About the secure architecture of a microservice-based computing system

The paper discusses a network-centric microservice architecture system in which all microservice computers are the same for simplicity. Each microservice computer may fail or receive malicious code. The maximum negative impact on the microservice computer is a calculation error and providing the consumer with the wrong result. The tasks of detecting failed microservice computers and detecting microservice computers with malicious code are considered. In solving the set tasks, elements of training are used. Correctly solved problems (conditions, source data, and correct answers) are accumulated in the memory of the control system. This means that one can restart any task with an already known correct result. At the same time, the article uses the ideas and results of the present authors to ensure information security while using metadata. Depending on the assumptions about the possible actions of malicious code, two classes of secure computing algorithms are built in the context of its possible impact on intermediate results in the flow of solved problems. The second class of algorithms works in the assumption that malicious code can correctly calculate the solution to the current problem with probability $p$ and introduce distortion into the result with probability $1-p$. The authors consider three types of distortions that malicious code can introduce and which allow one to either find the true solution accurately or with low probability of error.