Katkat
Back

Design a live video comments system

MetaSoftware Engineer · Onsite
MediumSoftware EngineerOnsiteSystem Design

Last updated

This problem tests your ability to design a high-throughput, real-time messaging system capable of handling sudden spikes in concurrent viewers and maintaining low latency for chat delivery.

Problem Statement

Design a live video comments system similar to those found on platforms like Facebook Live or Twitch. The system must support millions of concurrent viewers for a single broadcast while ensuring that comments are delivered to all users with minimal delay.

Functional Requirements

  • Users can post comments to a live broadcast.
  • Users can view a live feed of incoming comments.
  • The system should handle high-write throughput (many users commenting simultaneously).

Non-Functional Requirements

  • Latency: Comment delivery should be sub-500ms.
  • Scalability: The system must support 100k+ concurrent viewers per broadcast.
  • Reliability: Comments should be persisted eventually; occasional packet loss for live updates is acceptable.

Deliverables

  1. High-level architecture diagram and component interaction.
  2. API design (WebSocket/REST endpoints).
  3. Data model for storing comments.
  4. Strategy for handling "fan-out" to a massive audience.
  5. Approaches for dealing with comment moderation in real-time.

Hints

Hint 1

Consider the trade-offs between a pull-based polling mechanism and a push-based WebSocket approach for real-time delivery.

Hint 2

Think about using a message queue or a pub-sub model to decouple comment ingestion from the delivery service.

Hint 3

How can you partition the broadcast audience to avoid a single server bottleneck during the fan-out process?

Follow-up questions

  • How would you handle a "celebrity broadcast" where the comment rate exceeds 10,000 messages per second?
  • How would you ensure that a user who joins 5 minutes into the broadcast sees the history of comments without overloading the database?
  • If you needed to implement a real-time profanity filter, where would you place this logic to minimize latency?