Interaction Prediction and Anomaly Detection in a Microservices-based Telecommunication Platform

In microservice platforms with high number of users and heavy traffic, it is necessary to monitor the system, take quick action against errors and ensure the maintainability of the system. However, debugging on these platforms can take a long time. This difficulty arises from the need of understanding the behavior of microservices and detecting their interactions. In this study, which aims to increase the efficiency of DevOps engineers on the work/time unit, it was observed that providing microservice flows and interactions saves operations teams a significant amount of time during debugging. Accordingly, the study focused on microservice interactions and anomaly detection. Firstly, using the log patterns extracted from the microservice logs, different machine learning models were created to predict the previous and next microservices with which the current microservices interacted at a certain moment, and their performances were compared. Then, anomalous data were injected into the microservice logs, models were developed to detect these data and their performances were compared. In the experiments, unsupervised and supervised algorithms are used with 6 different datasets, and successful estimation results were obtained that can contribute positively to the debugging process.