hiveSQL语法及练习题整理(mysql)

目录

hiveSQL练习题整理:

第一题

第二题

第三题 

第四题

第五题

第六题 

第七题

第八题

第九题

第十题

第十一题

第十二题

hivesql常用函数:

hiveSQL常用操作语句(mysql)


hiveSQL练习题整理:

第一题

我们有如下的用户访问数据

userId  visitDate   visitCount

u01 2017/1/21   5

u02 2017/1/23   6

u03 2017/1/22   8

u04 2017/1/20   3

u01 2017/1/23   6

u01 2017/2/21   8

U02 2017/1/23   6

U01 2017/2/22   4

要求使用SQL统计出每个用户的累积访问次数,如下表所示:

用户id    月份  小计  累积

u01 2017-01 11  11

u01 2017-02 12  23

u02 2017-01 12  12

u03 2017-01 8   8

u04 2017-01 3   3



--建表

drop table if exists test_one;

create table test_one(

    userId string comment '用户id',

    visitDate string comment '访问日期',

    visitCount bigint comment '访问次数'

) comment '第一题'

row format delimited fields terminated by '\t';



--插入数据

insert into table test_one values('u01','2017/1/21',5);

insert into table test_one values('u02','2017/1/23',6);

insert into table test_one values('u03','2017/1/22',8);

insert into table test_one values('u04','2017/1/20',3);

insert into table test_one values('u01','2017/1/23',6);

insert into table test_one values('u01','2017/2/21',8);

insert into table test_one values('u02','2017/1/23',6);

insert into table test_one values('u01','2017/2/22',4);



--查询

select

    userId `用户id`,

    visitDate `月份`,

    sum_mn `小计`,

    sum(sum_mn) over(partition by userId rows between UNBOUNDED PRECEDING and current row) `累计`

from

(

    select

        t1.userId,

        t1.visitDate,

        sum(t1.visitCount) sum_mn

    from

    (

        select

            userId,

            --date_format(to_date(from_unixtime(UNIX_TIMESTAMP(visitDate,'yyyy/MM/dd'))),'yyyy-MM') visitDate,

            date_format(regexp_replace(visitdate,"/","-"),'yyyy-MM') visitDate,

            visitCount

        from test_one

    ) t1

    group by userId,visitDate

) t2;

第二题

有50W个京东店铺,每个顾客访客访问任何一个店铺的任何一个商品时都会产生一条访问日志,

访问日志存储的表名为Visit,访客的用户id为user_id,被访问的店铺名称为shop,请统计:

1)每个店铺的UV(访客数)

2)每个店铺访问次数top3的访客信息。输出店铺名称、访客id、访问次数



--建表

drop table if exists test_two;

create table test_two(

    shoop_name string COMMENT '店铺名称',

    user_id string COMMENT '用户id',

    visit_time string COMMENT '访问时间'

)

row format delimited fields terminated by '\t';



--插入数据

insert into table test_two values ('huawei','1001','2017-02-10');

insert into table test_two values ('icbc','1001','2017-02-10');

insert into table test_two values ('huawei','1001','2017-02-10');

insert into table test_two values ('apple','1001','2017-02-10');

insert into table test_two values ('huawei','1001','2017-02-10');

insert into table test_two values ('huawei','1002','2017-02-10');

insert into table test_two values ('huawei','1002','2017-02-10');

insert into table test_two values ('huawei','1001','2017-02-10');

insert into table test_two values ('huawei','1003','2017-02-10');

insert into table test_two values ('huawei','1004','2017-02-10');

insert into table test_two values ('huawei','1005','2017-02-10');

insert into table test_two values ('icbc','1002','2017-02-10');

insert into table test_two values ('jingdong','1006','2017-02-10');

insert into table test_two values ('jingdong','1003','2017-02-10');

insert into table test_two values ('jingdong','1002','2017-02-10');

insert into table test_two values ('jingdong','1004','2017-02-10');

insert into table test_two values ('apple','1001','2017-02-10');

insert into table test_two values ('apple','1001','2017-02-10');

insert into table test_two values ('apple','1001','2017-02-10');

insert into table test_two values ('apple','1002','2017-02-10');

insert into table test_two values ('apple','1002','2017-02-10');

insert into table test_two values ('apple','1005','2017-02-10');

insert into table test_two values ('apple','1005','2017-02-10');

insert into table test_two values ('apple','1006','2017-02-10');



--1)每个店铺的UV(访客数)

select

    shoop_name,

    count(*) shoop_uv

from test_two

group by shoop_name

order by shoop_uv desc;



--2)每个店铺访问次数top3的访客信息。输出店铺名称、访客id、访问次数

select

    shoop_name `商店名称`,

    user_id `用户id`,

    visit_time `访问次数`,

    rank_vis `忠诚排名`

from

    (

    select

        shoop_name,

        user_id,

        visit_time,

        row_number() over(partition by shoop_name order by visit_time desc) rank_vis



    from

    (

        select

            shoop_name,

            user_id,

            count(*) visit_time

        from test_two

        group by shoop_name,user_id

    ) t1

) t2

where rank_vis<=3;

第三题 


-- 已知一个表STG.ORDER,有如下字段:Date,Order_id,User_id,amount。

-- 请给出sql进行统计:数据样例:2017-01-01,10029028,1000003251,33.57。

-- 1)给出 2017年每个月的订单数、用户数、总成交金额。

-- 2)给出2017年11月的新客数(指在11月才有第一笔订单)



drop table if exists test_three_ORDER;

create table test_three_ORDER

(

    `Date` String COMMENT '下单时间',

    `Order_id` String COMMENT '订单ID',

    `User_id` String COMMENT '用户ID',

    `amount` decimal(10,2) COMMENT '金额'

)

row format delimited fields terminated by '\t';



--插入数据

insert into table test_three_ORDER values ('2017-10-01','10029011','1000003251',19.50);

insert into table test_three_ORDER values ('2017-10-03','10029012','1000003251',29.50);

insert into table test_three_ORDER values ('2017-10-04','10029013','1000003252',39.50);

insert into table test_three_ORDER values ('2017-10-05','10029014','1000003253',49.50);

insert into table test_three_ORDER values ('2017-11-01','10029021','1000003251',130.50);

insert into table test_three_ORDER values ('2017-11-03','10029022','1000003251',230.50);

insert into table test_three_ORDER values ('2017-11-04','10029023','1000003252',330.50);

insert into table test_three_ORDER values ('2017-11-05','10029024','1000003253',430.50);

insert into table test_three_ORDER values ('2017-11-07','10029025','1000003254',530.50);

insert into table test_three_ORDER values ('2017-11-15','10029026','1000003255',630.50);

insert into table test_three_ORDER values ('2017-12-01','10029027','1000003252',112.50);

insert into table test_three_ORDER values ('2017-12-03','10029028','1000003251',212.50);

insert into table test_three_ORDER values ('2017-12-04','10029029','1000003253',312.50);

insert into table test_three_ORDER values ('2017-12-05','10029030','1000003252',412.50);

insert into table test_three_ORDER values ('2017-12-07','10029031','1000003258',512.50);

insert into table test_three_ORDER values ('2017-12-15','10029032','1000003255',612.50);



-- 1)给出 2017年每个月的订单数、用户数、总成交金额。



select

    date_format(`date`,'yyyy-MM') `date`,

    count(*) `订单数`,

    count(distinct(user_id)) `用户数`,

    sum(amount) `总成交金额`

from test_three_ORDER

group by date_format(`date`,'yyyy-MM');



-- 2)给出2017年11月的新客数(指在11月才有第一笔订单)

select

    count(DISTINCT (t1.user_id))

from

(

    select

        user_id

    from test_three_ORDER

    where date_format(`date`,'yyyy-MM') = '2017-11'

    group by user_id

) t1

left join

(

select

    user_id

from test_three_ORDER

where date_format(`date`,'yyyy-MM') < '2017-11'

group by user_id

) t2

on t1.user_id = t2.user_id

where t2.user_id is null;

-- 第二种写法

select

    count(User_id) `11月新客数`

from

(

    SELECT

        User_id,

        Order_id,

        `Date`,

        LAG (`DATE`,1,0) over(partition by User_id order by `Date`) preOrder

    FROM

        test_three_ORDER

) t1

where date_format(`date`,'yyyy-MM')='2017-11' and preOrder=0;

第四题

-- 有一个5000万的用户文件(user_id,name,age),一个2亿记录的用户看电影的记录文件(user_id,url),

-- 根据年龄段观看电影的次数进行排序?

