Hive常用算子实现原理简述--MapReduce版

Posted 2023-04-24

参考技术A Hive中的常用算子包括distinct、join、group by、order by、distribute by、sort by、count等，这些操作符在SQL中使用起来很方便，能快速达到我们想要的效果，但是这些算子在底层是怎么实现的呢？

order by很容易想到执行原理，在一个reduce中将所有记录按值排序即可。因此order by在数据量大的情况下执行时间非常长，容易out of memory，非特殊业务需求一般不使用。distribute by也比较明显，根据hash值将distribute的值分发到不同的reduce。sort by是小号的order by，只负责将本reducer中的值排序，达到局部有序的效果。sort by和distribute by配合使用风味更佳，二者可以合并简写为cluster by。count则更加明晰，在combiner或reducer处按相同键累加值就能得到。

比较复杂的是distinct、join、group by，本文重点讨论这三个算子在MapReduce引擎中的大致实现原理。班门弄斧，抛砖引玉。

map阶段，将group by后的字段组合作为key，如果group by单字段那么key就一个。将group by之后要进行的聚合操作字段作为值，如要进行count，则value是1；如要sum另一个字段，则value就是该字段。

shuffle阶段，按照key的不同分发到不同的reducer。注意此时可能因为key分布不均匀而出现数据倾斜的问题。

reduce阶段，将相同key的值累加或作其他需要的聚合操作，得到结果。

对group by的过程讲解的比较清楚的是这篇文章 http://www.mamicode.com/info-detail-2292193.html 图文并茂，很生动。

实例如下图，对应语句是 select rank, isonline, count(*) from city group by rank, isonline;

如果group by出现数据倾斜，除去替换key为随机数、提前挑出大数量级key值等通用调优方法，适用于group by的特殊方法有以下几种：

（1）set hive.map.aggr=true，即开启map端的combiner，减少传到reducer的数据量，同时需设置参数hive.groupby.mapaggr.checkinterval 规定在 map 端进行聚合操作的条目数目。

（2）设置mapred.reduce.tasks为较大数量，降低每个reducer处理的数据量。

（3）set hive.groupby.skewindata=true，该参数可自动进行负载均衡。生成的查询计划会有两个 MR Job。第一个 MR Job 中，Map 的输出结果集合会随机分布到Reduce 中，每个 Reduce 做部分聚合操作，并输出结果，这样处理的结果是相同的 Group By Key有可能被分发到不同的 Reduce 中，从而达到负载均衡的目的；第二个 MR Job 再根据预处理的数据结果按照 Group ByKey 分布到 Reduce 中（这个过程可以保证相同的 Group By Key 被分布到同一个 Reduce中），最后完成最终的聚合操作。

Hive中有两种join方式：map join和common join

如果不显式指定map side join，或者没有达到触发自动map join的条件，那么会进行reduce端的join，即common join，这种join包含map、shuffle、reduce三个步骤。

（1）Map阶段

读取源表的数据，Map输出时候以Join on条件中的列为key，如果Join有多个关联键，则以这些关联键的组合作为key。Map输出的value为join之后所关心的(select或者where中需要用到的)列；同时在value中还会包含表的Tag信息，用于标明此value对应哪个表。然后按照key进行排序。

（2）Shuffle阶段

根据key的值进行hash,并将key/value按照hash值推送至不同的reduce中，这样确保两个表中相同的key位于同一个reduce中

（3）Reduce阶段

根据key的值完成join操作，期间通过Tag来识别不同表中的数据。

以下面的SQL为例，可用下图所示过程大致表达其join原理。

SELECT u.name, o.orderid FROM user u JOIN order o ON u.uid = o.uid;

关联字段是uid，因此以uid为map阶段的输出key，value为选取的字段name和标记源表的tag。shuffle阶段将相同key的键值对发到一起，reduce阶段将不同源表、同一key值的记录拼接起来，可能存在一对多的情况。

如果指定使用map join的方式，或者join的其中一张表小于某个体积（默认25MB），则会使用map join来执行。具体小表有多小，由参数 hive.mapjoin.smalltable.filesize 来决定。

