基于Hadoop生态圈的数据仓库实践 —— 进阶技术（三）

三、维度子集
有些需求不需要最细节的数据。例如更想要某个月而不是某天的记录。再比如相对于全部的销售数据，可能对某些特定状态的数据更感兴趣等。这些特定维度包含在从细节维度选择的行中，所以叫维度子集。维度子集比细节维度的数据少，因此更易使用，查询也更快。
本节中将准备两个特定维度，它们均取自现有的维度：月份维度（日期维度的子集），Pennsylvania州客户维度（客户维度的子集）。

1. 建立月份维度表
执行下面的脚本建立月份维度表。注意月份维度不包含promo_ind列，该列不适用月层次上，因为一个月中可能有多个促销期，而且并不是一个月中的每一天都是促销期。促销标记适用于天这个层次。

USE dw;

CREATE TABLE month_dim (
month_sk INT comment 'surrogate key',
month tinyint comment 'month',
month_name varchar(9) comment 'month name',
quarter tinyint comment 'quarter',
year smallint comment 'year'
)
comment 'month dimension table'
clustered by (month_sk) into 8 buckets
stored as orc tblproperties ('transactional'='true') ;

为了从日期维度同步导入月份维度，要把月份装载嵌入到日期维度的预装载脚本中。需要修改“建立数据仓库示例模型”里生成日期维度数据的脚本。下图显示了修改后的date_dim_generate.sh文件内容。



修改后的create_table_date_dim.sql文件内容如下：

-- 设置变量以支持事务
set hive.support.concurrency=true;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.txn.manager=org.apache.hadoop.hive.ql.lockmgr.DbTxnManager;
set hive.compactor.initiator.on=true;
set hive.compactor.worker.threads=1;

use dw;

-- 首次执行会建立日期维度临时表
create table if not exists date_dim_tmp (
date date comment 'date,yyyy-mm-dd',
month tinyint comment 'month',
month_name varchar(9) comment 'month name',
quarter tinyint comment 'quarter',
year smallint comment 'year',
promo_ind char(1) comment 'promotion index'
)
comment 'date dimension table'
row format delimited fields terminated by ','
stored as textfile;

-- 首次执行会建立日期维度表
create table if not exists date_dim (
date_sk int comment 'surrogate key',
date date comment 'date,yyyy-mm-dd',
month tinyint comment 'month',
month_name varchar(9) comment 'month name',
quarter tinyint comment 'quarter',
year smallint comment 'year',
promo_ind char(1) comment 'promotion index'
)
comment 'date dimension table'
clustered by (date_sk) into 8 buckets
stored as orc tblproperties ('transactional'='true');

-- 首次执行会建立月份维度表
CREATE TABLE month_dim (
month_sk INT comment 'surrogate key',
month tinyint comment 'month',
month_name varchar(9) comment 'month name',
quarter tinyint comment 'quarter',
year smallint comment 'year'
)
comment 'month dimension table'
clustered by (month_sk) into 8 buckets
stored as orc tblproperties ('transactional'='true');

新增的append_date.sql文件内容如下：

-- 设置变量以支持事务
set hive.support.concurrency=true;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.txn.manager=org.apache.hadoop.hive.ql.lockmgr.DbTxnManager;
set hive.compactor.initiator.on=true;
set hive.compactor.worker.threads=1;

use dw;

-- 向日期维度表追加数据
insert into date_dim
select row_number() over (order by date) + t2.sk_max,
t1.date,
t1.month,
t1.month_name,
t1.quarter,
t1.year,
t1.promo_ind
from (select * from date_dim_tmp) t1
cross join (select coalesce(max(date_sk),0) sk_max from date_dim) t2;

-- 向月份维度表追加数据
insert into month_dim
select row_number() over (order by t1.year,t1.month) + t2.sk_max,
t1.month,
t1.month_name,
t1.quarter,
t1.year
from (select distinct month, month_name, quarter, year FROM date_dim_tmp) t1
cross join (select coalesce(max(month_sk),0) sk_max from month_dim) t2;

对预装载主要做了以下三点修改：
生成CSV文件时去掉了代理键列，增加了促销期标记列。
生成date_dim.csv日期数据文件前先调用create_table_date_dim.sql脚本建表，新增一个append_date.sql脚本用于追加数据。
先向一个临时表date_dim_tmp中加载数据，然后在append_date.sql中处理从date_dim_tmp到date_dim的装载。之所以这样做有两个原因，一是考虑到后续可能需要追加日期，而不是重新生成所有数据，二是date_dim是一个ORC格式的二进制文件，不能直接从文本文件LOAD数据，只能从一个普通文本文件格式的表插入数据。
无论何时用修改后的脚本增加日期记录时，如果这个日期所在的月份没在月份维度中，那么该月份会被装载到月份维度中。下面测试一下日期和月份维度表数据的预装载。
（1）删除date_dim_tmp、date_dim、month_dim表

use dw;
drop table date_dim_tmp;
drop table date_dim;
drop table month_dim;

