RDS Multi-AZ vs Read Replica: When to Use Which?

Key Takeaway

Multi-AZ is synchronous replication for high availability (failover), Read Replica is asynchronous replication for read scaling. For production, use Multi-AZ for availability, and add Read Replicas for read-heavy workloads. They complement each other.

Multi-AZ vs Read Replica at a Glance

Comparison	Multi-AZ	Read Replica
Primary Purpose	High availability/Failover	Read scaling
Replication	Synchronous	Asynchronous
Read Traffic	❌ Not allowed	✅ Allowed
Auto Failover	✅ Supported	❌ Manual promotion
Region	Same region only	Same/Cross region
Count Limit	1-2 standby	Up to 15
Data Lag	None (synchronous)	Yes (asynchronous)
Additional Cost	Standby instance cost	Replica instance cost

Multi-AZ Deployment

What is Multi-AZ?

Multi-AZ automatically maintains a synchronous standby replica in a different Availability Zone. On primary failure, it automatically fails over to the standby.

Multi-AZ Architecture (Instance deployment):

AZ-a                          AZ-b
┌──────────────────┐         ┌──────────────────┐
│  Primary DB      │ ──────> │  Standby DB      │
│  (Read/Write)    │ Sync    │  (Standby only)  │
└──────────────────┘ Repl.   └──────────────────┘
         ↑
    Application connection
    (Single DNS endpoint)

Multi-AZ Deployment Types

1. Multi-AZ DB Instance (1 standby)

• Primary + 1 standby
• Standby cannot serve read traffic
• Failover time: 1-2 minutes

2. Multi-AZ DB Cluster (2 readers) - New

• Writer + 2 readers (distributed across 3 AZs)
• Readers can serve read traffic
• Failover time: ~35 seconds
• Supported: MySQL, PostgreSQL

Auto Failover Triggers

Auto failover to standby occurs in these situations:

• Primary instance failure
• Availability Zone failure
• Instance type change
• DB engine patching
• Manual failover (option during reboot)

Failover Process

1. Primary instance failure detected
   ↓
2. DNS record updated (same endpoint maintained)
   ↓
3. Standby promoted to new primary
   ↓
4. Application auto-reconnects (no DNS change needed)

Duration: RDS 1-2 min, Aurora ~30 sec

Multi-AZ Use Cases

• Production workloads: Minimize downtime essential
• Compliance: Data durability requirements
• Automated maintenance: Minimize downtime during patching
• AZ failure protection: Protection from single AZ failure

Read Replica

What is Read Replica?

Read Replica asynchronously replicates data to distribute read traffic. It reduces load on the primary DB and improves performance.

Read Replica Architecture:

┌──────────────────┐
│  Primary DB      │───────────────────────────────┐
│  (Read/Write)    │  Asynchronous replication     │
└──────────────────┘                               │
         ↑                    ↓                    ↓
    Write traffic       ┌─────────────┐    ┌─────────────┐
                        │ Read Replica │    │ Read Replica │
                        │ (Read-only)  │    │ (Read-only)  │
                        └─────────────┘    └─────────────┘
                              ↑                    ↑
                          Read traffic         Read traffic

Read Replica Key Features

Feature	Description
Max Count	RDS: 5, Aurora: 15
Region	Same region / Cross-region
Replication Lag	Milliseconds to seconds (async)
Separate Endpoint	Each replica has unique DNS
Can Be Promoted	Can become independent DB

Read Replica Usage Patterns

1. Read Scaling

Primary: INSERT, UPDATE, DELETE (writes)
Replica: SELECT (reads) - reports, analytics, search

2. Cross-Region Replication

Seoul Region (Primary) → Virginia Region (Replica)
- Reduce latency for global users
- Disaster recovery preparation

3. Promote to Independent DB

Test environment setup:
Primary → Create Read Replica → Promote → Independent DB (for testing)

Read Replica Promotion

You can promote a Read Replica to an independent DB instance:

• No longer replicates after promotion
• Becomes independent DB with read/write capability
• Use for DR via manual promotion

Promotion Scenario:

1. Primary DB fails
2. Manually promote Read Replica to independent DB
3. Update application connection string (new endpoint)
4. Resume service with new DB

Multi-AZ vs Read Replica Selection Criteria

Selection by Requirement

Requirement	Choice
Need auto failover	Multi-AZ
Need read scaling	Read Replica
Minimize downtime	Multi-AZ
Separate report/analytics queries	Read Replica
Replicate to different region	Read Replica
Minimize data loss	Multi-AZ (synchronous)

