Practice/Meta/Design Typing Indicator

Design Typing Indicator

Product DesignOptional

Problem Statement

Design a system that displays real-time viewer counts for live video streams, similar to what you see on YouTube Live, Twitch, or Instagram Live. When users join or leave a live stream, the viewer count should update for all participants within seconds. The system needs to handle popular streams with millions of concurrent viewers while maintaining accurate counts and efficient resource usage.

Your solution should account for the challenges of tracking ephemeral connections, dealing with network instability, preventing count manipulation, and efficiently broadcasting updates to massive audiences without overwhelming backend services or client devices.

Key Requirements

Functional

• Real-time count updates -- display current viewer count that updates within 2-3 seconds as users join/leave
• Join/leave detection -- accurately track when viewers connect to or disconnect from streams
• Historical metrics -- store peak viewer counts and view duration statistics for analytics
• Multi-platform support -- handle viewers across web, mobile apps, smart TVs, and embedded players

Non-Functional

• Scalability -- support up to 10 million concurrent viewers on a single popular stream
• Reliability -- maintain 99.9% accuracy in count despite network failures and edge cases
• Latency -- propagate count changes to all viewers within 3 seconds maximum
• Consistency -- eventual consistency is acceptable; exact counts less critical than approximate real-time updates

What Interviewers Focus On

Based on real interview experiences, these are the areas interviewers probe most deeply:

1. Connection Management and State Tracking

Understanding how to efficiently track millions of ephemeral connections without overloading your system is fundamental. Interviewers want to see how you handle the difference between "watching" and "connected" states, especially when users pause streams, switch tabs, or experience network hiccups.

Hints to consider:

• Heartbeat mechanisms to detect active viewers versus stale connections
• Tradeoffs between push-based and pull-based count updates
• Using probabilistic data structures like HyperLogLog for approximate counting at scale
• How to handle reconnection storms when network issues resolve

2. Aggregation and Broadcasting Strategy

With millions of viewers, you cannot simply broadcast every join/leave event to everyone. Interviewers look for sophisticated aggregation strategies that balance accuracy with system load.

Hints to consider:

• Batching updates over time windows (e.g., aggregate counts every 2 seconds)
• Hierarchical aggregation from edge servers to regional nodes to central coordinators
• Using pub/sub systems versus direct connections for update distribution
• Deciding when to show exact counts versus approximate ranges ("10K+ viewers")

3. Edge Cases and Failure Scenarios

Real-world systems face browser crashes, mobile app backgrounding, network partitions, and malicious actors. Strong candidates proactively address these scenarios.

Hints to consider:

• Timeout strategies for detecting disconnections versus temporary network issues
• Preventing bot farms from inflating viewer counts artificially
• Handling datacenter failures without losing count accuracy
• Dealing with clock skew across distributed components

4. Data Store Selection and Schema Design

The choice of database and how you model viewer sessions significantly impacts system performance and accuracy.

Hints to consider:

• Redis sorted sets for tracking active viewers with automatic expiration
• Time-series databases for storing historical view counts and analytics
• Differentiating between anonymous viewers and authenticated users
• Sharding strategies when single streams exceed one server's capacity

Suggested Approach

Step 1: Clarify Requirements

Start by confirming the scale and expectations. Ask about the maximum concurrent viewers for the most popular streams, whether exact counts are required or approximations are acceptable, and what latency is tolerable for count updates. Clarify if you need to support anonymous viewers versus only authenticated users, and whether you need to detect unique viewers or just concurrent connections. Confirm requirements around bot detection, historical analytics storage, and whether different viewer types (mobile vs web) should be tracked separately.