基于Hadoop生态圈的数据仓库实践 —— 进阶技术（三）

标签：

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‘) ;

-- 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‘);

-- 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;

• 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。

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

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

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

./on_demand.sh

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‘);

• 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。

-- 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;

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;

./regular_etl.sh

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

基于Hadoop生态圈的数据仓库实践 —— 进阶技术（三）

标签：

原文地址：http://blog.csdn.net/wzy0623/article/details/51941127