Towards Generating a Dataset for Failure Prediction in Microservices Applications

Microservices architecture has gained significant traction in modern software development due to its agility and scalability. However, using artificial intelligence models to effectively predict the failure of microservices workloads poses challenges, as there is a lack of comprehensive datasets that provide dedicated data to train these models. This paper presents a methodology for generating a dataset specifically designed for failure prediction tasks in microservices applications. We detail our approach for capturing relevant metrics, including the chosen workload generation method and data collection techniques. The paper then describes the characteristics of the generated dataset, including its size, granularity, and captured workload failure scenarios and patterns. Evaluation results demonstrate the utility of the obtained dataset in gaining insights into microservices performance. The aim of this work is to offer a new dataset with useful workload metrics, serving as a valuable resource for researchers and practitioners in the field.