--建表

--用户表

drop table if exists test_four_log;

create table test_four_user(

    user_id string COMMENT '用户ID',

    name string COMMENT '用户姓名',

    age int COMMENT '用户年龄'

)

row format delimited fields terminated by '\t';

--日志表

drop table if exists test_four_log;

create table test_four_log(

    user_id string COMMENT '用户ID',

    url string COMMENT '链接'

)

row format delimited fields terminated by '\t';

--插入数据

insert into table test_four_user values ('1','1',8);

insert into table test_four_user values ('2','2',45);

insert into table test_four_user values ('3','3',14);

insert into table test_four_user values ('4','4',18);

insert into table test_four_user values ('5','5',17);

insert into table test_four_user values ('6','6',19);

insert into table test_four_user values ('7','7',26);

insert into table test_four_user values ('8','8',22);

insert into table test_four_log values('1','111');

insert into table test_four_log values('2','111');

insert into table test_four_log values('3','111');

insert into table test_four_log values('4','111');

insert into table test_four_log values('5','111');

insert into table test_four_log values('6','111');

insert into table test_four_log values('7','111');

insert into table test_four_log values('8','111');

insert into table test_four_log values('1','111');

insert into table test_four_log values('2','111');

insert into table test_four_log values('3','111');

insert into table test_four_log values('4','111');

insert into table test_four_log values('5','111');

insert into table test_four_log values('6','111');

insert into table test_four_log values('7','111');

insert into table test_four_log values('8','111');

insert into table test_four_log values('1','111');

insert into table test_four_log values('2','111');

insert into table test_four_log values('3','111');

insert into table test_four_log values('4','111');

insert into table test_four_log values('5','111');

insert into table test_four_log values('6','111');

insert into table test_four_log values('7','111');

insert into table test_four_log values('8','111');



-- 根据年龄段观看电影的次数进行排序?

select

    age_size `年龄段`,

    count(*) `观影次数`

from

(

    select

    u.*,

    l.url,

    case

    when u.age >=0 and u.age <= 10 then '1-10'

    when u.age >=11 and u.age <= 20 then '11-20'

    when u.age >=21 and u.age <= 30 then '21-30'

    when u.age >=31 and u.age <= 40 then '31-40'

    when u.age >=41 and u.age <= 50 then '41-50'

    else '51-100'

    end age_size

    from

    test_four_user u join test_four_log l on u.user_id = l.user_id

) t1

group by age_size

order by `观影次数` desc;

第五题

-- 有日志如下,请写出代码求得所有用户和活跃用户的总数及平均年龄。(活跃用户指连续两天都有访问记录的用户)

-- 日期 用户 年龄

-- 11,test_1,23

-- 11,test_2,19

-- 11,test_3,39

-- 11,test_1,23

-- 11,test_3,39

-- 11,test_1,23

-- 12,test_2,19

-- 13,test_1,23



create table test_five_active(

    active_time string COMMENT '活跃日期',

    user_id string COMMENT '用户id',

    age int COMMENT '用户年龄'

)

row format delimited fields terminated by '\t';



insert into table test_five_active values ('11','test_1',11);

insert into table test_five_active values ('11','test_2',22);

insert into table test_five_active values ('11','test_3',33);

insert into table test_five_active values ('11','test_4',44);



insert into table test_five_active values ('12','test_3',33);

insert into table test_five_active values ('12','test_5',55);

insert into table test_five_active values ('12','test_6',66);



insert into table test_five_active values ('13','test_4',44);

insert into table test_five_active values ('13','test_5',55);

insert into table test_five_active values ('13','test_7',77);



-- 所有用户的总数及平均年龄

select

    count(*) sum_user,

    avg(age) avg_age

from

(

    select

        user_id,

        avg(age) age

    from test_five_active

    group by user_id

) t1;



-- 活跃人数的总数及平均年龄

select  -- 最外一层算出活跃用户的个数以及平均年龄

    count(*),

    avg(d.age)

from

(

    select -- 最后还需要以user_id分组,去重(防止某个用户在11,12号连续活跃,然后在14,15号又连续活跃,导致diff求出不一致,所以此用户会出现两次)

        c.user_id,

        c.age

    from

    (

        select -- 以用户和差值diff分组,看分组下的数据的个数是否大于等于2(连续两天登录),取出活跃用户的数据

            b.user_id,

            b.age,

            b.diff,

            count(*) flag

        from

        (

            select  -- 用活跃日期减去排名,求出差值,看差值是否相等,相等差值的数据肯定是连续活跃的数据

                a.active_time,

                a.user_id,

                a.age,

                a.rank_time,

                a.active_time-a.rank_time diff

            from

            (

                select  -- 以用户和活跃日期分组(去重,防止某个用户在同一天活跃多次),求出每个用户的活跃日期排名

                    active_time,

                    user_id,

                    age,

                    rank() over(partition by user_id order by active_time) rank_time



                from test_five_active

                group by active_time,user_id,age  

            ) a

        ) b

        group by b.user_id,b.age,b.diff

        having count(*) >=2

    ) c

    group by c.user_id,c.age

) d;

第六题 

请用sql写出所有用户中在今年10月份第一次购买商品的金额,

表ordertable字段(购买用户:userid,金额:money,购买时间:paymenttime(格式:2017-10-01),订单id:orderid)



create table test_six_ordertable

(

    `userid` string COMMENT '购买用户',

    `money` decimal(10,2) COMMENT '金额',

    `paymenttime` string COMMENT '购买时间',

    `orderid` string COMMENT '订单id'

)

row format delimited fields terminated by '\t';



--插入数据

insert into table test_six_ordertable values('1',1,'2017-09-01','1');

insert into table test_six_ordertable values('2',2,'2017-09-02','2');

insert into table test_six_ordertable values('3',3,'2017-09-03','3');

insert into table test_six_ordertable values('4',4,'2017-09-04','4');



insert into table test_six_ordertable values('3',5,'2017-10-05','5');

insert into table test_six_ordertable values('6',6,'2017-10-06','6');

insert into table test_six_ordertable values('1',7,'2017-10-07','7');

insert into table test_six_ordertable values('8',8,'2017-10-09','8');

insert into table test_six_ordertable values('6',6,'2017-10-16','60');

insert into table test_six_ordertable values('1',7,'2017-10-17','70');



-- 写出所有用户中在今年10月份第一次购买商品的金额

select

    userid,

    `money`,

    paymenttime,

    orderid

from

(

    select

        userid,

        `money`,

        paymenttime,

        orderid,

        rank() over(partition by userid order by paymenttime) rank_time

    from test_six_ordertable

    where date_format(paymenttime,'yyyy-MM') = '2017-10'

) a

where rank_time=1;


 

第七题

--现有图书管理数据库的三个数据模型如下:



--图书(数据表名:BOOK)

--序号    字段名称     字段描述       字段类型

--1       BOOK_ID     总编号         文本

--2       SORT        分类号         文本

--3       BOOK_NAME   书名           文本

--4       WRITER      作者           文本

--5       OUTPUT      出版单位       文本

--6       PRICE       单价           数值(保留小数点后2位)



