The article explores innovative directions in database architecture design, advocating that most transactional databases should adopt a single-threaded and aggressive sharding strategy. Traditional multi-threaded databases face multiple challenges including transaction management, serialization, race conditions, and locking mechanisms, leading to sharp performance degradation under high loads and becoming bottlenecks for system scalability. In contrast, single-threaded sharded databases eliminate the complexity of concurrent writes through a design where each shard has only one write thread, achieving perfect transaction serialization. Although sharding introduces challenges such as cross-shard queries, transactions, and migrations, the author believes these obstacles can be overcome with proper tool support. ScyllaDB and VoltDB have demonstrated the value of this architecture in high-scalability scenarios, but the author points out that its value extends beyond large-scale systems to ordinary web applications. The article emphasizes that single-threaded sharded databases offer conceptual simplicity and predictability, and while they require early sharding planning, they significantly simplify development and maintenance.
