This paper systematically reviews the evolution of the architecture and core technology paths of distributed cloud databases. It first points out that traditional centralized databases struggle to meet the demands for high concurrency, elasticity, and low cost in a cloud computing environment. Cloud-native databases effectively break through performance bottlenecks through paradigms such as the separation of computation and storage, and “the log is the database”. For representative systems, the paper analyzes Aurora's Quorum storage optimization, PolarDB-SCC's RDMA-accelerated strong consistent read, and CockroachDB's hybrid logical clock and Commit Wait protocol mechanisms and trade-offs in achieving linear consistency. The paper further discusses how key technologies such as remote memory pools, HTAP architecture (dynamic format conversion, heterogeneous replication, unified table storage), and intelligent scheduling mechanisms in Serverless architecture (such as Moneyball predictive warm-up and logical pause) enhance system elasticity and throughput. Additionally, the overview provides a detailed summary of the three-stage evolution path of distributed cloud databases in terms of hardware-software co-design: from software adaptation to general-purpose hardware, transitioning to RDMA-based offloading of critical path hardware, and ultimately moving towards a new hardware acceleration stage based on CXL, achieving memory pooling and microsecond-level data synchronization. The paper provides a systematic reference and technical direction for building a new generation of high-performance, highly available distributed cloud databases.