Applications performing operations on stored data engage in two fundamental types: write operations, which record facts, and read operations, which answer questions. While a single database can manage both on modest hardware, high traffic necessitates optimization. Initially, slow pages might be fixed with indexes. As load increases, caching becomes necessary, followed by read replicas for reporting traffic. These read path optimizations, while effective, introduce complexities. For instance, a user updating a profile might still see the old value after a reload, a side effect of data copies not being synchronized with the source. This highlights how read path fixes can inadvertently impact the write path's behavior.

This article delves into read and write path operations and techniques. It explores why fast reads and correct writes often demand opposing data structures, and how precomputation and duplication underpin read optimizations. The discussion covers two distinct definitions of consistency and the bugs arising from their conflation. Strategies like indexing, denormalization, caching, read replicas, materialized views, purpose-built read stores, fan-out on write versus read, and CQRS are examined. For each, the article analyzes its sync mechanism, staleness window, and characteristic failure mode. The unique challenges of write-heavy systems, where optimization decisions are reversed, are also addressed.