Microservice architecture is the latest trend in the design and development of software systems based on modularization. In a microservice-based system, each microservice acts as an independent system; whereas, processing a request may require communication between two or more microservices. Therefore, increasing the complexity and size of microservice-based systems and increasing the communication between microservices increase the probability of runtime failures in such systems. On the other hand, failure detection in microservice-based systems faces challenges due to complex communications, frequent updates, dynamicity at runtime, and complex log management. In a microservice-based system, distributed tracing data is used to overcome the challenge of complex communication and dynamicity at runtime. Distributed tracing data can be used to obtain call paths between microservices when responding to a request. The purpose of this study is to improve the performance of the process of failure detection in terms of increasing accuracy and reducing the false-positive rate to detect failure in the microservices’ call paths using distributed tracing data. For this purpose, we selected a basic method among the existing failure detection methods, and according to the basic method, we designed a process in three phases. The process phases of the proposed method are 1) training and evaluating the failure detection model, 2) online failure detection, and 3) updating the failure detection model. The evaluation results of the proposed method show that with 98% Accuracy and a false-positive rate of 0.02, we achieved a significant improvement over the basic method.

Keywords