腾讯云MaxCompute网站用户访问数据分析从零到实践：完整技术指南

{"type":"doc","content":[{"type":"heading","attrs":{"id":"d934c5fb-f001-4041-9ae6-adb5c74ab21d","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"1. 引言：为什么选择MaxCompute分析网站用户访问数据"}]},{"type":"paragraph","attrs":{"id":"708c4ae6-7447-4aaa-8732-af1f39e3e29b","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"网站用户访问数据是互联网企业最核心的数据资产之一。每一笔用户请求都会在服务器日志中留下记录，这些日志蕴含着页面浏览量、独立访客数、用户来源地域、访问终端类型、用户行为路径等丰富信息。通过对这些数据的深度分析，企业可以精准把握用户偏好、优化产品功能、提升转化率，从而实现精细化运营。"}]},{"type":"paragraph","attrs":{"id":"472d80c1-95bb-47c3-8c12-e0c7131f9e97","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"然而，网站访问日志的数据量往往非常庞大，一个日均PV百万级的中型网站，一天产生的原始日志就可能达到数十GB。传统的关系型数据库难以承载如此规模的数据存储与计算，而腾讯云MaxCompute正是为这种海量数据场景而生的云原生大数据计算服务。MaxCompute支持EB级数据存储、提供标准SQL分析能力、按量付费的弹性计费模式，并且与WeData数据开发平台无缝集成，能够一站式完成从数据集成、数据开发到运维调度的全流程工作。"}]},{"type":"paragraph","attrs":{"id":"507bad9c-69f6-4a25-adbb-8b1b0f6e042d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"本文将以一个完整的网站用户访问数据分析项目为载体，从零开始讲解如何利用MaxCompute WeData完成数据集成、数仓建模、指标计算、性能优化和可视化展示的全链路操作。"}]},{"type":"paragraph","attrs":{"id":"a72cfce7-8edf-4dc0-8d6b-26d7c73a509a","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"需要先登录腾讯云控制台，点击："},{"type":"text","marks":[{"type":"link","attrs":{"href":"https://cloud.tencent.com/act/cps/redirect?redirect=37484&cps_key=c6c11b1aa9ecc74cc31e4c2c4c34cae9","target":"_blank","rel":"noopener noreferrer nofollow","class":null}}],"text":"腾讯云控制台"}]},{"type":"heading","attrs":{"id":"46828807-ffda-4d34-9db4-2e5bdb2dd12c","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"2. 环境准备与数据集成"}]},{"type":"heading","attrs":{"id":"b47a597a-ce6a-46f4-ae8a-b04a071f9e75","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"2.1 开通MaxCompute与WeData"}]},{"type":"paragraph","attrs":{"id":"31eacb2d-b286-48d0-a97e-d1e4ef37f49f","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"在开始分析之前，首先需要完成MaxCompute和WeData的开通与项目创建。MaxCompute是计算引擎，WeData是数据开发与运维平台，两者通常配合使用。"}]},{"type":"paragraph","attrs":{"id":"7dfd11fb-f279-4894-8cc6-73abd89066de","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"开通MaxCompute时，推荐选择靠近业务源数据的地域，规格类型选择标准计算资源。首次使用可以选用按量付费模式，无需预购资源，用多少付多少。开通后，需要新建MaxCompute项目，建议按照开发和生产环境隔离的标准模式，分别创建两个项目，例如生产环境项目名为workshop_prod，开发环境项目名为workshop_dev。数据类型建议选择2.0数据类型，以获得更丰富的SQL类型支持。"}]},{"type":"paragraph","attrs":{"id":"50bbca0a-390a-4e00-95c2-c75808aadb9e","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"WeData方面，需要创建一个标准模式的工作空间，并购买Serverless资源组用于数据同步和任务调度。"}]},{"type":"heading","attrs":{"id":"316f1585-c116-420c-815e-402a62e926e1","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"2.2 数据集成：将网站日志同步至MaxCompute"}]},{"type":"paragraph","attrs":{"id":"649c1c8e-4d13-4aa2-bae9-9fa5874735e7","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"网站访问日志通常以两种形式存在：一种是直接写入日志文件，存储在云服务器或对象存储COS中；另一种是通过日志服务CLS实时采集。无论哪种形式，WeData数据集成模块都提供了成熟的同步方案。"}]},{"type":"paragraph","attrs":{"id":"2dfe84e9-24ed-4f87-9d06-52e695fbbe08","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"场景一：从COS同步日志文件"}]},{"type":"paragraph","attrs":{"id":"5e8bb2ad-5e3d-4c8a-9784-282f6f351df5","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"如果网站日志以文件形式存储在COS，可以通过WeData的离线同步节点，将COS中的日志文件同步至MaxCompute的ODS原始日志表。同步时需要注意：源端COS文件需确认其格式和字段分隔符；目标端MaxCompute表建议按日期分区，便于后续按天查询和管理；同步任务可配置为每天定时执行，实现日志的T 1分析。"}]},{"type":"paragraph","attrs":{"id":"d6edd3da-4284-4970-ac92-87541a3e6237","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"场景二：从日志服务CLS实时同步"}]},{"type":"paragraph","attrs":{"id":"007684a9-c6b1-4658-bd7b-1b6fae07a5b3","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"如果网站已接入日志服务CLS进行日志采集，可以通过CLS的投递功能直接将日志数据投递到MaxCompute。在CLS控制台中配置投递任务，选择目标MaxCompute项目和表，即可实现日志数据从采集到存储的无缝流转。"}]},{"type":"paragraph","attrs":{"id":"30ad9de5-4c16-418a-bc24-2dbfae52a7ba","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"场景三：通过DataHub实时接入"}]},{"type":"paragraph","attrs":{"id":"6782d357-a8bd-474d-ad81-caf76cfba2bd","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"对于需要实时接入的场景，可以通过DataHub服务将流式数据实时或准实时同步到MaxCompute。DataHub DataConnector支持将Topic中的数据同步到MaxCompute、COS、ElasticSearch等多种云产品。"