（2）执行预装载，生成从2000年1月1日到2010年12月31日的日期数据

./date_dim_generate.sh 2000-01-01 2010-12-31

这次执行维度表都是新建的，生成的日期和月份维度数据如下图所示。





（3）再次执行预装载，生成从2011年1月1日到2020年12月31日的日期数据

./date_dim_generate.sh 2011-01-01 2020-12-31

这次执行是向已有的维度表中追加日期，生成的日期和月份维度数据如下图所示。





（4）执行上一节建立的on_demand.sh脚本文件，更新促销标记列，数据恢复原状。

./on_demand.sh

2. 建立PA客户维度
月份维度是一个上卷维度，它包含基本维度的上层数据。而特定维度子集是选择基本维度的一个特定子集。执行下面的脚本建立特定维度表，并导入Pennsylvania (PA)客户维度子集数据。

USE dw;

CREATE TABLE pa_customer_dim (
customer_sk int comment 'surrogate key',
customer_number int comment 'number',
customer_name varchar(50) comment 'name',
customer_street_address varchar(50) comment 'address',
customer_zip_code int comment 'zipcode',
customer_city varchar(30) comment 'city',
customer_state varchar(2) comment 'state',
shipping_address varchar(50) comment 'shipping_address',
shipping_zip_code int comment 'shipping_zip_code',
shipping_city varchar(30) comment 'shipping_city',
shipping_state varchar(2) comment 'shipping_state',
version int comment 'version',
effective_date date comment 'effective date',
expiry_date date comment 'expiry date'
)
CLUSTERED BY (customer_sk) INTO 8 BUCKETS
STORED AS ORC TBLPROPERTIES ('transactional'='true');

注意，PA客户维度子集与月份维度子集有两点区别：
pa_customer_dim表和customer_dim表有完全相同的列，而month_dim不包含date_dim表的日期列。
pa_customer_dim表的代理键就是客户维度的代理键，而month_dim表里的月份维度代理键并不来自日期维度。
3. 修改定期装载regular_etl.sql文件
需要增加对PA客户维度的处理，这里只是在装载完customer_dim后简单重载PA客户维度数据，修改后的regular_etl.sql文件内容如下：

-- 设置变量以支持事务
set hive.support.concurrency=true;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.txn.manager=org.apache.hadoop.hive.ql.lockmgr.DbTxnManager;
set hive.compactor.initiator.on=true;
set hive.compactor.worker.threads=1;

USE dw;

-- 设置SCD的生效时间和过期时间
SET hivevar:cur_date = CURRENT_DATE();
SET hivevar:pre_date = DATE_ADD(${hivevar:cur_date},-1);
SET hivevar:max_date = CAST('2200-01-01' AS DATE);

-- 设置CDC的上限时间
INSERT OVERWRITE TABLE rds.cdc_time SELECT last_load, ${hivevar:cur_date} FROM rds.cdc_time;

-- 装载customer维度
-- 设置已删除记录和地址相关列上SCD2的过期，用<=>运算符处理NULL值。
UPDATE customer_dim
SET expiry_date = ${hivevar:pre_date}
WHERE customer_dim.customer_sk IN
(SELECT a.customer_sk
FROM (SELECT customer_sk,
customer_number,
customer_street_address,
customer_zip_code,
customer_city,
customer_state,
shipping_address,
shipping_zip_code,
shipping_city,
shipping_state
FROM customer_dim WHERE expiry_date = ${hivevar:max_date}) a LEFT JOIN
rds.customer b ON a.customer_number = b.customer_number
WHERE b.customer_number IS NULL OR
(  !(a.customer_street_address <=> b.customer_street_address)
OR !(a.customer_zip_code <=> b.customer_zip_code)
OR !(a.customer_city <=> b.customer_city)
OR !(a.customer_state <=> b.customer_state)
OR !(a.shipping_address <=> b.shipping_address)
OR !(a.shipping_zip_code <=> b.shipping_zip_code)
OR !(a.shipping_city <=> b.shipping_city)
OR !(a.shipping_state <=> b.shipping_state)
));