Combined Usage (Recommended)

Production environments commonly use both:

Recommended Architecture:

              Multi-AZ (High Availability)
         ┌───────────────────────┐
         │                       │
    AZ-a │        AZ-b           │
┌─────────────┐  ┌─────────────┐ │
│  Primary    │→ │  Standby    │ │
│  (R/W)      │  │  (failover) │ │
└─────────────┘  └─────────────┘ │
         │                       │
         └───────────────────────┘
         │
         │ Async replication
         ↓
┌─────────────┐  ┌─────────────┐
│ Read Replica│  │ Read Replica│
│ (read-only) │  │ (read-only) │
└─────────────┘  └─────────────┘
    AZ-a              AZ-c

Read Replica (Read Scaling)

Scenario-Based Solutions

Scenario 1: Production DB High Availability

Requirement: 24/7 operation, minimize downtime
Solution: Multi-AZ deployment
Reason: Auto failover recovers within 1-2 minutes

Scenario 2: Read Load Distribution

Requirement: Heavy SELECT queries, report generation
Solution: Add Read Replicas
Reason: Distribute read traffic to replicas

Scenario 3: Global User Support

Requirement: Reduce latency for US, Europe users
Solution: Cross-region Read Replicas
Reason: Serve reads from replicas in each region

Scenario 4: Disaster Recovery (DR)

Requirement: Recovery plan for region failure
Solution: Cross-region Read Replica + manual promotion
Reason: Promote replica in different region if source region fails

Scenario 5: Separate Analytics Queries

Requirement: Run analytics queries without affecting OLTP
Solution: Run analytics on Read Replica
Reason: No impact on primary DB performance

Aurora Read Replicas

Aurora handles replication differently from standard RDS:

Aurora vs RDS Read Replica

Item	RDS Read Replica	Aurora Read Replica
Replication Method	Log-based async	Shared storage
Replication Lag	Seconds to minutes	Milliseconds
Max Count	5	15
Auto Failover	❌	✅ (Aurora cluster)
Endpoint	Individual endpoints	Reader endpoint (auto load balancing)

Aurora Cluster Endpoints

Aurora Endpoints:

1. Cluster Endpoint (Writer)
   → For write operations, connects to current writer

2. Reader Endpoint
   → For read operations, auto load balances across Read Replicas

3. Instance Endpoint
   → Direct connection to specific instance

SAA-C03 Exam Focus Points

1. ✅ Multi-AZ Purpose: High availability, auto failover (NOT read scaling)
2. ✅ Read Replica Purpose: Read scaling, manual promotion
3. ✅ Replication Method: Multi-AZ = Synchronous, Read Replica = Asynchronous
4. ✅ Failover: Multi-AZ = Auto (DNS maintained), Read Replica = Manual (endpoint changes)
5. ✅ Cross-Region: Only Read Replica supports different regions
6. ✅ Aurora: Read Replicas also support auto failover

Frequently Asked Questions

Q: Can I send read traffic to Multi-AZ standby?

No. Multi-AZ DB Instance standby is for failover only and cannot be directly connected. Use Read Replicas for read distribution. However, Multi-AZ DB Cluster reader instances can serve read traffic.

Q: Can I configure Multi-AZ for a Read Replica?

Yes. You can configure a Read Replica itself as Multi-AZ to increase replica availability. Useful for cross-region Read Replica DR scenarios.

Q: Is there data loss during Multi-AZ failover?

No. Multi-AZ uses synchronous replication so standby always has the latest data. However, in-progress transactions may be rolled back during failover.

Q: How much is Read Replica replication lag?

Typically milliseconds to a few seconds. Varies by network bandwidth, write load, and replica instance performance. Monitor with CloudWatch's ReplicaLag metric.

Q: What's the cost for cross-region Read Replica?

Data transfer costs apply. Outbound data transfer charges apply for cross-region replication. Same-region replication is free.

Q: Does promoting a Read Replica stop replication?

Yes. The promoted instance becomes an independent DB with no replication relationship to the original. Create a new Read Replica to restore replication.

---

Related Posts

• Aurora vs RDS: Which Should You Choose?
• DynamoDB Global Tables
• DR Strategies (Backup/Restore, Pilot Light, Warm Standby, Active-Active)

---

References

• AWS RDS Multi-AZ Documentation
• AWS RDS Read Replica Documentation
• RDS Multi-AZ Features
• RDS Read Replica Features