With the rapid development of supercomputers, the scale of computing clusters is increasing, and their computing performance is moving from quadrillion to exascale. However, with the increase of the failure frequency of the system nodes and the decrease of the average time between failures, the stability and reliability of the supercomputer system are greatly affected. Logs are an important component to record the system running status and ensure the smooth running of the system. In addition, logs help to discover and rectify system faults. However, with the increase of system scale and complexity, the log generated by nodes also increases exponentially. Node performance degradation is not easy to detect, and the efficiency and accuracy of anomaly diagnosis are faced with challenges. In view of the above problems, this paper studies the node anomaly diagnosis technology of supercomputers, and designs a node anomaly diagnosis system to help diagnose node anomalies in supercomputers, so as to effectively predict possible anomalies.