-- 处理customer_street_addresses列上SCD2的新增行
INSERT INTO customer_dim
SELECT
ROW_NUMBER() OVER (ORDER BY t1.customer_number) + t2.sk_max,
t1.customer_number,
t1.customer_name,
t1.customer_street_address,
t1.customer_zip_code,
t1.customer_city,
t1.customer_state,
t1.shipping_address,
t1.shipping_zip_code,
t1.shipping_city,
t1.shipping_state,
t1.version,
t1.effective_date,
t1.expiry_date
FROM
(
SELECT
t2.customer_number customer_number,
t2.customer_name customer_name,
t2.customer_street_address customer_street_address,
t2.customer_zip_code customer_zip_code,
t2.customer_city customer_city,
t2.customer_state customer_state,
t2.shipping_address shipping_address,
t2.shipping_zip_code shipping_zip_code,
t2.shipping_city shipping_city,
t2.shipping_state shipping_state,
t1.version + 1 version,
${hivevar:pre_date} effective_date,
${hivevar:max_date} expiry_date
FROM customer_dim t1
INNER JOIN rds.customer t2
ON t1.customer_number = t2.customer_number
AND t1.expiry_date = ${hivevar:pre_date}
LEFT JOIN customer_dim t3
ON t1.customer_number = t3.customer_number
AND t3.expiry_date = ${hivevar:max_date}
WHERE (!(t1.customer_street_address <=> t2.customer_street_address)
OR  !(t1.customer_zip_code <=> t2.customer_zip_code)
OR  !(t1.customer_city <=> t2.customer_city)
OR  !(t1.customer_state <=> t2.customer_state)
OR  !(t1.shipping_address <=> t2.shipping_address)
OR  !(t1.shipping_zip_code <=> t2.shipping_zip_code)
OR  !(t1.shipping_city <=> t2.shipping_city)
OR  !(t1.shipping_state <=> t2.shipping_state)
)
AND t3.customer_sk IS NULL) t1
CROSS JOIN
(SELECT COALESCE(MAX(customer_sk),0) sk_max FROM customer_dim) t2;

-- 处理customer_name列上的SCD1
-- 因为hive的update的set子句还不支持子查询，所以这里使用了一个临时表存储需要更新的记录，用先delete再insert代替update
-- 因为SCD1本身就不保存历史数据，所以这里更新维度表里的所有customer_name改变的记录，而不是仅仅更新当前版本的记录
DROP TABLE IF EXISTS tmp;
CREATE TABLE tmp AS
SELECT
a.customer_sk,
a.customer_number,
b.customer_name,
a.customer_street_address,
a.customer_zip_code,
a.customer_city,
a.customer_state,
a.shipping_address,
a.shipping_zip_code,
a.shipping_city,
a.shipping_state,
a.version,
a.effective_date,
a.expiry_date
FROM customer_dim a, rds.customer b
WHERE a.customer_number = b.customer_number AND !(a.customer_name <=> b.customer_name);
DELETE FROM customer_dim WHERE customer_dim.customer_sk IN (SELECT customer_sk FROM tmp);
INSERT INTO customer_dim SELECT * FROM tmp;

-- 处理新增的customer记录
INSERT INTO customer_dim
SELECT
ROW_NUMBER() OVER (ORDER BY t1.customer_number) + t2.sk_max,
t1.customer_number,
t1.customer_name,
t1.customer_street_address,
t1.customer_zip_code,
t1.customer_city,
t1.customer_state,
t1.shipping_address,
t1.shipping_zip_code,
t1.shipping_city,
t1.shipping_state,
1,
${hivevar:pre_date},
${hivevar:max_date}
FROM
(
SELECT t1.* FROM rds.customer t1 LEFT JOIN customer_dim t2 ON t1.customer_number = t2.customer_number
WHERE t2.customer_sk IS NULL) t1
CROSS JOIN
(SELECT COALESCE(MAX(customer_sk),0) sk_max FROM customer_dim) t2;

-- 重载PA客户维度
TRUNCATE TABLE pa_customer_dim;
INSERT INTO pa_customer_dim
SELECT
customer_sk
, customer_number
, customer_name
, customer_street_address
, customer_zip_code
, customer_city
, customer_state
, shipping_address
, shipping_zip_code
, shipping_city
, shipping_state
, version
, effective_date
, expiry_date
FROM customer_dim
WHERE customer_state = 'PA' ;

-- 装载product维度
-- 设置已删除记录和product_name、product_category列上SCD2的过期
UPDATE product_dim
SET expiry_date = ${hivevar:pre_date}
WHERE product_dim.product_sk IN
(SELECT a.product_sk
FROM (SELECT product_sk,product_code,product_name,product_category
FROM product_dim WHERE expiry_date = ${hivevar:max_date}) a LEFT JOIN
rds.product b ON a.product_code = b.product_code
WHERE b.product_code IS NULL OR (a.product_name <> b.product_name OR a.product_category <> b.product_category));

