Self-adaptive master node active election method based on LSTM (Long Short Term Memory) model

Abstract

The invention discloses a self-adaptive master node active election method based on an LSTM (Long Short Term Memory) model, which is used for solving the problem of performance bottleneck caused by overhigh load of a master node in the traditional Raft by introducing an LSTM neural network model to dynamically select a cluster master node. When the system is initialized, the election model node is selected and is responsible for collecting the database state, the working load information and the physical resource use condition of each node in the cluster. Inputting the collected data into the LSTM model, capturing the time sequence change of the node state, and predicting the probability that each node becomes a main node; and when a certain node has the highest probability and obtains agreement of most nodes, master node switching is completed through a safe right handover process. According to the method, the load problem is avoided in advance by dynamically sensing the cluster state, and meanwhile, the election robustness is improved by adopting a majority decision-making mechanism. Compared with a traditional passive triggering election method, the method is more real-time and flexible, is suitable for a complex and changeable distributed environment, and effectively improves the system performance and stability.

Images