Race condition. Another name: data race. A programming error in a multitask system when the system's work depends on the order in which code sections are executed. Race condition is a classical Heisenbug. Race condition occurs when several threads of a multithread application try to simultaneously get access to the data, meanwhile at least one thread performs saving. Race condition can lead to unpredictable results and they are often difficult to detect. Sometimes race condition's consequences occur only after a large period of time and in some other part of the application. Besides, such errors are very difficult to reproduce. To avoid race condition, synchronization methods are used allowing you to properly arrange operations executed by different threads. Here is an example. Assume one thread execute operation x = x + 3 over a shared variable x, while another thread executes operation x = x + 5. These operations for each thread are actually divided into three separate suboperations: to read x from memory, to increase x and then write x in memory. Depending on mutual order of suboperations executed by the threads the final value of x variable can excess the original one by 3, 5 or 8.
If you develop a parallel application on the basis of OpenMP we recommend you to pay attention to VivaMP static analyzer which can detect some errors relating to race conditions at the very stage of writing the code and this can help you significantly reduce the time of debugging parallel algorithms.
- Wikipedia. "Race condition"
- Wikipedia. "Heisenbug"
- Rahul V. Patil and Boby George. Tools And Techniques to Identify Concurrency Issues