Everything you don’t need to know about Amazon Aurora DSQL: Part 2 – Shallow view

This article explains Amazon Aurora DSQL's distributed architecture and how design decisions impact application compatibility and performance.

• Aurora DSQL is a serverless, PostgreSQL-compatible distributed SQL database for OLTP workloads
• Active-active deployment with multi-Region support provides 99.99% single-Region and 99.999% multi-Region availability
• Architecture uses loosely coupled components: Query Processors, Adjudicators, Journals, Crossbar, and Storage nodes
• Implements optimistic concurrency control with snapshot isolation, eliminating locks and cross-AZ read latency
• Write transactions incur two round-trip times of cross-Region latency only at commit time
• Each connection gets dedicated compute resource, preventing slow clients from affecting others
• Components scale independently based on workload demands without maintenance windows
• Requires avoiding hot key writes and hot key ranges for optimal performance
• Does not support explicit database locks or more than two read-write Regions
• PostgreSQL feature subset supported; gap narrows over time but some features differ due to distributed nature

Understanding Aurora DSQL's architecture helps developers design applications that leverage its distributed capabilities while avoiding common pitfalls like hot keys and lock-dependent logic.