数仓学习---数仓开发之ADS层

本文涉及的产品
实时数仓Hologres,5000CU*H 100GB 3个月
智能开放搜索 OpenSearch行业算法版,1GB 20LCU 1个月
实时计算 Flink 版,5000CU*H 3个月
简介: 数仓学习---数仓开发之ADS层

image.png

ADS层:Application Data Service

image.png

image.png

1.ADS层保存的数据是最终的统计结果,无需做进一步的计算
  --不需要列式存储,也不需要snaapy压缩
2.统计结果的目的是对外提供服务,所以表不会最终数据的存储位置
   --需要将表中的数据同步到第三方存储(MySQL)
   --ADS层的表最好是行式存储:tsv(DataX)
   --压缩格式采用gzip
3.统计结果的数据量不会太多,ADS层的表无需分区设计
4.表的设计
   --ODS层:表的结构依托于数据源的数据结构
   --DIM层:遵循维度建模的维度表的设计理念(维度越丰富越好)
   --DWD层:遵循维度模型的事实表的设计概念(粒度越细越好)
   --ADS层:客户要啥加啥

基础概念

维度:分析数据的角度
颗粒度:分析数据的详细程度
统计周期:统计的时候,数据统计时间范围(窗口)
统计粒度:分析数据的具体角度,称为统计粒度(站在哪一个角度统计数据)
指标:客户想要的一个结果数值
--2022-06-08(最近一周):02,03,04,05,06,07,08
--2022-06-09(最近一周):03,04,05,06,07,08,09
--2022-06-10(最近一周):04,05,06,07,08,09,10(类似于spark的滑动窗口)

各品牌商品下单统计

--统计行为:下单
--分析的角度:品牌
--指标:下单数量,下单人数
--建表语句
  --统计时间:以获取数据的那一天为准
ROP TABLE IF EXISTS ads_order_stats_by_tm;
CREATE EXTERNAL TABLE ads_order_stats_by_tm
(
    `dt`                      STRING COMMENT '统计日期',
    `recent_days`             BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `tm_id`                   STRING COMMENT '品牌ID',
    `tm_name`                 STRING COMMENT '品牌名称',
    `order_count`             BIGINT COMMENT '下单数',
    `order_user_count`        BIGINT COMMENT '下单人数'
) COMMENT '各品牌商品下单统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_order_stats_by_tm/';

数据装载

--统计指标分析
   --将复杂的指标的分解成小的,简单的指标
   --指标分析
     --1.原子指标(最简单的统计值)
         --业务过程(行为)+   度量值+           聚合逻辑
            下单     +      (order_id)次数+count()
     --2.派生指标
         --基于原子指标,增加其他的条件,角度之类
          --派生指标=原子指标+统计周期(最近一天)+业务限定+统计粒度
          --统计周期和业务限定其实都是数据筛选条件,但是不一样
              --统计周期一般指的就是数据时间范围(分区字段的过滤,过滤文件夹)
              --业务限定一般指的就是数据约束条件(数据字段的过滤,过滤文件)
select
  count(distanct order_id)
from dwd_trade_order_detail_inc


select
  count(distanct order_id)
from dwd_trade_order_detail_inc
where dt>data_sub('2022-06-08',0) and dt<='2022-06-08'
and 品牌='华为'
group byu 品牌
select
  count(distanct order_id)
from dwd_trade_order_detail_inc
where dt>data_sub('2022-06-08',6) and dt<='2022-06-08'

派生指标

派生指标=原子指标+统计周期+业务限定+统计粒度

最近一天各省份手机品类订单总额=订单总额+最近一天+品类为手机+省份

                                                     定义业务过程、度量值和聚合逻辑        

                                                     限定统计范围,相当与SQL中的where

                                                     定义统计粒度,相当于SQL中Group by

--最近1天各个品牌下单表,下单人数统计
   --统计周期其实就是数据的时间范围,一般会在where字句中添加dt条件
   --业务限定其实就是数据约束,一般会在where子句中添加业务条件
   --统计粒度其实就是数据的分析维度,一般会在group by 子句中添加维度字段
-- 当数据中的Hash表结构为空时抛出类型转换异常,禁用相应优化即可
set hive.mapjoin.optimized.hashtable=false;
insert overwrite table ads_sku_cart_num_top3_by_cate
select * from ads_sku_cart_num_top3_by_cate
union
select
    '2022-06-08' dt,
    category1_id,
    category1_name,
    category2_id,
    category2_name,
    category3_id,
    category3_name,
    sku_id,
    sku_name,
    cart_num,
    rk
from
(
    select
        sku_id,
        sku_name,
        category1_id,
        category1_name,
        category2_id,
        category2_name,
        category3_id,
        category3_name,
        cart_num,
        rank() over (partition by category1_id,category2_id,category3_id order by cart_num desc) rk
    from
    (
        select
            sku_id,
            sum(sku_num) cart_num
        from dwd_trade_cart_full
        where dt='2022-06-08'
        group by sku_id
    )cart
    left join
    (
        select
            id,
            sku_name,
            category1_id,
            category1_name,
            category2_id,
            category2_name,
            category3_id,
            category3_name
        from dim_sku_full
        where dt='2022-06-08'
    )sku
    on cart.sku_id=sku.id
)t1
where rk<=3;
-- 优化项不应一直禁用,受影响的SQL执行完毕后打开
set hive.mapjoin.optimized.hashtable=true;

各品类商品下单统计


统计周期

统计粒度

指标

说明

最近1、7、30

品类

下单数

最近1、7、30

品类

下单人数

DROP TABLE IF EXISTS ads_order_stats_by_cate;
CREATE EXTERNAL TABLE ads_order_stats_by_cate
(
    `dt`                      STRING COMMENT '统计日期',
    `recent_days`             BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `category1_id`            STRING COMMENT '一级品类ID',
    `category1_name`          STRING COMMENT '一级品类名称',
    `category2_id`            STRING COMMENT '二级品类ID',
    `category2_name`          STRING COMMENT '二级品类名称',
    `category3_id`            STRING COMMENT '三级品类ID',
    `category3_name`          STRING COMMENT '三级品类名称',
    `order_count`             BIGINT COMMENT '下单数',
    `order_user_count`        BIGINT COMMENT '下单人数'
) COMMENT '各品类商品下单统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_order_stats_by_cate/';


insert overwrite table ads_order_stats_by_cate
select * from ads_order_stats_by_cate
union
select
*
from
(--最近一天
select
  '2022-06-08',
  1,
 `category1_id`           ,
    `category1_name`         ,
    `category2_id`           ,
    `category2_name`         ,
    `category3_id`           ,
    `category3_name`         ,

count(distinct order_id) as     order_count   ,
count(distinct user_id) as order_user_count
from(
  select
  order_id,
  user_id,
  sku_id
  from
  dwd_trade_order_detail_inc
  where dt='2022-06-08'
)od
left join(
  select
     category1_id,
     category2_id,
     category3_id,
  category1_name,
  category2_name,
  category3_name,
          
  from
  dim_sku_full
  where dt='2022-06-08'
  
)sku on od.sku_id=sku.id

group by `category1_id`           ,
    `category1_name`         ,
    `category2_id`           ,
    `category2_name`         ,
    `category3_id`           ,
    `category3_name`         
union all
--最近七天
select
  '2022-06-08',
  7,
 `category1_id`           ,
    `category1_name`         ,
    `category2_id`           ,
    `category2_name`         ,
    `category3_id`           ,
    `category3_name`         ,

count(distinct order_id) as     order_count   ,
count(distinct user_id) as order_user_count
from(
  select
  order_id,
  user_id,
  sku_id
  from
  dwd_trade_order_detail_inc
  where dt>=data_sub('2022-06-08',6) and dt<='2022-06-08'
)od
left join(
  select
     category1_id,
     category2_id,
     category3_id,
  category1_name,
  category2_name,
  category3_name,
          
  from
  dim_sku_full
  where dt='2022-06-08'   #维度表不用改,我们没有逻辑删除,数据不会少
  
)sku on od.sku_id=sku.id

group by `category1_id`           ,
    `category1_name`         ,
    `category2_id`           ,
    `category2_name`         ,
    `category3_id`           ,
    `category3_name` 
union all
--最近30天
select
  '2022-06-08',
  30,
 `category1_id`           ,
    `category1_name`         ,
    `category2_id`           ,
    `category2_name`         ,
    `category3_id`           ,
    `category3_name`         ,

count(distinct order_id) as     order_count   ,
count(distinct user_id) as order_user_count
from(
  select
  order_id,
  user_id,
  sku_id
  from
  dwd_trade_order_detail_inc
  where dt>=data_sub('2022-06-08',29) and dt<='2022-06-08'
)od
left join(
  select
     category1_id,
     category2_id,
     category3_id,
  category1_name,
  category2_name,
  category3_name,
          
  from
  dim_sku_full
  where dt='2022-06-08'   #维度表不用改,我们没有逻辑删除,数据不会少
  
)sku on od.sku_id=sku.id

group by `category1_id`           ,
    `category1_name`         ,
    `category2_id`           ,
    `category2_name`         ,
    `category3_id`           ,
    `category3_name` )t

性能如果提升?