}]},{"type":"heading","attrs":{"id":"34957432-c2c4-433b-992e-b79401f531bd","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"2.3 建表：定义数据模型"}]},{"type":"paragraph","attrs":{"id":"801b1259-8fcb-4c3e-bc9c-428c336bdb51","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"在数据同步之前，需要先在MaxCompute中创建目标表。以下是一张典型的网站访问日志ODS表建表语句："}]},{"type":"codeBlock","attrs":{"id":"aa78b8b3-931d-4087-8611-8c1f5fa8db48","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"CREATE TABLE IF NOT EXISTS ods_raw_log_d (nip STRING COMMENT '客户端IP地址',nts BIGINT COMMENT '访问时间戳',nmethod STRING COMMENT '请求方法 GET/POST',nurl STRING COMMENT '请求URL',nstatus BIGINT COMMENT 'HTTP状态码',nbytes BIGINT COMMENT '响应字节数',nreferer STRING COMMENT '来源页面',nuser_agent STRING COMMENT '用户袋里字符串',ncookie_id STRING COMMENT 'Cookie标识',nsession_id STRING COMMENT '会话ID'n) PARTITIONED BY (dt STRING COMMENT '日期分区 yyyyMMdd');n"}]},{"type":"paragraph","attrs":{"id":"cf143370-661b-4b67-9794-1feee30e7f68","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"分区字段dt用于按天组织数据，既可以提升查询效率，也便于数据生命周期管理。"}]},{"type":"heading","attrs":{"id":"c729e126-1d96-4eb2-9a92-771f3c4922e4","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"3. 数仓分层建模"}]},{"type":"paragraph","attrs":{"id":"acb06597-e89a-48d3-b780-7bd30c9d8c67","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"网站用户访问数据分析遵循经典的四层数仓建模架构：ODS（操作数据存储层）、DWD（明细数据仓库层）、DWS（汇总数据仓库层）和ADS（应用数据层）。"}]},{"type":"heading","attrs":{"id":"e7921cb2-12e6-4197-a2c0-d4d38e96cdfb","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"3.1 ODS层：原始日志表"}]},{"type":"paragraph","attrs":{"id":"bac1c799-6716-4d57-8783-e541e1f0d7b4","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"ODS层是数据仓库的源头，存储从业务系统直接同步过来的原始数据，不做任何清洗和转换。上文中创建的ods_raw_log_d表即属于ODS层。ODS层的数据与源系统保持一致，保留了最完整的日志明细，便于后续的数据回溯和问题排查。"}]},{"type":"heading","attrs":{"id":"9ecc0e6f-d154-486d-931a-9ea144eecbaf","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"3.2 DWD层：明细数据仓库表"}]},{"type":"paragraph","attrs":{"id":"d99c78e5-70fb-4696-a26d-2afc0b9246a4","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"DWD层对ODS层的原始数据进行清洗、标准化和维度退化处理。主要操作包括：过滤无效数据、解析User-Agent获取设备类型和浏览器信息、解析IP获取地域信息、将时间戳转换为可读日期时间格式等。"}]},{"type":"paragraph","attrs":{"id":"cf25bbdf-ae73-4f7c-8e4d-3a2fdc538dae","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"以下是将ODS层数据加工到DWD层的SQL示例："}]},{"type":"codeBlock","attrs":{"id":"51a5fc18-f420-4a0b-9fc6-259d3e5799f1","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"INSERT OVERWRITE TABLE dwd_access_log_d PARTITION (dt = '${bizdate}')nSELECTnip,nFROM_UNIXTIME(ts) AS access_time,nmethod,nurl,nstatus,nbytes,nreferer,nuser_agent,nGET_JSON_OBJECT(user_agent, '$.device_type') AS device_type,nGET_JSON_OBJECT(user_agent, '$.browser') AS browser,nSUBSTR(ip, 1, INSTR(ip, '.') - 1) AS ip_segment,n-- 使用MaxCompute内置函数或UDF解析IP地域nip_to_region(ip) AS region,ncookie_id,nsession_id,n-- 提取URL中的页面路径nREGEXP_EXTRACT(url, '^(https?://[^/] )?([^?] )', 2) AS page_pathnFROM ods_raw_log_dnWHERE dt = '${bizdate}'nAND status < 500-- 过滤服务端错误nAND url IS NOT NULLnAND LENGTH(TRIM(url)) > 0;n"}]},{"type":"paragraph","attrs":{"id":"9d44cbbb-720a-42ed-a64c-b93e99d97e3f","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"通过DWD层的加工，数据从"脏乱差"的原始日志变成了结构清晰、字段规范的明细数据，为后续的汇总分析奠定了坚实基础。"}]},{"type":"heading","attrs":{"id":"4cfa7a3a-0957-4ea3-8c1e-76e576fcabfe","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"3.3 DWS层：汇总数据仓库表"}]},{"type":"paragraph","attrs":{"id":"8eb541a0-ab8e-4782-ac8c-f089722e4b12","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"DWS层是在DWD层基础上，按照分析主题进行预聚合的轻度汇总层。这一层通常按照天、小时等粒度，对用户、页面、地域等维度进行聚合计算，以提升查询性能。"}]},{"type":"paragraph","attrs":{"id":"3aabf492-df92-4ba4-8990-693113b07867","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"以下是一张按天和页面汇总的DWS表建表语句："}]},{"type":"codeBlock","attrs":{"id":"7ea8e6a2-6278-4c82-8c21-ac4843dcbb9b","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"CREATE TABLE IF NOT EXISTS dws_page_stat_d (npage_path STRING COMMENT '页面路径',npv BIGINT COMMENT '页面浏览量',nuv BIGINT COMMENT '独立访客数',navg_response_time DOUBLE COMMENT '平均响应时间',ntotal_bytes BIGINT COMMENT '总流量字节数'n) PARTITIONED BY (dt STRING COMMENT '日期分区');n"}]},{"type":"heading","attrs":{"id":"e1cd171c-848a-4fc7-80a5-87366d27c22c","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"3.4 ADS层：应用数据层"}]},{"type":"paragraph","attrs":{"id":"c1f8ade7-187e-4693-b8b5-9f2c6052777d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"ADS层是面向具体应用场景的数据服务层，直接为报表、仪表盘和数据分析师提供数据。