Anomalies Detection in Cloud-Native Architectures: Background, State-of-the-Art, and Research Directions

Cloud-native architectures, which utilize containerization, microservices, and dynamic orchestration, offer scalable, resilient, and flexible applications. However, their inherently dynamic and distributed nature introduces complexities, making them susceptible to anomalies. Machine learning (ML) approaches are crucial for identifying patterns that signal potential anomalies in cloud-native architectures. This article reviews ML-based anomaly detection approaches for cloud-native architecture, proposing a taxonomy and focusing on ML techniques, anomaly types, data characteristics, and evaluation aspects. We identify six key requirements for effective anomaly detection in cloud-native architectures and analyze the reviewed approaches against these criteria. Through this analysis, we provide insights into the state of the field and uncover research gaps. Our findings reveal that existing approaches often fail to meet adaptability, distributed operation, and interpretability requirements in cloud-native architectures. Future research should focus on developing anomaly detection methods that adapt seamlessly to dynamic cloud-native architectures and provide interpretable knowledge to improve operational reliability.