--减少join

--可以减少数据量

--减少重复量

--优化思路

--将最近1天的数据保存在一张表中

 --最近7天和最近30天数据从最近1天的统计表中获取数据,进行进一步计算

通过预聚合中取数据

--ODS(x)

--DIM(x)

--DWD(x)

--DWS(OK)

--预聚合,保存中间计算结果

--中间计算结果表不是最终结果表

--需要进一步的计算

     --存储方式:列式存储

--压缩方式:snappy

--ADS(x)

DROP TABLE IF EXISTS dws_order_stats_by_cate;
CREATE EXTERNAL TABLE dws_order_stats_by_tm_1d
(
  
    `tm_id`,                  STRING COMMENT '品牌ID',
    `tm_name`                STRING COMMENT '品牌名称',
    `order_count`             BIGINT COMMENT '下单数',
    `order_user_count`        BIGINT COMMENT '下单人数'
) COMMENT '各品类商品下单统计'
  partition by(`dt` String)
  stored as orc
  location '/warehourse/gmall/dws/dws_order_stats_by_tm_1d/'
  tblproperties('orc.compress'='snappy');

--将最近1天的统计结果保存在1d表中
insert overwriter table dws_order_stats_by_tm_1d partition(dt='2022-06-08')
select
`tm_id`,
`tm_name`,

count(distinct order_id) as     `order_count_1d`   ,
count(distinct user_id) as `order_user_count`
from(
  select
  order_id,
  user_id,
  sku_id
  from
  dwd_trade_order_detail_inc
  where dt='2022-06-08'
)od
left join(
  select
    id,tm_id,tm_name
          
  from
  dim_sku_full
  where dt='2022-06-08'
  
)sku on od.sku_id=sku.id

group by tm_id,
tm_name;
--从1d表获取最近1天,最近7天,最近30天数据,保存到ADS层的表中
insert overwrite table ads_order_stats_by_tm
select * from ads_order_status_by_tm
union 
select *
from(--最近1天
select
  '2022-06-08',
  1,
  tm_id,
  tm_name,
  order_count_1d,
  order_user_count_1d

from dws_order_stats_by_tm_1d
where dt='2022-06-08'
union all
--最近七天
select
  '2022-06-08',
  7,
  tm_id,
  tm_name,
  sum(order_count_1d),
  sum(order_user_count_1d)
from dws_order_stats_by_tm_1d
where dt>=date_sub('2022-06-08',6) and dt<='2022-06-08'
group by tm_id,
  tm_name
union all
--最近30天
select
  '2022-06-08',
  7,
  tm_id,
  tm_name,
  sum(order_count_1d),
  sum(order_user_count_1d)
from dws_order_stats_by_tm_1d
where dt>=date_sub('2022-06-08',29) and dt<='2022-06-08'
group by tm_id,
  tm_name

)t

ADS层-各品类商品下单统计--SQL优化

--当前的问题:读了三次相同的数据,使用三个不同的场景

--解决思路:读了一次数据,使用三个场景(炸裂)

炸裂【1,2,3】=》【1,2,3】,【1,2,3】,【1,2,3】

扁平化【1,2,3】=》1,2,3

select * from (select 'a'  name) t lateral view explode(array(1,2,3)) tmp as name1

where name1!=7

1.获取时间范围最大的数据集

2.将查询的数据集在内存中进行炸裂操作(三份),变成多份,增加标记

3.将炸裂后的数据进行筛选,保存有效数据

4.将过滤的后的数据按照标记进行分组,然后进行统计

insert overwrite table ads_order_stats_by_tm
select * from ads_order_status_by_tm
union 
select
  '2022-06-08',
  days,
  tm_id,
  tm_name,
  sum(order_count_1d),
  sum(order_user_count_1d)
from dws_order_stats_by_tm_1d lateral view explode(`array`(1,7,30)) tmp as days
where dt>=date_sub('2022-06-08',days-1) and dt<='2022-06-08'
group by days,tm_id,
  tm_name

新增下单用户统计

需求说明如下。

统计周期

指标

说明

最近1、7、30

新增下单人数

以前这个用户没有下过订单,最近1天,第一回下订单,称为新增用户下单用户

DROP TABLE IF EXISTS ads_new_order_user_stats;
CREATE EXTERNAL TABLE ads_new_order_user_stats
(
    `dt`                   STRING COMMENT '统计日期',
    `recent_days`          BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `new_order_user_count` BIGINT COMMENT '新增下单人数'
) COMMENT '新增下单用户统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_new_order_user_stats/';
--新增:以前没做过,今天第一回
  --判断1天前用户没有出现过
 insert overwrite table ads_new_order_user_stats
 select * from ads_order_new_order_user_stats
 union
 select
 *from(select 
   '2022-06-08',
   1,
  count(distinct user_id)
from dwd_trade_order_detail_inc
where dt='2022-06-08'
and user_id not in(
  select user_id
  from dwd_trade_order_detail_inc
  where dt<'2022-06-08'
)
union all
select 
   '2022-06-08',
   7,
  count(distinct user_id)
from dwd_trade_order_detail_inc
where dt>=date_sub('2022-06-08',6) and dt<='2022-06-08'
and user_id not in(
  select user_id
  from dwd_trade_order_detail_inc
  where dt<date_sub('2022-06-07',6) 
)
union all
select 
'2022-06-08',
  30
  count(distinct user_id)
from dwd_trade_order_detail_inc
where dt>=date_sub('2022-06-08',29)  and dt<='2022-06-08'
and user_id not in(
  select user_id
  from dwd_trade_order_detail_inc
  where dt<date_sub('2022-06-07',29)
))t

--需求:统计最近1天,7天,30天新增注册用户

DROP TABLE IF EXISTS ads_new_reg_user_stats;
CREATE EXTERNAL TABLE ads_new_order_user_stats
(
    `dt`                   STRING COMMENT '统计日期',
    `recent_days`          BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `new_reg_user_count` BIGINT COMMENT '新增注册人数'
) COMMENT '新增下单用户统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_new_reg_user_stats/';
insert overwrite table ads_new_new_user_status
select * from ads_new_reg_user_stats
union
select *from(
select
 '2022-06-08',
 1,
 count(*)
 from dwd_user_register_inc
 where dt>='2022-06-08'
 union all
 select
 '2022-06-08',
 7,
 count(*)
 from dwd_user_register_inc
 where date_sub(dt>='2022-06-08',6) and dt<=2022-06-08'
 union all
 select
 '2022-06-08',
 30,
 count(*)
 from dwd_user_register_inc
 where date_sub(dt>='2022-06-08',29) and dt<=2022-06-08')t

--需求:统计最近1天,7天,30天新增下单用户(难)

  --新增下单(用户的第一次下单)

--需求:统计最近1天,7天,30天新增注册用户(简单)

  --新增注册(用户的第一次注册)

DROP TABLE IF EXISTS dws_first_order_stats;
CREATE EXTERNAL TABLE dws_first_order_stats
(
    `user_id`             String comment '用户ID',
    `first_order_date`     Bigint common '用户首次下单时间'
) COMMENT '新增下单用户统计'
   partition by(`dt` String)
  stored as orc
  location '/warehourse/gmall/dws/dws_first_order_stats/'
  tblproperties('orc.compress'='snappy');
  --取8号
  insert overwrite table dws_first_order_stats partition (dt='2022-06-08')
  select
    user_id,
    min(date_id)
  from
  dwd_trade_order_detail_inc
  group by user_id
   --取9号
     insert overwrite table dws_first_order_stats partition (dt='2022-06-09')
  select
    user_id,
    min(date_id)
  from
  dwd_trade_order_detail_inc
  group by user_id

nd

image.png

image.png

各渠道流量统计(流量主题)

需求说明如下

统计周期

统计粒度

指标

说明

最近1/7/30

渠道

访客数

统计访问人数

最近1/7/30

渠道

会话平均停留时长

统计会话平均停留时长

最近1/7/30

渠道

会话平均浏览页面数

统计会话平均浏览页面数

最近1/7/30

渠道

会话总数

统计会话总数

最近1/7/30

渠道

跳出率

只有一个页面的会话的比例

DROP TABLE IF EXISTS ads_traffic_stats_by_channel;
CREATE EXTERNAL TABLE ads_traffic_stats_by_channel
(
    `dt`               STRING COMMENT '统计日期',
    `recent_days`      BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `channel`          STRING COMMENT '渠道',
    `uv_count`         BIGINT COMMENT '访客人数',
    `avg_duration_sec` BIGINT COMMENT '会话平均停留时长,单位为秒',
    `avg_page_count`   BIGINT COMMENT '会话平均浏览页面数',
    `sv_count`         BIGINT COMMENT '会话数',
    `bounce_rate`      DECIMAL(16, 2) COMMENT '跳出率'
) COMMENT '各渠道流量统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_traffic_stats_by_channel/';

