| With the rapid development of the Internet industry,the frequent interaction of software makes the task of development and operation and maintenance heavy.The industry solves the problem by realizing operation and maintenance automation.The idea of DevOps(the combination of Development and Operations)is to emphasize the automation of operation and maintenance work and realize the automation of operation and maintenance in the form of a unified environment,so as to relieve the pressure of operation and maintenance and development.Container technology can ensure the consistency of the environment,and is the best choice to implement DevOps.In container technology,the combination of Docker and Kubernetes to achieve operation and maintenance automation has become the preferred way for major enterprises to deploy business.The built-in elastic scaling strategy of Kubernetes adopts the elastic scaling based on the threshold value.By monitoring the resource indicators of the business and comparing them with the threshold value,it can achieve scaling.It belongs to reactive scaling.Elastic scalability is divided into two stages: capacity expansion and capacity reduction.Responsive scaling in the capacity expansion stage is easy to cause resource waste and untimely response.If you scale in advance by predicting the future threshold in advance,this predictive scaling can avoid the problem of untimely response.Most of the existing prediction models use traditional algorithms to predict the load,and the prediction accuracy is not high;Responsive scaling will not cause the problem of untimely response in the capacity reduction phase.However,the factors for choosing to delete the pod during capacity reduction are not comprehensive,and the load balance state of the cluster may be affected.In addition to the original state priority capacity reduction rule,the factor of adding the cluster load state should also be considered.Based on the above analysis,this thesis studies the built-in elastic scaling strategy of Kubernetes,and designs and implements a DevOps cloud platform on this basis.The main work is as follows:1.In order to further improve the accuracy of load prediction,a hybrid multi-step prediction model SSA Bi LG is proposed.First,the initial data is decomposed into multiple trend components after singular spectrum analysis(SSA).Then,an automatic encoder network structure of bidirectional long short neural network(Bi LSTM)and bidirectional gated cycle unit(Bi GRU)is designed.Finally,multiple trend components are input into the corresponding automatic encoder for prediction,The prediction results can be obtained by summing the prediction values.A comparative experiment was conducted on the public dataset,and the experimental results show that SSA Bi LG can further improve the prediction accuracy.2.In order to optimize the performance of the elastic expansion and contraction phase,and prevent resource waste and untimely response,an elastic expansion strategy SHPA is proposed to design and implement the expansion and contraction phases respectively.In the capacity expansion phase,predictive capacity expansion is used instead of reactive capacity expansion,including three parts: data collection,load forecasting and capacity expansion optimization.The load forecasting model uses SSA Bi LG.In the capacity reduction phase,according to the original state first capacity reduction strategy,load balancing factors are designed to optimize the load balancing state of the cluster after capacity reduction.Then,the two stages of capacity expansion and capacity reduction are tested respectively.The experiment shows that SHPA can solve the problems of resource waste and untimely response.3.Based on the above research and in combination with the background of a company’s "DevOps Cloud Platform Based on Kubernetes" project,a DevOps cloud platform is designed and implemented.Kubernetes uses SHPA management to scale and realize operation and maintenance automation.The platform is designed and implemented according to software engineering development specifications.First,we will analyze the requirements of the cloud platform to solve the problems of code redundancy,operation and maintenance difficulties and untimely abnormal positioning.We will design the overall architecture and database according to the requirements analysis.The architecture design includes the technical architecture and functional architecture.The functional architecture includes three modules: code generation module,project and application management module and log alarm module;Then detailed design and implementation are carried out.Finally,black box test,performance test and compatibility test are carried out for the cloud platform,and the test results meet the online requirements.The cloud platform has been deployed on the company’s online server for one year,with normal functions,fast response speed and good compatibility. |
PDF Full Text Request