Research On Service Monitoring Based On Microservice Business Platform

With the increasing demand of users for application services,the traditional monolithic architecture can no longer support the system load pressure brought by the increase in user access.Microservice architecture has become the mainstream architecture for highly concurrent systems.While the microservice architecture brings many advantages,it also leads to the problem of difficult operation and maintenance.An efficient service monitoring system can greatly reduce the difficulty of operation and maintenance by minimizing the loss of resources.The topic selected in this article comes from enterprise scientific research projects.The research on service monitoring related to the microservice-based ISMP(Integrated Service Management Platform)system includes service node monitoring and service call chain monitoring.For service node monitoring,this paper proposes two container load forecasting models for service node load forecasting from two aspects:univariate forecasting and multivariable forecasting.The first model obtains a set of Intrinsic Mode Functions through Ensemble Empirical Mode Decomposition of the load data,and then performs clustering by K-Shape algorithm to establish a neural network-based load prediction model for each class.The output is merged to get the final prediction result.The second model uses convolutional neural networks to mine the correlation of various load data,and then uses the LSTM Encoder-Decoder with attention mechanism to perform load prediction.The verification on the open source dataset shows that the proposed container load prediction algorithm has higher prediction accuracy than traditional algorithms such as LSTM.In addition,this paper proposes a low consumption service call chain tracing protocol based on the characteristics of the ISMP system.The system's service tracing messages are merged during runtime,reducing the resource overhead caused by service tracing.In the system test part,the proposed protocol is compared with the open source tracing tool Zipkin through experiments,which verifies the effectiveness of the service call chain tracing protocol.Finally,this paper designs and implements a service monitoring system based on the two proposed service node prediction models and service call chain protocols,which are actually applied in the service monitoring of the Integrated Service Management Platform system.