1、Apache Doris 在虎牙业务监控场景下的实践陈仕明虎牙 数据平台负责人Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2

2、023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023目录2.技术选型及设计3.关键技术实现1.OLAP 在虎牙的场景应用4.未来规划Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit As

3、ia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023“虎牙直播”诞生20148月10日，虎牙公司正式注册成立深

4、耕游戏直播，转战移动互联网2016虎牙直播月活用户突破1亿5月11日纽交所成功上市，股票代码HUYA布局海外市场，发行海外产品Nimo TV2018全年营收突破100亿Nimo TV海外月活突破3000万腾讯控股虎牙2020旗下产品2023布局新赛道Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 202

5、3Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023以电竞+游戏直播为核心 构建多元内容生态体系正能量美食星秀户外体育二次元音乐电竞游戏截至2023年3月共播出官方赛事182个，共7570场，总直播时长9084

6、0小时Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summi

7、t Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023OLAP 在虎牙的场景应用1Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summi

8、t Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023垂直的数据产品 敏捷简便，即报即用开放的数据场景 场景定制数据模型 深度优化性能FlinkHive、SparkDorisKafkaBI报表用户画像漏斗&留存中间件服务监控音视频链路监

9、控技术业务分析业务故障排查建模预计算数据应用上报即用虎牙 OLAP 的应用服务场景查询次数业务需求数量50万3万Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Do

10、ris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023业务监控的典型场景案例面向时间线的应用面向产品研发的业务分析音视频全链路故障监控&排障微服务调用监控基础运维监控中间件监控技术开发的产品分析中台产品的运营分析推流/拉流链路监控及切换大主播开播质量监控及排查P2P效果分析更敏捷：无需建模，开发门槛更低，开发效率更高更高吞吐：分钟粒度数据，每天万亿

11、规模延时更低：分钟内延时，否则可能误判/漏判模型复杂：维度更多，聚合方式更灵活计算复杂：存在关联分析场景，拆分集群较难数据量更大：明细数据排障跟BI相比跟 Metric 监控相比Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit

12、Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023使用方式数据接入即刻使用配图监控分析即报即用，无需建模Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023

13、Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit A

14、sia 2023技术选型及设计2Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 20

15、23Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023系统要求&选型 计算复杂性：单表查询，多表关联，细粒度分组 性能：万亿写入，分钟内延时强大的分析性能 索引 物化视图的预计算、sql rewrite平台可透明优化 便捷的扩缩容 集群跨机房 租户和场景的隔离运维简单 数据最终一致性 事务可见性数据可靠Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2

16、023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summi

17、t Asia 2023Doris Summit Asia 2023系统方案KafkaDoris(主)flink sink(明细)Trinopush downconvert rulerflink sink(时间维度降维)Doris(备)核心数据，异地机房冗余容灾 部署：主备集群，机房容灾 写入：Flink写入，提前按需时间降维(不使用Doris物化视图)查询：trino接入，向上屏蔽GrafanaTableauAPI核心特点Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 202

18、3Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit

19、Asia 2023落地情况数据规模Table5K5-7K 亿/天节点200+查询延时查询量50W 次3秒 98%实时10秒-2分钟Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit As

20、ia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023关键技术实现3Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023

21、Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023系统建设的核心挑战 监控场景：分钟内延迟 S赛事：峰值暴涨2-3倍高可靠

22、保障01 避免场景滥用 隔离性运营&治理02 迁过来：历史包袱太重 牵出去：应对未来风险迁移03Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summ

23、it Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023容灾 多副本：保障存储和计算的可靠 机房级热备：核心数据按需冗余，查询路由复用备集群算力，提高备用集群资源利用率赛事临时扩容 提前云上算力扩容，node group标注云上节点 预先table list，标记云上存储副本，云上分担计算压力 事后15%上限比例进行缩容，IO影响较大集群高可靠Doris Summi

24、t Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Dor

25、is Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023查询greenplum，kylin，impala，kudu，druid，clickhouse，doris/sr，各种云。“敏捷”开发的几万SQL将来如何升级到其他方案？无压力的引入新技术对用户透明，OneSQL&OneOLAP？TrinoHiveESDB集成计算集成数据TrinoHiveESDB场景下最优引擎粘合剂tablescan下推下推“执行树”OneSQL，但性能较低 Catalog不同，存储位置对用户不透明 逻辑元数据

26、，存储透明化 引擎各司其职，性能有保障物化视图Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit

27、Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023查询物化：mv根据场景可选择不同的存储引擎，如Doris、ES、Phoenix，“一张表”支撑各类场景，SQL Auto RewritePushdown：将SQL分支“尽可能”的pushdown到下层OLAP引擎计算，保障计算性能跨源：和离线数仓、DB进行关联分析，不仅异构，还能异地数据湖：虚拟数据湖，集成数据分析更高效，成本更低Trino Doris Con

28、nectorDoris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Su

29、mmit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Trino Doris ConnectorPushdown:算子：JOIN、UNION、TOPN、ORDER BY、AGGR、FILTER、WINDOW 函数：Trino内置函数、UDF、UDAF，逐步覆盖 Exp：部分JSON Exp，Regular Exp，Lambda Exp，Date Exp Hint：可透传给doris落地效果:实际业务场景：查询次数50万每天，单次查询耗时增加20m

30、s，SQL完全pushdown覆盖率99.5%TPC-H基准测试：与直接查询doris性能相同跨源查询，尽量分别pushdown到底层OLAP计算，最终再trino中汇总Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia

31、 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Trino Doris ConnectorTrino SQL 下推的瓶颈在哪里？l算子？支持l函数?不支持Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023

32、Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit A

33、sia 2023Doris Summit Asia 2023Trino Doris Connector社区兼容性:目前尚不兼容，但社区在往该方向发展注册登记 Ruler注册函数转换Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit

34、 Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023业务迁移Grafana on Clickhouse(3万+图表)Grafana on Trino转换工具Grafana MetaClickhouse SQLConvert ToolTrino SQLGrafana Meta截图对

35、比人工修正替换Trino SQL能力扩展:clickhouse常用函数：如uniqif，sumif等 别名：where，group中可使用select中别名，减少SQL代码量 Any Join耗时半年!Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023D

36、oris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023SSD、HDD分层存储，加快IO速度硬件索引、排序、分区，减少 IO 加载存储使用Doris物化或预先在flink中预计算预计算使用colocation Join优化特定的表关联分析执行计划bucket、h

37、int提高计算并行度并行计算Doris 查询优化Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit

38、 Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023 计算资源队列，隔离监控查询和分析查询影响 全文检索，加速点查场景速度 存算分离，和虎牙离在线混部算力打通，提供更充裕的临时算力2.0版本升级 异步物化，替代目前flink预计算 数据建模方式的改变：后置建模物化视图未来规划Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Do

39、ris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asi

40、a 2023Doris Summit Asia 2023获取更多社区动态与最佳实践Doris Summit 峰会官网：doris- Doris Summit 峰会回放：https:/ Doris 官网：doris.apache.orgApache Doris GitHub： Doris 官方平台：Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023Doris Summit Asia 2023