The modern system is becoming more and more complex in scale and structure. Mastering the operation status is crucial to ensure the stable and reliable operation of the system. Log anomaly detection is the critical means of system state monitoring and anomaly response. However, the characteristics of complex log data structure, large amount of data and hidden abnormal behavior patterns bring new challenges to efficient and automated log anomaly detection. This paper summarized the basic framework of log anomaly detection, including log collection and filtering, log parsing, feature extraction and anomaly detection. We have reviewed the relevant technologies and methods involved in each link. In particular, various deep learning detection models in recent years are analyzed, such as the use of recurrent neural network and convolutional neural network to capture the context information of log sequences, the use of generative adversarial network to make up for the deficiency of abnormal data, and the training of federated learning between different systems. We hope that our work can help beginners understand log anomaly detection and relevant experts keep abreast of the latest research trends.