在dws_traffic_session_page_view_1d中,我们对每个会话的页面访问情况做了汇总,本节基于该表统计。

ADS层报表数据量不大,分区意义不大,通过dt字段区分统计日期即可。每日装载将历史报表数据取出,与当日统计结果union后写回该表即可。要注意,为了避免数据重复,此处要用union去重。

insert overwrite table ads_traffic_stats_by_channel
--为了防止全表覆盖,所以我们需要把旧的表先取出来
select * from ads_traffic_stats_by_channel
union
select * from
(--最近1天
select '2022-06-08',1,channel,count(distinct mid_id),avg(during_time/1000),
avg(page_count_1d),count(session_id),
sum(if(page_count=1,1,0))/count(session_id)
from dws_traffic_session_page_view_1d
where dt='2022-06-08'
group by channel
union all
--最近七天
select '2022-06-08',7,channel,count(distinct mid_id),avg(during_time/1000),
avg(page_count_1d),count(session_id),
sum(if(page_count=1,1,0))/count(session_id)
from dws_traffic_session_page_view_1d
where dt>=date_sub('2022-06-08',6) and dt<='2022-06-08'
group by channel
union all
select '2022-06-08',30,channel,count(distinct mid_id),avg(during_time/1000),
avg(page_count_1d),count(session_id),
sum(if(page_count=1,1,0))/count(session_id)
from dws_traffic_session_page_view_1d
where dt>=date_sub('2022-06-08',29) and dt<='2022-06-08'
group by channel)t

SQL优化

--当前的问题:读了三次相同的数据,使用三个不同的场景

--解决思路:读了一次数据,使用三个场景(炸裂)

炸裂【1,2,3】=》【1,2,3】,【1,2,3】,【1,2,3】

扁平化【1,2,3】=》1,2,3

select * from (select 'a'  name) t lateral view explode(array(1,2,3)) tmp as name1

where name1!=7

1.获取时间范围最大的数据集

2.将查询的数据集在内存中进行炸裂操作(三份),变成多份,增加标记

3.将炸裂后的数据进行筛选,保存有效数据

4.将过滤的后的数据按照标记进行分组,然后进行统计

insert overwrite table ads_traffic_stats_by_channel
--为了防止全表覆盖,所以我们需要把旧的表先取出来
select * from ads_traffic_stats_by_channel
--炸裂函数
select '2022-06-08',days,channel,count(distinct mid_id),avg(during_time/1000),
avg(page_count_1d),count(session_id),
sum(if(page_count=1,1,0))/count(session_id)
from dws_traffic_session_page_view_1d  lateral view explode(array(1,7,30)) tmp as days 
where dt>=date_sub('2022-06-08',days-1) and dt<='2022-06-08'
group by channel,days

路径分析

用户路径分析,顾名思义,就是指用户在APP或网站中的访问路径。为了衡量网站优化的效果或营销推广的效果,以及了解用户行为偏好,时常要对访问路径进行分析。

用户访问路径的可视化通常使用桑基图。如下图所示,该图可真实还原用户的访问路径,包括页面跳转和页面访问次序。

桑基图需要我们提供每种页面跳转的次数,每个跳转由source/target表示,source指跳转起始页面,target表示跳转终到页面。

image.png

建表语句
DROP TABLE IF EXISTS ads_page_path;
CREATE EXTERNAL TABLE ads_page_path
(
    `dt`          STRING COMMENT '统计日期',
    `source`      STRING COMMENT '跳转起始页面ID',
    `target`      STRING COMMENT '跳转终到页面ID',
    `path_count`  BIGINT COMMENT '跳转次数'
) COMMENT '页面浏览路径分析'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_page_path/';
insert overwrite table ads_page_path
select * from ads_page_path
union
select '2022-06-08',last_page_id source,
page_id target,
count(*)
from dwd_traffic_page_view_inc
where dt='2022-06-08'
group by last_page_id,page_id

问题:

桑基图不允许出现环状图形:可以在页面前增加流程标记,区分页面跳转顺序

我们需要统计页面离开的情况

当前统计的是页面路径跳转,前提条件是同一个会话

在同一个会话中,给每个页面额外增加一个标记,用于表示跳转后的页面,如果跳转后面的页面为null,表示离开

insert overwrite table ads_page_path
select * from ads_page_path
select
  '2022-06-08',
  source,
  target,
  count(*)
from
(  select 
    concat('step-',rn,':',source) source,
    concat('step-',(rn1),':',target)  target,
 target
 from(
    select page_id,source,
lead(page_id,1,'out') over(partition by session_id order by view_time) target,
 row_number() over(partition by session_id order by view_time) rn
from dwd_traffic_page_view_inc
where dt='2022-06-08'

)t1
)t
group by source,target

image.png


用户变动统计

该需求包括两个指标,分别为流失用户数和回流用户数,以下为对两个指标的解释说明。

image.png

image.png

流失用户数:用户的末次登录时间是7天前当天=(8-7
回流用户数:首先要今天登录和上一次登录时间为7天前<(8-7)
上一次登录其实就是昨天统计的末次登录
DROP TABLE IF EXISTS ads_user_change;
CREATE EXTERNAL TABLE ads_user_change
(
    `dt`               STRING COMMENT '统计日期',
    `user_churn_count` BIGINT COMMENT '流失用户数',
    `user_back_count`  BIGINT COMMENT '回流用户数'
) COMMENT '用户变动统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_user_change/';

--统计日期和业务日期的区别

   --统计日期(dt):数据在哪一天统计的

#统计流失数据的数量
select
  count(*) user_churn_count
from dws_user_login_td
where dt='2022-06-08' and login_date_last=date_sub('2022-06-08',7)
#回流用户
insert overwrite table ads_user_change
select * from ads_user_change
union
select
  '2022-06-08',
  user_churn_count,
  user_back_count
from(
    select
  1 a,
  count(*)  user_churn_count
  from dws_user_login_td
where dt='2022-06-08'
  where dt='2022-06-08' and login_date_last=date_sub('2022-06-08',7)
)churn
join(
  select 1 b, count(*)  user_back_count
from
  #把两个昨天和今天时间和在一起
(select
user_id,
 '2022-06-08'
from dws_user_login_td
where dt='2022-06-08'
and login_date_last='2022-06-08')new
join(
  select user_id
from dws_user_user_login_td
where dt=date_sub('2022-06-08',1)  #昨天统计的
)old on new.user_id=old.user_id
where datediff(new.login_date,old.login_date_last)>7
)back
on churn.a=back.b

用户留存率

留存分析一般包含新增留存和活跃留存分析。

新增留存分析是分析某天的新增用户中,有多少人有后续的活跃行为。活跃留存分析是分析某天的活跃用户中,有多少人有后续的活跃行为。

留存分析是衡量产品对用户价值高低的重要指标。

此处要求统计新增留存率,新增留存率具体是指留存用户数与新增用户数的比值,例如2022-06-08新增100个用户,1日之后(2022-06-09)这100人中有80个人活跃了,那2022-06-08的1日留存数则为80,2022-06-08的1日留存率则为80%

要求统计每天的1至7日留存率,如下图所示。

image.png

用户留存率

2022-06-08

zhangsan    reg

lisi               reg

wangwu       reg

2022-06-09(1日留存率33.33%)

zhangsan       login

lisi                 login

2022-06-10(2日留存率 66.66%)

zhangsan    reg

lisi               reg

zhangsan    reg

反推

2022-06-08(1日留存率:100%,2日留存率:100%)

zhangsan    login

lisi               login

2022-06-07

zhangsan       reg

2022-06-06

lisi               reg

规律:获取1天前注册用户数量+获取1天注册且当天用户登录用户数量

获取2天前注册用户数量+获取2天注册且当天用户登录用户数量

.。。。。

用户登录历史至今表(td)