这一层的数据通常已经完成了多维度的聚合和计算，查询响应速度快。例如，面向运营报表的每日核心指标汇总表、面向用户画像的标签宽表等都属于ADS层。"}]},{"type":"heading","attrs":{"id":"8886c79b-e724-4add-97a1-b0d4d4965959","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"4. 核心分析场景与SQL实现"}]},{"type":"heading","attrs":{"id":"15738ada-c505-4559-ade2-da524c5da03e","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"4.1 PV和UV统计"}]},{"type":"paragraph","attrs":{"id":"892b49d0-7119-46aa-917f-cdf55b4bd14d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"PV和UV是衡量网站流量的两项最基本指标。PV统计的是页面被访问的总次数，UV统计的是访问网站的不重复用户数。"}]},{"type":"paragraph","attrs":{"id":"2f24efb7-a1e6-43cc-92a3-697a909bb47c","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"按天统计全站PV和UV："}]},{"type":"codeBlock","attrs":{"id":"2dcaa700-9781-48fa-b0ac-0ed1a41db5ef","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"SELECTndt,nCOUNT(*) AS pv,nCOUNT(DISTINCT cookie_id) AS uvnFROM dwd_access_log_dnWHERE dt BETWEEN '20260101' AND '20260131'nGROUP BY dtnORDER BY dt;n"}]},{"type":"paragraph","attrs":{"id":"8c42433e-cf31-41e9-9e5b-19c9d1b5cd0f","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"按页面路径统计PV和UV："}]},{"type":"codeBlock","attrs":{"id":"911137e7-a8be-4976-a653-c815921c7d31","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"SELECTnpage_path,nCOUNT(*) AS pv,nCOUNT(DISTINCT cookie_id) AS uvnFROM dwd_access_log_dnWHERE dt = '20260115'nGROUP BY page_pathnORDER BY pv DESCnLIMIT 20;n"}]},{"type":"paragraph","attrs":{"id":"cedd390a-6be7-42d4-9b03-03cfc2085bee","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"按终端类型分别统计："}]},{"type":"codeBlock","attrs":{"id":"0b886a97-e525-42da-804a-dba028603557","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"SELECTndevice_type,nCOUNT(*) AS pv,nCOUNT(DISTINCT cookie_id) AS uvnFROM dwd_access_log_dnWHERE dt = '20260115'nGROUP BY device_type;n"}]},{"type":"heading","attrs":{"id":"9af408b2-f997-4264-9075-c74e8edc0b0e","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"4.2 漏斗转化分析"}]},{"type":"paragraph","attrs":{"id":"7d1c3b3c-ad94-4ac1-829d-b5d59444aa92","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"漏斗模型是通过产品各项数据的转化率来判断产品运营情况的工具，通过各阶段数据的转化来判断产品在哪一个环节出了问题。以电商网站为例，典型的用户转化路径为：浏览首页 → 搜索商品 → 查看详情 → 加入购物车 → 提交订单 → 支付成功。"}]},{"type":"paragraph","attrs":{"id":"e45978a9-3f9a-4061-a8ed-45d1c5d0b669","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"以下是一个漏斗分析的SQL实现："}]},{"type":"codeBlock","attrs":{"id":"01de8ec8-5418-448c-8c9a-92acb629c33e","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"WITH funnel_data AS (nSELECTncookie_id,nsession_id,ndt,n-- 定义各阶段的触发条件nMAX(CASE WHEN page_path = '/index' THEN 1 ELSE 0 END) AS step1_home,nMAX(CASE WHEN page_path LIKE '/search/%' THEN 1 ELSE 0 END) AS step2_search,nMAX(CASE WHEN page_path LIKE '/product/%' THEN 1 ELSE 0 END) AS step3_detail,nMAX(CASE WHEN page_path = '/cart' THEN 1 ELSE 0 END) AS step4_cart,nMAX(CASE WHEN page_path = '/checkout' THEN 1 ELSE 0 END) AS step5_checkout,nMAX(CASE WHEN page_path = '/pay/success' THEN 1 ELSE 0 END) AS step6_paynFROM dwd_access_log_dnWHERE dt = '20260115'nAND session_id IS NOT NULLnGROUP BY cookie_id, session_id, dtn)nSELECTnCOUNT(DISTINCT CASE WHEN step1_home = 1 THEN cookie_id END) AS step1_users,nCOUNT(DISTINCT CASE WHEN step2_search = 1 THEN cookie_id END) AS step2_users,nCOUNT(DISTINCT CASE WHEN step3_detail = 1 THEN cookie_id END) AS step3_users,nCOUNT(DISTINCT CASE WHEN step4_cart = 1 THEN cookie_id END) AS step4_users,nCOUNT(DISTINCT CASE WHEN step5_checkout = 1 THEN cookie_id END) AS step5_users,nCOUNT(DISTINCT CASE WHEN step6_pay = 1 THEN cookie_id END) AS step6_users,n-- 计算转化率nROUND(COUNT(DISTINCT CASE WHEN step2_search = 1 THEN cookie_id END) * 100.0 / nNULLIF(COUNT(DISTINCT CASE WHEN step1_home = 1 THEN cookie_id END), 0), 2) AS rate1_2,nROUND(COUNT(DISTINCT CASE WHEN step3_detail = 1 THEN cookie_id END) * 100.0 / nNULLIF(COUNT(DISTINCT CASE WHEN step2_search = 1 THEN cookie_id END), 0), 2) AS rate2_3,nROUND(COUNT(DISTINCT CASE WHEN step4_cart = 1 THEN cookie_id END) * 100.0 / nNULLIF(COUNT(DISTINCT CASE WHEN step3_detail = 1 THEN cookie_id END), 0), 2) AS rate3_4,nROUND(COUNT(DISTINCT CASE WHEN step5_checkout = 1 THEN cookie_id END) * 100.0 / nNULLIF(COUNT(DISTINCT CASE WHEN step4_cart = 1 THEN cookie_id END), 0), 2) AS rate4_5,nROUND(COUNT(DISTINCT CASE WHEN step6_pay = 1 THEN cookie_id END) * 100.0 / nNULLIF(COUNT(DISTINCT CASE WHEN step5_checkout = 1 THEN cookie_id END), 0), 2) AS rate5_6nFROM funnel_data;n"}]},{"type":"paragraph","attrs":{"id":"b55bfb2d-f97e-41d3-b85b-06dd1451addc","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"通过漏斗分析，运营团队可以清晰地看到用户在哪个环节流失最严重，从而有针对性地进行优化。"}]},{"type":"heading","attrs":{"id":"b2b434c7-4746-4f77-ab77-37c435263df5","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"4.3 用户留存计算"}]},{"type":"paragraph","attrs":{"id":"a1b76309-c468-4b44-b4b1-1624c132a52d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"用户留存是衡量产品用户粘性的关键指标。日留存、周留存和月留存是三种最常用的留存计算方式。"}]},{"type":"paragraph","attrs":{"id":"9414194f-b863-4e3d-9723-f752b3d04039","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"计算次日留存率："}]},{"type":"codeBlock","attrs":{"id":"165f4601-7ba3-4ce0-8602-2d15c4b64f2d","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"WITH first_visit AS (nSELECTncookie_id,nMIN(dt) AS first_dtnFROM dwd_access_log_dnWHERE dt BETWEEN '20260101' AND '20260131'nGROUP BY cookie_idn),nreturn_visit AS (nSELECT DISTINCTncookie_id,ndtnFROM dwd_access_log_dnWHERE dt BETWEEN '20260101' AND '20260207'n)nSELECTnf.first_dt,nCOUNT(DISTINCT f.cookie_id) AS new_users,nCOUNT(DISTINCT r.cookie_id) AS retained_users,nROUND(COUNT(DISTINCT r.cookie_id) * 100.0 / nNULLIF(COUNT(DISTINCT f.cookie_id), 0), 2) AS retention_ratenFROM first_visit fnLEFT JOIN return_visit r nON f.cookie_id = r.cookie_id nAND r.dt = DATE_ADD(f.first_dt, 1)nGROUP BY f.first_dtnORDER BY f.first_dt;n"}]},{"type":"heading","attrs":{"id":"306d2338-4ce6-4f25-98d3-ef12d6c5b886","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"4.4 地域分布分析"}]},{"type":"paragraph","attrs":{"id":"7da6d3a0-91ad-4ce0-8042-dab5c8acbff5","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"了解用户的地域分布有助于制定区域化的运营策略和资源配置。通过在DWD层解析IP地址获取地域信息，可以轻松实现地域维度的分析。"}]},{"type":"codeBlock","attrs":{"id":"51b46040-4b51-474f-92df-e1de140eb7c5","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"SELECTnregion,nCOUNT(*) AS pv,nCOUNT(DISTINCT cookie_id) AS uv,nCOUNT(DISTINCT session_id) AS sessionsnFROM dwd_access_log_dnWHERE dt = '20260115'nAND region IS NOT NULLnGROUP BY regionnORDER BY uv DESCnLIMIT 20;n"}]},{"type":"heading","attrs":{"id":"806ec579-2763-4505-9028-4e5d8f9aca3f","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"4.5 用户画像标签开发"}]},{"type":"paragraph","attrs":{"id":"f4354df5-fd56-4feb-b475-55883bda2562","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"用户画像是基于用户历史行为数据构建的用户特征集合。以下是一个简单的用户标签开发示例，计算用户的访问频次、偏好页面类型和活跃时段："}]},{"type":"codeBlock","attrs":{"id":"33923db7-ad6f-4514-b1de-d0cdea162e83","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"CREATE TABLE IF NOT EXISTS ads_user_profile_d ASnSELECTncookie_id,nCOUNT(*) AS total_visits,nCOUNT(DISTINCT dt) AS active_days,n-- 偏好页面类型：出现次数最多的页面一级分类nMAX(page_category) AS preferred_category,n-- 活跃时段：访问次数最多的时段nMAX(hour_slot) AS active_hour,n-- 设备偏好nMAX(device_type) AS preferred_device,n-- 最近访问时间nMAX(access_time) AS last_visit_timenFROM (nSELECTncookie_id,ndt,naccess_time,nSPLIT(page_path, '/')[1] AS page_category,ndevice_type,nCONCAT(CAST(HOUR(access_time) AS STRING), ':00-', n CAST(HOUR(access_time) 1 AS STRING), ':00') AS hour_slot,nROW_NUMBER() OVER (PARTITION BY cookie_id, page_path ORDER BY access_time DESC) AS rnnFROM dwd_access_log_dnWHERE dt BETWEEN '20260101' AND '20260131'n) tnWHERE rn = 1nGROUP BY cookie_id;n"}]},{"type":"heading","attrs":{"id":"26d8fc2c-7014-4eaf-a4e0-fc9aba2d5ba5","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"5. 