Fan-out on Write vs Fan-out on Read

---

🧠 Core Idea

• Fan-out on Write → Pay cost during write
• Fan-out on Read → Pay cost during read

👉 Trade-off: Write Amplification vs Read Amplification

Fan-out on Write (Push Model)

How it works

• When a user creates a post:

  • System pushes it to all followers’ feeds

• Feed is precomputed and stored per user

Pros

• Fast read latency (feed already ready)
• Simple read queries

Cons

• High write cost (1 post → millions of writes)
• Hot shard / partition issues
• Storage duplication (same post stored many times)
• Partial failures → inconsistent feeds

Key Challenges

• Write amplification
  • overload
• Fault tolerance (retry, idempotency)
  • in case multiple write fails
• Ordering consistency
• Storage explosion

Best Use Case

• Read-heavy systems
• Users with moderate follower count
• Low latency feed requirement

---

Fan-out on Read (Pull Model)

How it works

• Post stored once
• On feed request:
  • Fetch posts from all followed users
  • Merge + sort + rank

Pros

• Cheap writes
• No duplication
• Scales well for high-follower users

Cons

• Slow reads (aggregation required)
• Complex merging logic
• High CPU/memory usage
• Cache invalidation challenges

Key Challenges

• Read amplification
• Merge & ranking complexity
• Pagination across multiple sources
• Cold start latency

Best Use Case

• Write-heavy systems
• Users following fewer accounts
• Need for fresh data

Hybrid Approach (Industry Standard)

Strategy

• Normal users → Fan-out on Write
• High-follower users → Fan-out on Read

Flow

• On post:

  • Normal → push to feeds
  • Celebrity → store once

• On read:

  • Fetch precomputed feed
  • Merge celebrity posts dynamically

---

Advanced Concepts

Async Fan-out

Post → Queue → Workers → Feed updates

Consistency Model

• Eventual consistency is acceptable
• Slight delays in feed updates are normal

Backpressure Handling

• Drop or delay low-priority updates
• Use queues to smooth spikes

---

⚖️ Quick Comparison

Factor	Fan-out on Write	Fan-out on Read
Write Cost	High	Low
Read Cost	Low	High
Latency (Read)	Fast	Slower
Scalability	Hard (writes)	Hard (reads)
Storage	High	Low