--读者(数据表名:READER)

--序号      字段名称        字段描述    字段类型

--1       READER_ID       借书证号         文本

--2       COMPANY         单位            文本

--3       NAME            姓名             文本

--4       SEX             性别          文本

--5       GRADE           职称          文本

--6       ADDR            地址          文本

--

--借阅记录(数据表名:BORROW LOG)

--序号      字段名称        字段描述        字段类型

--1        READER_ID      借书证号            文本

--2       BOOK_D          总编号             文本

--3       BORROW_ATE      借书日期            日期




--(1)创建图书管理库的图书、读者和借阅三个基本表的表结构。请写出建表语句。

--图书

create table test_seven_BOOK

(

    BOOK_ID String COMMENT '总编号',

    SORT String COMMENT '分类号',

    BOOK_NAME String COMMENT '书名',

    WRITER String COMMENT '作者',

    OUTPUT String COMMENT '出版单位',

    PRICE decimal(10,2) COMMENT '单价'

)

row format delimited fields terminated by '\t';



--读者

create table test_seven_READER

(  

    READER_ID String COMMENT '借书证号',

    COMPANY String COMMENT '单位',

    NAME String COMMENT '姓名',

    SEX String COMMENT '性别',

    GRADE String COMMENT '职称',

    ADDR String COMMENT '地址'

)

row format delimited fields terminated by '\t';



--借阅记录

create table test_seven_BORROW_LOG

(  

    READER_ID String COMMENT '借书证号',

    BOOK_D String COMMENT '总编号',

    BORROW_ATE date COMMENT '借书日期'

)

row format delimited fields terminated by '\t';



-- 插入数据

insert into table test_seven_book values ('1001','A1','Java','James Gosling','sun','11');

insert into table test_seven_book values ('1002','A2','linux','Linus Benedict Torvalds','sun','22');

insert into table test_seven_book values ('1003','A3','Java3','James Gosling3','sun3','33');

insert into table test_seven_book values ('1004','A4','Java4','James Gosling4','sun4','44');

insert into table test_seven_book values ('1005','B1','Java5','James Gosling5','sun','55');

insert into table test_seven_book values ('1006','C1','Java6','James Gosling6','sun5','66');

insert into table test_seven_book values ('1007','D1','Java7','James Gosling7','sun6','77');

insert into table test_seven_book values ('1008','E1','Java8','James Gosling4','sun3','88');

insert into table test_seven_reader values ('7','buu',decode(binary('李大帅'),'utf-8'),'man','lay1','beijing4');

insert into table test_seven_reader values ('2','buu2','苏大强','man','lay2','beijing2');

insert into table test_seven_reader values ('3','buu2','李二胖','woman','lay3','beijing3');

insert into table test_seven_reader values ('4','buu3','王三涛','man','lay4','beijing4');

insert into table test_seven_reader values ('5','buu4','刘四虎','woman','lay5','beijing1');

insert into table test_seven_reader values ('6','buu','宋冬野','woman','lay6','beijing5');

insert into table test_seven_borrow_log values ('1','1002','2019-06-01');

insert into table test_seven_borrow_log values ('1','1003','2019-06-02');

insert into table test_seven_borrow_log values ('1','1006','2019-06-03');

insert into table test_seven_borrow_log values ('2','1001','2019-06-04');

insert into table test_seven_borrow_log values ('3','1002','2019-06-05');

insert into table test_seven_borrow_log values ('4','1005','2019-06-06');

insert into table test_seven_borrow_log values ('5','1003','2019-06-06');

insert into table test_seven_borrow_log values ('3','1006','2019-06-07');

insert into table test_seven_borrow_log values ('2','1003','2019-06-03');

insert into table test_seven_borrow_log values ('3','1008','2019-06-03');

insert into table test_seven_borrow_log values ('1','1002','2019-06-04');

--(2)找出姓李的读者姓名(NAME)和所在单位(COMPANY)。

select name,company from test_seven_reader where name like '李%';

--(3)查找“高等教育出版社”的所有图书名称(BOOK_NAME)及单价(PRICE),结果按单价降序排序。

select BOOK_NAME,PRICE from test_seven_book order by PRICE desc;

--(4)查找价格介于10元和20元之间的图书种类(SORT)出版单位(OUTPUT)和单价(PRICE),结果按出版单位(OUTPUT)和单价(PRICE)升序排序。

select SORT,OUTPUT,PRICE from test_seven_book where PRICE between 10 and 20 order by OUTPUT,PRICE asc;



--(5)查找所有借了书的读者的姓名(NAME)及所在单位(COMPANY)。

select

    rd.name,

    rd.COMPANY

from

(

select

    READER_ID

from test_seven_borrow_log

group by READER_ID

) t1

join

test_seven_reader rd

on t1.READER_ID = rd.READER_ID;

--(6)求”科学出版社”图书的最高单价、最低单价、平均单价。

select

    max(PRICE) max,

    min(PRICE) min,

    avg(PRICE) avg

from

test_seven_book;

--(7)找出当前至少借阅了2本图书(大于等于2本)的读者姓名及其所在单位。

select

   rd.READER_ID,

   rd.name,

   rd.COMPANY

from

(

    select

        READER_ID,

        count(*) num

    from test_seven_BORROW_LOG

    group by READER_ID

    having count(*) >= 2

) t1

join

test_seven_reader rd

on t1.READER_ID = rd.READER_ID;



--(8)考虑到数据安全的需要,需定时将“借阅记录”中数据进行备份,

-- 请使用一条SQL语句,在备份用户bak下创建与“借阅记录”表结构完全一致的数据表BORROW_LOG_BAK.

--井且将“借阅记录”中现有数据全部复制到BORROW_l0G_BAK中。

create table BORROW_LOG_BAK

(

    READER_ID String COMMENT '借书证号',

    BOOK_D String COMMENT '总编号',

    BORROW_ATE date COMMENT '借书日期'

)

as select * from test_seven_BORROW_LOG;



--(9)现在需要将原Oracle数据库中数据迁移至Hive仓库,

--请写出“图书”在Hive中的建表语句(Hive实现,提示:列分隔符|;数据表数据需要外部导入:分区分别以month_part、day_part 命名)

create table test_seven_book_oracle (

    book_id string COMMENT '总编号',

    sort string COMMENT '分类号',

    book_name string COMMENT '书名',

    writer string COMMENT '作者',

    output string COMMENT '出版单位',

    price decimal(10,2) COMMENT '单价'

)

PARTITIONED BY (month string,day string)

row format delimited fields terminated by '|';



--(10)Hive中有表A,现在需要将表A的月分区 201505 中 user_id为20000的user_dinner字段更新为bonc8920,

--  其他用户user_dinner字段数据不变,请列出更新的方法步骤。

--(Hive实现,提示:Hlive中无update语法,请通过其他办法进行数据更新)

create table tmp_A as select * from A where user_id<>20000 and month_part=201505;

insert into table tmp_A partition(month_part=’201505’) values(20000,其他字段,bonc8920);

insert overwrite table A partition(month_part=’201505’) select * from tmp_A where month_part=201505;



第八题

-- 有一个线上服务器访问日志格式如下(用sql答题)

--        时间                   接口                           ip地址

-- 2016-11-09 11:22:05    /api/user/login                  110.23.5.33

-- 2016-11-09 11:23:10    /api/user/detail                 57.3.2.16

-- .....