性能优化最佳实践"}]},{"type":"heading","attrs":{"id":"88427be5-74d2-41bb-9cbf-90cc78f4a4d4","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"5.1 分区设计优化"}]},{"type":"paragraph","attrs":{"id":"b499be61-9b02-4417-877a-5abe56ea5a4e","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"合理的设计分区是MaxCompute性能优化的第一原则。对于网站访问日志，最常用的分区策略是按日期分区。但当日志量极大时，还可以考虑二级分区，例如按日期 小时进行分区，以进一步提升查询效率。"}]},{"type":"paragraph","attrs":{"id":"fe1ba021-9cda-4f20-ac5b-d01f91ced096","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"需要注意，分区列取值不均匀可能导致数据倾斜。如果发现某个分区数据量远大于其他分区，可以考虑调整分区粒度，例如从按天分区调整为按小时分区，或增加随机分桶来平衡数据分布。"}]},{"type":"heading","attrs":{"id":"ecd337fb-16bb-4a28-8c80-ca8ef44ceaad","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"5.2 数据倾斜调优"}]},{"type":"paragraph","attrs":{"id":"42c126f9-0496-4324-9575-e18261da3fce","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"数据倾斜是大数据计算中最常见的性能问题之一，表现为某些Reducer处理的数据量远大于其他Reducer，导致整体任务执行时间被拖长。"}]},{"type":"paragraph","attrs":{"id":"0d0fab77-2b04-4435-a28a-be723cf4c953","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"常见的数据倾斜场景与解决方案："}]},{"type":"bulletList","attrs":{"id":"cff8d8eb-daee-4233-ae92-22dcdfe0212e","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"6a9560a6-9b7f-4cd0-a24a-364d537635a0"},"content":[{"type":"paragraph","attrs":{"id":"3fb28b8e-9de7-4371-bf40-29b989a16432","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"Group By倾斜："},{"type":"text","text":"当某个分组键的值分布极不均匀时（例如某个热门页面的访问量占比极高），可以在Group By之前先进行随机打散，或者使用参数"},{"type":"text","marks":[{"type":"code"}],"text":"set odps.sql.groupby.skewindata=true;"},{"type":"text","text":"开启倾斜优化。"}]}]},{"type":"listItem","attrs":{"id":"283505c5-8e61-4824-bb33-1f9a3110f08b"},"content":[{"type":"paragraph","attrs":{"id":"b9761d0c-a846-437b-8699-e178b2b19191","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"Join倾斜："},{"type":"text","text":"当大表和小表Join时，可以使用Map Join提示"},{"type":"text","marks":[{"type":"code"}],"text":"/* MAPJOIN(small_table) */"},{"type":"text","text":"将小表加载到内存中，避免Shuffle阶段的数据倾斜。"}]}]},{"type":"listItem","attrs":{"id":"b390c002-bcb6-4aa9-828d-451505178354"},"content":[{"type":"paragraph","attrs":{"id":"f3eb792d-41ec-4d67-aec7-25f0188e31be","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"Distinct Count倾斜："},{"type":"text","text":"对于"},{"type":"text","marks":[{"type":"code"}],"text":"COUNT(DISTINCT)"},{"type":"text","text":"操作，如果去重字段的基数很高且分布不均，可以使用"},{"type":"text","marks":[{"type":"code"}],"text":"approx_distinct"},{"type":"text","text":"近似函数替代精确去重，或者分两步计算。"}]}]}]},{"type":"heading","attrs":{"id":"92d637cb-e1ca-4c18-8624-1d12e9fdc9be","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"5.3 长周期指标计算优化"}]},{"type":"paragraph","attrs":{"id":"2130953c-a07c-45ee-aae2-8352a9e72f5e","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"计算近30天、近90天的累计指标时，如果每次都是从明细表中重新计算，性能开销极大。优化思路是构建中间汇总表。"}]},{"type":"paragraph","attrs":{"id":"f2af4d41-3e51-48e5-8bf5-85b37babf52f","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"例如，计算商品近30天的访客数，可以每天构建一个cookie_id粒度的日汇总表，然后基于日汇总表计算30天的累计去重，而不是每次都扫描30天的明细数据："}]},{"type":"codeBlock","attrs":{"id":"5dc3a720-4325-4372-9f19-b5443692ecf9","language":"javascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"javascript"},"content":[{"type":"text","text":"-- 日汇总表：每天每个cookie_id的访问汇总nCREATE TABLE IF NOT EXISTS dws_user_daily_summary_d (ncookie_id STRING,npage_count BIGINT,nsession_count BIGINT,ntotal_bytes BIGINTn) PARTITIONED BY (dt STRING);nn-- 基于日汇总表计算近30天累计UVnSELECTnCOUNT(DISTINCT cookie_id) AS uv_30dnFROM dws_user_daily_summary_dnWHERE dt >= DATE_ADD(CURRENT_DATE(), -30)nAND dt <= CURRENT_DATE();n"}]},{"type":"heading","attrs":{"id":"c738f44f-2d16-421c-aeab-8083ad11c6b9","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"5.4 SQL查询优化技巧"}]},{"type":"bulletList","attrs":{"id":"b178e1bc-ce68-4042-a41c-cc67d78c7950","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"d0ed8412-78cb-428b-91d2-1666ff00b3a5"},"content":[{"type":"paragraph","attrs":{"id":"8c131627-3514-453f-afb3-5a6b801a4c3a","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"先过滤后关联："},{"type":"text","text":"在Join操作前，先通过子查询或CTE对数据进行过滤，减少参与Join的数据量。"