一位开发者在Hacker News分享了在数据库事务和提交管理中存在的关键代码架构问题[1]。这些问题包括隐藏的手动提交、数据库模型泄露以及数据丢失风险[1]。作者通过实际代码示例揭示了这些问题的根源，并强调了将数据库抽象层作为事务唯一所有者的重要性[1]。
该文章提出了数据库操作的核心原则，包括不应在代码中零散地注入数据库会话或事务[1]、不应在数据库层之外传递数据库模型对象[1]、不应进行手动提交尤其是在使用上下文管理器或装饰器时[1]，以及不应将涉及多个写入操作的代码分散在多个辅助函数中[1]。作者建议严禁在数据库层外部执行事务、提交、查询等任何数据库操作[1]。
为了自动检测和防止这些违规行为，文章介绍了两种工具化方案：通过抽象语法树（AST）分析或flake8插件来禁止手动提交调用[1]，以及在持续集成/持续部署流程中使用大语言模型检查数据库模型是否被不当返回[1]。
A software developer has shared practical guidance on database transaction and commit management, highlighting frequent architectural mistakes in codebases [1]. The author emphasizes several core principles to prevent data loss and maintain code integrity, beginning with the fundamental rule: do not scatter database sessions or transactions throughout code [1].
The guidance identifies multiple anti-patterns developers should avoid [1]. These include passing database models in and out of the database layer, which creates improper coupling [1]; manually committing transactions, especially when using context managers or decorators that should handle this automatically [1]; splitting atomic multi-write operations across helper functions rather than keeping them in a single function [1]; and performing transaction, commit, or query operations outside the designated database layer [1]. The author stresses that the database abstraction layer should be the sole owner of transaction control [1].
To enforce these principles, the author proposes automated detection methods [1]. These include using Abstract Syntax Tree (AST) testing or flake8 linter plugins to prohibit manual commit calls in restricted code areas [1], and deploying large language models within CI/CD pipelines to detect when database models are being improperly returned from the database layer [1].