A Log-Guided Adaptive Framework for Concurrency Bug Detection in Distributed Systems

Distributed cloud systems are widely used in modern applications such as online services, largescale data processing, and microservice-based platforms. These systems consist of multiple distributed nodes that communicate with each other through asynchronous messages. While this architecture improves scalability and performance, it also introduces complex concurrency challenges. One of the most critical issues in such systems is the occurrence of concurrency bugs, which arise when multiple processes or nodes access hared resources or exchange messages in unexpected orders. These bugs are difficult to detect because they often appear only under specific message inter leaving’s or timing conditions, making them rare, nondeterministic, and hard to reproduce during testing. Traditional approaches such as systematic testing, stress testing, and model checking attempt to detect concurrency bugs by exploring all possible message orderings or execution paths. Although these techniques can identify many potential issues, they suffer from the state explosion problem, where the number of possible message interleaving’s grows exponentially as the system size increases. As a result, these methods require significant computational resources and time, making them impractical for large-scale distributed systems operating in real-time cloud environments. To address these limitations, this paper proposes a Log-Guided Adaptive Concurrency Detection System (LG-ACDS) that improves the efficiency of concurrency bug detection by leveraging runtime execution logs. Instead of exhaustively exploring all possible message sequences, LG-ACDS analyzes historical system logs to understand actual message interactions between distributed components. The system mines the logs to identify accessrelated message pairs and extract relevant identifiers that correlate communication events across nodes. By applying ID-based correlation and happen-before analysis, LG-ACDS determines which message orderings have already been executed and which feasible orderings remain insufficiently tested. Based on this analysis, the proposed system adaptively prioritizes high-risk message pairs that are more likely to produce concurrency-related failures. These message sequences are then selectively triggered and monitored during runtime to expose hidden bugs. This targeted approach significantly reduces the computational overhead associated with brute-force testing while still maintaining high detection effectiveness. Experimental evaluation demonstrates that LGACDScanefficientlydetectcriticalconcurrency bugs with minimal runtime overhead. The system successfully identifies problematic message interactions that traditional testing techniques may overlook due to scalability constraints. By combining log mining, adaptive triggering, and concurrency analysis, LG-ACDS provides a practical and scalable solution for detecting concurrency issues in distributed cloud systems. Therefore, the proposed approach is well-suited for modern large-scale distributed applications that require reliable and efficient de bugging mechanisms in dynamic cloud environments.