Fan-out on write vs fan-out on read in HLD - Design Approaches Compared

Design: Fan-out Messaging System

Design the message delivery mechanism focusing on fan-out strategies. Out of scope: message content creation, user authentication, and UI design.

Functional Requirements

FR1: Deliver messages from one sender to many receivers efficiently

FR2: Support real-time or near real-time message delivery

FR3: Handle up to 1 million users with 100,000 concurrent active users

FR4: Ensure message delivery latency under 200ms for 99th percentile

FR5: Allow users to read their message feeds on demand

Non-Functional Requirements

NFR1: System availability of 99.9% uptime

NFR2: Support horizontal scaling as user base grows

NFR3: Minimize storage and compute costs

NFR4: Handle spikes in message volume gracefully

Think Before You Design

Questions to Ask

❓ Question 1

❓ Question 2

❓ Question 3

❓ Question 4

❓ Question 5

Key Components

Message queue or broker

Datastore for messages and user feeds

Cache layer for fast reads

Worker services for processing fan-out

API servers for client interaction

Design Patterns

Fan-out on write (push updates to receivers on message arrival)

Fan-out on read (store messages once, fan-out when receivers read)

Caching strategies

Batch processing for fan-out

Load balancing and rate limiting

Reference Architecture

Client
  |
  v
API Servers
  |
  v
Message Queue / Broker
  |
  +-------------------+
  |                   |
Fan-out Workers    Datastore (Messages, User Feeds)
  |                   |
  +-------------------+
          |
          v
       Cache
          |
          v
        Clients (Receivers)

Components

API Servers

REST/HTTP servers

Receive messages from senders and serve message feeds to receivers

Message Queue / Broker

Kafka or RabbitMQ

Buffer incoming messages and decouple producers from consumers

Fan-out Workers

Microservices

Process messages and distribute them to receivers depending on fan-out strategy

Datastore

NoSQL database (e.g., Cassandra, DynamoDB)

Store messages and user feeds for persistence

Cache

Redis or Memcached

Serve frequently accessed message feeds with low latency

Request Flow

1. Sender sends message to API Server

2. API Server publishes message to Message Queue

3. Fan-out Workers consume message from queue

4. For fan-out on write: Workers push message copies to each receiver's feed in Datastore

5. For fan-out on read: Workers store message once; feeds are generated on demand

6. Receivers request their feeds from API Server

7. API Server checks Cache; if miss, reads from Datastore

8. API Server returns messages to receivers

Database Schema

Entities: - User: user_id (PK), name - Message: message_id (PK), sender_id (FK User), content, timestamp - UserFeed (for fan-out on write): user_id (FK User), message_id (FK Message), read_status Relationships: - One User can send many Messages (1:N) - UserFeed stores many messages per user (1:N) For fan-out on read, UserFeed may be virtual or generated on demand by querying Messages from followed users.

Scaling Discussion

Bottlenecks

Fan-out on write: High write amplification when sender has many receivers

Fan-out on read: High read latency and compute when many receivers read simultaneously

Message Queue throughput limits

Datastore write/read capacity

Cache size and eviction policies

Solutions

Use batch writes and asynchronous processing to reduce write load in fan-out on write

Implement caching and pre-computed feeds to reduce read latency in fan-out on read

Partition message queue topics by sender or user groups to increase throughput

Use scalable NoSQL databases with horizontal partitioning

Employ multi-level caching and TTL strategies to optimize cache usage

Interview Tips

Time: Spend 10 minutes clarifying requirements and constraints, 15 minutes designing architecture and data flow, 10 minutes discussing scaling and trade-offs, 10 minutes answering questions.

Explain difference between fan-out on write and fan-out on read with examples

Discuss trade-offs in latency, storage, and compute costs

Describe components and their roles clearly

Address scaling challenges and mitigation strategies

Show understanding of real-world constraints and user behavior