Autoencoder-based Anomaly Detection in Microservices using Distributed Tracing

In the fast-evolving realm of software development, microservice architecture has become a pivotal strategy for managing modern business tasks. While offering benefits like scalability and independence, ensuring the reliability of microservice systems is crucial for sustained customer satisfaction and business success. This paper introduces AnoTraceAE, an unsupervised anomaly detection framework tailored for microservice applications. Using distributed tracing, AnoTraceAE employs span and trace embeddings with a Convolutional Autoencoder to identify deviations from normal behavior in microservice systems. The model’s efficacy is demonstrated through experiments on the TrainTicket dataset, showcasing superior performance across key metrics compared to existing models. AnoTraceAE proves versatile, robustly addressing the challenges of anomaly detection in microservice architectures, highlighting its effectiveness in accurately identifying anomalies and minimizing false positives.