 --userid+first+last

#要求统计每天的1至7日留存率
DROP TABLE IF EXISTS ads_user_retention;
CREATE EXTERNAL TABLE ads_user_retention
(
    `dt`              STRING COMMENT '统计日期',
    `create_date`     STRING COMMENT '用户新增日期',
    `retention_day`   INT COMMENT '截至当前日期留存天数',
    `retention_count` BIGINT COMMENT '留存用户数量',
    `new_user_count`  BIGINT COMMENT '新增用户数量',
    `retention_rate`  DECIMAL(16, 2) COMMENT '留存率'
) COMMENT '用户留存率'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_user_retention/';
2)数据装载
--1日留存率
select 
  '2022-06-08',
  date_sub('2022-06-08',1),
  1,
  sum(if(login_date_last='2022-06-08',1,0)) retention_count,
  count(*) new_user_count,
   sum(if(login_date_last='2022-06-08',1,0))/ count(*)*100
from
dws_user_user_login_td
where dt='2022-06-08'
and login_date_first=date_sub('2022-06-08',1) #注册里面包含着登录
--2日留存率
select 
  '2022-06-08',
  date_sub('2022-06-08',2),
  2,
  sum(if(login_date_last='2022-06-08',1,0)) retention_count,
  count(*) new_user_count,
   sum(if(login_date_last='2022-06-08',1,0))/ count(*)*100
from
dws_user_user_login_td
where dt='2022-06-08'
and login_date_first=date_sub('2022-06-08',2) #注册里面包含着登录
  insert overwrite table ads_user_retention
  select * from ads_user_retention
  union
  select user_id,
              '2022-06-08',
                login_date_first,
                 datediff('2022-06-08', login_date_first) retention_day,
                   sum(if(login_date_last = '2022-06-08', 1, 0)) retention_count,
       count(*) new_user_count,
       cast(sum(if(login_date_last = '2022-06-08', 1, 0)) / count(*) * 100 as decimal(16, 2)) retention_rate
         from dws_user_user_login_td
         where dt = '2022-06-08'
           and login_date_first >= date_sub('2022-06-08', 7)
           and login_date_first < '2022-06-08'
           group by login——date_first

用户新增活跃统计

本需求对活跃的定义是限定时间范围内有登录记录,可以通过限制末次登录日期在最近1/7/30日筛选活跃用户。注册日期等价于首次登录日期,可以根据首次登录日期筛选新增用户。

统计周期

指标

指标说明

最近1、730

新增用户数

最近1、730

活跃用户数

DROP TABLE IF EXISTS ads_user_stats;
CREATE EXTERNAL TABLE ads_user_stats
(
    `dt`                STRING COMMENT '统计日期',
    `recent_days`       BIGINT COMMENT '最近n日,1:最近1日,7:最近7日,30:最近30日',
    `new_user_count`    BIGINT COMMENT '新增用户数',
    `active_user_count` BIGINT COMMENT '活跃用户数'
) COMMENT '用户新增活跃统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_user_stats/';
insert overwrite table ads_user_stats
select * from
ads_user_stats
union
select * from(
  --最近一天
select 
'2022-06-08',
1
  sum(if( login_date_first=date_sub('2022-06-08',1),1,0)) new_user_count,
  count(*) active_user_count
from dws_user_login_td
where dt='2022-06-08'
and login_date_last=date_sub('2022-06-08',1)  #昨天登录的用户数量
union all
--最近七天
select 
'2022-06-08',
7
  sum(if( login_date_firs>=date_sub('2022-06-08',6),1,0)) new_user_count,
  count(*) active_user_count
from dws_user_login_td
where dt='2022-06-08'
and login_date_last>=date_sub('2022-06-08',6) and login_date_last<='2022-06-08'  七天之类的用户数量
union all
--最近30天
select 
'2022-06-08',
30
  sum(if( login_date_first>=date_sub('2022-06-08',29),1,0)) new_user_count,
  count(*) active_user_count
from dws_user_login_td
where dt='2022-06-08'
and login_date_last>=date_sub('2022-06-08',29) and login_date_last<='2022-06-08'  七天之类的用户数量
)t

SQL优化

--当前的问题:读了三次相同的数据,使用三个不同的场景

--解决思路:读了一次数据,使用三个场景(炸裂)

炸裂【1,2,3】=》【1,2,3】,【1,2,3】,【1,2,3】

扁平化【1,2,3】=》1,2,3

select * from (select 'a'  name) t lateral view explode(array(1,2,3)) tmp as name1

where name1!=7

1.获取时间范围最大的数据集

2.将查询的数据集在内存中进行炸裂操作(三份),变成多份,增加标记

3.将炸裂后的数据进行筛选,保存有效数据

4.将过滤的后的数据按照标记进行分组,然后进行统计

insert overwrite table ads_user_stats
select * from
ads_user_stats
union
select * from(
  select 
'2022-06-08',
days
  sum(if( login_date_first>=date_sub('2022-06-08',days-1),1,0)) new_user_count,
  count(*) active_user_count
from dws_user_login_td lateral view explode(array(1,7,30)) as days
where dt='2022-06-08'
and login_date_last>=date_sub('2022-06-08',days-1) and login_date_last<='2022-06-08'  七天之类的用户数量
group by days
)t

用户行为漏斗分析

漏斗分析是一个数据分析模型,它能够科学反映一个业务流程从起点到终点各阶段用户转化情况。由于其能将各阶段环节都展示出来,故哪个阶段存在问题,就能一目了然。

image.png

统计周期

指标

说明

最近1 

首页浏览人数

最近1 

商品详情页浏览人数

最近1 

加购人数

最近1 

下单人数

最近1 

支付人数

支付成功人数

DROP TABLE IF EXISTS ads_user_action;
CREATE EXTERNAL TABLE ads_user_action
(
    `dt`                STRING COMMENT '统计日期',
    `home_count`        BIGINT COMMENT '浏览首页人数',
    `good_detail_count` BIGINT COMMENT '浏览商品详情页人数',
    `cart_count`        BIGINT COMMENT '加购人数',
    `order_count`       BIGINT COMMENT '下单人数',
    `payment_count`     BIGINT COMMENT '支付人数'
) COMMENT '用户行为漏斗分析'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_user_action/';
  ---效率不高  
 select
   count(*) home_count
 from dws_traffic_page_visitor_page_view_1d
 where dt='2022-06-08'
 and page_id='home'
   ---效率不高  
  select
   count(*) good_detail_count
 from dws_traffic_page_visitor_page_view_1d
 where dt='2022-06-08'
 and page_id='good_detail'
 --优化,融合,读一次数据表
 insert overwrite table ads_user_action
 select * from ads_user_action
 union
( select 
    `2022-06-08`             ,
    `home_count`       ,
    `good_detail_count`,
    `cart_count`       ,
    `order_count`      ,
    `payment_count`    
 from(
  select
  1,a,
  sum(if(page_id ='home',1,0)) home_count,
  sum(if(page_id='good_detail',1,0)) good_detail_count
 from   dws_traffic_page_visitor_page_view_1d
 where dt='2022-06-08'
 and(page_id='home' or page_id='good_detail')
 )pv

 join (
  --加购人数统计
 select
  1 b,
  count(user_id)  cart_count
 from dws_trade_user_cart_add_1d
 where dt='2022-06-08'
 )cart
on pv.a=cart.b
join( select 
   1,c
 count(user_id) order_count
 from dws_trade_user_1d
 where dt='2022-06-08') oi pv.a=oi.c
join(
    select 
   1,d
 count(user_id) payment_count
 from dws_trade_user_1d
 where dt='2022-06-08
)pay on pay.a=pay.d)t
 

 
 

新增下单用户统计

需求说明如下。

统计周期

指标

说明

最近1、7、30

新增下单人数

建表语句
DROP TABLE IF EXISTS ads_new_order_user_stats;
CREATE EXTERNAL TABLE ads_new_order_user_stats
(
    `dt`                   STRING COMMENT '统计日期',
    `recent_days`          BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `new_order_user_count` BIGINT COMMENT '新增下单人数'
) COMMENT '新增下单用户统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_new_order_user_stats/';
insert overwrite table ads_new_order_user_stats
select * from ads_new_order_user_stats
union
select * from (
select 
  '2022-06-08',
  1,
  count(user_id)
  from dws_trade_user_order_td
  where dt='2022-06-08'
  and order_date_first='2022-06-08'
union all  
select 
  '2022-06-08',
  7,
  count(user_id)
from dws_trade_user_order_td
where dt='2022-06-08'
and order_date_first>=data_sub('2022-06-08',6) and order_date_first<='2022-06-08'
union all
select 
  '2022-06-08',
  30,
  count(user_id)
from dws_trade_user_order_td
where dt='2022-06-08'
and order_date_first>=data_sub('2022-06-08',29) and order_date_first<='2022-06-08
)t

优化

select 
  '2022-06-08',
  days,
  count(user_id)
from dws_trade_user_order_td lateral view explode(array(1,7,30)) tmp as days
where dt='2022-06-08'
and order_date_first>=data_sub('2022-06-08',days-1) and order_date_first<='2022-06-08
group by days

