Everything you don’t need to know about Amazon Aurora DSQL: Part 3 – Transaction processing

This article explains transaction processing in Amazon Aurora DSQL, covering read-only and read-write transaction workflows in its active-active distributed architecture.

• Read-only transactions use snapshot isolation based on transaction start time (τ-start)
• Commit time equals start time for read-only transactions, guaranteeing consistency
• Read-write transactions spool writes locally and check for conflicts at commit
• Early abort mechanism detects newer data versions during execution
• Adjudicators verify write conflicts and assign commit timestamps (τ-commit)
• Hybrid two-phase and warp-style commit algorithm minimizes round-trip times
• Each database key owned by exactly one adjudicator globally
• SELECT FOR UPDATE manages write skew in read-write transactions

Aurora DSQL uses sophisticated distributed transaction mechanisms combining snapshot isolation, conflict detection, and coordinated adjudication to maintain consistency across regions while minimizing communication overhead.