Replica Auto Scaling
Before scaling: After scaling:
┌────────┐ Writer Endpoint ┌────────┐ Writer Endpoint
│ Writer │◀──────────────── │ Writer │◀────────────────
└────────┘ └────────┘
┌────────┐ Reader Endpoint ┌────────┐
│Reader 1│◀──────────────── │Reader 1│
├────────┤ (high CPU!) ├────────┤ Reader Endpoint
│Reader 2│ │Reader 2│◀────────────────
└────────┘ ├────────┤ (auto scaled)
│Reader 3│
├────────┤
│Reader 4│
└────────┘
- Set up auto scaling policies based on CPU utilization or connection count
- New replicas are automatically added to the reader endpoint
- Scale from 1 to 15 read replicas
Custom Endpoints
┌────────┐ Writer Endpoint
│ Writer │◀────────────────
└────────┘
┌──────────────┐
│ db.r3.large │ ── Regular reader endpoint (not used after custom)
│ db.r3.large │
└──────────────┘
┌──────────────────┐
│ db.r5.2xlarge │ ── Custom Endpoint (analytical queries)
│ db.r5.2xlarge │
└──────────────────┘
| Feature | Detail |
|---|---|
| Purpose | Route specific workloads to specific replica subsets |
| Use case | Run analytical queries on larger, more powerful instances |
| Reader endpoint | Not typically used after defining custom endpoints |
| Best practice | Define multiple custom endpoints for different workload types |
Aurora Serverless
| Feature | Detail |
|---|---|
| Scaling | Automated instantiation + auto-scaling based on actual usage |
| Best for | Infrequent, intermittent, or unpredictable workloads |
| Capacity planning | Not needed |
| Billing | Pay per second of Aurora instance usage |
| Architecture | Client → Proxy Fleet (managed by Aurora) → Aurora instances |
┌────────┐ ┌─────────────┐ ┌──────────────────┐
│ Client │────▶│ Proxy Fleet │────▶│ Aurora Instances │
│ │ │ (managed) │ │ (auto-created) │
└────────┘ └─────────────┘ └──────────────────┘
When you can't predict your database workload. No need to provision capacity in advance.
Aurora Serverless path
Global Aurora
| Feature | Detail |
|---|---|
| Primary region | All reads and writes |
| Secondary regions | Up to 5 read-only regions |
| Replication lag | < 1 second cross-region |
| Read replicas per secondary | Up to 16 |
| Disaster recovery RTO | < 1 minute |
┌─────────────────┐ ┌─────────────────┐
│ us-east-1 │ < 1s │ eu-west-1 │
│ (Primary) │────────▶│ (Secondary) │
│ Read + Write │ repl. │ Read Only │
│ │ │ (up to 16 │
│ │ │ replicas) │
└─────────────────┘ └─────────────────┘
If you see "less than 1 second to replicate data across regions" → think Global Aurora.
Global Aurora
Aurora Machine Learning
| Service | Use Case |
|---|---|
| SageMaker | Any ML model (fraud detection, product recommendations) |
| Amazon Comprehend | Sentiment analysis |
┌─────────┐ SQL query ┌────────┐ data ┌──────────────┐
│ App │─────────────▶│ Aurora │────────▶│ ML Service │
│ │ │ │◀────────│ (SageMaker/ │
│ │◀─────────────│ │ predict │ Comprehend) │
│ │ results │ │ └──────────────┘
└─────────┘ └────────┘
- No ML experience needed — use via SQL interface
- Use cases: fraud detection, ads targeting, sentiment analysis, product recommendations
Questions & Answers
Q1: What is Aurora Replica Auto Scaling?
Answer:
A feature that automatically adds read replicas when CPU utilization or connection count exceeds a threshold. New replicas are automatically added to the reader endpoint.
Q2: What are Aurora Custom Endpoints?
Answer:
Endpoints that route traffic to a specific subset of Aurora replicas. Used when you have different instance sizes and want to direct analytical queries to more powerful instances.
Q3: What happens to the reader endpoint after defining custom endpoints?
Answer:
The reader endpoint is generally not used anymore. You would use multiple custom endpoints for different workload types instead.
Q4: What is Aurora Serverless?
Answer:
An Aurora mode with automated instantiation and auto-scaling based on actual usage. Clients connect through a proxy fleet managed by Aurora. Pay per second. No capacity planning needed.
Q5: When should you use Aurora Serverless?
Answer:
For infrequent, intermittent, or unpredictable workloads where you can't predict database usage patterns.
Q6: How many secondary regions can Global Aurora have?
Answer:
Up to 5 secondary read-only regions, each with up to 16 read replicas.
Q7: What is the cross-region replication lag for Global Aurora?
Answer:
Less than 1 second. This is a key exam hint — if you see this phrase, think Global Aurora.
Q8: What is the RTO for Global Aurora disaster recovery?
Answer:
Less than 1 minute. You promote a secondary region to become the new primary.
Q9: What ML services does Aurora integrate with?
Answer:
- SageMaker — any ML model (fraud detection, product recommendations)
- Amazon Comprehend — sentiment analysis All accessible via SQL queries.
Q10: Do you need ML experience to use Aurora Machine Learning?
Answer:
No. Aurora handles the integration. You just run SQL queries, and Aurora sends data to the ML service and returns predictions as query results.
Check yourself
Quiz
Flashcards
Click the card (or focus it), then use ← → to move and Space to flip.