这个工作太神奇了,阅读 Understanding, detecting and localizing partial failures in large system software 的时候,在思考怎样检测 silent semantic violation,论文里说一个难点就是不知道正确的语义是什么,我想到也许可以用 LLM 推测。完全没想到可以用论文如此简洁的方式推测。
论文的思路很简单,从系统的 regression test 入手。尽管这些 test 通常是真的特定的 bug 的,但这些 test 仍然蕴含了系统的语义。论文要做的就是从 regression test 中推导出这些语义,并在运行时检测系统是否违背了语义。
![Featured image for [Paper Note] ALPS An Adaptive Learning, Priority OS Scheduler for Serverless Functions](/posts/alps-an-adaptive-learning-priority-os-scheduler-for-serverless-functions/images/alps-figure-2.png)
![Featured image for [Paper Note] Demystifying and Checking Silent Semantic Violations in Large Distributed Systems](/posts/demystifying-and-checking-silent-semantic-violations-in-large-distributed-systems/images/demystifying-and-checking-silent-semantic-violations-in-large-distributed-systems-architecture.png)
![Featured image for [Paper Note] Understanding, detecting and localizing partial failures in large system software](/posts/understanding-detecting-and-localizing-partial-failures-in-large-system-software/images/understanding-detecting-and-localizing-partial-failures-in-large-system-software-figure-4.png)
![Featured image for [Paper Note] Finding a needle in haystack Facebook's photo storage](/posts/finding-a-needle-in-haystack-facebooks-photo-storage/images/screenshot_20230709_175418.png)