Hive0.7之前，需要使用hint提示 /*+ mapjoin(table) */才会执行MapJoin,否则执行Common Join，但在0.7版本之后，默认自动会转换Map Join，由参数 hive.auto.convert.join 来控制，默认为true。

以下图为例说明map join如何执行，该图来自 http://lxw1234.com/archives/2015/06/313.htm ，博主是一个水平深厚又乐于分享的前辈，图片水印上也有其网址。

yarn会启动一个Local Task（在客户端本地执行的Task）--Task A，负责扫描小表b的数据，将其转换成一个HashTable的数据结构，并写入本地的文件中，之后将该文件加载到DistributeCache中。

接下来是Task B，该任务是一个没有Reduce的MR，启动MapTasks扫描大表a,在Map阶段，根据a的每一条记录去和DistributeCache中b表对应的HashTable关联，并直接输出结果。

由于MapJoin没有Reduce，所以由Map直接输出结果文件，有多少个Map Task，就有多少个结果文件。

distinct一般和group by同时出现。

当distinct一个字段时，将group by的字段和distinct的字段组合在一起作为map输出的key，value设置为1，同时将group by的字段定为分区键，这可以确保相同group by字段的记录都分到同一个reducer，并且map的输入天然就是按照组合key排好序的。根据分区键将记录分发到reduce端后，按顺序取出组合键中的distinct字段，这时distinct字段也是排好序的。依次遍历distinct字段，每找到一个不同值，计数器就自增1，即可得到count distinct结果。例如下面的SQL语句，过程可以下图示意。

我暂时没有理解这是怎么实现的，别人写的也没有看明白。有善良的学富五车的大佬指点一下吗？

Hive基础练习一

下面是hive基本练习，持续补充中。

简述Hive工作原理

hive是基于hadoop，可以管理hdfs上的数据的工具，它本质上是执行MapReduce程序，只是使用了类sql语句更加方便开发，hive驱动器会将类sql语句转换成MapReduce的task来执行，因此执行速度会比较慢。

hive的核心是驱动器，它可以连接sql和hdfs，将sql转换成了MapReduce任务，驱动器主要包括：

（1）解析器：解析sql语句，划分为不同的stage

（2）编译器：将不同阶段的stage编译，变成一个个的MR任务

（3）优化器：对逻辑执行进行优化

（4）执行器：将sql里的逻辑任务转换为hdfs的物理任务，hive的执行器就是MapReduce

hive 内部表和外部表区别

内部表：主要用在数据仓库层（DW层），为个人独自占有，如果删除表格，对应的原始数据也将删除，但是对其他人没有影响。

外部表：主要用在源数据层（ODS层），删除表格不会删除对应的数据。

在建表时，如果是内部表，不需要使用external关键字，外部表需要使用external关键字。

创建表格导入数据练习1

战狼2,吴京:吴刚:卢婧姗,2017-08-16
大话西游,周星驰:吴孟达,1995-09-01
哪吒,吕艳婷:瀚墨,2019-07-26
使徒行者2,张家辉:古天乐:吴镇宇,2019-08-07
鼠胆英雄,岳云鹏:佟丽娅:田雨:袁弘,2019-08-02

