Aiming at the practical problems of violent fluctuations in high-frequency data and high-dimensional heterogeneity of risk factors in the financial market, this paper constructs a distributed real-time financial risk assessment system based on the MapReduce architecture. By decomposing the VaR and ES models into parallel computing tasks, factor segmentation and preprocessing are completed in the Map phase, aggregation and weighted compression are performed in the Reduce phase, and dynamic updates of risk indicators are achieved in combination with HDFS. The system demonstrates superior assessment accuracy and latency control capabilities in real financial data cluster tests, verifying the applicability and deployment value of the proposed algorithm in a multi-source financial environment.