}]}]},{"type":"listItem","attrs":{"id":"b42819c3-3d63-4227-97b5-890304bb6cb5"},"content":[{"type":"paragraph","attrs":{"id":"8445df16-6168-4761-bfc3-f84af7b0bb46","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"避免全表扫描："},{"type":"text","text":"始终在WHERE条件中使用分区字段进行裁剪。"}]}]},{"type":"listItem","attrs":{"id":"d31c151f-df9d-4274-9b5a-a94d088fcadb"},"content":[{"type":"paragraph","attrs":{"id":"cc113251-4b2f-4aa1-baae-8d373a9a08c1","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"合理使用索引："},{"type":"text","text":"对于频繁查询的字段，可以考虑创建聚簇索引或使用Clustering Key优化。"}]}]},{"type":"listItem","attrs":{"id":"10e412ab-8dce-4302-a65d-b8ead67a44a9"},"content":[{"type":"paragraph","attrs":{"id":"44712f92-76f3-4a22-98ba-34ed617da182","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"使用CUBE/ROLLUP："},{"type":"text","text":"对于多维度汇总需求，使用GROUP BY CUBE或ROLLUP可以一次计算多个维度的组合汇总，减少重复扫描。"}]}]}]},{"type":"heading","attrs":{"id":"0dab0a82-6686-466f-bffd-0cb5158603b6","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"6. 调度配置与数据质量监控"}]},{"type":"heading","attrs":{"id":"ceace50f-943d-4a37-a84d-aed26351c666","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"6.1 WeData调度配置"}]},{"type":"paragraph","attrs":{"id":"f5be5a9f-5540-41bf-9dd1-3b2ebcff53fa","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"WeData提供了完善的调度能力，支持分钟、小时、天、周、月等多种调度周期。在网站访问数据分析场景中，典型的调度配置如下："}]},{"type":"bulletList","attrs":{"id":"d25867d8-689a-4d7e-a715-cca817c9ab2e","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"4434bf0d-2cea-4967-a391-7b84057a59d4"},"content":[{"type":"paragraph","attrs":{"id":"80d73bb2-2f48-49ad-98a3-8650fd94180d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"ODS层同步任务："},{"type":"text","text":"每天凌晨1点执行，从COS或CLS同步前一天的日志数据。"}]}]},{"type":"listItem","attrs":{"id":"dddcef02-b6e2-437d-90ab-074a0f9a546a"},"content":[{"type":"paragraph","attrs":{"id":"6690ee94-8799-47e2-80ce-d8492db522b5","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"DWD层加工任务："},{"type":"text","text":"每天凌晨2点执行，依赖ODS层任务完成。"}]}]},{"type":"listItem","attrs":{"id":"c1d04c90-2ef8-415b-b1b7-e342b6db643d"},"content":[{"type":"paragraph","attrs":{"id":"04701648-504f-4c98-b7f6-71b408d5d53c","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"DWS层汇总任务："},{"type":"text","text":"每天凌晨3点执行，依赖DWD层任务完成。"}]}]},{"type":"listItem","attrs":{"id":"d1a1a2f6-41b0-4521-94c1-30b9cf94094b"},"content":[{"type":"paragraph","attrs":{"id":"089937cd-e614-4aa2-9f9c-7dbef4d1da8d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"ADS层报表任务："},{"type":"text","text":"每天凌晨4点执行，依赖DWS层任务完成。"}]}]}]},{"type":"paragraph","attrs":{"id":"f026e571-976a-46d5-b692-8de3191c5691","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"通过配置任务间的上下游依赖关系，可以确保数据按照正确的顺序流转。同时，WeData支持设置任务的生效时间范围，超过有效期后任务将不再自动调度。"}]},{"type":"heading","attrs":{"id":"f6dae713-aba6-4b84-9dd3-e62bed157a53","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"6.2 数据质量监控"}]},{"type":"paragraph","attrs":{"id":"cc3f9bc9-4f33-4029-b03c-b729f3eec0f9","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"数据质量是数据分析的生命线。在WeData中，可以通过配置数据质量规则来监控数据的完整性、准确性和及时性："}]},{"type":"bulletList","attrs":{"id":"f6011fca-9085-4603-a6d5-6a216c4aab2a","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"6ea4e323-dde3-416f-8f5d-811b07136f33"},"content":[{"type":"paragraph","attrs":{"id":"97df17c0-4904-4435-b2aa-4edbdfa3dec9","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"完整性检查："},{"type":"text","text":"检查ODS层数据是否完整同步，例如对比源端记录数和目标端记录数。"}]}]},{"type":"listItem","attrs":{"id":"e1e2517a-8ebd-4902-a186-a9ba6b24da2d"},"content":[{"type":"paragraph","attrs":{"id":"aa89bebc-ff44-4876-833a-fb6008d3426e","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"一致性检查："},{"type":"text","text":"检查DWD层加工后的关键字段是否存在空值或异常值。"}]}]},{"type":"listItem","attrs":{"id":"0b6fad92-5023-4bcf-a867-94623e6863bd"},"content":[{"type":"paragraph","attrs":{"id":"3956c351-bedd-4c89-9b5d-f904cd3a7ef7","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"趋势监控："},{"type":"text","text":"监控每日PV、UV等核心指标的环比变化，发现异常波动时及时告警。"}]}]}]},{"type":"heading","attrs":{"id":"09add319-af74-4f47-846e-343b82755d3a","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"7. 