最近7日内连续3日下单用户数

DROP TABLE IF EXISTS ads_order_continuously_user_count;
CREATE EXTERNAL TABLE ads_order_continuously_user_count
(
    `dt`                            STRING COMMENT '统计日期',
    `recent_days`                   BIGINT COMMENT '最近天数,7:最近7天',
    `order_continuously_user_count` BIGINT COMMENT '连续3日下单用户数'
) COMMENT '最近7日内连续3日下单用户数统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_order_continuously_user_count/';
dws_trade_user_order_1d
数据规律:1.连续三天的基本条件应该数据量大于等于3
2.连续的第三条数据的时间减去当前时间应该为2
3.一个用户可能会重复出现三次所以要去重
insert overwrite table ads_order_continuously_user_count
select * from  ads_order_continuously_user_count
union
select 
  '2022-06-08',
  7,
  count(distinct user)
from
(
  select
  user_id
  datediff(lead(dt,2,'9999-12-31') over (partition by user_id order by dt),dt) diff
from
dws_trade_user_order_1d
where dt>=date_sub('2022-06-08',1) and dt<='2022-06-08'
)t
where diff=2

方法二

select
    count(distinct(user_id))
from
(
    select
        user_id
    from
    (
        select
            user_id,
            date_sub(dt,rank() over(partition by user_id order by dt)) diff
        from dws_trade_user_order_1d
        where dt>=date_add('2022-06-08',-6)
    )t1
    group by user_id,diff
    having count(*)>=3
)t2;

方法三

insert overwrite table ads_order_continuously_user_count
select * from  ads_order_continuously_user_count
union
select 
  '2022-06-08',
  7,
  count(distinct user)
from(
select
  user_id,
  lead(dt,2,'999-12-31) over (partition by user_id order by dt) now,
  date_add(dt,2)  ss
from
    dws_trade_user_order_1d   
)t
where now=ss
       

如果断一天也算连续(例如2022-06-05,2022-06-06,2022-06-08),应该如何计算?

select
    count(distinct(user_id))
from
(
    select
        user_id,
        datediff(lead(dt,2,'9999-12-31') over(partition by user_id order by dt),dt) diff
    from dws_trade_user_order_1d
    where dt>=date_add('2022-06-08',-6)
)t1
where diff<=3;

商品主题

最近30日各品牌复购率

统计周期

统计粒度

指标

说明

最近30

品牌

复购率

重复购买人数占购买人数比例

--复购率
--衍生指标(比例)
  --N派生指标
      --重复购买人数(下单(次数>1)人数) /购买人数(下单人数)
--数据来源:
    --DWS
    --DWD(下单)+DIM(SKU)
    --ODS(X)
    
    
DROP TABLE IF EXISTS ads_repeat_purchase_by_tm;
CREATE EXTERNAL TABLE ads_repeat_purchase_by_tm
(
    `dt`                  STRING COMMENT '统计日期',
    `recent_days`       BIGINT COMMENT '最近天数,30:最近30天',
    `tm_id`              STRING COMMENT '品牌ID',
    `tm_name`            STRING COMMENT '品牌名称',
    `order_repeat_rate` DECIMAL(16, 2) COMMENT '复购率'
) COMMENT '最近30日各品牌复购率统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_repeat_purchase_by_tm/';

粒度转化

先把我们的粒度从商品变成我们的品牌,在从品牌去掉我们的user_id,做统计

先取得某个用户对商品下单几次

--粒度变化
  --数据源粒度:user+sku
  --统计粒度:sku
    --此时对另外一个粒度的数据统计不需要去重操作
  --数据源粒度:user+sku
  --统计粒度:user+tm
  --粒度如果变粗,那么数据就需要进一步聚合
  --nd表的问题
      --nd表计算时可能出现的问题:30天有数据,但是7天没有数据,但是7天和3天会融合成一条
insert overwrite table ads_repeat_purchase_by_tm
select * from ads_repeat_purchase_by_tm
union
select
    '2022-06-08',
    30,
    tm_id,
    tm_name,
  sum(if(order_count>1,1,0))/count(user_id)*100
   --cast(sum(if(order_count>=2,1,0))/sum(if(order_count>=1,1,0)) as decimal(16,2))
from(select
  user_id,
  tm_id,
  tm_name,
  sum(order_count_30d)
  count(distinct user_id)
from dsw_trade_user_sku_order_nd
where dt='2022-06-08'
group by tm_id,tm_name,user_id)t
group by tm_id,tm_name

各品牌商品下单统计

需求说明如下

统计周期

统计粒度

指标

说明

最近1、7、30

品牌

下单数

最近1、7、30

品牌

下单人数

--DWD+DIM=>1d
DROP TABLE IF EXISTS ads_order_stats_by_tm;
CREATE EXTERNAL TABLE ads_order_stats_by_tm
(
    `dt`                      STRING COMMENT '统计日期',
    `recent_days`             BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `tm_id`                   STRING COMMENT '品牌ID',
    `tm_name`                 STRING COMMENT '品牌名称',
    `order_count`             BIGINT COMMENT '下单数',
    `order_user_count`        BIGINT COMMENT '下单人数'
) COMMENT '各品牌商品下单统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_order_stats_by_tm/';
insert overwrite table ads_order_stats_by_tm
select * from ads_order_stats_by_tm
union
 --最近1天
    --数据源粒度:user+sku
    --统计粒度:tm
 select
    '2022-06-08',
    1,
    tm_id,
    tm_name,
    sum(order_count_1d),
    count(distinct user_id)
 from dws_trade_sku_order_1d
 where dt='2022-06-08'
 group by tm_id,tm_name
 union all
 --最近七天
 select 
    '2022-06-08',
    7,
    tm_id,
    tm_name,
    sum(order_count_7d),
    count(distinct user_id)
 from 
 dws_trade_user_sku_order_nd
 where dt='2022-06-08'
 and oreder_count_7d>0
  group by tm_id,tm_name
  union all
   select 
    '2022-06-08',
    30,
    tm_id,
    tm_name,
    sum(order_count_30d),
    count(distinct user_id)
 from 
 dws_trade_user_sku_order_nd
 where dt='2022-06-08'
  and oreder_count_30d>0
  group by tm_id,tm_name

优化

  insert overwrite table ads_order_stats_by_tm
select * from ads_order_stats_by_tm
union
 ( select 
    '2022-06-08',
    30,
    tm_id,
    tm_name,
    sum(order_count_30d),
    count(distinct user_id)
 from 
  (select days,tm_id,tm_name,
  if(days=7,order_count_7d,order_count_30d) order_count
  from  dws_trade_user_sku_order_nd lateral view explode(array(7,30)) tmp as days
   where dt='2022-06-08')t
where order_count>0
  group by tm_id,tm_name,days)t

各品类商品下单统计

需求说明如下

统计周期

统计粒度

指标

说明

最近1、7、30

品类

下单数

最近1、7、30

品类

下单人数

DROP TABLE IF EXISTS ads_order_stats_by_cate;
CREATE EXTERNAL TABLE ads_order_stats_by_cate
(
    `dt`                      STRING COMMENT '统计日期',
    `recent_days`             BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `category1_id`            STRING COMMENT '一级品类ID',
    `category1_name`          STRING COMMENT '一级品类名称',
    `category2_id`            STRING COMMENT '二级品类ID',
    `category2_name`          STRING COMMENT '二级品类名称',
    `category3_id`            STRING COMMENT '三级品类ID',
    `category3_name`          STRING COMMENT '三级品类名称',
    `order_count`             BIGINT COMMENT '下单数',
    `order_user_count`        BIGINT COMMENT '下单人数'
) COMMENT '各品类商品下单统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_order_stats_by_cate/';
    
insert overwrite table ads_order_stats_by_cate
select * from ads_order_stats_by_cate
union
select
    '2022-06-08' dt,
    recent_days,
    category1_id,
    category1_name,
    category2_id,
    category2_name,
    category3_id,
    category3_name,
    order_count,
    order_user_count
from
(
    select
        1 recent_days,
        category1_id,
        category1_name,
        category2_id,
        category2_name,
        category3_id,
        category3_name,
        sum(order_count_1d) order_count,
        count(distinct(user_id)) order_user_count
    from dws_trade_user_sku_order_1d
    where dt='2022-06-08'
    group by category1_id,category1_name,category2_id,category2_name,category3_id,category3_name
    union all
    select
        recent_days,
        category1_id,
        category1_name,
        category2_id,
        category2_name,
        category3_id,
        category3_name,
        sum(order_count),
        count(distinct(if(order_count>0,user_id,null)))
    from
    (
        select
            recent_days,
            user_id,
            category1_id,
            category1_name,
            category2_id,
            category2_name,
            category3_id,
            category3_name,
            case recent_days
                when 7 then order_count_7d
                when 30 then order_count_30d
            end order_count
        from dws_trade_user_sku_order_nd lateral view explode(array(7,30)) tmp as recent_days
        where dt='2022-06-08'
    )t1
    group by recent_days,category1_id,category1_name,category2_id,category2_name,category3_id,category3_name
)odr;