创建表格，导入数据。

# 建表
0: jdbc:hive2://node01:10000> create table movie_info(moviename string,actors array<string>,showtime string) row format delimited by ',' collection items teminated by ':';
Error: Error while compiling statement: FAILED: ParseException line 1:100 cannot recognize input near 'by' '','' 'collection' in serde properties specification (state=42000,code=40000)
0: jdbc:hive2://node01:10000> create table movie_info(moviename string,actors array<string>,showtime string) row format delimited fields terminated by ',' collection items teminated by ':';
Error: Error while compiling statement: FAILED: ParseException line 1:142 mismatched input 'teminated' expecting TERMINATED near 'items' in table row format's column separator (state=42000,code=40000)
0: jdbc:hive2://node01:10000> create table movie_info(moviename string,actors array<string>,showtime string) row format delimited fields terminated by ',' collection items terminated by ':';
INFO  : Compiling command(queryId=hadoop_20191115220202_01acb251-d8e2-46c5-bf20-5b354d6d2923): create table movie_info(moviename string,actors array<string>,showtime string) row format delimited fields terminated by ',' collection items terminated by ':'
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:null, properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115220202_01acb251-d8e2-46c5-bf20-5b354d6d2923); Time taken: 0.035 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115220202_01acb251-d8e2-46c5-bf20-5b354d6d2923): create table movie_info(moviename string,actors array<string>,showtime string) row format delimited fields terminated by ',' collection items terminated by ':'
INFO  : Starting task [Stage-0:DDL] in serial mode
INFO  : Completed executing command(queryId=hadoop_20191115220202_01acb251-d8e2-46c5-bf20-5b354d6d2923); Time taken: 0.286 seconds
INFO  : OK
No rows affected (0.359 seconds)
# 加载数据
0: jdbc:hive2://node01:10000> load data local inpath '/kkb/install/hivedatas/move_info.txt' into table movie_info;
INFO  : Compiling command(queryId=hadoop_20191115220505_1c2ea52f-37cf-43d2-85b0-e2b54049aa33): load data local inpath '/kkb/install/hivedatas/move_info.txt' into table movie_info
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:null, properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115220505_1c2ea52f-37cf-43d2-85b0-e2b54049aa33); Time taken: 0.061 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115220505_1c2ea52f-37cf-43d2-85b0-e2b54049aa33): load data local inpath '/kkb/install/hivedatas/move_info.txt' into table movie_info
INFO  : Starting task [Stage-0:MOVE] in serial mode
INFO  : Loading data to table db_hive.movie_info from file:/kkb/install/hivedatas/move_info.txt
INFO  : Starting task [Stage-1:STATS] in serial mode
INFO  : Table db_hive.movie_info stats: [numFiles=1, totalSize=235]
INFO  : Completed executing command(queryId=hadoop_20191115220505_1c2ea52f-37cf-43d2-85b0-e2b54049aa33); Time taken: 0.645 seconds
INFO  : OK
No rows affected (0.726 seconds)
# 查询加载后结果
0: jdbc:hive2://node01:10000> select * from movie_info;
INFO  : Compiling command(queryId=hadoop_20191115220505_0e4cbd35-3f1c-44d8-8a03-2b7dcd7ab03e): select * from movie_info
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:movie_info.moviename, type:string, comment:null), FieldSchema(name:movie_info.actors, type:array<string>, comment:null), FieldSchema(name:movie_info.showtime, type:string, comment:null)], properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115220505_0e4cbd35-3f1c-44d8-8a03-2b7dcd7ab03e); Time taken: 0.09 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115220505_0e4cbd35-3f1c-44d8-8a03-2b7dcd7ab03e): select * from movie_info
INFO  : Completed executing command(queryId=hadoop_20191115220505_0e4cbd35-3f1c-44d8-8a03-2b7dcd7ab03e); Time taken: 0.0 seconds
INFO  : OK
+-----------------------+--------------------------+----------------------+--+
| movie_info.moviename  |    movie_info.actors     | movie_info.showtime  |
+-----------------------+--------------------------+----------------------+--+
| 战狼2                   | ["吴京","吴刚","卢婧姗"]        | 2017-08-16           |
| 大话西游                  | ["周星驰","吴孟达"]            | 1995-09-01           |
| 哪吒                    | ["吕艳婷","瀚墨"]             | 2019-07-26           |
| 使徒行者2                 | ["张家辉","古天乐","吴镇宇"]      | 2019-08-07           |
| 鼠胆英雄                  | ["岳云鹏","佟丽娅","田雨","袁弘"]  | 2019-08-02           |
+-----------------------+--------------------------+----------------------+--+
5 rows selected (0.177 seconds)

3.1 查询出每个电影的第二个主演

0: jdbc:hive2://node01:10000> select moviename,actors[1] from movie_info;
INFO  : Compiling command(queryId=hadoop_20191115220909_702e25ac-bbf9-4fb3-a39e-44156379fcf3): select moviename,actors[1] from movie_info
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:moviename, type:string, comment:null), FieldSchema(name:_c1, type:string, comment:null)], properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115220909_702e25ac-bbf9-4fb3-a39e-44156379fcf3); Time taken: 0.134 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115220909_702e25ac-bbf9-4fb3-a39e-44156379fcf3): select moviename,actors[1] from movie_info
INFO  : Completed executing command(queryId=hadoop_20191115220909_702e25ac-bbf9-4fb3-a39e-44156379fcf3); Time taken: 0.0 seconds
INFO  : OK
+------------+------+--+
| moviename  | _c1  |
+------------+------+--+
| 战狼2        | 吴刚   |
| 大话西游       | 吴孟达  |
| 哪吒         | 瀚墨   |
| 使徒行者2      | 古天乐  |
| 鼠胆英雄       | 佟丽娅  |
+------------+------+--+