-- 2016-11-09 23:59:40    /api/user/login                  200.6.5.166

-- 求11月9号下午14点(14-15点),访问api/user/login接口的top10的ip地址



create table test_eight_serverlog

(

    server_time string COMMENT '时间',

    server_api  string comment '接口',

    server_ip string COMMENT 'ip地址'

)

row format delimited fields terminated by '\t';



insert into table test_eight_serverlog values ('2016-11-09 11:22:05','/api/user/login','110.23.5.33');

insert into table test_eight_serverlog values ('2016-11-09 11:23:10','/api/user/detail','57.3.2.16');

insert into table test_eight_serverlog values ('2016-11-09 14:59:40','/api/user/login','200.6.5.161');

insert into table test_eight_serverlog values ('2016-11-09 14:22:05','/api/user/login','110.23.5.32');

insert into table test_eight_serverlog values ('2016-11-09 14:23:10','/api/user/detail','57.3.2.13');

insert into table test_eight_serverlog values ('2016-11-09 14:59:40','/api/user/login','200.6.5.164');

insert into table test_eight_serverlog values ('2016-11-09 14:59:40','/api/user/login','200.6.5.165');

insert into table test_eight_serverlog values ('2016-11-09 14:22:05','/api/user/login','110.23.5.36');

insert into table test_eight_serverlog values ('2016-11-09 14:23:10','/api/user/detail','57.3.2.17');

insert into table test_eight_serverlog values ('2016-11-09 14:59:40','/api/user/login','200.6.5.168');

insert into table test_eight_serverlog values ('2016-11-09 14:59:40','/api/user/login','200.6.5.168');

insert into table test_eight_serverlog values ('2016-11-09 14:22:05','/api/user/login','110.23.5.32');

insert into table test_eight_serverlog values ('2016-11-09 14:23:10','/api/user/detail','57.3.2.13');

insert into table test_eight_serverlog values ('2016-11-09 14:59:40','/api/user/login','200.6.5.164');

insert into table test_eight_serverlog values ('2016-11-09 15:22:05','/api/user/login','110.23.5.33');

insert into table test_eight_serverlog values ('2016-11-09 15:23:10','/api/user/detail','57.3.2.16');

insert into table test_eight_serverlog values ('2016-11-09 15:59:40','/api/user/login','200.6.5.166');




select

    server_ip,

    count(*) visit_time

from test_eight_serverlog

where date_format(server_time,'yyyy-MM-dd HH')='2016-11-09 14'

and server_api = '/api/user/login'

group by server_ip

order by visit_time desc;

第九题


-- 有一个充值日志表如下:

-- CREATE TABLE `credit log`

-- (

--     `dist_id` int(11)DEFAULT NULL COMMENT '区组id',

--     `account` varchar(100)DEFAULT NULL COMMENT '账号',

--     `money` int(11) DEFAULT NULL COMMENT '充值金额',

--     `create_time` datetime DEFAULT NULL COMMENT '订单时间'

-- )ENGINE=InnoDB DEFAUILT CHARSET-utf8

-- 请写出SQL语句,查询充值日志表2015年7月9号每个区组下充值额最大的账号,要求结果:

-- 区组id,账号,金额,充值时间

--建表

create table test_nine_credit_log(

    dist_id string COMMENT '区组id',

    account string COMMENT '账号',

    `money` decimal(10,2) COMMENT '充值金额',

    create_time string COMMENT '订单时间'

)

row format delimited fields terminated by '\t';



--插入数据

insert into table test_nine_credit_log values ('1','11',100006,'2019-01-02 13:00:01');

insert into table test_nine_credit_log values ('1','12',110000,'2019-01-02 13:00:02');

insert into table test_nine_credit_log values ('1','13',102000,'2019-01-02 13:00:03');

insert into table test_nine_credit_log values ('1','14',100300,'2019-01-02 13:00:04');

insert into table test_nine_credit_log values ('1','15',100040,'2019-01-02 13:00:05');

insert into table test_nine_credit_log values ('1','18',110000,'2019-01-02 13:00:02');

insert into table test_nine_credit_log values ('1','16',100005,'2019-01-03 13:00:06');

insert into table test_nine_credit_log values ('1','17',180000,'2019-01-03 13:00:07');




insert into table test_nine_credit_log values ('2','21',100800,'2019-01-02 13:00:11');

insert into table test_nine_credit_log values ('2','22',100030,'2019-01-02 13:00:12');

insert into table test_nine_credit_log values ('2','23',100000,'2019-01-02 13:00:13');

insert into table test_nine_credit_log values ('2','24',100010,'2019-01-03 13:00:14');

insert into table test_nine_credit_log values ('2','25',100070,'2019-01-03 13:00:15');

insert into table test_nine_credit_log values ('2','26',100800,'2019-01-02 15:00:11');



insert into table test_nine_credit_log values ('3','31',106000,'2019-01-02 13:00:08');

insert into table test_nine_credit_log values ('3','32',100400,'2019-01-02 13:00:09');

insert into table test_nine_credit_log values ('3','33',100030,'2019-01-02 13:00:10');

insert into table test_nine_credit_log values ('3','34',100003,'2019-01-02 13:00:20');

insert into table test_nine_credit_log values ('3','35',100020,'2019-01-02 13:00:30');

insert into table test_nine_credit_log values ('3','36',100500,'2019-01-02 13:00:40');

insert into table test_nine_credit_log values ('3','37',106000,'2019-01-03 13:00:50');

insert into table test_nine_credit_log values ('3','38',100800,'2019-01-03 13:00:59');



--查询充值日志表2019年1月2号每个区组下充值额最大的账号,要求结果:区组id,账号,金额,充值时间

select

    aaa.dist_id,

    aaa.account,

    aaa.`money`,

    aaa.create_time,

    aaa.money_rank

from

(

    select

        dist_id,

        account,

        `money`,

        create_time,

        dense_rank() over(partition by dist_id order by `money` desc) money_rank   -- dense_rank最完美,因为不仅可以求第一多,而且还可以求第二多,第三多...

    from test_nine_credit_log

    where date_format(create_time,'yyyy-MM-dd') = '2019-01-02'

) aaa

where money_rank = 1;



-- 第二种写法,不用开窗函数

with

tmp_max_money as(

    select

        dist_id,

        max(`money`) max

    from test_nine_credit_log

    where date_format(create_time,'yyyy-MM-dd')='2019-01-02'

    group by dist_id

)

select

    cl.dist_id dist_id,cl.account acount,cl.money money,cl.create_time create_time

from test_nine_credit_log cl

left join tmp_max_money mm

on cl.dist_id=mm.dist_id

where cl.money=mm.max and date_format(create_time,'yyyy-MM-dd')='2019-01-02';

第十题


-- 有一个账号表如下,请写出SQL语句,查询各自区组的money排名前十的账号(分组取前10)

-- CREATE TABIE `account`

-- (

--     `dist_id` int(11)

--     DEFAULT NULL COMMENT '区组id',

--     `account` varchar(100)DEFAULT NULL COMMENT '账号' ,

--     `gold` int(11)DEFAULT NULL COMMENT '金币'

--     PRIMARY KEY (`dist_id`,`account_id`),

-- )ENGINE=InnoDB DEFAULT CHARSET-utf8

-- 替换成hive表

drop table if exists `test_ten_account`;

create table `test_ten_account`(

    `dist_id` string COMMENT '区组id',

    `account` string COMMENT '账号',

    `gold` bigint COMMENT '金币'

)

row format delimited fields terminated by '\t';



insert into table test_ten_account values ('1','11',100006);

insert into table test_ten_account values ('1','12',110000);

insert into table test_ten_account values ('1','13',102000);

insert into table test_ten_account values ('1','14',100300);

insert into table test_ten_account values ('1','15',100040);

insert into table test_ten_account values ('1','18',110000);

insert into table test_ten_account values ('1','16',100005);

insert into table test_ten_account values ('1','17',180000);



insert into table test_ten_account values ('2','21',100800);

insert into table test_ten_account values ('2','22',100030);

insert into table test_ten_account values ('2','23',100000);

insert into table test_ten_account values ('2','24',100010);

insert into table test_ten_account values ('2','25',100070);

insert into table test_ten_account values ('2','26',100800);



insert into table test_ten_account values ('3','31',106000);

insert into table test_ten_account values ('3','32',100400);

insert into table test_ten_account values ('3','33',100030);

insert into table test_ten_account values ('3','34',100003);

insert into table test_ten_account values ('3','35',100020);

insert into table test_ten_account values ('3','36',100500);

insert into table test_ten_account values ('3','37',106000);

insert into table test_ten_account values ('3','38',100800);



select

    dist_id,

    account,

    gold,

    gold_rank

from

(

    select

        `dist_id`,

        `account`,

        `gold`,

        dense_rank() over(partition by dist_id order by gold desc) gold_rank

    from test_ten_account

) tmp

where gold_rank <= 3;

