Ref: https://learn.cantrill.io/courses/1820301/lectures/41301428

• Summary
• Contents:

Amazon Aurora - Overview

• 🔧 DB engine designed by AWS
  • Officially part of Amazon RDS…
  • 💡 …but better to consider Aurora as its own, distinct product, different from RDS
    • Features and architecture are radically different from RDS
• Architecture:
  • Two compute paradigms:
    1. Aurora Provisioned → customer deploys cluster of instances
    2. Aurora Serverless → cluster and capacity managed by AWS, customer sets ACUs
  • Storage: shared cluster volume → high performance, improved availability

Amazon Aurora - Architecture

Aurora Provisioned DB Instances Architecture - DB Instance Cluster

• 🔧 Aurora Cluster = 1 primary & 0+ (read) replicas
  • Up to 15 replicas
  • DB instances (primary & replicas) can be in same or different AZs
  • Replicas can be used for reads → 👍 read scalability
  • Replicas can replace primary in failovers → 👍 improved availability
• 💡 Like RDS Multi-AZ cluster deployments, Aurora grants both scalability & availability benefits (but better!). Other RDS modes require addition of separate RRs.
  • ❗ RDS Multi-AZ cluster deployments only have 2 RRs natively!

Aurora Storage Architecture - Cluster Volume

• ‼️ NO local storage!! (unlike RDS)
• 🔧 An Aurora cluster uses a cluster volume
  • Shared storage → 👍 faster provisioning
    • Available to all instances (primary & replicas) within a cluster
    • Replicas added/removed without provisioning/removing storage → fast & efficient
    • Very fast failovers (storage modifications not needed)
  • SSD based → high IOPS, low latency → 👍 high performance storage by default
    • Magnetic storage not an option
  • Max size = 128 TiB
  • 6 storage nodes/replicas across multiple AZs → 👍 improved availability
    • All cluster instances can access any storage node
    • Data loss risk is minimal → high resilience
    • Aurora automatically detects and repairs storage disk failures in any node
    • 💡 Less need for point-in-time or snapshot restores to recover from disk failures
• Data writes from primary synchronously replicated to all 6 storage nodes
  • ‼️ Replication happens at the storage level!
    • No extra resources consumed by primary or replicas during replication process

Aurora - Access Architecture

• Access/Connection to cluster via endpoints (DNS addresses)
  • 💡 Very similar to access architecture in RDS Multi-AZ cluster deployments
• 🔧 Main endpoints
  • Cluster endpoint
    • Always points to primary instance (reads, writes & admin operations)
  • Reader endpoint
    • If 0 replicas → points to primary
    • If 1+ replicas → points to an available instance (primary or replica)
    • Load balances reads across all of the available instances (usually replicas)
    • Endpoint updated with creation/deletion of replicas to load balance across all replicas