Research And Optimization Of Container Technology In Large-Scale Environment

In recent years,the container technology represented by Docker,with its excellent characteristics such as lightweight and agile,is rapidly replacing virtual machines as a new generation of underlying computing models.On this basis,CNCF's huge technical ecology centered on the container orchestration engine Kubernetes is steadily promoting the evolution of the traditional architecture of the Internet to the cloud native architecture.However,in the face of extreme and complex production environments in large-scale scenarios,the new generation architecture with container technology as the core still has many shortcomings.This thesis has carried out in-depth research on some of the core issues and provide practical and effective solutions for the preliminary exploration of the cloud native evolution of the architecture under large-scale scenarios.Containerization is the premise of architectural evolution and subsequent research work,although the container's "single process model" fits perfectly into the future trend of application architecture microservices.However,its invasiveness to the application makes it impossible to land on large-scale scenes with complex and diverse inventory applications.Therefore,this thesis proposes a new "rich container process model",which has zero intrusion to the existing application architecture and operation&maintenance system,thus enabling rapid containerization of large-scale inventory services.After full containerization,in the face of large-scale container deployment,the traditional container-image distribution method based on master-slave mode will become the bottleneck of the system and may further affect the overall stability of the system.Therefore,this thesis deeply studies the evolution of image distribution mode from master-slave mode to P2P mode,and designs a new image distribution system with good performance and scalability based on BitTorrent protocol,which better avoids the adverse effects of image related problems on the system.However,in a large-scale environment,a single container runtime deployment often cannot completely cover a complex and diverse application scenario.Therefore,multiple container runtimes need to be deployed in a hybrid manner to achieve full coverage of the scenario.However,due to a series of factors,the establishment of existing container standards and the construction of the above-mentioned ecology have obvious bias to the container runtime based on operating system virtualization,and the network aspect is particularly prominent.Therefore,this thesis proposes a method of Layer 2 traffic forwarding through a multi-level relay bridge,which intelligently implements the compatibility of the container network standard CNI and the various container network schemes built on it in a variety of container runtimes,thereby constructing a heterogeneous network.Used to support hybrid deployments of multi-container runtimes.After the container-level technology is complete,in a large-scale scenario,the container engine can only fully release the technical dividend brought by containerization if it is effectively coordinated with the container orchestration system.However,there is a big difference between the two in terms of design concept and implementation.Therefore,this thesis studies and solves some of the key issues of compatibility between the two,thus ensuring that the powerful functions of the underlying container engine can be passed to the upper-level scheduling system through the container runtime interface CRI in a standardized manner.