第十一题


-- 1)有三张表分别为会员表(member)销售表(sale)退货表(regoods)

-- (1)会员表有字段memberid(会员id,主键)credits(积分);

-- (2)销售表有字段memberid(会员id,外键)购买金额(MNAccount);

-- (3)退货表中有字段memberid(会员id,外键)退货金额(RMNAccount);

-- 2)业务说明:

-- (1)销售表中的销售记录可以是会员购买,也可是非会员购买。(即销售表中的memberid可以为空)

-- (2)销售表中的一个会员可以有多条购买记录

-- (3)退货表中的退货记录可以是会员,也可是非会员4、一个会员可以有一条或多条退货记录

-- 查询需求:分组查出销售表中所有会员购买金额,同时分组查出退货表中所有会员的退货金额,

--  把会员id相同的购买金额-退款金额得到的结果更新到表会员表中对应会员的积分字段(credits)

-- 建表

--会员表

drop table if exists test_eleven_member;

create table test_eleven_member(

    memberid string COMMENT '会员id',

    credits bigint COMMENT '积分'

)

row format delimited fields terminated by '\t';

--销售表

drop table if exists test_eleven_sale;

create table test_eleven_sale(

    memberid string COMMENT '会员id',

    MNAccount decimal(10,2) COMMENT '购买金额'

)

row format delimited fields terminated by '\t';

--退货表

drop table if exists test_eleven_regoods;

create table test_eleven_regoods(

    memberid string COMMENT '会员id',

    RMNAccount decimal(10,2) COMMENT '退货金额'

)

row format delimited fields terminated by '\t';



insert into table test_eleven_member values('1001',0);

insert into table test_eleven_member values('1002',0);

insert into table test_eleven_member values('1003',0);

insert into table test_eleven_member values('1004',0);

insert into table test_eleven_member values('1005',0);

insert into table test_eleven_member values('1006',0);

insert into table test_eleven_member values('1007',0);



insert into table test_eleven_sale values('1001',5000);

insert into table test_eleven_sale values('1002',4000);

insert into table test_eleven_sale values('1003',5000);

insert into table test_eleven_sale values('1004',6000);

insert into table test_eleven_sale values('1005',7000);

insert into table test_eleven_sale values('1004',3000);

insert into table test_eleven_sale values('1002',6000);

insert into table test_eleven_sale values('1001',2000);

insert into table test_eleven_sale values('1004',3000);

insert into table test_eleven_sale values('1006',3000);

insert into table test_eleven_sale values(NULL,1000);

insert into table test_eleven_sale values(NULL,1000);

insert into table test_eleven_sale values(NULL,1000);

insert into table test_eleven_sale values(NULL,1000);



insert into table test_eleven_regoods values('1001',1000);

insert into table test_eleven_regoods values('1002',1000);

insert into table test_eleven_regoods values('1003',1000);

insert into table test_eleven_regoods values('1004',1000);

insert into table test_eleven_regoods values('1005',1000);

insert into table test_eleven_regoods values('1002',1000);

insert into table test_eleven_regoods values('1001',1000);

insert into table test_eleven_regoods values('1003',1000);

insert into table test_eleven_regoods values('1002',1000);

insert into table test_eleven_regoods values('1005',1000);

insert into table test_eleven_regoods values(NULL,1000);

insert into table test_eleven_regoods values(NULL,1000);

insert into table test_eleven_regoods values(NULL,1000);

insert into table test_eleven_regoods values(NULL,1000);



-- 分组查出销售表中所有会员购买金额,同时分组查出退货表中所有会员的退货金额,

-- 把会员id相同的购买金额-退款金额得到的结果更新到表会员表中对应会员的积分字段(credits)

with

tmp_member as

(  

    select memberid,sum(credits) credits

    from test_eleven_member

    group by memberid

),

tmp_sale as

(

    select memberid,sum(MNAccount) MNAccount

    from test_eleven_sale

    group by memberid

),

tmp_regoods as

(

    select memberid,sum(RMNAccount) RMNAccount

    from test_eleven_regoods

    group by memberid

)

insert overwrite table test_eleven_member

select

    t1.memberid,

    sum(t1.creadits)+sum(t1.MNAccount)-sum(t1.RMNAccount) credits

from

(

    select

        memberid,

        credits,

        0 MNAccount,

        0 RMNAccount

    from tmp_member

    union all

    select

        memberid,

        0 credits,

        MNAccount,

        0 RMNAccount

    from tmp_sale

    union all

    select

        memberid,

        0 credits,

        0 MNAccount,

        RMNAccount

    from tmp_regoods

) t1

where t1.memberid is not NULL

group by t1.memberid

---------------------第2种写法-用left join--------------------------

insert overwrite table test_eleven_member

select

    t3.memberid,

    sum(t3.credits) credits

from

(

    select

        t1.memberid,

        t1.MNAccount - NVL(t2.RMNAccount,0) credits

    from

    (

        select

            memberid,

            sum(MNAccount) MNAccount

        from test_eleven_sale

        group by memberid

    ) t1

    left join

    (

        select

            memberid,

            sum(RMNAccount) RMNAccount

        from test_eleven_regoods

        group by memberid

    )t2

    on t1.memberid = t2.memberid

    where t1.memberid is not NULL



    union all



    select

        memberid,

        credits

    from test_eleven_member

) t3

group by t3.memberid;

第十二题

--现在有三个表student(学生表)、course(课程表)、score(成绩单),结构如下:

--建表

create table test_twelve_student

(

    id bigint comment '学号',

    name string comment '姓名',

    age bigint comment '年龄'

)

row format delimited fields terminated by '\t';



create table test_twelve_course

(

    cid string comment '课程号,001/002格式',

    cname string comment '课程名'

)

row format delimited fields terminated by '\t';



Create table test_twelve_score

(

    id bigint comment '学号',

    cid string comment '课程号',

    score bigint comment '成绩'

)

row format delimited fields terminated by '\t';




--插入数据

insert into table test_twelve_student values (1001,'wsl1',21);

insert into table test_twelve_student values (1002,'wsl2',22);

insert into table test_twelve_student values (1003,'wsl3',23);

insert into table test_twelve_student values (1004,'wsl4',24);

insert into table test_twelve_student values (1005,'wsl5',25);



insert into table test_twelve_course values ('001','math');

insert into table test_twelve_course values ('002','English');

insert into table test_twelve_course values ('003','Chinese');

insert into table test_twelve_course values ('004','music');



insert into table test_twelve_score values (1001,'004',10);

insert into table test_twelve_score values (1002,'003',21);

insert into table test_twelve_score values (1003,'002',32);

insert into table test_twelve_score values (1004,'001',43);

insert into table test_twelve_score values (1005,'003',54);

insert into table test_twelve_score values (1001,'002',65);

insert into table test_twelve_score values (1002,'004',76);

insert into table test_twelve_score values (1003,'002',77);

insert into table test_twelve_score values (1001,'004',48);

insert into table test_twelve_score values (1002,'003',39);




--其中score中的id、cid,分别是student、course中对应的列请根据上面的表结构,回答下面的问题

--1)请将本地文件(/home/users/test/20190301.csv)文件,加载到分区表score的20190301分区中,并覆盖之前的数据

load data local inpath '/home/users/test/20190301.csv' overwrite into table test_twelve_score partition(event_day='20190301');

