

Functional Requirements

• Event ingestion: Accept events from multiple sources (web, mobile, APIs, IoT)
• Real-time aggregation: Compute counts, sums, averages, percentiles in real-time
• Windowed computations: Aggregate over time windows (last 1 min, 5 min, 1 hour, 24 hours)
• Dimensional analysis: Slice data by any combination of dimensions (country, device, campaign)
• Live dashboards: Charts update in real-time without page refresh
• Alerting: Trigger alerts when metrics cross thresholds
• Historical queries: Query historical data for comparisons and trends

Non-Functional Requirements

• Ingestion throughput: 500,000 events per second
• Dashboard latency: Data visible on dashboard within 5 seconds of event occurrence
• Query latency: Ad-hoc queries return in under 500ms
• Data freshness: Aggregates are at most 5 seconds stale
• Scale: Support 1,000 concurrent dashboard viewers