Research On Multi-index Integration Expansion And Contraction Prediction And Service Failure Handling Method

Since the birth of container technology,it has quickly become an indispensable component of the microservice architecture system.Correspondingly,cloud platforms based on container technology have become more and more widely used,which not only improves the utilization rate of cluster resources,but also reduces maintenance costs.At present,major cloud vendors at home and abroad have successively carried out a large number of experiments and applications in the field of container platforms,and various container cloud systems have sprung up to major communities.Among them,the Kubernetes container cluster platform has attracted the most attention.At the beginning of its birth,it has overcome all obstacles and continuously defeated its opponents,and it has become the standard in the industry.With the widespread application of container clusters,the elastic scaling technology that takes the expansion and contraction of containers as the actual performance becomes more and more important.The better the scalability strategy,the more stable and reliable the cluster can be guaranteed.In order to improve the utilization level of cluster resources,existing research focuses on the prediction of cluster load changes.These researches are extremely important for improving the scalability of container clusters,but there are still some shortcomings to a certain extent.First of all,the existing container cluster prediction methods usually only consider a single load index data.In a real container cluster environment,it is either difficult to guarantee the prediction accuracy,or multiple prediction models are set to adapt to multiple index scenarios,which consumes a large number of clusters resources,and generate some invalid calculations.In response to this problem,this paper proposes a high-impact factor index selection algorithm based on load distance,which fully integrates multiple indexes and calculates the actual type of service,and improves the multi-index batch modeling and forecasting method to an efficient mode with only one effective calculation.Secondly,the existing research on load prediction algorithms has low prediction accuracy when facing load jitters and complex and changeable data scenarios.In response to this problem,this paper proposes a VMD_F-BiLSTM prediction model,which fully combines the non-stationary data processing capabilities of the VMD signal decomposition algorithm and the prediction capabilities of the BiLSTM model.Finally,the existing research on elastic scaling technology lacks service failure handling capabilities,and it is difficult to avoid abnormal index data caused by service failures,which leads to a serious decrease in prediction accuracy,which in turn makes container clusters perform wrong elastic adjustments.In response to this problem,this paper proposes an automatic and elastic processing algorithm for service failures,which captures the number of successful and failed cluster access requests in real time,dynamically judges the running status of cluster services,and dynamically expands services whose running status reaches the fault judgment threshold.In addition,in order to make the above algorithms better adapt to cluster online scenarios,this paper designs and implements cluster resource index prediction mechanism for online scenarios and an automatic scaling mechanism for resource index prediction.In summary,this paper carries out experiments based on Google cluster tracking data and real Kubernetes container cluster environment.The experimental results show that the algorithm in this paper can accurately realize the load prediction of cluster multi-index integration,and has the ability to shield and handle fault services,further improve the prediction accuracy,realize the efficient utilization of resources and maintain the stability of the cluster.