可视化展示"}]},{"type":"paragraph","attrs":{"id":"0ac156ba-2d66-491f-939d-b2ee911efadf","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"完成数据计算后，需要通过可视化的方式将分析结果呈现给业务人员。MaxCompute可以与多种BI工具无缝集成："}]},{"type":"bulletList","attrs":{"id":"9145f4ba-78c1-488d-867d-08eec74f3989","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"97222cf0-ccdd-49de-9e3a-4e547b9c3120"},"content":[{"type":"paragraph","attrs":{"id":"7e837f0d-a521-466a-a1b5-2bcf112e35f2","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"Quick BI："},{"type":"text","text":"腾讯云自研的BI工具，支持直接连接MaxCompute数据源，拖拽式生成报表和仪表盘。"}]}]},{"type":"listItem","attrs":{"id":"02fcd15a-3a03-4fb7-bc98-07b285017efb"},"content":[{"type":"paragraph","attrs":{"id":"26466368-caed-4db0-932f-2f9d3f07258f","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"DataV："},{"type":"text","text":"适用于大屏展示场景，支持丰富的可视化组件。"}]}]},{"type":"listItem","attrs":{"id":"8d34e872-9f56-48f7-9a49-5f943a7e16e1"},"content":[{"type":"paragraph","attrs":{"id":"f85c8e89-7595-4414-9bb1-47ccbf61691a","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"第三方BI工具："},{"type":"text","text":"通过JDBC/ODBC接口，MaxCompute可以与Tableau、Power BI等主流BI工具对接。"}]}]}]},{"type":"paragraph","attrs":{"id":"695c2f28-929c-4ba7-a423-7cb35cf1b437","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"典型的网站运营仪表盘可以包含以下图表：每日PV/UV趋势图、TOP页面排行榜、用户地域分布地图、漏斗转化图、用户留存曲线等。"}]},{"type":"heading","attrs":{"id":"e188c3c8-d3de-49e4-b075-b6f60a80b062","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"8. 成本优化策略"}]},{"type":"heading","attrs":{"id":"7d59aaf0-f6e2-4a1b-8545-f79e3883201c","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"8.1 存储成本优化"}]},{"type":"paragraph","attrs":{"id":"8cc561fc-8786-470c-8b18-43640b6af757","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"MaxCompute的存储费用按实际存储量计费。为控制存储成本，可以采取以下措施："}]},{"type":"bulletList","attrs":{"id":"c9e19fce-2c74-47e6-b565-667befab3c2e","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"8264df0f-e2d2-48d0-ab08-a148e97342c7"},"content":[{"type":"paragraph","attrs":{"id":"0d24d46c-205c-4a00-acd0-05704f50c26a","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"生命周期管理："},{"type":"text","text":"为ODS层和DWD层的明细表设置生命周期，例如保留最近90天的数据，超过90天的历史数据自动删除或归档到低频存储。"}]}]},{"type":"listItem","attrs":{"id":"b04f7938-2831-43fc-9ba7-e2f7893f9d6e"},"content":[{"type":"paragraph","attrs":{"id":"401f4efe-4ca0-49b0-a66f-07e8579de0ba","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"数据压缩："},{"type":"text","text":"MaxCompute默认对数据进行列式存储和压缩，但在建表时可以指定更高效的压缩算法。"}]}]},{"type":"listItem","attrs":{"id":"87a5c929-c0b0-4095-9b1f-081e245c4d51"},"content":[{"type":"paragraph","attrs":{"id":"b88d4059-88e7-4787-aca9-f3bba80a66db","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"合理分层："},{"type":"text","text":"将明细数据存储在ODS/DWD层，汇总数据存储在DWS/ADS层，避免在DWS/ADS层存储大量明细数据。"}]}]}]},{"type":"heading","attrs":{"id":"a61ae583-797e-4a93-be09-701be9e7c364","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"8.2 计算成本优化"}]},{"type":"paragraph","attrs":{"id":"bcafc1f1-268b-41bd-bb5a-5e06b6db73e1","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"MaxCompute的计算费用按实际使用的计算资源量计费。优化计算成本的核心思路是减少不必要的计算："}]},{"type":"bulletList","attrs":{"id":"37bd9435-9324-474e-9292-761e53e97698","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"0c8fefe7-8ce8-4646-9976-033b204d0680"},"content":[{"type":"paragraph","attrs":{"id":"f9d65c41-80eb-415e-9344-014b7595fd95","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"分区裁剪："},{"type":"text","text":"在查询时始终使用分区字段进行过滤，避免全表扫描。"}]}]},{"type":"listItem","attrs":{"id":"27fa2275-b620-44e3-9ba7-835776cb9dd4"},"content":[{"type":"paragraph","attrs":{"id":"51e7cad3-44bf-4aa9-a06b-99d24df8f948","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"列裁剪："},{"type":"text","text":"只SELECT需要的字段，避免SELECT *。"}]}]},{"type":"listItem","attrs":{"id":"41275714-00c4-4263-9b35-a4e2b0aac994"},"content":[{"type":"paragraph","attrs":{"id":"21b391e7-6cec-4876-a691-47d985ebcf59","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"预聚合："},{"type":"text","text":"在DWS层进行充分的预聚合，减少ADS层查询时的计算量。"