Because of the uncertainty of multithreading, not sequential, each run don't know where is wrong, hard to find, is relying on the support multithreading logging system? Or is there a better way