The system will make mistakes in "extreme scenarios". The reason may be the complexity of the interactive environment (network, IO, concurrency) itself, and more importantly, the logic of the program itself is not rigorous enough.
For example, when updating data, there is a problem with transaction management. After calling its system interface fails, there is no corresponding remedy. Data inconsistency occurs between redis and DB, etc.
This kind of bug cannot be verified by functional testing. Generally, it needs to be reproduced after a large number of requests after going online.
-------------
Solve the problem: Introduce code walkthrough to discover in time the scenes where the function implementation is not fully considered