Design And Implement Of Failover On MPP Cluster Database

With the degree of social information gradually improve,People deals with the explosive growth of the data in the daily life.And Now,We really have entered the era of “big data”.Data,has penetrated into every industry and business fields;Data,become an more and more important factor in production.People's demand for the mining and use of massive data,Promote the innovation and development of database software technology.MPP Cluster DataBase is a distributed parallel database cluster,which is developed on the basis of the data of Nothing Shared.MPP Cluster DataBase has high iperformance,high availability,high scalability,and can provide high performance and cost data management for the large scale data management.However,with the expansion of the scale of the cluster,in the face of tens of hundreds of nodes of the cluster environment,7*24 hours of uninterrupted service requirements,the system of the original based on the “ two phase commit strategy cluster distributed transactions”,it`s high available defects no doubt exposed.This paper describes the original product architecture,based on the introduction of failover(failure transfer)mechanism,to make up for the original cluster distributed transaction submitted defects,the perfect solution to the product in the case of abnormal conditions may lead to data integrity is destroyed,the problem of inconsistent data,making MPP Cluster Database products more stable and reliable.In this paper,the system architecture design,detailed design,encoding implementation and testing of a step by step to introduce the mechanism from the analysis design to the final implementation of the whole process,the final will be summarized and the future work of the future will be further work.