各品类商品购物车存量Top3

--各品类商品购物车
--各品类商品
--购物车存量
    --周期快照事实表
实现思路:
1.(统计粒度(维度)+维度)(度量值)销量统计出来    =     (品类+存量)销量统计出来
2.将相同的统计粒度数据分在一个组中(数据不能减少)(开窗)
3.将组内的统计结果数据进行排序(开窗中的order by)
4.将组内排序的结果取前N条(row_number>=N)
DROP TABLE IF EXISTS ads_sku_cart_num_top3_by_cate;
CREATE EXTERNAL TABLE ads_sku_cart_num_top3_by_cate
(
    `dt`             STRING COMMENT '统计日期',
    `category1_id`   STRING COMMENT '一级品类ID',
    `category1_name` STRING COMMENT '一级品类名称',
    `category2_id`   STRING COMMENT '二级品类ID',
    `category2_name` STRING COMMENT '二级品类名称',
    `category3_id`   STRING COMMENT '三级品类ID',
    `category3_name` STRING COMMENT '三级品类名称',
    `sku_id`         STRING COMMENT 'SKU_ID',
    `sku_name`       STRING COMMENT 'SKU名称',
    `cart_num`       BIGINT COMMENT '购物车中商品数量',
    `rk`             BIGINT COMMENT '排名'
) COMMENT '各品类商品购物车存量Top3'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_sku_cart_num_top3_by_cate/';
2)数据装载
select 
'2022-06-08',
category1_id,
   category1_name,category2_id,category2_name,category3_id,category3_name,sku_id,sku_name
from(
select
category1_id,
   category1_name,category2_id,category2_name,category3_id,category3_name,sku_id,sku_name
rank() over (partition by category1_id,category2_id,category3_id order by cart_num)
from(
  select 
  category1_id,
   category1_name,category2_id,category2_name,category3_id,category3_name,sku_id,sku_name,
sum(sku_num) cart_num
from
(
  select
  sku_id
from dwd_trade_cart_full
where dt='2022-06-08'
) cart
left join(
     select
   id,
  category1_id,category1_name,category2_id,category2_name,category3_id,category3_name,
  from dim_sku_full
  where dt='2022-06-08'
)sku  on cart.sku_id=sku.id
group by 
 category1_id,category1_name,category2_id,category2_name,category3_id,category3_name,sku_id
)t
)rr
where rk<=3

各品牌商品收藏次数Top3

1)建表语句

DROP TABLE IF EXISTS ads_sku_favor_count_top3_by_tm;
CREATE EXTERNAL TABLE ads_sku_favor_count_top3_by_tm
(
    `dt`          STRING COMMENT '统计日期',
    `tm_id`       STRING COMMENT '品牌ID',
    `tm_name`     STRING COMMENT '品牌名称',
    `sku_id`      STRING COMMENT 'SKU_ID',
    `sku_name`    STRING COMMENT 'SKU名称',
    `favor_count` BIGINT COMMENT '被收藏次数',
    `rk`          BIGINT COMMENT '排名'
) COMMENT '各品牌商品收藏次数Top3'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_sku_favor_count_top3_by_tm/';
1.(品牌+商品,收藏次数) 统计
2.按照品类分组
3.按照次数进行排序(降序)
4.将排序后的数据取3名
--粒度的变化
  --数据源粒度:sku
  --统计粒度:sku
select
  '2022-06-08',
  'tm_id',
'tm_name',
'sku_id',
'sku_name',
'favor_add_couunt_1d',
rk
(select
 'tm_id',
'tm_name',
'sku_id',
'sku_name',
'favor_add_couunt_1d',
   favor_add_cont_1d,
   rank() over (partition by tm_id order by favor_add_count_1d desc) rk
from dws_interation_sku_facor_add_1d
where dt='2022-06-08')t
where rk<=3
group by tm_id,tm_name,sku_id,sku_name

下单到支付时间间隔平均值

具体要求:最近1日完成支付的订单的下单时间到支付时间的时间间隔的平均值

DROP TABLE IF EXISTS ads_order_to_pay_interval_avg;
CREATE EXTERNAL TABLE ads_order_to_pay_interval_avg
(
    `dt`                        STRING COMMENT '统计日期',
    `order_to_pay_interval_avg` BIGINT COMMENT '下单到支付时间间隔平均值,单位为秒'
) COMMENT '下单到支付时间间隔平均值统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_order_to_pay_interval_avg/';
--时间间隔1平均值:avg(datediff(支付时间,下单时间))
--datediff:用于天的时间的差值,不能用于时分秒的差值
1.时间差问题
2.累计快照事实表分区策略:
--事实表会以流程中最后一个行为的时间作为分区字段值

insert overwrite table ads_order_to_pay_interval_avg
select * from ads_order_to_pay_interval_avg
union
select
    '2022-06-08',
    cast(avg(to_unix_timestamp(payment_time)-to_unix_timestamp(order_time)) as bigint)
from dwd_trade_trade_flow_acc
where dt in ('9999-12-31','2022-06-08')#有可能收货也有可能收货
and payment_date_id='2022-06-08';#支付成功的时间

各省份交易统计

需求说明如下。

统计周期

统计粒度

指标

说明

最近1、7、30

省份

订单数

最近1、7、30

省份

订单金额

1)建表语句
DROP TABLE IF EXISTS ads_order_by_province;
CREATE EXTERNAL TABLE ads_order_by_province
(
    `dt`                 STRING COMMENT '统计日期',
    `recent_days`        BIGINT COMMENT '最近天数,1:最近1天,7:最近7天,30:最近30天',
    `province_id`        STRING COMMENT '省份ID',
    `province_name`      STRING COMMENT '省份名称',
    `area_code`          STRING COMMENT '地区编码',
    `iso_code`           STRING COMMENT '旧版国际标准地区编码,供可视化使用',
    `iso_code_3166_2`    STRING COMMENT '新版国际标准地区编码,供可视化使用',
    `order_count`        BIGINT COMMENT '订单数',
    `order_total_amount` DECIMAL(16, 2) COMMENT '订单金额'
) COMMENT '各省份交易统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_order_by_province/';
insert overwrite table ads_order_by_province
select * from ads_order_by_province
union
select * from (
    select
 '2022-06-08' ,
    1,
    province_id,
    province_name,
    area_code,
    iso_code,
    iso_3166_2  `iso_code_3166_2`,
    `order_count_1d`,
    `order_total_amount_1d`
from dws_trade_province_order_1d
where dt='2022-06-08'
union all
select
 '2022-06-08' ,
    7,
    province_id,
    province_name,
    area_code,
    iso_code,
    iso_3166_2  `iso_code_3166_2`,
    `order_count_7d`,
    `order_total_amount_7d`
from dws_trade_province_order_nd
where dt='2022-06-08'
union all
select
 '2022-06-08' ,
    30,
    province_id,
    province_name,
    area_code,
    iso_code,
    iso_3166_2  `iso_code_3166_2`,
    `order_count_30d`,
    `order_total_amount_30d`
from dws_trade_province_order_nd
where dt='2022-06-08'
)t

优化

(4)代码实现
insert overwrite table ads_order_by_province
select * from ads_order_by_province
union
select *
from(select
    '2022-06-08' dt,
    1 recent_days,
    province_id,
    province_name,
    area_code,
    iso_code,
    iso_3166_2,
    order_count_1d,
    order_total_amount_1d
from dws_trade_province_order_1d
where dt='2022-06-08'
union
select
 '2022-06-08' ,
    days,
    province_id,
    province_name,
    area_code,
    iso_code,
    iso_3166_2  `iso_code_3166_2`,
    if(days=7,`order_count_7d`,`order_count_30d`),
    if(days=7,`order_total_amount_7d`,`order_total_amount_30d`)
from dws_trade_province_order_nd lateral view explode(array(7,30)) tmp as days
where dt='2022-06-08')t

优惠券使用统计

需求说明如下。

统计周期

统计粒度

指标

说明

最近1日

优惠券

使用次数

支付才算使用

最近1日

优惠券

使用人数

支付才算使用

--数据源:dws_tool_user_coupon_coupon_used_1d
--粒度:user+coupon
DROP TABLE IF EXISTS ads_coupon_stats;
CREATE EXTERNAL TABLE ads_coupon_stats
(
    `dt`              STRING COMMENT '统计日期',
    `coupon_id`       STRING COMMENT '优惠券ID',
    `coupon_name`     STRING COMMENT '优惠券名称',
    `used_count`      BIGINT COMMENT '使用次数',
    `used_user_count` BIGINT COMMENT '使用人数'
) COMMENT '优惠券使用统计'
    ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
    LOCATION '/warehouse/gmall/ads/ads_coupon_stats/';
insert overwrite table ads_coupon_stats
select * from ads_coupon_stats
union
select
    '2022-06-08' dt,
    coupon_id,
    coupon_name,
    cast(sum(used_count_1d) as bigint),
    cast(count(*) as bigint)
from dws_tool_user_coupon_coupon_used_1d
where dt='2022-06-08'
group by coupon_id,coupon_name;

