数仓ADS层指标计算案例分享

ADS层数据往往是最终的结果指标数据，在大屏展示，或者实时流处理时候使用，通过下面两个例子来练习业务大屏展示sql该怎么写。

1. 会员分析案例

1.1 数据准备

表结构如下，其中此表是dws层以天为维度的会员表，比如每天的会员信息汇总，

use dws;
drop table if exists dws.dws_member_start_day;
create table dws.dws_member_start_day(
`device_id` string, -- 设备id，来区分用户
`uid` string, -- uid
`app_v` string,
`os_type` string,
`language` string,
`channel` string,
`area` string,
`brand` string
) COMMENT '会员日启动汇总'
partitioned by(dt string)
stored as parquet;

1.2 会员指标计算

沉默会员的定义：只在安装当天启动过App，而且安装时间是在7天前；
流失会员的定义：最近30天未登录的会员

1.2.1 如何计算沉默会员数

-- 关键点在于只启动一次
-- 拿到只启动一次的会员，后面再过滤安装时间是再7天前的，使用sum窗口函数
SELECT count(*)
FROM(SELECT device_id,sum(device_id) OVER (PARTITION BY device_id) AS sum_num,dtFROM dws.dws_member_start_day) tmp
WHERE dt <= date_add(CURRENT_DATE, -7)AND sum_num=1

1.2.2 如何计算流失会员数

-- 关键点在于最近一次登录
-- 拿到会员最近一次登录时间，并用row_number来过滤
SELECT count(*)
FROM(SELECT device_id,dt,row_number() OVER (PARTITION BY device_idORDER BY dt DESC) roFROM dws.dws_member_start_day) tmp
WHERE ro=1AND dt >= date_add(CURRENT_DATE, -30)

2. 核心交易案例

2.1 数据准备

给定一个每日订单维度表，表结构如下：

-- 订单明细
DROP TABLE IF EXISTS dwd.dwd_trade_orders;
create table dwd.dwd_trade_orders(
`orderId`    int,
`orderNo`   string,
`userId`    bigint,
`status`    tinyint,
`productMoney` decimal,
`totalMoney`  decimal,
`payMethod`   tinyint,
`isPay`     tinyint,
`areaId`    int,
`tradeSrc`   tinyint,
`tradeType`   int,
`isRefund`   tinyint,
`dataFlag`   tinyint,
`createTime`  string,
`payTime`   string,
`modifiedTime` string,
`start_date`  string,
`end_date`   string
) COMMENT '订单事实拉链表'
partitioned by (dt string)
STORED AS PARQUET;

其中，订单状态 -3 用户拒收；-2未付款的订单；-1用户取消；0 待发货；1配送中；2用户确认收货，订单有效标志；-1 删除；1 有效。

数据预处理，在明细事实拉链表处理时不太方便，可以做一张中间表，dws_trade_orders_day 其表结构和加工如下：

DROP TABLE IF EXISTS dws.dws_trade_orders_day;CREATE TABLE IF NOT EXISTS dws.dws_trade_orders_day(day_dt string COMMENT '日期：yyyy-MM-dd',day_cnt decimal commnet '日订单笔数',day_sum decimal COMMENT '日订单总额') COMMENT '日订单统计表';SELECT dt,count(*) cnt,sum(totalMoney) sm
FROM(SELECT DISTINCT orderid,dt,totalMoneyFROM dwd.dwd_trade_ordersWHERE status >= 0AND dataFlag = '1') tmp
GROUP BY dt;INSERT OVERWRITE TABLE dws.dws_trade_orders_day
SELECT dt,count(*) cnt,sum(totalMoney) sm
FROM(SELECT DISTINCT orderid,dt,totalMoneyFROM dwd.dwd_trade_ordersWHERE status >= 0AND dataFlag = '1') tmp
GROUP BY dt;SELECT *
FROM dws.dws_trade_orders_day
WHERE day_dt BETWEEN '2020-01-01' AND '2020-12-31';