-- 处理product_name、product_category列上SCD2的新增行
INSERT INTO product_dim
SELECT
ROW_NUMBER() OVER (ORDER BY t1.product_code) + t2.sk_max,
t1.product_code,
t1.product_name,
t1.product_category,
t1.version,
t1.effective_date,
t1.expiry_date
FROM
(
SELECT
t2.product_code product_code,
t2.product_name product_name,
t2.product_category product_category,
t1.version + 1 version,
${hivevar:pre_date} effective_date,
${hivevar:max_date} expiry_date
FROM product_dim t1
INNER JOIN rds.product t2
ON t1.product_code = t2.product_code
AND t1.expiry_date = ${hivevar:pre_date}
LEFT JOIN product_dim t3
ON t1.product_code = t3.product_code
AND t3.expiry_date = ${hivevar:max_date}
WHERE (t1.product_name <> t2.product_name OR t1.product_category <> t2.product_category) AND t3.product_sk IS NULL) t1
CROSS JOIN
(SELECT COALESCE(MAX(product_sk),0) sk_max FROM product_dim) t2;

-- 处理新增的product记录
INSERT INTO product_dim
SELECT
ROW_NUMBER() OVER (ORDER BY t1.product_code) + t2.sk_max,
t1.product_code,
t1.product_name,
t1.product_category,
1,
${hivevar:pre_date},
${hivevar:max_date}
FROM
(
SELECT t1.* FROM rds.product t1 LEFT JOIN product_dim t2 ON t1.product_code = t2.product_code
WHERE t2.product_sk IS NULL) t1
CROSS JOIN
(SELECT COALESCE(MAX(product_sk),0) sk_max FROM product_dim) t2;

-- 装载order维度
INSERT INTO order_dim
SELECT
ROW_NUMBER() OVER (ORDER BY t1.order_number) + t2.sk_max,
t1.order_number,
t1.version,
t1.effective_date,
t1.expiry_date
FROM
(
SELECT
order_number order_number,
1 version,
order_date effective_date,
'2200-01-01' expiry_date
FROM rds.sales_order, rds.cdc_time
WHERE entry_date >= last_load AND entry_date < current_load ) t1
CROSS JOIN
(SELECT COALESCE(MAX(order_sk),0) sk_max FROM order_dim) t2;

-- 装载销售订单事实表
INSERT INTO sales_order_fact
SELECT
order_sk,
customer_sk,
product_sk,
date_sk,
order_amount,
order_quantity
FROM
rds.sales_order a,
order_dim b,
customer_dim c,
product_dim d,
date_dim e,
rds.cdc_time f
WHERE
a.order_number = b.order_number
AND a.customer_number = c.customer_number
AND a.order_date >= c.effective_date
AND a.order_date < c.expiry_date
AND a.product_code = d.product_code
AND a.order_date >= d.effective_date
AND a.order_date < d.expiry_date
AND to_date(a.order_date) = e.date
AND a.entry_date >= f.last_load AND a.entry_date < f.current_load ;

-- 更新时间戳表的last_load字段
INSERT OVERWRITE TABLE rds.cdc_time SELECT current_load, current_load FROM rds.cdc_time;

4. 测试
（1）执行下面的SQL脚本往客户源数据里添加一个PA的客户和四个OH的客户。

use source;
insert into customer
( customer_name, customer_street_address, customer_zip_code,
customer_city, customer_state, shipping_address,
shipping_zip_code, shipping_city, shipping_state)
values
('PA Customer', '1111 Louise Dr.', '17050',
'Mechanicsburg', 'PA', '1111 Louise Dr.',
'17050', 'Mechanicsburg', 'PA'),
('Bigger Customers', '7777 Ridge Rd.', '44102',
'Cleveland', 'OH', '7777 Ridge Rd.',
'44102', 'Cleveland', 'OH'),
('Smaller Stores', '8888 Jennings Fwy.', '44102',
'Cleveland', 'OH', '8888 Jennings Fwy.',
'44102', 'Cleveland', 'OH'),
('Small-Medium Retailers', '9999 Memphis Ave.', '44102',
'Cleveland', 'OH', '9999 Memphis Ave.',
'44102', 'Cleveland', 'OH'),
('OH Customer', '6666 Ridge Rd.', '44102',
'Cleveland', 'OH', '6666 Ridge Rd.',
'44102','Cleveland', 'OH') ;

commit;

（2）执行定期装载并查看结果。
使用下面的命令执行定期装载。

./regular_etl.sh

使用下面的查询验证结果。

use dw;
select customer_name, customer_state, effective_date, expiry_date from pa_customer_dim;

结果是pa_customer_dim表增加了19条记录，如下图所示。



本节通过月份维度和PA客户维度的例子说明了两种典型维度子集的实现。