Monitoring And Analysis Of Service Invocation Link For Microservice System

As a new software architecture style of large-scale and complex applications,microservice architecture was gradually being widely accepted and popularized.The application system based on microservice architecture(hereinafter referred to as microservice system)had the characteristics of decentralization and componentization.However,with the gradual increase of the scale of the microservice system,the system composition changed more and more frequently,and the system complexity was higher and higher,which leaded to the lack of reliability guarantee of the microservice system.There were many new problems in performance optimization and operation and maintenance management,such as difficulty in fault location and troubleshooting,unclear correlation between services and so on.Microservice invocation link monitoring and analysis system were a kind of microservice basic support system proposed to deal with the above problems.It was used to record all requests of microservice system and a series of microservice invocation information involved behind the requests(i.e.Microservice invocation link data),and provided analysis function based on this kind of data to support the management,operation and maintenance of microservice system.However,due to the characteristics of real-time,distribution,high redundancy and complex association of microservice invocation link data,there were great difficulties and challenges in the core functions of microservice invocation link data collection and association analysis.Therefore,focusing on the above challenges,based on the investigation and analysis of the existing service invocation link monitoring system and related work,this paper studied the contents and key technologies such as dynamic data collection and association mining of microservice invocation link,which were carried out from the following three aspects:1.A dynamic data acquisition method of microservice invocation link based on event driven and pipeline sampling was proposed.Firstly,the pipeline sampling was carried out for the link data of different nodes that met the predefined event characteristics,that was,the data acquisition server collected the same link data for all nodes only when a node generated event defined data;At the same time,aiming at the inconsistency of data generation rate of different nodes,multi-threaded streaming data processing and data synchronization technology based on time window were adopted to realize data collection and transmission of different nodes;Finally,for the link data of each node that arrived at the server in different order,the whole link data was synchronized and summarized through timing alignment.Experiments show that the proposed method has higher accuracy and efficiency for link data including specific events such as abnormal and slow response,compared with full acquisition and sampling acquisition methods.2.A service dependency mining algorithm based on invocation link data was proposed.The business process of the algorithm was divided into four steps: data aggregation,service dependency set aggregation counting,service partial dependency mining and combined dependency mining.Firstly,the link logs distributed in different nodes were aggregated by using big data technology,and the service dependencies were aggregated.Through the calculation of support,the partial dependencies of services were found.A method based on the combination of frequent itemset calculation and candidate generation was designed to realize the discontinuous service combination dependency extraction.Experiments show that this method can complete the above service dependency mining process in a short time,and according to the algorithm to mine the combined dependencies for service deployment,the total time of service invocation links reduce,and the number of abnormal links significantly reduce.3.Design and implement a microservice invocation link monitoring and correlation analysis system integrating design invocation link,buried point,acquisition,monitoring and analysis.The system was based on Istio as the underlying architecture of microservices,Zikpin as the embedded point collection tool,and provided a clear visual interface and accurate indicators for link monitoring and analysis,so as to ensure the reliability of link analysis.