3.2 查询每部电影有几名主演

0: jdbc:hive2://node01:10000> select moviename,size(actors) as actorcount from movie_info;
INFO  : Compiling command(queryId=hadoop_20191115221010_1b46f1ce-12bf-406d-9cfc-395df7f26816): select moviename,size(actors) as actorcount from movie_info
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:moviename, type:string, comment:null), FieldSchema(name:actorcount, type:int, comment:null)], properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115221010_1b46f1ce-12bf-406d-9cfc-395df7f26816); Time taken: 0.075 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115221010_1b46f1ce-12bf-406d-9cfc-395df7f26816): select moviename,size(actors) as actorcount from movie_info
INFO  : Completed executing command(queryId=hadoop_20191115221010_1b46f1ce-12bf-406d-9cfc-395df7f26816); Time taken: 0.001 seconds
INFO  : OK
+------------+-------------+--+
| moviename  | actorcount  |
+------------+-------------+--+
| 战狼2        | 3           |
| 大话西游       | 2           |
| 哪吒         | 2           |
| 使徒行者2      | 3           |
| 鼠胆英雄       | 4           |
+------------+-------------+--+

3.3主演里面包含古天乐的电影

# 需要使用lateral view
0: jdbc:hive2://node01:10000> Select t.moviename,t.actor from
. . . . . . . . . . . . . . > (
. . . . . . . . . . . . . . > select moviename,actor from movie_info lateral view explode(actors)temp as actor
. . . . . . . . . . . . . . > ) t
. . . . . . . . . . . . . . > Where t.actor='古天乐';
INFO  : Compiling command(queryId=hadoop_20191115222525_532eeeb8-06db-4828-9e6d-73594e31877e): Select t.moviename,t.actor from
(
select moviename,actor from movie_info lateral view explode(actors)temp as actor
) t
Where t.actor='古天乐'
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:t.moviename, type:string, comment:null), FieldSchema(name:t.actor, type:string, comment:null)], properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115222525_532eeeb8-06db-4828-9e6d-73594e31877e); Time taken: 0.225 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115222525_532eeeb8-06db-4828-9e6d-73594e31877e): Select t.moviename,t.actor from
(
select moviename,actor from movie_info lateral view explode(actors)temp as actor
) t
Where t.actor='古天乐'
INFO  : Completed executing command(queryId=hadoop_20191115222525_532eeeb8-06db-4828-9e6d-73594e31877e); Time taken: 0.0 seconds
INFO  : OK
+--------------+----------+--+
| t.moviename  | t.actor  |
+--------------+----------+--+
| 使徒行者2        | 古天乐      |
+--------------+----------+--+

创建表格导入数据练习2

1,张三,18:male:北京
2,李四,29:female:上海
3,杨朝来,22:male:深圳?
4,蒋平,34:male:成都
5,唐灿华,25:female:哈尔滨
6,马达,17:male:北京
7,赵小雪,23:female:杭州
8,薛文泉,26:male:上海
9,丁建,29:male:北京