}]}]},{"type":"listItem","attrs":{"id":"c2f67a17-39e9-4e6c-80d3-0ba1e74f822d"},"content":[{"type":"paragraph","attrs":{"id":"f92ca6f5-e976-4241-b3a9-e16a09d395dc","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"错峰调度："},{"type":"text","text":"将计算密集型任务安排在业务低峰期执行，利用闲时计算资源。"}]}]}]},{"type":"heading","attrs":{"id":"1278e67b-3b25-474b-b146-81a2d6179c51","textAlign":"inherit","indent":0,"level":3,"isHoverDragHandle":false},"content":[{"type":"text","text":"8.3 内网免流量"}]},{"type":"paragraph","attrs":{"id":"08af3e31-9c7c-4236-b445-f3a0a2bd96c1","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"如果网站日志存储在COS，而MaxCompute与COS在同一个地域，数据读取走内网通道可以免收流量费用。在配置数据同步任务时，务必确认数据源和目标端在同一地域。"}]},{"type":"heading","attrs":{"id":"74237971-d12f-4f44-b6a2-8c680adb193a","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"9. 总结"}]},{"type":"paragraph","attrs":{"id":"ab70d92f-8736-4a2f-b353-ec627ff0b003","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"本文系统讲解了如何利用腾讯云MaxCompute对网站用户访问数据进行全链路分析。从环境准备、数据集成、数仓分层建模，到PV/UV统计、漏斗转化分析、用户留存计算、地域分布分析等核心分析场景，再到数据倾斜调优、长周期指标优化、分区设计等性能最佳实践，以及调度配置、数据质量监控、可视化展示和成本优化，形成了一套完整的可落地的技术方案。"}]},{"type":"paragraph","attrs":{"id":"dabd6925-90da-416a-9cc7-258e487de0f5","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"通过MaxCompute WeData的组合，数据开发工程师和数据分析师可以快速搭建起企业级的网站用户行为分析平台，将海量的日志数据转化为驱动业务决策的数据洞察。"}]},{"type":"heading","attrs":{"id":"c8154e6e-76fd-4853-bf7d-4069192369fa","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"常见问题解答"}]},{"type":"paragraph","attrs":{"id":"2e1b6107-45f0-4254-81b0-f17f78674442","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"问1：MaxCompute和传统数据库有什么区别？"}]},{"type":"paragraph","attrs":{"id":"0c7439fa-0372-463c-b7cc-ebfe8afe4553","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"答：MaxCompute是云原生的大数据计算服务，专为海量数据场景设计，支持EB级数据存储和PB级数据计算。与传统数据库相比，MaxCompute采用分布式架构，计算和存储分离，支持按量付费的弹性计费模式，适合处理TB级别以上的网站访问日志数据。"}]},{"type":"paragraph","attrs":{"id":"330e325a-65ad-4e1d-aae6-1508d1eeb773","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"问2：网站访问日志数据量很大，同步到MaxCompute需要多长时间？"}]},{"type":"paragraph","attrs":{"id":"b84d3361-7c8a-4de8-82d1-33243a746eff","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"答：同步时间取决于数据量大小和网络带宽。对于日均数十GB的日志数据，使用WeData的离线同步功能通常在1-2小时内可以完成同步。如果使用CLS直接投递到MaxCompute，可以实现准实时同步，延迟在分钟级别。"}]},{"type":"paragraph","attrs":{"id":"64d70dc4-ae90-49f4-a65e-4f8a26357e52","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"问3：MaxCompute的SQL和标准SQL有什么差异？"}]},{"type":"paragraph","attrs":{"id":"e0ca41ad-5206-4981-9aef-594ee45d5ffa","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"答：MaxCompute SQL兼容大部分标准SQL语法，但也有一些差异。例如，MaxCompute不支持事务操作、不支持UPDATE和DELETE（可通过INSERT OVERWRITE实现数据更新）、对某些函数的行为可能与标准SQL略有不同。建议在开发前仔细阅读MaxCompute SQL官方文档。"}]},{"type":"paragraph","attrs":{"id":"94311a98-4732-4aa5-ac14-64e995860da3","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"问4：如何处理数据倾斜问题？"}]},{"type":"paragraph","attrs":{"id":"900088a0-47bf-4d43-84d1-99514ab6756f","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"答：数据倾斜的常见解决方案包括：开启倾斜优化参数"},{"type":"text","marks":[{"type":"code"}],"text":"set odps.sql.groupby.skewindata=true;"},{"type":"text","text":"、使用Map Join优化小表与大表的关联、对倾斜键进行加盐打散后二次聚合、合理设计分区键避免数据分布不均等。具体采用哪种方案需要根据实际的倾斜场景来判断。"}]},{"type":"paragraph","attrs":{"id":"9a797bb7-cf28-4e26-ad69-21a04ecede2d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"问5：网站访问日志分析一般需要保留多长时间的数据？"}]},{"type":"paragraph","attrs":{"id":"f19bf780-e69e-42db-a9c1-18b1ff9bcefb","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"答：这取决于业务需求。通常，ODS层的原始日志建议保留30-90天，以便进行问题回溯和数据校验；DWD层的明细数据建议保留90-180天；DWS层的汇总数据可以根据需要保留更长时间（如1-3年），用于长期趋势分析。可以通过配置表的生命周期来自动管理数据的保留和清理。"}]},{"type":"paragraph","attrs":{"id":"ead8a103-5528-4895-9e64-c8d871c1b86d","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"问6：MaxCompute的费用构成是怎样的？如何控制成本？"}]},{"type":"paragraph","attrs":{"id":"670c5e1a-c975-4982-ad87-e2dd3fe6fb95","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"答：MaxCompute的费用主要包括存储费用和计算费用两部分。存储费用按实际存储量计费，计算费用按实际使用的计算资源量计费。控制成本的核心方法包括：合理设计分区和生命周期管理、在查询时进行分区裁剪和列裁剪、在DWS层进行充分的预聚合减少重复计算、将计算任务安排在业务低峰期执行等。"}]}]}","createTime":1783404076,"ext":{"closeTextLink":0,"comment_ban":0,"description":"","focusRead":0},"favNum":0,"html":"","isOriginal":0,"likeNum":0,