--2)查出平均成绩大于60分的学生的姓名、年龄、平均成绩

select

    stu.name,

    stu.age,

    t1.avg_score

from

test_twelve_student stu

join

(

    select

        id,

        avg(score) avg_score

    from test_twelve_score

    group by id

) t1

on t1.id = stu.id

where avg_score > 60;

--3)查出没有'001'课程成绩的学生的姓名、年龄

select

    stu.name,

    stu.age

from

test_twelve_student stu

join

(

    select

        id

    from test_twelve_score

    where cid != 001

    group by id

) t1

on stu.id = t1.id;

--4)查出有'001'\'002'这两门课程下,成绩排名前3的学生的姓名、年龄

select

    stu.name,

    stu.age

from

(

    select

        id,

        cid,

        score,

        rank() over(partition by cid order by score desc) ran

    from

    test_twelve_score

    where cid = 001 or cid = 002

) t1

join test_twelve_student stu

on t1.id = stu.id

where ran <= 3;




--5)创建新的表score_20190317,并存入score表中20190317分区的数据

create table score_20190317

as select * from test_twelve_score where dt = '20190317';

--6)如果上面的score_20190317score表中,uid存在数据倾斜,请进行优化,查出在20190101-20190317中,学生的姓名、年龄、课程、课程的平均成绩

select

    stu.name,

    stu.age,

    cou.cname,

    t1.avg_score



from

(

    select

        id,

        cid,

        avg(score) avg_score

    from test_twelve_score

    group by id,cid

    where dt >= '20190101' and dt <= '20190317'

) t1

left join test_twelve_student stu on t1.id = stu.id

left join test_twelve_course cou on t1.cid = cou.cid

--7)描述一下union和union all的区别,以及在mysql和HQL中用法的不同之处?



union会对数据进行排序去重,union all不会排序去重。

HQL中要求union或union all操作时必须保证select 集合的结果相同个数的列,并且每个列的类型是一样的。



--8)简单描述一下lateral view语法在HQL中的应用场景,并写一个HQL实例

-- 比如一个学生表为:

-- 学号  姓名  年龄  成绩(语文|数学|英语)

-- 001   张三  16     90,80,95

-- 需要实现效果:

-- 学号  成绩

-- 001 90

-- 001 80

-- 001 95




create table student(

`id` string,

`name` string,

`age` int,

`scores` array

)

row format delimited fields terminated by '\t'

collection items terminated by ',';




select

    id,

    score

from

student lateral view explode(scores) tmp_score as score;

hivesql常用函数:

-------------------------Hive  SQL  常用函数  ----------------------------

--常用日期函数
--unix_timestamp:返回当前或指定时间的时间戳	        select unix_timestamp();  select unix_timestamp('2008-08-08 08:08:08'); 
--from_unixtime:将时间戳转为日期格式                 select from_unixtime(1218182888);
--current_date:当前日期                  select current_date();
--current_timestamp:当前的日期加时间     select current_timestamp();
--to_date:抽取日期部分                   select to_date('2008-08-08 08:08:08');   select to_date(current_timestamp());
--year:获取年                            select year(current_timestamp());
--month:获取月                           select month(current_timestamp());
--day:获取日                             select DAY(current_timestamp());
--hour:获取时                            select HOUR(current_timestamp());
--minute:获取分                          select minute(current_timestamp());
--second:获取秒                          select SECOND(current_timestamp());
--weekofyear:当前时间是一年中的第几周    select weekofyear(current_timestamp());  select weekofyear('2020-01-08');
--dayofmonth:当前时间是一个月中的第几天  select dayofmonth(current_timestamp());  select dayofmonth('2020-01-08');
--months_between: 两个日期间的月份       select months_between('2020-07-29','2020-06-28');
--add_months:日期加减月                  select add_months('2020-06-28',1);
--datediff:两个日期相差的天数            select datediff('2019-03-01','2019-02-01');   select datediff('2020-03-01','2020-02-01');
--date_add:日期加天数                    select date_add('2019-02-28',1);   select date_add('2020-02-28',1);
--date_sub:日期减天数                    select date_sub('2019-03-01',1);   select date_sub('2020-03-01',1);
--last_day:日期的当月的最后一天          select last_day('2020-02-28');   select last_day('2019-02-28');
--date_format() :格式化日期   日期格式:'yyyy-MM-dd hh:mm:ss'   select date_format('2008-08-08 08:08:08','yyyy-MM-dd hh:mm:ss');  


--常用取整函数
--round: 四舍五入     select round(4.5);     
--ceil:  向上取整     select ceil(4.5);
--floor: 向下取整     select floor(4.5);
--
--常用字符串操作函数
--upper: 转大写         select upper('abcDEFg');
--lower: 转小写         select lower('abcDEFg');
--length: 长度          select length('abcDEFg');
--trim:  前后去空格     select length('   abcDEFg    ');  select length(trim('   abcDEFg    '));
--lpad: 向左补齐,到指定长度   select lpad('abc',11,'*');
--rpad:  向右补齐,到指定长度  select rpad('abc',11,'*');  
--substring: 剪切字符串         select substring('abcdefg',1,3);     select rpad(substring('13843838438',1,3),11,'*');
--regexp_replace: SELECT regexp_replace('100-200', '(\\d+)', 'num');   select regexp_replace('abc d e f',' ','');
--	使用正则表达式匹配目标字符串,匹配成功后替换!
--
--集合操作
--size: 集合中元素的个数
--map_keys: 返回map中的key
--map_values: 返回map中的value         select size(friends),map_keys(children),map_values(children) from person;
--array_contains: 判断array中是否包含某个元素     select array_contains(friends,'lili') from person;
--sort_array: 将array中的元素排序         select sort_array(split('1,3,4,5,2,6,9',','));   
--                                         select sort_array(split('a,d,g,b,c,f,e',','));

--------------------常用日期函数
--返回时间戳
select  unix_timestamp();--返回当前时间到1970年1月1号的时间戳(经过了多少秒)
select  unix_timestamp("1970-01-01 00:00:05");--指定时间的时间戳
--时间戳转日期
select from_unixtime(5);
--当前日期
select current_date();
--当前的日期加时间
select current_timestamp();
--抽取日期部分
select to_date(current_timestamp());
select to_date('2008-08-08 08:08:08');
--获取年月日、时分秒  (注意,必须满足日期和时间的格式才能识别)
select year(current_timestamp()),month (current_timestamp()),day(current_timestamp()),
hour(current_timestamp()),minute (current_timestamp()), second(current_timestamp());
--当前时间或指定时间是一年中的第几周 、 一个月中的第几天
select weekofyear(current_timestamp());
select weekofyear('2008-08-08 08:08:08');
select dayofmonth(CURRENT_date());
select dayofmonth('2008-08-08 08:08:08');
--两个日期间的月份    两个日期见相差的天数
select months_between('2008-08-08','2008-09-08');
select datediff('2008-09-08','2008-08-08');
--日期加减月、   加减天
select add_months('2008-08-08',1);
select date_add('2008-08-08',1);
select date_sub('2008-08-08',1);
--日期的当月的最后一天 
select last_day('2008-08-08');
--格式化日期   日期格式:'yyyy-MM-dd hh:mm:ss'  把日期转化为SQL能够识别的格式
select date_format('2008-08-08 08:08:08','yyyy-MM-dd hh:mm:ss');


--------------------------------常用取整函数          具体的使用看需求
--四舍五入
select round(4.6);
--向上取整
select ceil(4.01);
--向下取整
select floor(4.99);

