Methodology And Practice Of Container Network Observability Based On EBPF

One of the biggest challenges faced by distributed systems is observability.With the development of cloud-native,technologies such as microservices,Service Mesh and Kubernetes make it easier to deliver software at scale,but the problem troubleshooting and fault location of massive microservices make observability more challenging,and the current observability technologies under cloud-native system have many shortcomings in facing the above problems.In this paper,we conduct an in-depth study on the related issues,and introduce kernel technology from the perspective of container networks to design and implement an observable system,providing an efficient solution for improving the observability of applications in cloud-native architecture.The lack of kernel observability in the cloud-native environment makes the performance analysis of cloud-native applications also face bottlenecks.The operability,performance and security of current invasive network performance analysis and problem location solutions are difficult to guarantee,and it is difficult to collect fine-grained data in the cloud-native environment.Therefore,this paper designs and implements a non-intrusive container network analysis system based on eBPF,which does not require any kernel and application code modification and is transparent to user applications.At the same time,different network protocols for different applications are well supported within the framework,achieving network protocol independence.Secondly,it provides fine-grained network performance analysis from threaded processes,container applications,Kubernetes platform to different topology levels related to system calls,while in terms of performance,it can achieve millions of data throughputs per second,with a maximum impact on platform network performance of only 1.2% increase.After obtaining fine-grained container network performance analysis data,in the face of complex distributed network link relationships,complex request paths and routes in the container environment,distributed network problem diagnosis based on container network performance analysis can use fine-grained data to locate the specific links in the network links where performance problems occur,providing a solution for locating and diagnosing problems in multi-link container networks for cloud-native applications.Finally,this paper introduces a typical case of container network observability analysis system applied in clusters,and provides a technical solution for the implementation of observability analysis system in practical scenarios such as cloud database and microservice performance diagnosis.