Research On Live Migration Technology Of Web Application Container

At present,there are many scenarios where web applications are used in academia and industry,and their operation depends on the environment.In order to facilitate deployment and management,Docker is usually used to make web application code and configuration files into Docker images,packaged and deployed on the server.In the process of largescale deployment of web applications using containers,the unbalanced server load or the need for update and maintenance often have to stop the service,and there is no live migration solution suitable for web applications.In order to allow users to have a better user experience when use web applications,an live migration mechanism for web application containers is proposed,and an live migration system is implemented for two different storage modes.In the pre-migration stage,using the characteristics of container image hierarchical management,a scheme of dynamically reloading images and incremental matching files is designed,and static files are placed in the pre-migration stage for migration,which reduces the unnecessary loss when initialize container on the destination host.In the migration phase,Checkpoint and Restore in Userspace(CRIU)iterative transfer migration is used to add and overwrite the memory mirror file on the destination host side,which reduces the repeated transmission of redundant data and shortens the downtime of the server.At the same time,use the HAProxy load balancer and the back-end backup application designed to keep responding to client requests.Finally,in the recovery phase,the container on the destination host uses the memory state image file to quickly start and restore to the state before the container on the source host freezes.At the same time,HAProxy forwards the client request in the waiting queue to the destination host to resume the response.The web application live migration system was tested to verify the feasibility of the network migration program,and the performance was analyzed.Using dynamic reloading,incremental matching,and iterative migration methods to reduce the amount of data file migration,an average reduction of more than 50%.Under the condition of constant network bandwidth,server downtime is shortened by more than 60%,which is significantly improved container migration efficiency.In the case of sequential increase in network bandwidth,compared with the unoptimized method,it still maintains a performance improvement of more than 60%.