Amazon Aurora DSQL The Definitive Guide (for . .) (Wendy L. Wise)（Z-Library）

# Amazon Aurora DSQL: The Definitive Guide ## 【One-Line Pitch】 A practical, hands-on guide to building on Amazon Aurora DSQL—AWS's serverless, multi-Region distributed SQL database—covering architecture, optimistic concurrency control, PostgreSQL compatibility, and production patterns. Essential reading for backend engineers and architects evaluating or adopting DSQL for resilient, scalable applications. ## 【Book Arc】 - **Opening (~0%–9%)**: Introduces the distributed database landscape—why single-machine databases get outgrown, the consistency-vs-availability tradeoff, and how Spanner, CockroachDB, YugabyteDB, and TiDB approach distributed SQL. Sets up why DSQL's serverless model is a fundamentally different approach. - **Early (~9%–24%)**: Explains the common machinery of distributed SQL databases—data partitioning, replication with Raft/Paxos consensus, and cross-partition transactions—and compares major players (Spanner, CockroachDB, YugabyteDB, TiDB) with their tradeoffs. Establishes why primary key design and consistency choices matter. - **Early–Middle (~24%–33%)**: Positions DSQL within the landscape, highlighting its separated-component architecture (relay, compute, transaction log, storage) that enables true serverless scaling. Introduces optimistic concurrency control (OCC) as DSQL's defining difference from pessimistic locking systems. - **Middle (~33%–42%)**: Covers what DSQL is not—a drop-in PostgreSQL replacement. Details unsupported features (stored procedures, triggers, foreign keys, advisory locks, temp tables) and provides a decision framework for choosing between DSQL, DynamoDB, Aurora, RDS, and self-managed distributed databases. - **Middle (~42%–52%)**: Moves to hands-on cluster creation—prerequisites, IAM permissions, and the dramatically simplified provisioning experience (no instance classes, storage, or replicas to configure). Introduces infrastructure-as-code options (Terraform, CloudFormation, Python). ## 【Key Takeaways】 - **Distributed databases exist because single machines get outgrown** (Early): A single PostgreSQL or MySQL instance handles most workloads fine, but horizontal scaling, regional resilience, and ACID transactions at scale require distributed systems. Understanding this spectrum is the first architectural decision. - **Consistency vs. availability is the fundamental tradeoff** (Early): During network partitions, distributed databases must choose—refuse writes to keep data correct (consistency) or serve potentially conflicting data (availability). DSQL, Spanner, CockroachDB, and YugabyteDB choose consistency; DynamoDB and Cassandra lean toward availability. - **Primary key design determines distribution quality** (Early): Sequential integer keys create hot spots—all new writes hit the same partition. UUID or composite keys that spread writes across partitions are almost always better in distributed databases. - **DSQL's serverless model separates components that other databases fuse** (Early–Middle): Instead of monolithic nodes handling queries, storage, and replication, DSQL separates relay/connectivity, compute, transaction log, and storage into independently scaling multi-tenant components across three Availability Zones. - **Optimistic concurrency control is DSQL's defining architectural choice** (Early–Middle): Transactions don't acquire locks; they proceed optimistically and abort at commit time if conflicts are detected. Application code must handle retries—this is non-negotiable and a significant mental shift from PostgreSQL's locking model. - **DSQL is deliberately not a drop-in PostgreSQL replacement** (Middle): No stored procedures, triggers, foreign keys, advisory locks, or temporary tables. These are intentional architectural decisions—foreign keys would require cross-partition coordination on every insert, conflicting with DSQL's performance goals. - **Choosing the right database means knowing when DSQL is wrong** (Middle): DynamoDB wins for single-digit millisecond key-value access; Aurora for full PostgreSQL/MySQL compatibility; RDS for specific engines (Oracle, SQL Server) or maximum configuration control. DSQL fits serverless + distributed SQL + multi-Region needs. - **DSQL dramatically simplifies provisioning** (Middle): No instance classes, storage allocation, replicas, or parameter groups. A cluster is an endpoint and a few settings—but the decisions you do make carry more weight. ## 【Reading Tips】 - **Skim Chapter 1's landscape comparison** (~9%–24%) if you're already familiar with distributed databases; the decision framework table near the end is the key takeaway. - **Deep-read the OCC sections** (~33% and Chapter 6) if you're coming from PostgreSQL—the retry-handling mental model shift is the hardest adjustment for most teams. - **Pay close attention to the "What DSQL Is Not" section** (~39%–42%): knowing unsupported features upfront saves painful migration surprises later. - **Follow along with the hands-on cluster creation** (~48%–52%) if you have AWS access; the GitHub repo contains all code examples. - **Note that this is an early-release edition**—chapters 4–16 (schema design, OCC in practice, multi-Region, observability, cost optimization, testing) are listed but unavailable in this sample. ## 【Coverage Limits】 This guide covers the opening chapters (distributed database landscape, DSQL architecture, cluster creation) but does not cover the unavailable chapters on schema design, OCC patterns, multi-Region architectures, observability, cost optimization, testing strategies, or adoption paths. ##