------------------------------常用字符串操作函数
--转为大写
select upper('sdadsadASSS');
--转为小写
SELECT lower('AAAAASASDDDA');
--求字符串的长度
SELECT length('sdadasdasd');
--把字符串前后的空格去掉    字符串中间的空格去不掉,需要使用替换了
SELECT trim('    woshi   haoren     ');
--向左、右补齐,到指定长度       l表示左  r表示右   pad 填补
SELECT lpad('abc',8,'*');
SELECT rpad('abc',8,'*');
--剪切字符串       从哪开始剪,剪切的长度是多少
SELECT SUBSTRING('12345678',2,5); 
select rpad(substring('13843838438',1,3),11,'*');
--使用正则表达式匹配目标字符串,匹配成功后替换!
SELECT regexp_replace('a      b  c  -12 32',' ','');--去掉所有的空
select replace('d d d',' ','-');


------------------------------集合操作
show tables;
desc test;
SELECT * from test limit 10;
--集合中元素的个数         不含有struct ,而且struct也不属于集合
SELECT size(friends),size(children)
FROM test;
--返回map中的key,返回map中的value
SELECT map_keys(children),map_values(children)
from test;
--将array数组中的元素排序
SELECT sort_array(split('1,3,4,2,6,4,3,8',','));
SELECT sort_array(split('b,a,ss,a,z,w,f,z',','));











hiveSQL常用操作语句(mysql)
 

1.建库语句:
	CREATE DATABASE [IF NOT EXISTS] database_name
	[COMMENT database_comment]
	[LOCATION hdfs_path]
	[WITH DBPROPERTIES (property_name=property_value, ...)];
		例:
			create DATABASE if NOT EXISTS hive_db2
			comment "my first database"
			location "/hive_db2"
2.库的修改:
	alter database hive_db2 set DBPROPERTIES ("createtime"="2018-12-19");
3.库的删除
	drop database db_hive cascade if exists; -- 删除存在表的数据库
3.建表语句:
	CREATE [EXTERNAL] TABLE [IF NOT EXISTS] table_name 
	[(col_name data_type [COMMENT col_comment], ...)] 
	[COMMENT table_comment] 
	[PARTITIONED BY (col_name data_type [COMMENT col_comment], ...)] 
	[CLUSTERED BY (col_name, col_name, ...) 
	[SORTED BY (col_name [ASC|DESC], ...)] INTO num_buckets BUCKETS] 
	[ROW FORMAT row_format] 
	DELIMITED [FIELDS TERMINATED BY char] [COLLECTION ITEMS TERMINATED BY char]
			[MAP KEYS TERMINATED BY char] [LINES TERMINATED BY char] 
	   | SERDE serde_name [WITH SERDEPROPERTIES (property_name=property_value, property_name=property_value, ...)]
	[STORED AS file_format] Textfile
	[LOCATION hdfs_path]
	[TBLPROPERTIES (property_name=property_value, ...)]
	[AS select_statement]
		例1:
			create table student2(id int COMMENT "xuehao", name string COMMENT "mingzi")
			COMMENT "xueshengbiao"
			ROW format delimited
			fields terminated by '\t'
			STORED as Textfile
			location '/student2' -- 直接加载该目录下的数据文件到表中
			TBLPROPERTIES ("createtime"="2018-12-18");
		例2:
			create table student(id int, name string)
			row format delimited
			fields terminated by '\t';
			load data local inpath '/opt/module/datas/student.txt' into table student;
		例3:
			create table student4 like student2; -- 仅复制表结构
4.导入数据语句
	4.1 不加local则导入hdfs上文件,但会剪贴原文件,local本地仅粘贴
			load data [local] inpath '/opt/module/datas/student.txt' 
			[overwrite] into table student [partition (partcol1=val1,…)];
	4.2 创建表并导入数据(依据以存在的表)
			create table student6 as select * from student; -- 仅导入数据,不会导入其他细节属性 --被创建表不能是外部表 -- 被创建表不支持分区及分桶
	4.3 覆盖插入
			insert overwrite table student3 select * from student;
	4.4 插入带分区的表
			insert into table stu_par partition(month = '08') select id ,name from stu_par where month = '09';
	4.5 将单表中数据导入多表
			from student
			insert into table student4 select *
			insert into table student5 select *;
	4.6 多分区导入单表
			from stu_par
			insert into table stu_par partition(month = '06')
			select id ,name where month = '08'
			insert into table stu_par partition(month = '07')
			select id,name where month = '10';
5.表的修改操作
	5.1 修改表的属性
			alter table stu_ex set TBLPROPERTIES ('EXTERNAL' = 'TRUE');
	5.2 重命名表名
			alter table student4 rename to student3;
	5.3 修改表的serde属性(序列化和反序列化)
		alter table table_name set serdepropertyes('field.delim'='\t');
6.列的更新操作
	6.1 修改列语法
		ALTER TABLE table_name CHANGE [COLUMN] col_old_name col_new_name column_type [COMMENT col_comment] [FIRST|AFTER column_name]
	6.2 增加或替换列语法
		ALTER TABLE table_name ADD|REPLACE COLUMNS (col_name data_type [COMMENT col_comment], ...) 
		例1:
			增加列:alter table student2 add COLUMNS (score double);
		例2:
			修改列:alter table student2 CHANGE COLUMN score score int AFTER id;
		例3:
			替换列(全部替换):alter table student2 replace COLUMNS (id int, name string);
7.带有分区的表
	7.0 查看分区
			show partitions table_name;
	7.1 创建单个分区
			create table stu_par(id int, name string)
			partitioned by (month string)
			ROW format delimited
			FIELDS terminated by '\t';
		-- 错误示例
			create table stu_par2(id int, name string)
			partitioned by (id int)
			ROW format delimited
			FIELDS terminated by '\t';  错!!!!(不能以数据库字段作为分区)
		-- 加载数据到指定分区(分区不存在则自动创建)
			load data local inpath '/opt/module/datas/student.txt' into table stu_par partition(month = '12');
			load data local inpath '/opt/module/datas/student.txt' into table stu_par partition(month = '11');

