Research And Practice Of Multi-dimensional Container Monitoring For Kubernetes

With the development of information technology,cloud computing and container technology are gradually maturing.Enterprises deploy tens of thousands of business containers to the cloud platform to provide services for users,and use Kubernetes(referred to as K8s)for container orchestration management and resource allocation.In order to ensure the regular operation of K8 s clusters and their business systems,it is essential to use container monitoring.Compared with the traditional host monitoring,K8 s container monitoring has the characteristics of a large amount of data and frequent changes.Due to the short life cycle of the container,it needs higher requirements for real-time monitoring.At present,the container monitoring system represented by Prometheus generally collects monitoring metrics by deploying agents in the cluster through containers,which will more or less consume resources in K8 s clusters.In practical application scenarios,enterprise users often deploy multiple K8 s clusters.However,the existing container monitoring tools are not suitable for multi-cluster monitoring.On the other hand,when the service container in a K8 s cluster is abnormal,how to infer the abnormal resource metrics from the network traffic is still mainly based on the long-term experience of professionals,and there is still a lack of intuitive and effective tools to help the maintainers quickly locate the cause of the abnormal service container in order to ensure the efficient availability of business applications better.To solve the above problems,the paper proposes a multi-dimensional monitoring scheme for K8 s containers,and realizes the resource monitoring of the core components of multiple K8 s clusters from the four dimensions of CPU,memory,network and file storage based on the existing commercial monitoring sytem.The paper also obtains the correlation between network flow metrics and resource metrics through the correlation analysis in order to realize the multi-dimensional monitoring at the container level.First,the paper analyzes the monitoring architecture and the security mechanism of K8 s,and puts forward a simplified cluster configuration method to realize the management of the monitored clusters.Users can modify the cluster configuration according to their needs.Second,according to the requirements of multi-cluster monitoring,the paper designs a metrics collection method which does not need agents and adapts to the cluster size.The method can regularly collect the monitoring information of multiple core components in the cluster by accessing the K8 s cluster API server and provide users with multi-level resource metrics,which help users judge whether the container is abnormal.Third,the paper analyzes the correlation between resource metrics and network flow metrics of the service containers and gives qualitative correlation conclusions by calculating the strength of the correlation coefficient of the service container in different network environments.Finally,the paper realizes the multi-dimensional monitoring prototype system.On the premise of providing various resource metrics of core components in the cluster,the monitoring system provides the correlation coefficient and metrics with strong correlation to users as the core metrics of multi-dimensional monitoring.The paper also shows the function of the multi-dimensional container monitoring system,and verifies the low consumption of K8 s cluster resources by the monitoring system through performance tests.The multi-dimensional container monitoring can reflect the status of the container from multiple dimensions,which is convenient for users to predict exceptions and set alarms according to the actual situation,and provides help for K8 s cluster operation and maintenance.