ChronoGraph is a graph-structured, multivariate time series forecasting dataset built from real-world microservices. Each node is a service emitting a multivariate stream of system-level performance metrics that capture CPU, memory, and network usage patterns, while directed edges represent inter-service dependencies. The primary challenge is predicting future values of these signals at the service level. Furthermore, ChronoGraph provides expert-annotated event windows with anomaly detection labels, enabling the evaluation of anomaly detection methods and the robustness of predictions during outages. Compared to existing benchmarks in the industrial control system or transportation and air quality domains, ChronoGraph uniquely combines (i) multivariate time series, (ii) explicit, machine-readable dependency graphs, and (iii) anomaly labels that align with real-world events. It reports baseline results that include prediction models, pretrained time series-based models, and standard anomaly detectors. ChronoGraph provides a realistic benchmark for studying structure-aware prediction and event-aware evaluation in microservice systems.