数据装载脚本

hadoop102的/home/atguigu/bin目录下创建dws_to_ads.sh

vim dws_to_ads.sh 
!/bin/bash

APP=gmall
# 如果是输入的日期按照取输入日期;如果没输入日期取当前时间的前一天
if [ -n "$2" ] ;then
    do_date=$2
else 
    do_date=`date -d "-1 day" +%F`
fi

ads_coupon_stats="
insert overwrite table ${APP}.ads_coupon_stats
select * from ${APP}.ads_coupon_stats
union
select
    '$do_date' dt,
    coupon_id,
    coupon_name,
    cast(sum(used_count_1d) as bigint),
    cast(count(*) as bigint)
from ${APP}.dws_tool_user_coupon_coupon_used_1d
where dt='$do_date'
group by coupon_id,coupon_name;
"
ads_new_order_user_stats="
insert overwrite table ${APP}.ads_new_order_user_stats
select * from ${APP}.ads_new_order_user_stats
union
select
    '$do_date' dt,
    recent_days,
    count(*) new_order_user_count
from ${APP}.dws_trade_user_order_td lateral view explode(array(1,7,30)) tmp as recent_days
where dt='$do_date'
and order_date_first>=date_add('$do_date',-recent_days+1)
group by recent_days;
"
ads_order_by_province="
insert overwrite table ${APP}.ads_order_by_province
select * from ${APP}.ads_order_by_province
union
select
    '$do_date' dt,
    1 recent_days,
    province_id,
    province_name,
    area_code,
    iso_code,
    iso_3166_2,
    order_count_1d,
    order_total_amount_1d
from ${APP}.dws_trade_province_order_1d
where dt='$do_date'
union
select
    '$do_date' dt,
    recent_days,
    province_id,
    province_name,
    area_code,
    iso_code,
    iso_3166_2,
    case recent_days
        when 7 then order_count_7d
        when 30 then order_count_30d
    end order_count,
    case recent_days
        when 7 then order_total_amount_7d
        when 30 then order_total_amount_30d
    end order_total_amount
from ${APP}.dws_trade_province_order_nd lateral view explode(array(7,30)) tmp as recent_days
where dt='$do_date';
"
ads_order_continuously_user_count="
insert overwrite table ${APP}.ads_order_continuously_user_count
select * from ${APP}.ads_order_continuously_user_count
union
select
    '$do_date',
    7,
    count(distinct(user_id))
from
(
    select
        user_id,
        datediff(lead(dt,2,'9999-12-31') over(partition by user_id order by dt),dt) diff
    from ${APP}.dws_trade_user_order_1d
    where dt>=date_add('$do_date',-6)
)t1
where diff=2;
"
ads_order_stats_by_cate="
insert overwrite table ${APP}.ads_order_stats_by_cate
select * from ${APP}.ads_order_stats_by_cate
union
select
    '$do_date' dt,
    recent_days,
    category1_id,
    category1_name,
    category2_id,
    category2_name,
    category3_id,
    category3_name,
    order_count,
    order_user_count
from
(
    select
        1 recent_days,
        category1_id,
        category1_name,
        category2_id,
        category2_name,
        category3_id,
        category3_name,
        sum(order_count_1d) order_count,
        count(distinct(user_id)) order_user_count
    from ${APP}.dws_trade_user_sku_order_1d
    where dt='$do_date'
    group by category1_id,category1_name,category2_id,category2_name,category3_id,category3_name
    union all
    select
        recent_days,
        category1_id,
        category1_name,
        category2_id,
        category2_name,
        category3_id,
        category3_name,
        sum(order_count),
        count(distinct(if(order_count>0,user_id,null)))
    from
    (
        select
            recent_days,
            user_id,
            category1_id,
            category1_name,
            category2_id,
            category2_name,
            category3_id,
            category3_name,
            case recent_days
                when 7 then order_count_7d
                when 30 then order_count_30d
            end order_count
        from ${APP}.dws_trade_user_sku_order_nd lateral view explode(array(7,30)) tmp as recent_days
        where dt='$do_date'
    )t1
    group by recent_days,category1_id,category1_name,category2_id,category2_name,category3_id,category3_name
)odr;
"
ads_order_stats_by_tm="
insert overwrite table ${APP}.ads_order_stats_by_tm
select * from ${APP}.ads_order_stats_by_tm
union
select
    '$do_date' dt,
    recent_days,
    tm_id,
    tm_name,
    order_count,
    order_user_count
from
(
    select
        1 recent_days,
        tm_id,
        tm_name,
        sum(order_count_1d) order_count,
        count(distinct(user_id)) order_user_count
    from ${APP}.dws_trade_user_sku_order_1d
    where dt='$do_date'
    group by tm_id,tm_name
    union all
    select
        recent_days,
        tm_id,
        tm_name,
        sum(order_count),
        count(distinct(if(order_count>0,user_id,null)))
    from
    (
        select
            recent_days,
            user_id,
            tm_id,
            tm_name,
            case recent_days
                when 7 then order_count_7d
                when 30 then order_count_30d
            end order_count
        from ${APP}.dws_trade_user_sku_order_nd lateral view explode(array(7,30)) tmp as recent_days
        where dt='$do_date'
    )t1
    group by recent_days,tm_id,tm_name
)odr;
"
ads_order_to_pay_interval_avg="
insert overwrite table ${APP}.ads_order_to_pay_interval_avg
select * from ${APP}.ads_order_to_pay_interval_avg
union
select
    '$do_date',
    cast(avg(to_unix_timestamp(payment_time)-to_unix_timestamp(order_time)) as bigint)
from ${APP}.dwd_trade_trade_flow_acc
where dt in ('9999-12-31','$do_date')
and payment_date_id='$do_date';
"
ads_page_path="
insert overwrite table ${APP}.ads_page_path
select * from ${APP}.ads_page_path
union
select
    '$do_date' dt,
    source,
    nvl(target,'null'),
    count(*) path_count
from
(
    select
        concat('step-',rn,':',page_id) source,
        concat('step-',rn+1,':',next_page_id) target
    from
    (
        select
            page_id,
            lead(page_id,1,null) over(partition by session_id order by view_time) next_page_id,
            row_number() over (partition by session_id order by view_time) rn
        from ${APP}.dwd_traffic_page_view_inc
        where dt='$do_date'
    )t1
)t2
group by source,target;
"
ads_repeat_purchase_by_tm="
insert overwrite table ${APP}.ads_repeat_purchase_by_tm
select * from ${APP}.ads_repeat_purchase_by_tm
union
select
    '$do_date',
    30,
    tm_id,
    tm_name,
    cast(sum(if(order_count>=2,1,0))/sum(if(order_count>=1,1,0)) as decimal(16,2))
from
(
    select
        user_id,
        tm_id,
        tm_name,
        sum(order_count_30d) order_count
    from ${APP}.dws_trade_user_sku_order_nd
    where dt='$do_date'
    group by user_id, tm_id,tm_name
)t1
group by tm_id,tm_name;
"
ads_sku_cart_num_top3_by_cate="
set hive.mapjoin.optimized.hashtable=false;
insert overwrite table ${APP}.ads_sku_cart_num_top3_by_cate
select * from ${APP}.ads_sku_cart_num_top3_by_cate
union
select
    '$do_date' dt,
    category1_id,
    category1_name,
    category2_id,
    category2_name,
    category3_id,
    category3_name,
    sku_id,
    sku_name,
    cart_num,
    rk
from
(
    select
        sku_id,
        sku_name,
        category1_id,
        category1_name,
        category2_id,
        category2_name,
        category3_id,
        category3_name,
        cart_num,
        rank() over (partition by category1_id,category2_id,category3_id order by cart_num desc) rk
    from
    (
        select
            sku_id,
            sum(sku_num) cart_num
        from ${APP}.dwd_trade_cart_full
        where dt='$do_date'
        group by sku_id
    )cart
    left join
    (
        select
            id,
            sku_name,
            category1_id,
            category1_name,
            category2_id,
            category2_name,
            category3_id,
            category3_name
        from ${APP}.dim_sku_full
        where dt='$do_date'
    )sku
    on cart.sku_id=sku.id
)t1
where rk<=3;
set hive.mapjoin.optimized.hashtable=true;
"
ads_sku_favor_count_top3_by_tm="
insert overwrite table ${APP}.ads_sku_favor_count_top3_by_tm
select * from ${APP}.ads_sku_favor_count_top3_by_tm
union
select
    '$do_date' dt,
    tm_id,
    tm_name,
    sku_id,
    sku_name,
    favor_add_count_1d,
    rk
from
(
    select
        tm_id,
        tm_name,
        sku_id,
        sku_name,
        favor_add_count_1d,
        rank() over (partition by tm_id order by favor_add_count_1d desc) rk
    from ${APP}.dws_interaction_sku_favor_add_1d
    where dt='$do_date'
)t1
where rk<=3;
"
ads_traffic_stats_by_channel="
insert overwrite table ${APP}.ads_traffic_stats_by_channel
select * from ${APP}.ads_traffic_stats_by_channel
union
select
    '$do_date' dt,
    recent_days,
    channel,
    cast(count(distinct(mid_id)) as bigint) uv_count,
    cast(avg(during_time_1d)/1000 as bigint) avg_duration_sec,
    cast(avg(page_count_1d) as bigint) avg_page_count,
    cast(count(*) as bigint) sv_count,
    cast(sum(if(page_count_1d=1,1,0))/count(*) as decimal(16,2)) bounce_rate
from ${APP}.dws_traffic_session_page_view_1d lateral view explode(array(1,7,30)) tmp as recent_days
where dt>=date_add('$do_date',-recent_days+1)
group by recent_days,channel;
"
ads_user_action="
insert overwrite table ${APP}.ads_user_action
select * from ${APP}.ads_user_action
union
select
    '$do_date' dt,
    home_count,
    good_detail_count,
    cart_count,
    order_count,
    payment_count
from
(
    select
        1 recent_days,
        sum(if(page_id='home',1,0)) home_count,
        sum(if(page_id='good_detail',1,0)) good_detail_count
    from ${APP}.dws_traffic_page_visitor_page_view_1d
    where dt='$do_date'
    and page_id in ('home','good_detail')
)page
join
(
    select
        1 recent_days,
        count(*) cart_count
    from ${APP}.dws_trade_user_cart_add_1d
    where dt='$do_date'
)cart
on page.recent_days=cart.recent_days
join
(
    select
        1 recent_days,
        count(*) order_count
    from ${APP}.dws_trade_user_order_1d
    where dt='$do_date'
)ord
on page.recent_days=ord.recent_days
join
(
    select
        1 recent_days,
        count(*) payment_count
    from ${APP}.dws_trade_user_payment_1d
    where dt='$do_date'
)pay
on page.recent_days=pay.recent_days;
"
ads_user_change="
insert overwrite table ${APP}.ads_user_change
select * from ${APP}.ads_user_change
union
select
    churn.dt,
    user_churn_count,
    user_back_count
from
(
    select
        '$do_date' dt,
        count(*) user_churn_count
    from ${APP}.dws_user_user_login_td
    where dt='$do_date'
    and login_date_last=date_add('$do_date',-7)
)churn
join
(
    select
        '$do_date' dt,
        count(*) user_back_count
    from
    (
        select
            user_id,
            login_date_last
        from ${APP}.dws_user_user_login_td
        where dt='$do_date'
        and login_date_last = '$do_date'
    )t1
    join
    (
        select
            user_id,
            login_date_last login_date_previous
        from ${APP}.dws_user_user_login_td
        where dt=date_add('$do_date',-1)
    )t2
    on t1.user_id=t2.user_id
    where datediff(login_date_last,login_date_previous)>=8
)back
on churn.dt=back.dt;
"
ads_user_retention="
insert overwrite table ${APP}.ads_user_retention
select * from ${APP}.ads_user_retention
union
select '$do_date' dt,
       login_date_first create_date,
       datediff('$do_date', login_date_first) retention_day,
       sum(if(login_date_last = '$do_date', 1, 0)) retention_count,
       count(*) new_user_count,
       cast(sum(if(login_date_last = '$do_date', 1, 0)) / count(*) * 100 as decimal(16, 2)) retention_rate
from (
         select user_id,
                login_date_last,
                login_date_first
         from ${APP}.dws_user_user_login_td
         where dt = '$do_date'
           and login_date_first >= date_add('$do_date', -7)
           and login_date_first < '$do_date'
     ) t1
group by login_date_first;
"
ads_user_stats="
insert overwrite table ${APP}.ads_user_stats
select * from ${APP}.ads_user_stats
union
select '$do_date' dt,
       recent_days,
       sum(if(login_date_first >= date_add('$do_date', -recent_days + 1), 1, 0)) new_user_count,
       count(*) active_user_count
from ${APP}.dws_user_user_login_td lateral view explode(array(1, 7, 30)) tmp as recent_days
where dt = '$do_date'
  and login_date_last >= date_add('$do_date', -recent_days + 1)
group by recent_days;
"


