Design Realtime Comment for Newsfeed

July 2, 2023 7:00 PM PDT

---

This document outlines the system design for a realtime commenting feature for a newsfeed application. The design focuses on handling a high volume of comments and ensuring low latency for users. Key considerations include scalability, data consistency, and the choice of database technologies.

Requirements

• Functionality:

  • Realtime commenting similar to Facebook.
  • Comments should appear below the post.
  • Support for one level of comments, with optional privileged comments.

• User Metrics:

  • Monthly Active Users (MAU): 1 billion
  • Posts per minute: 100 million
  • Comments per minute: 600,000

API Design

• Publish Comment: Publish_comment(uid, ...)
• View Comments: View_comment(post_id, ...)
• View Comments by User: View_comment_by_user(uid, self_uid, auth_token)

Non-Functional Requirements

1. High availability (99.999%)
2. Resilience in comment publishing
3. Scalability
4. Eventual consistency (no need for linearizability or transactions)
5. Real-time delivery (within 1 second)

Performance Estimates

• Queries Per Second (QPS): 600,000 comments per minute translates to approximately 10,000 QPS.
• Daily Comments: 1.08 billion comments per day.
• Data Size: Assuming 1 KB per comment, approximately 10 TB of data per day (after 50% compression, around 5 TB).

High-Level Design

• Database Options:

  • Realtime databases like leader-follower or multi-level databases.
  • Quorum-based systems such as Dynamo or Cassandra.

• Database Type:

  • NoSQL is preferred (document or column store) for better locality and performance.
  • Comments can be stored in a post table with JSON format for flexibility.

Conflict Resolution

• Use a version vector system to handle concurrent comments.
• Old versions can be deleted once combined.

Real-Time Delivery Mechanism

• Users maintain a socket connection with the server.
• Comments can be streamed using Kafka or similar technologies.
• Comments are published to a stream and read from an in-memory datastore for fast access.

Throughput Calculation

• For an app with 1 billion users, reading every 3 seconds:
  • Estimated read throughput: 4.5 billion reads per second.
  • Scale by partitioning across 1,000 to 2,000 servers.
  • Each server would handle approximately 4.5 million reads per second.

Conclusion

The design focuses on achieving high throughput and low latency for a large-scale commenting system. By leveraging NoSQL databases and real-time streaming technologies, the system can efficiently handle the expected load while maintaining a responsive user experience.