Hive II

Hive

数据仓库。OLAP(online analyze process)hdfs元数据关系型数据中。

Hive执行流程

cli交互driverdriver通过编译器进行编译(语法解析和语义解析)编译器查询metastore进行编译，生成计划。执行计划返回driver，driver提交执行引擎，执行引擎再提交作业给hadoop，hadoop返回结果直至client。tool,hadoop mr.

技术

hivehiveserver2beeline:

分区表 :分区就是目录。
桶表 :桶是文件。
内部表 :删除全删
外部表 :只删除表结构，表结构在metastore中。

结构化数据 ：

tinyint
smallint
int
bigint
float
double
decimal

array [,,,,]
struct {“”,12,}
named struct {“key1”:”v1”,”k2”,”v2”}
map {1:”“,,,,}
union {a:[]}

split()函数

explode()

炸裂函数,表生成函数。炸开的是array和map.

cross

优化手段

mapjoin

select /*+MAPJOIN()*/

创建表完整语法

CREATE TABLE employee    (    name string,    arr ARRAY<string>,    struc STRUCT<sex:string,age:int>,    map1 MAP<string,int>,    map2 MAP<string,ARRAY<string>>    )    ROW FORMAT DELIMITED        FIELDS TERMINATED BY '|'                //默认\001        COLLECTION ITEMS TERMINATED BY ','      //默认\002        MAP KEYS TERMINATED BY ':'              //默认\003        LINES TERMINATED BY '\n'                //行结束符        STORED AS TEXTFILE;                     //

Map端连接

    //连接暗示/*+ MAPJOIN(employee) */    SELECT /*+ MAPJOIN(employee) */ c.* FROM custs c CROSS JOIN orders o WHERE c.id  <> o.cid;    //通过设置自动map端连接转换，实现map连接    set hive.auto.convert.join=true    SELECT c.* FROM custs c CROSS JOIN orders o WHERE c.id  <> o.cid;

load data local inpath //上传
load data inpath //移动

load命令可以向分区表加载数据,无法加入指定的桶中.
桶表使用insert into..

//在shell中直接执行hive命令
$>hive -e “select * from mydb.custs”

//在shell命令行下执行hive的脚本
$>hive -f hive.sql

//导出表数据,导出元数据和数据本身
$hive>export table mydb.custs to ‘/user/centos/custs.dat’

//order by,全局排序,使用一个reduce实现全排序，强烈推荐数量用limit配合使用。
//对于每个reduce只要查询limit的数量即可。
$hive>select * from custs limit 3 ;

//设置mapred模式为严格模式，
//1.order by时必须使用limit限制结果数量
//2.如果分区表，必须指定分区。
hive>sethive.mapred.mode=stricthive>set hive.mapred.mode = nonstrict

//sort by是指map端排序
//在每个reduce中按照指定字段排序(asc|desc)
//如果mapred.reduce.tasks=1,等价于order by
//order by时始终使用一个reduce
// 没有hive前缀，是hadoop属性
$hive>set mapred.reduce.tasks=2

//distribute by等价mr的分区过程，按照指定字段进行分区,
//按照哪个列分发，必须出现在select语句中。
$hive>select * from orders distribute by cid sort by prices desc ;

//cluster by是快捷方式，如果使用同一个字段进行distribute和sort，
//可以使用该方式。
$hive>select * from orders cluster by cid ;

排序总结

1.order by    全局排序2.sort by    reduce内排序3.distribute by    分区，决定记录按哪个字段分配到分区。4.cluster by    distribute by x sort by x ;

//函数
//size()提取数据大小
$hive>select size(arr) from emp ;

//是否包含指定元素
$hive>select array_contains(arr,”xx”) from emp ;

//查看所有函数
show functions ;

//
desc formatted function array_contains ;
select current_database() ;
select current_user() ;
select current_date() ;

//case when == switch case
SELECT CASE WHEN length(name) <= 4 THEN ‘Short’ ELSE ‘Long’ END as xx FROM emp ;

