Logo
Published on

Primary-Replica vs Peer-to-Peer Replication - System Design Guide

  • Primary-Replica ? Centralized writes (one leader)
  • Peer-to-Peer ? Distributed writes (no leader)

?? Script

�Primary-Replica has one writer for consistency. Peer-to-Peer allows all nodes to write for availability and scale.�


?? 2. Primary�Replica (Leader-Based)

? What it does

  • One Primary (Leader) handles all writes
  • Replicas follow and sync data
  • Reads ? replicas, Writes ? primary

?? Key Insight

?? Prioritizes consistency + simplicity


??? Architecture Signals (Use this when)

  • Need strong consistency
  • System is read-heavy (80�90% reads)
  • Want simple conflict-free design
  • Example: databases, banking, user systems

? Problems

  • Single point of failure (primary)
  • Replication lag ? stale reads
  • Write scaling is limited

?? FAANG Q&A

Q1: How to fix primary failure? ?? Leader election (e.g., failover using ZooKeeper/Raft)

Q2: Sync vs Async replication?

  • Sync ? strong consistency, slower
  • Async ? faster, eventual consistency

Q3: Bottleneck? ?? Writes (only one node)


?? Script

�I use primary-replica when I need strong consistency and predictable behavior. Writes go to a single leader, and replicas scale reads.�


?? 3. Peer-to-Peer (Leaderless / Multi-Leader)

? What it does

  • All nodes can read + write
  • Data replicated across peers
  • No central authority

?? Key Insight

?? Prioritizes availability + scalability


??? Architecture Signals (Use this when)

  • Need high availability (no downtime)
  • Global systems (multi-region writes)
  • Massive scale (billions of users)
  • Can tolerate eventual consistency

? Problems

  • Conflict resolution required
  • Complex (versioning, vector clocks)
  • Hard to debug

?? FAANG Q&A

Q1: What happens on write conflict? ?? Use:

  • Last Write Wins (LWW)
  • Vector clocks
  • CRDTs

Q2: Why used in Cassandra/Dynamo? ?? No single failure point + global scale.

Q3: Tradeoff? ?? Data may be temporarily inconsistent.


?? Script

�I use peer-to-peer when I need high availability and global scalability. Multiple nodes can accept writes, and conflicts are resolved asynchronously.�


?? 4. Key Differences (Interview Table)

Aspect Primary-Replica Peer-to-Peer
Writes One node All nodes
Reads Replicas Any node
Architecture Centralized Decentralized
Consistency Strong Eventual
Availability Lower High
Complexity Simple Complex
Conflict handling ? Not needed ? Required

?? Script

�Primary-replica simplifies consistency by centralizing writes, while peer-to-peer distributes writes for scalability but introduces conflicts.�


?? 5. Real Architecture Decision (FAANG Level)

? Hybrid is common

User DB ? Primary-Replica (strong consistency)
Analytics / Logs ? Peer-to-Peer (high scale)

?? Why?

  • Critical data ? Primary-Replica
  • High-scale data ? Peer-to-Peer

?? FAANG Q&A

Q: Why not use peer-to-peer everywhere? ?? Conflict resolution overhead + complexity.

Q: Why not only primary-replica? ?? Doesn�t scale writes globally.


?? Script

�In real systems, I combine both. Critical data uses primary-replica for consistency, while large-scale distributed systems use peer-to-peer for availability.�


?? 6. Strong Signals vs Weak Signals

?? Choose Primary-Replica (Strong Signals)

  • �Banking / payments�
  • �Strong consistency required�
  • �Read-heavy workload�
  • �Simple system design�

?? Choose Peer-to-Peer (Strong Signals)

  • �Global users�
  • �High availability required�
  • �Multi-region writes�
  • �Eventual consistency acceptable�

?? Weak Signals

  • �Distributed system ? use peer-to-peer� ?
  • �Scaling ? always peer-to-peer� ?

?? 7. CAP Theorem Mapping (IMPORTANT)

  • Primary-Replica ? CP (Consistency + Partition tolerance)
  • Peer-to-Peer ? AP (Availability + Partition tolerance)

?? FAANG Q&A

Q: Why peer-to-peer is AP? ?? Keeps system running even during network partitions.

Q: Example issue? ?? Same user updates profile in two regions ? conflict.


?? Script

�Primary-replica favors consistency, while peer-to-peer favors availability under partition.�


?? 8. Extra FAANG Insights (Added)

?? 1. Leader Election (Primary systems)

  • Raft / Paxos

?? 2. Quorum Reads/Writes (Peer systems)

  • R + W > N ensures consistency

?? 3. Eventual Consistency Models

  • Read repair
  • Anti-entropy

?? 4. Multi-Leader (Hybrid)

  • Between both models (used in global DBs)

?? Final Ultra-Short Summary

?? Golden Line

�Primary-Replica gives consistency through a single writer. Peer-to-Peer gives scalability by allowing everyone to write.�


If you want, I can next connect this with SQL vs NoSQL + API Gateway decisions into one complete system design cheat sheet (FAANG-ready).