Data race detection method and system, cluster and medium

The invention provides a data race detection method, which comprises the following steps of: for a candidate instruction pair comprising a calling instruction of an undefined application program interface (API), obtaining a memory access attribute of the undefined API, and then performing data race detection based on an operand of an instruction in the candidate instruction pair and the memory access attribute, and obtaining a detection result used for representing whether the instructions in the candidate instruction pair generate data competition or not. Therefore, the problem that data race detection is difficult to perform on undefined APIs in the prior art is solved, the accuracy of data race detection is improved, and the false alarm rate is reduced. 本申请提供了一种数据竞争检测方法，包括如下步骤：对于包括对未定义的应用程序接口API的调用指令的候选指令对，通过获取未定义的API的内存访问属性，然后基于候选指令对中的指令的操作数和上述内存访问属性进行数据竞争检测，获得用于表征候选指令对中的指令是否产生数据竞争的检测结果。如此解决了相关技术难以对未定义的API进行数据竞争检测的问题，提高了数据竞争检测的准确度，降低了误报率。