//倒序字符串
select reverse(“12345”) ;
SELECT reverse(split(reverse(‘/home/user/employee.txt’),’/’)[0])

//创建数组对象
select array(1,1,1,2,3,4) ;

//collect_set()聚合函数，对结果集进行聚合，返回一个集合。
SELECT collect_set(work_place[0]) AS flat_workplace0 FROM employee;

//虚列，内置列
select INPUT__FILE__NAME from emp ;

//事务,hive 0.13.0之后完全支持行级acid事务处理。
//所有事务都是自动提交，并且存储文件只能是orc文件,而且只能在桶表中使用。
1.设置相关属性
SET hive.support.concurrency = true;
SET hive.enforce.bucketing = 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;

2.显式事务命令:
SHOW TRANSACTIONS;

3.操作语法
INSERT INTO TABLE tablename [PARTITION (partcol1[=val1], partcol2[=val2]…)] VALUES values_row [, values_row …];
UPDATE tablename SET column = value [, column = value…] [WHERE expression]
DELETE FROM tablename [WHERE expression]

4.创建表时，使用桶表，orc文件格式，支持事务的属性
create table tx(id int ,name string , age int)
clustered by (id) into 2 buckets
stored as orc
TBLPROPERTIES(‘transactional’=’true’);

5.执行操作
insert into tx(id,name,age) values(1,’tom’,2) ;
update tx set name = ‘tomas’ where id = 1 ;
delete from tx where id =1 ;

面向行存储

结构数据。select name from orders ;磁盘寻址非线性。

面向列存储

线性的。orc

数据聚合与采样

count()sum()avg()max()min()//查询每个customer的订单数select cid,count(*) from orders group by cid ;//错，select字段必须出现在group by中。select cid,orderno ,count(*) from group by cid  ;//去重集合select cid,collect_set(price) from group by cid  ;//select中出现多个聚合函数select cid,max(price),min(price) from group by cid  ;//coalesce返回第一个非空参数SELECT sum(coalesce(sex_age.age,0)) AS age_sum,sum(if(sex_age.sex = 'Female',sex_age.age,0))AS female_age_sum FROM employee;//不允许嵌套聚合,一下语句错误SELECT avg(count(*)) AS row_cnt ;//如果使用count + distinct组合，mapred.reduce.tasks属性失效，使用//使用一个reduce进行，类似于order by//map端聚合，预聚合，消耗更多内存。默认falseset hive.map.aggr=true

高级聚合

GROUPING SETS.group by + union all //查询每个cust的订单数select count(*) from orders group by cid ;select count(*) from orders group by orderno ;//group + union allselect count(*) from orders group by cid union all select count(*) from orders group by orderno ;//group by :指定按照哪些字段分组，//grouping sets : 以上字段集如何组合。select count(*) from orders group by cid,orderno grouping sets(cid,orderno,()) ;//

rollup

rollup扩展了group by，rollup比grouping sets多了一层聚合(n + 1)。GROUP BY a,b,c WITH ROLLUPGROUP BY a,b,c GROUPING SETS ((a,b,c),(a,b),(a),())select cid,orderno,count(*) GROUP BY cid,orderno GROUPING SETS ((cid,orderno),(cid),())

cube

扩展了grouping sets，做了各种条件的组合，不做排序。//代金券//vip//6点等价于GROUP BY a,b,c GROUPING SETS ((a,b,c),(a,b),(b,c),(a,c),(a),(b),(c),())

聚合条件

having,用于在组内过滤。//使用havingselect cid , max(price) mx from orders group by cid having mx > 100.1 ; //嵌套子查询select t.cid , t.mx from (select cid , max(price) mx from orders group by cid) t where t.mx > 100.1 ;

分析函数

0.11之后支持的，扫描多个输入的行计算每行的结果。通常和OVER, PARTITION BY, ORDER BY,windowing配合使用。和传统分组结果不一样，传统结果没组中只有一个结果(max)。分析函数的结果会出现多次，和每条记录都连接输出。Function (arg1,..., argn) OVER ([PARTITION BY <...>] [ORDER BY <....>] [<window_clause>])

