Aurora

• Proprietary database technology (AWS-owned)
• Compatible with Postgres and MySQL drivers
• Cloud optimized (5x performance over MySQL)
• Shared Storage Volume that grows automatically up to 128TB (at 10GB steps)
• 20% more expensive

Write modes

• One writer + Multiple readers
• One writer + Multiple readers (parallel query)
• Multiple writers
• Serverless: pay per usage only!

Replication & Disaster Recovery

• Master (write) + Up to 15 Read replicas with faster sync

• Automatic failover. Failover is immediate if multi-master is enabled

• Cross Region Replication: useful for disaster recovery

• Aurora Global Database
• 1 Primary Region (RW)
• 5 Secondary Regions (RO) - with up to 15 replicas in each
• Less than 1 second lag

• RTO (recovery time objective) of 1 min. For disaster recovery

• Priority Tier

• Each Read Replica is associated with a priority tier (0-15)
• In the event of a failover, the Read Replica that has the highest priority will be promoted
• If two or more Replicas share the same priority, the replica with largest size is used

Endpoints & Autoscaling

• There is a writer endpoint and reader endpoint (load balanced)
• The read replicas can be auto-scaled and accessed through a proxy fleet

• Some read instances can also be configured with a custom endpoint (e.g., for special analytical queries)
• It's a good practice to define many custom endpoints for different workload types (query only a subset of the instances)

Machine Learning

• ML-based predictions using SQL language
• Services
• Amazon SageMaker: any model
• Amazon Comprehend: sentiment analysis
• Use cases
• Fraud detection
• Ads targeting
• Sentiment analysis
• Product recommendations