case $1 in
    "ads_coupon_stats" )
        hive -e "$ads_coupon_stats"
    ;;
    "ads_new_order_user_stats" )
        hive -e "$ads_new_order_user_stats"
    ;;
    "ads_order_by_province" )
        hive -e "$ads_order_by_province"
    ;;
    "ads_order_continuously_user_count" )
        hive -e "$ads_order_continuously_user_count"
    ;;
    "ads_order_stats_by_cate" )
        hive -e "$ads_order_stats_by_cate"
    ;;
    "ads_order_stats_by_tm" )
        hive -e "$ads_order_stats_by_tm"
    ;;
    "ads_order_to_pay_interval_avg" )
        hive -e "$ads_order_to_pay_interval_avg"
    ;;
    "ads_page_path" )
        hive -e "$ads_page_path"
    ;;
    "ads_repeat_purchase_by_tm" )
        hive -e "$ads_repeat_purchase_by_tm"
    ;;
    "ads_sku_cart_num_top3_by_cate" )
        hive -e "$ads_sku_cart_num_top3_by_cate"
    ;;
    "ads_sku_favor_count_top3_by_tm" )
        hive -e "$ads_sku_favor_count_top3_by_tm"
    ;;
    "ads_traffic_stats_by_channel" )
        hive -e "$ads_traffic_stats_by_channel"
    ;;
    "ads_user_action" )
        hive -e "$ads_user_action"
    ;;
    "ads_user_change" )
        hive -e "$ads_user_change"
    ;;
    "ads_user_retention" )
        hive -e "$ads_user_retention"
    ;;
    "ads_user_stats" )
        hive -e "$ads_user_stats"
    ;;
    "all" )
        hive -e "$ads_coupon_stats$ads_new_order_user_stats$ads_order_by_province$ads_order_continuously_user_count$ads_order_stats_by_cate$ads_order_stats_by_tm$ads_order_to_pay_interval_avg$ads_page_path$ads_repeat_purchase_by_tm$ads_sku_cart_num_top3_by_cate$ads_sku_favor_count_top3_by_tm$ads_traffic_stats_by_channel$ads_user_action$ads_user_change$ads_user_retention$ads_user_stats"
    ;;

增加脚本执行权限

[atguigu@hadoop102 bin]$ chmod +x dws_to_ads.sh

(4)脚本用法

[atguigu@hadoop102 bin]$ dws_to_ads.sh all 2022-06-08
相关实践学习
AnalyticDB MySQL海量数据秒级分析体验
快速上手AnalyticDB MySQL,玩转SQL开发等功能!本教程介绍如何在AnalyticDB MySQL中,一键加载内置数据集,并基于自动生成的查询脚本,运行复杂查询语句,秒级生成查询结果。
阿里云云原生数据仓库AnalyticDB MySQL版 使用教程
云原生数据仓库AnalyticDB MySQL版是一种支持高并发低延时查询的新一代云原生数据仓库,高度兼容MySQL协议以及SQL:92、SQL:99、SQL:2003标准,可以对海量数据进行即时的多维分析透视和业务探索,快速构建企业云上数据仓库。 了解产品 https://www.aliyun.com/product/ApsaraDB/ads
相关文章
|
5月前
|
存储 数据采集 JavaScript
深入理解数仓开发(一)数据技术篇之日志采集
深入理解数仓开发(一)数据技术篇之日志采集
|
5月前
|
消息中间件 关系型数据库 Kafka
深入理解数仓开发(二)数据技术篇之数据同步
深入理解数仓开发(二)数据技术篇之数据同步
|
5月前
|
数据挖掘
离线数仓(十)【ADS 层开发】(2)
离线数仓(十)【ADS 层开发】
|
5月前
|
定位技术
离线数仓(九)【DWS 层开发】(3)
离线数仓(九)【DWS 层开发】
|
5月前
|
存储
离线数仓(九)【DWS 层开发】(2)
离线数仓(九)【DWS 层开发】
|
5月前
|
SQL 分布式计算 Java
离线数仓(八)【DWD 层开发】(5)
离线数仓(八)【DWD 层开发】
|
4月前
|
存储 DataWorks Java
DataWorks产品使用合集之开发离线数仓时,需要多个工作空间的情况有哪些
DataWorks作为一站式的数据开发与治理平台,提供了从数据采集、清洗、开发、调度、服务化、质量监控到安全管理的全套解决方案,帮助企业构建高效、规范、安全的大数据处理体系。以下是对DataWorks产品使用合集的概述,涵盖数据处理的各个环节。
|
5月前
|
消息中间件 存储 Kafka
Flink 实时数仓(二)【ODS 层开发】
Flink 实时数仓(二)【ODS 层开发】