文章背景与核心概要
现代 DevOps 基础设施管理主要依赖命令行接口（CLI）/API 脚本以及基础设施即代码（IaC），然而在真实的云环境中进行测试往往面临高昂的成本、安全风险和时间消耗。虽然云模拟器（API 级别的模拟程序）提供了一种本地测试解决方案，但它们的构建和维护极其困难，需要人工去解读复杂且不断演进的云服务文档。
为了解决这一痛点，本文介绍了 CloudEmu——一种利用神经符号代码综合（neurosymbolic code synthesis）技术自动构建云模拟器的新方法。通过将大语言模型（LLM）的自然语言理解能力与云特定的符号抽象相结合，CloudEmu 最大程度地减少了幻觉现象并确保了高精度。该系统利用真实的云环境作为预言机（oracle），进行持续的测试、修复和对齐。评估表明，CloudEmu 在性能和覆盖率上均超越了由庞大工程团队历时十年开发的行业知名工具 LocalStack。
Automated Synthesis of Cloud Emulators
arXiv: 2608.23842
Date: August 24, 2026
Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
arXiv: 2608.23842
Date: August 24, 2026
Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
Summary
DevOps infrastructure management—relying on CLI/API scripts and Infrastructure-as-Code (IaC)—is currently hindered by the high cost, safety risks, and time requirements of testing against live cloud environments. While cloud emulators (API-level mocks) offer a local testing solution, they are notoriously difficult to build and maintain, requiring manual interpretation of complex, evolving cloud documentation.
CloudEmu is a novel, automated approach that constructs these emulators by leveraging neurosymbolic code synthesis. By combining the natural language understanding of Large Language Models (LLMs) with cloud-specific symbolic abstractions, CloudEmu minimizes hallucinations and ensures high precision. The system utilizes the actual cloud as an oracle for continuous testing, repair, and alignment. Evaluation demonstrates that CloudEmu surpasses the performance and coverage of LocalStack, a tool developed by a large engineering team over a decade.
Summary
DevOps infrastructure management—relying on CLI/API scripts and Infrastructure-as-Code (IaC)—is currently hindered by the high cost, safety risks, and time requirements of testing against live cloud environments. While cloud emulators (API-level mocks) offer a local testing solution, they are notoriously difficult to build and maintain, requiring manual interpretation of complex, evolving cloud documentation.
CloudEmu is a novel, automated approach that constructs these emulators by leveraging neurosymbolic code synthesis. By combining the natural language understanding of Large Language Models (LLMs) with cloud-specific symbolic abstractions, CloudEmu minimizes hallucinations and ensures high precision. The system utilizes the actual cloud as an oracle for continuous testing, repair, and alignment. Evaluation demonstrates that CloudEmu surpasses the performance and coverage of LocalStack, a tool developed by a large engineering team over a decade.
Authors
- Archit Bhatnagar
- Zhenning Yang
- Sarah McClure
- Yiming Qiu
- Sylvia Ratnasamy
- Ang Chen
Authors
- Archit Bhatnagar
- Zhenning Yang
- Sarah McClure
- Yiming Qiu
- Sylvia Ratnasamy
- Ang Chen
Access Paper
Access Paper
License
License
Submission History
- v1: Mon, 24 Aug 2026 21:29:52 UTC (929 KB)
Submission History
- v1: Mon, 24 Aug 2026 21:29:52 UTC (929 KB)