AWS re:Invent 2025 - Deep dive into Amazon Aurora DSQL and its architecture (DAT439)

Deep Dive into Amazon Aurora DSQL Architecture

Overview of Amazon Aurora DSQL

• Amazon Aurora DSQL is a distributed implementation of a relational database, built on top of PostgreSQL
• It aims to bridge the worlds of relational databases and distributed databases, offering a fully serverless, scalable, and highly available database service

Key Components of DSQL Architecture

1. Query Processor (QP): Responsible for parsing, planning, and executing SQL queries
   • Receives SQL statements from applications and interprets them
   • Performs all DML operations ephemerally in memory until commit
2. Journal: Distributed transaction log that provides durability guarantees
   • Writes transactions to at least two Availability Zones in a single region
   • Replicates transactions to at least one other AWS region in a multi-region setup
3. Storage: Custom-built storage service that learns about committed transactions from the Journal and updates its local view

Consistency and Concurrency Control

• DSQL provides strong consistency through a time-based synchronization protocol
  • Query Processors use the EC2 Time Sync service and AWS Clockbound to get linearizable timestamps
  • Storage implements multi-version concurrency control to return results for the requested timestamp
• DSQL is active-active, allowing any Query Processor to update data
  • The Adjudicator service implements optimistic concurrency control to detect and resolve conflicts

Automatic Scaling and Availability

• DSQL automatically scales reads by creating clones of Storage based on recent snapshots
• It can also scale out the Journal layer through transparent sharding
• DSQL is designed to be highly available, with the ability to survive the loss of an Availability Zone or even an entire AWS region

Connection Management in DSQL

• DSQL provides a shared endpoint that any cluster can use, leveraging TLS Server Name Indication (SNI) to identify the target cluster
• Connections are handled by a Relay service that authenticates and authorizes the client, then routes the connection to an available Query Processor
• Each Query Processor runs in its own hardware-isolated microVM, providing strong security and isolation between connections

Connection Pooling and Placement

• DSQL maintains a warm pool of pre-initialized Query Processors to enable fast connection establishment
• Connections are assigned to a per-cluster connection pool, allowing DSQL to automatically scale out capacity as needed
• The Placement service continuously monitors the fleet and intelligently assigns new connections to healthy Query Processors

Pricing and Cost Optimization

• DSQL uses an activity-based pricing model, charging for compute (QP time), reads, and writes
• There is no need to provision capacity or manage connection pools - DSQL scales automatically based on usage
• Customers only pay for the resources they consume, with no idle costs when the application is not actively using the database

Key Takeaways

• DSQL provides a highly secure, scalable, and available database service with a focus on simplicity for developers
• The architecture leverages hardware-level isolation, distributed transaction management, and intelligent connection routing to deliver a seamless experience
• DSQL's activity-based pricing model aligns costs with actual usage, making it easy to scale up or down without over-provisioning