		-- 合并分区查询结果
			select * from stu_par where month = '11'
			union
			select * from stu_par where month = '12';
	7.2 增加多个分区
		alter table stu_par add partition (month = '08') partition(month='07');
	7.3 删除多个分区
		alter table stu_par drop partition(month='08'),partition(month='09');
	7.4 创建多级分区
		create table stu_par2(id int, name string)
		partitioned  by (month string, day string)
		row format delimited
		FIELDS terminated by '\t';
	7.5 导入数据到多级分区
		load data local inpath '/opt/module/datas/student.txt' into table stu_par2 
		partition (month='12',day='19');
	7.6 向多级分区增加分区
		alter table stu_par2 add partition(month = '12', day = '17');
	7.7 查询多级分区中的数据
		select * from stu_par2 where day = '18';
	7.8 修复分区(也可以使用添加分区的语句)
		msck repair table dept_partition2;
8.创建外部表(删除表不会删除表中数据,仅删除表的元数据)
	create external table stu_ex2(id int, name string)
	ROW format delimited
	FIELDS terminated by '\t'
	location '/student';
	8.1 外部表与内部表的转换
		alter table stu_ex set TBLPROPERTIES ('EXTERNAL' = 'TRUE');
9.数据的导出
	9.1 导出同时格式化(不加local则导出到hdfs)
		insert overwrite local directory '/opt/module/datas/student'
		row format delimited
		fields terminated by '\t'
		select * from student;
	9.2 hadoop命令导出到本地
		dfs -get /user/hive/warehouse/student/month=201709/000000_0 /opt/module/datas/export/student3.txt;
	9.3 shell命令导出
		hive -f/-e 执行语句或者脚本 > file	-- -f跟文件,-e跟执行语句
	9.4 export仅可以导出到hdfs,常用于hdfs集群hive表迁徙
		export table default.student to '/user/hive/warehouse/export/student'; -- 同时会导出表的元数据
10.数据的导入(仅能导入export导出的数据,因为需要获取表的元数据)
	import table table_name from 'export导出数据的路径';
11.清除表中数据
	truncate table student; -- 只能删除管理表,不能删除外部表中数据
12.Like、RLike:RLike可以使用java的正则表达式
13.group by及having的使用 -- hive中对于使用group by后查询字段仅限group by的字段及聚合函数
	select deptno, avg(sal) avg_sal from emp group by deptno having avg_sal > 2000;
14.mapreduce的join	
	14.1 mapreduce中的reducejoin特点:在mapper阶段进行数据关联标记,在reducer阶段进行数据聚合
	14.2 mapreduce中的mapjoin特点:将小表加载到内存中,在mapper阶段根据内存中的数据对大表进行数据处理,没有reduce阶段
15.HQL的join
	15.1 仅支持等值连接不支持非等值连接
		例:不支持select * from A left join B on A.id != B.id;
	15.2 不支持在on条件中使用‘or’
	15.3 每个join都会启动一个mapreduce任务,但hive默认开启mapreduce优化
		关闭mapreduce优化:set hive.auto.convert.join=false;
16.order by
	会进行全局排序,则reduce数量被看作1个,效率低下
17.sort by -- 局部排序
	对于每个mapreduce各分区进行局部排序,分区中的数据随机给定
18.distribute by
	18.1 即mapreduce中自定义分区操作,hql书写规则:先分区后排序  
	18.2 distribute by的分区规则是根据分区字段的hash码与reduce的个数进行模除后,余数相同的分到一个区。
19.cluster by
	当distribute和sort字段相同时可用cluster进行替代,默认正序,单不支持desc倒序
20.分桶 -- 分桶表的数据需要通过子查询的方式导入
	20.1 开启分桶的设置
		set hive.enforce.bucketing=true;
	20.2 分桶表的创建
		create table stu_buck(id int, name string)
		clustered by(id) 
		into 4 buckets
		row format delimited fields terminated by '\t';
	20.3 分桶的规则
		用分桶字段的hash值与桶的个数求余,来决定数据存放在那个桶,
	20.4 分桶与分区区别
		a. 分桶结果在表的目录下存在多个分桶文件
		b. 分区是将数据存放在表所在目录不同文件路径下
		c. 分区针对是数据存储路径,分桶针对的是数据文件,分桶可以在分区的基础粒度细化
21.分桶的抽样	
	21.1 抽样语法 -- 必须 x<=y
		select * from table_name tablesample(bucket x out of y on bucketKey); -- on bucketKey可不写
	21.2 抽样规则
		a. y用来决定抽样比例,必须为bucket数的倍数或者因子,
			例:bucket数为4时,当y=2时,4/2=2,则抽取两个桶的数据,具体抽取哪个桶由x决定
		b. x用来决定抽取哪个桶中的数据
			例1:当bucket=4, y=4, x=2时,则需要抽取的数据量为bucket/y=1个桶,抽取第x桶的数据
			例2:当bucket=4, y=2, x=2时,则需要抽取的数据量为bucket/y=2个桶,抽取第x桶和第x+y桶的数据
			例3:当bucket=12, y=3, x=2时,抽bucket/y=4个桶,抽取第x桶和第x+2y桶的数据
22.NVL函数
	NVL(column_name, default_cvalue),如果该行字段值为null,则返回default_value的值
23.CONCAT_WS()函数
	使用规则:concat_ws(separator, [string | array(string)]+)
	例:select concat_ws('_', 'www', array('achong','com')) 拼接结果:www_achong_com
24.COLLECT_SET(col)函数
	使用规则:仅接受基本数据类型,将字段去重汇总,并返回array类型
	例(行转列):表结构
			name	xingzuo     	  blood
			孙悟空	  白羊座	        A
			大海	  射手座	        A
			宋宋	  白羊座	        B
			猪八戒	  白羊座	        A
			凤姐	  射手座	        A
		需求:把星座和血型一样的人归类到一起
			射手座,A            大海|凤姐
			白羊座,A            孙悟空|猪八戒
			白羊座,B            宋宋
		查询语句:
			SELECT CONCAT_WS(',', xingzuo, blood), CONCAT_WS('|', COLLECT_SET(NAME))
			FROM xingzuo
			GROUP BY xingzuo, blood
25.EXPLODE(爆炸函数)及LATERAL_VIEW)(侧写函数)
	25.1 explode:将列中的array或者map结构拆分成多行 -- 一般需结合lateral_view使用
	25.2 lateral_view: LATERAL VIEW udtf(expression) 表别名 AS 列别名
	例(行转列)
		select movie, category_name
		from  movie_info 
		lateral view explode(category) table_tmp as category_name;
26.开窗函数 -- 常结合聚合函数使用,解决即需要聚合前的数据又需要聚合后的数据展示
	26.1 语法:UDAF() over (PARTITION By col1,col2 order by col3 窗口子句(rows between .. and ..)) AS 列别名
			(partition by .. order by)可替换为(distribute by .. sort by ..)
	26.2 over(): 指定分析数据窗口大小
	26.3 窗口子句 -- 先分区在排序然后接rows限定执行窗口
		26.3.01 n PRECEDING:往前n行数据
		26.3.02 n FOLLOWING:往后n行数据
			例:select name, orderdate, cost, sum(cost) over(
																partition by name 
																order by orderdate 
																rows between 1 PRECEDING and 1 FOLLOWING
															) from business;
		26.3.03 CURRENT ROW:当前行
		26.3.04 UNBOUNDED PRECEDING 表示从前面的起点
		26.3.05 UNBOUNDED FOLLOWING表示到后面的终点
			例:select name, orderdate, cost, sum(cost) over(
																partition by name 
																order by orderdate 
																rows between CURRENT ROW and UNBOUNDED FOLLOWING
															) from business;
27.LAG(col,n,default_val):往前第n行数据
28.LEAD(col,n, default_val):往后第n行数据
	例:select name, orderdate, cost, lag(orderdate, 1, 'null') over(partition by name order by orderdate)
		from business; -- 即获取前1行的orderDate数据
29.ntile(n):把有序分区中的行分为n组,每组编号从1开始 -- 分组规则详见:ntile的分组规则.sql
	例:select name,orderdate,cost, ntile(5) over(order by orderdate) num from business
30.Rank函数
	rank() 出现相同排序时,总数不变
	dense_rank() 出现相同排序时,总数减少
	row_number() 不会出现相同排序




sql执行顺序
	from... where...group by... having.... select ... order by...
	
hql执行顺序	
	from … where … group by … having … select … order by … 或
	from … on … join … where … group by … having … select … distinct … order by … limit
	存在开窗函数时,起码在order by之前执行

	
例题1:-- 集合类型数据导入
	{
		"name": "songsong",
		"friends": ["bingbing" , "lili"] ,       //列表Array, 
		"children": {                      //键值Map,
			"xiao song": 18 ,
			"xiaoxiao song": 19
		}
		"address": {                      //结构Struct,
			"street": "hui long guan" ,
			"city": "beijing" 
		}
	}

	基于上述数据结构,我们在Hive里创建对应的表,并导入数据。
	1.1 格式化数据为:
		songsong,bingbing_lili,xiao song:18_xiaoxiao song:19,hui long guan_beijing
		yangyang,caicai_susu,xiao yang:18_xiaoxiao yang:19,chao yang_beijing
	1.2 建表语句:
		create table test(name string, 
		friends array,
		children map,
		address struct)
		row format delimited
		fields terminated by ','
		collection items terminated by '_'
		map keys terminated by ':';
	1.3 数据写入语句
		load data local inpath '/opt/module/datas/test.txt' into table test;
	1.4 查询语句
		select friends[0] friend,children['xiao song'] age,address.city from test where name = 'songsong';

你可能感兴趣的:(hive,大数据,mysql,hive)