SELECT  name, dept_num, salary,            COUNT(*) OVER (PARTITION BY dept_num) AS row_cnt,            SUM(salary) OVER(PARTITION BY dept_num ORDER BY dept_num) AS deptTotal,            SUM(salary) OVER(ORDER BY dept_num) AS runningTotal1,             SUM(salary) OVER(ORDER BY dept_num, name rows unbounded preceding) AS runningTotal2            FROM employee_contract            ORDER BY dept_num, name;

//宏观使用cid排序整个数据集，在分区内按照id降序排列。

SELECT  id, orderno, price,cid ,    COUNT(*) OVER (PARTITION BY cid) AS cnt  ,     min(price) over (partition by orderno order by id desc) FROM orders ORDER BY cid;    //    SELECT  id, orderno, price,cid ,    min(price) over (partition by orderno) FROM orders ORDER BY cid;    //order by每条记录内取.    SELECT  id, orderno, price,cid ,    min(price) over (order by price desc) FROM orders ORDER BY cid;    //分区都是独立分区，不是嵌套再分区    SELECT  id, orderno, price,cid ,    COUNT(*) OVER (PARTITION BY cid) AS cnt  ,     min(price) over (partition by orderno) FROM orders ORDER BY cid;    //分区内排序    SELECT  id, orderno, price,cid ,    min(price) over (partition by cid order by price desc) FROM orders ORDER BY cid;    //rank    SELECT  id, orderno, price,cid ,    RANK() OVER (PARTITION BY cid ORDER BY price) FROM orders ORDER BY cid;    //dense_rank    SELECT  id, orderno, price,cid ,    dense_rank() over (partition by cid) FROM orders ORDER BY cid;    //row_number()    SELECT  id, orderno, price,cid ,    row_number() over (partition by cid) FROM orders ORDER BY cid;    //CUME_DIST:    //PERCENT_RANK    currow-1 / totalrow - 1    1: 1 - 1 / 3 - 1 = 0    2: 2 - 1 / 3 - 1 = 0.5     3: 3 - 1 / 3 - 1 = 1    //NTILE:

CREATE TABLE employee(name string,dept_num int,salary float )ROW FORMAT DELIMITEDFIELDS TERMINATED BY ','LINES TERMINATED BY '\n'STORED AS TEXTFILE;

SELECT name, dept_num, salary,RANK()       OVER (PARTITION BY dept_num ORDER BY salary) AS rank,DENSE_RANK() OVER (PARTITION BY dept_num ORDER BY salary) AS dense_rank, ROW_NUMBER() OVER () AS row_num,ROUND((CUME_DIST() OVER (PARTITION BY dept_num ORDER BY salary)), 2) AS cume_dist,PERCENT_RANK() OVER(PARTITION BY dept_num ORDER BY salary) AS percent_rank,NTILE(4) OVER(PARTITION BY dept_num ORDER BY salary) AS ntileFROM employee ORDER BY dept_num;

CUME_DIST:累加分布

current row_num/ total rows,如果重复行，都取相同末尾行的行号。例如:1:  2 / 3 = 0.671:  2 / 3 = 0.672:  3 / 3 = 11:  1 / 3 = 0.332:  3 / 3 = 12:  3 / 3 = 11:  3 / 3 = 11:  3 / 3 = 11:  3 / 3 = 1

percent_rank

currentrow - 1 / totalrow - 1类似于cume_dist,但是提取相同rank的首行行号。1:  1 - 1 / 3 - 1 = 01:  1 - 1 / 3 - 1 = 02:  3 - 1 / 3 - 1 = 1 1:  1 - 1 / 3 - 1 = 02:  2 - 1 / 3 - 1 = 0.52:  2 - 1 / 3 - 1 = 0.5 

NTile

对每条记录分配桶的编号，桶的个数.指定桶的数。