2.2 指标1，统计2020年每个季度的销售订单笔数、订单总额

先创建ads指标表：dws_trade_orders_quarter

DROP TABLE IF EXISTS dws.dws_trade_orders_quarter;CREATE TABLE IF NOT EXISTS dws.dws_trade_orders_quarter(YEAR string COMMENT '年份',QUARTER string COMMENT '季度',cnt decimal COMMENT '订单总笔数',SUM decimal COMMENT '订单总额') COMMENT '季度订单统计表';INSERT OVERWRITE TABLE dws.dws_trade_orders_quarter WITH tmp AS(SELECT substr(day_dt, 0, 4) YEAR,CASE WHEN substr(dat_dt, 6, 2)="01"OR substr(dat_dt, 6, 2)="02"OR substr(day_dt, 6, 2)="03" THEN "1" WHEN substr(dat_dt, 6, 2)="04"OR substr(dat_dt, 6, 2)="05"OR substr(day_dt, 6, 2)="06" THEN "2" WHEN substr(dat_dt, 6, 2)="07"OR substr(dat_dt, 6, 2)="08"OR substr(day_dt, 6, 2)="09" THEN "3" WHEN substr(dat_dt, 6, 2)="10"OR substr(dat_dt, 6, 2)="11"OR substr(day_dt, 6, 2)="12" THEN "4" AS QUARTER day_cnt,day_sumFROM dws.dws_trade_orders_day)
SELECT YEAR,QUARTER,sum(day_cnt),sum(day_sum)
FROM tmp
GROUP BY YEAR QUARTER;

2.3 统计2020年每个月的销售订单笔数、订单总额

先创建ads指标表：dws_trade_orders_month

DROP TABLE IF EXISTS dws.dws_trade_orders_month;CREATE TABLE IF NOT EXISTS dws.dws_trade_orders_month(yearstring COMMENT '年份',MONTH string COMMENT '月份',month_cnt decimal COMMENT '月订单总笔数',month_sum decimal COMMENT '月订单总额') COMMENT '月订单统计表';INSERT OVERWRITE TABLE dws.dws_trade_orders_month WITH tmp AS(SELECT substr(day_dt, 0, 4) YEAR,sunstr(day_dt, 6, 2) MONTH,day_cnt,day_sumFROM dws.dws_trade_orders_day)
SELECT YEAR,MONTH,sum(day_cnt) month_cnt,sum(day_sum) month_sum
FROM tmp
GROUP BY YEAR,MONTH;

2.4 统计2020年每周（周一到周日）的销售订单笔数、订单总额

创建ads层指标表：dws_trade_orders_week利用到日期函数```weekofyear

DROP TABLE IF EXISTS dws.dws_trade_orders_week;
CREATE TABLE IF NOT EXISTS dws.dws_trade_orders_week(YEAR string COMMENT '年份',WEEK string COMMENT '一年中的第几周',week_cnt decimal COMMENT '周订单总笔数',week_sum decimal COMMENT '周订单总额') COMMENT '周订单统计表';INSERT OVERWRITE TABLE dws.dws_trade_orders_week
SELECT substr(day_dt, 0, 4) YEAR,weekofyear(day_dt) WEEK,sum(day_cnt),sum(day_sum)
FROM dws.dws_trade_orders_day
GROUP BY substr(day_dt, 0, 4) YEAR,weekofyear(day_dt) WEEK;

2.5 统计2020年国家法定节假日、休息日、工作日的订单笔数、订单总额

创建日期信息维表：dim_day_info 并录入节假日信息数据（数据每年都不一样，需要国务院通知的公告，所以定期手动维护）

drop table if exists dim.dim_day_info;
create table if not exists dim.dim_day_info(day_dt string comment '日期',is_holidays int comment '节假日标识：0不是 1是',is_workday int comment '工作日标识 0不是 1是'
) comment '日期信息表';-- 统计2020节假日的订单笔数，订单总额SELECT nvl(sum(day_cnt), 0) nvl(sum(day_sum), 0)
FROM dws.dws_trade_orders_day A
LEFT JOIN dim.dim_day_info B ON A.day_dt = B.day_dt
WHERE B.is_holiday = 1;-- 统计2020年休息日的订单笔数，订单总额SELECT nvl(sum(day_cnt), 0) nvl(sum(day_sum), 0)
FROM dws.dws_trade_orders_day A
LEFT JOIN dim.dim_day_info B ON A.day_dt = B.day_dt
WHERE B.is_workday = 0;-- 统计2020节工作日的订单笔数，订单总额SELECT nvl(sum(day_cnt), 0) nvl(sum(day_sum), 0)
FROM dws.dws_trade_orders_day A
LEFT JOIN dim.dim_day_info B ON A.day_dt = B.day_dt
WHERE B.is_workday = 1;

作者：Hoult丶吴邪

来源：大数据真有意思