您好,欢迎访问三七文档
当前位置:首页 > IT计算机/网络 > 其它相关文档 > 构建可扩展可演进大数据架构-毕洪宇
构建可扩展可演进⼤大数据架构GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC毕洪宇Agenda•背景介绍及⺫⽬目标•数据平台•数据应⽤用•团队•总结GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC关于饿了么16亿元+08年10年12年14年16年交易2年份800600400200城+数年份700+.均交易210城+008年10年12年14年16年日均交易116亿元+日0单+500万+覆.城市700+GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCbackground•••越来越多的场景需要数据驱动决策•⼈人难招业务快速发展GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC数据需求越来越强motivation••可扩展,为持续演进提供基础GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC可演进,形成闭环数据平台•TechArchitecture:•数据服务•数据引擎•数据基础架构Persistenttfm*DataInfrastructureadhoct!m*DataServiceGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCDataEngine查询平台••运⾏行状态⼤大屏•保护机制•笛卡尔积,分区检查•资源检查•splitsizetuning多数据源,查询引擎切换GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC数据视图引擎••SQL即接⼝口SQL即报GITC表GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC报表类型•••细粒度权限•细粒度开关多控件图表⽀支持GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC可拖拽式开发SQL⽣生成图表GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCdrag&drop⽅方式GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC查询引擎选型•MMR与MPP的区别•分析型•交互型GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCOLAPengine:kylin•商业常品对接⽅方便•基于查询热度构建CubeGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC⼤大数据量存储引擎选型•MySQL+DAL:核⼼心调⽤用路径•Cassandra:⾮非核⼼心调⽤用路径•Redis-Cluster:⾼高速k-vstore•Elasticsearch:多维度聚合查询GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC调度引擎•并⾏行抽取,⽀支持分库分表•⽀支持多种数据库推送:MySQL/HBase/Cassandra/Redis•将guideline融⼊入到⼯工具中:⾃自动依赖•关键路径性能分析•所有任务情况可回溯GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC数据应⽤用••算法应⽤用实时业务GITC看GITCG板ITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC实时业务看板GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC计算链路•解析复⽤用•HAMySQLNginxBinlogParserFlumeKa8aStormMySQLRedisClusterDataPipelineGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC•扩展存储:•⼿手动分⽚片-三层架构-⾃自动分⽚片••本地化聚合•多流异步合并实时计算Storm可扩展优化实时存储Redis-luster实时监控集群性能监控容量-析水位系统扩展计算:GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC计算领域CAP•成本•⾼高吞吐低时延•精准度GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC算法应⽤用•个性化美⾷食推荐•红包营销•物流调度GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC实验平台•驱动算法快速迭代算法验平台GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCabkey-point•流量分配•••流量shuffle•样本量•异常监控multi-layer请求复⽤用GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCmulti-layer使⽤用正交的hash函数团队⼯工具••jira•workflow:bug管理,敏捷看板,变更管理,任务管理•report⼯工具:资源图,to-dolist,变更⽇日报等;wiki:知识共享,SOP,故障报告管理;GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCsoftskill•沟通:会议必须可落地&&结果确认需落地&&信息分类••failforward:DERPSceneProtectionEscala%onRemedia%onPreven%onGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITDCeGItTeCcGI%TCoGnITCGITClet’sdatatalk:e.g.alertreport总结•nomeasure,noimprovement•••automateeverything•everythingmayfail•conway’slawbalancedesign,thenoptimisationGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCsyncattherighttime–MarkZhang“ThewaytodoreallybigthingsistodoreallysmallthingsGITCGITCGITCGITCaGITnCGdITCGgITCrGoITCwGITCthGITeCGmITCGIbTCiGgITCgGITeCGrI.T”CGITCGITCGITCGITC
本文标题:构建可扩展可演进大数据架构-毕洪宇
链接地址:https://www.777doc.com/doc-5230935 .html