创建表格，加载数据。

# 创建表格
0: jdbc:hive2://node01:10000> create table dept(id int,name string,info struct<age:int,gender:string,city:string>) row format delimited fields terminated by ',' collection items terminated by ':';
INFO  : Compiling command(queryId=hadoop_20191115223434_1d5c9e49-c4e7-4930-938d-4a5823486099): create table dept(id int,name string,info struct<age:int,gender:string,city:string>) row format delimited fields terminated by ',' collection items terminated by ':'
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:null, properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115223434_1d5c9e49-c4e7-4930-938d-4a5823486099); Time taken: 0.008 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115223434_1d5c9e49-c4e7-4930-938d-4a5823486099): create table dept(id int,name string,info struct<age:int,gender:string,city:string>) row format delimited fields terminated by ',' collection items terminated by ':'
INFO  : Starting task [Stage-0:DDL] in serial mode
INFO  : Completed executing command(queryId=hadoop_20191115223434_1d5c9e49-c4e7-4930-938d-4a5823486099); Time taken: 0.086 seconds
INFO  : OK
No rows affected (0.131 seconds)
0: jdbc:hive2://node01:10000> desc dept;
INFO  : Compiling command(queryId=hadoop_20191115223434_270cf471-d325-40ed-af27-4afff5a6aabf): desc dept
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:col_name, type:string, comment:from deserializer), FieldSchema(name:data_type, type:string, comment:from deserializer), FieldSchema(name:comment, type:string, comment:from deserializer)], properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115223434_270cf471-d325-40ed-af27-4afff5a6aabf); Time taken: 0.1 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115223434_270cf471-d325-40ed-af27-4afff5a6aabf): desc dept
INFO  : Starting task [Stage-0:DDL] in serial mode
INFO  : Completed executing command(queryId=hadoop_20191115223434_270cf471-d325-40ed-af27-4afff5a6aabf); Time taken: 0.042 seconds
INFO  : OK
+-----------+--------------------------------------------+----------+--+
| col_name  |                 data_type                  | comment  |
+-----------+--------------------------------------------+----------+--+
| id        | int                                        |          |
| name      | string                                     |          |
| info      | struct<age:int,gender:string,city:string>  |          |
+-----------+--------------------------------------------+----------+--+
3 rows selected (0.169 seconds)
# 加载数据
0: jdbc:hive2://node01:10000> load data local inpath '/kkb/install/hivedatas/dept.txt' overwrite into table dept;
INFO  : Compiling command(queryId=hadoop_20191115223535_d81bba57-47cc-473b-be36-fe32053922b8): load data local inpath '/kkb/install/hivedatas/dept.txt' overwrite into table dept
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:null, properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115223535_d81bba57-47cc-473b-be36-fe32053922b8); Time taken: 0.026 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115223535_d81bba57-47cc-473b-be36-fe32053922b8): load data local inpath '/kkb/install/hivedatas/dept.txt' overwrite into table dept
INFO  : Starting task [Stage-0:MOVE] in serial mode
INFO  : Loading data to table db_hive.dept from file:/kkb/install/hivedatas/dept.txt
INFO  : Starting task [Stage-1:STATS] in serial mode
INFO  : Table db_hive.dept stats: [numFiles=1, totalSize=240]
INFO  : Completed executing command(queryId=hadoop_20191115223535_d81bba57-47cc-473b-be36-fe32053922b8); Time taken: 0.303 seconds
INFO  : OK
# 查询数据
0: jdbc:hive2://node01:10000> select * from dept;
INFO  : Compiling command(queryId=hadoop_20191115223737_b10807f4-8f05-4b3b-aa33-4e0bc8c408f6): select * from dept
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:dept.id, type:int, comment:null), FieldSchema(name:dept.name, type:string, comment:null), FieldSchema(name:dept.info, type:struct<age:int,gender:string,city:string>, comment:null)], properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115223737_b10807f4-8f05-4b3b-aa33-4e0bc8c408f6); Time taken: 0.054 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115223737_b10807f4-8f05-4b3b-aa33-4e0bc8c408f6): select * from dept
INFO  : Completed executing command(queryId=hadoop_20191115223737_b10807f4-8f05-4b3b-aa33-4e0bc8c408f6); Time taken: 0.0 seconds
INFO  : OK
+----------+------------+--------------------------------------------+--+
| dept.id  | dept.name  |                 dept.info                  |
+----------+------------+--------------------------------------------+--+
| 1        | 张三         | {"age":18,"gender":"male","city":"北京"}     |
| 2        | 李四         | {"age":29,"gender":"female","city":"上海"}   |
| 3        | 杨朝来        | {"age":22,"gender":"male","city":"深圳?"}    |
| 4        | 蒋平         | {"age":34,"gender":"male","city":"成都"}     |
| 5        | 唐灿华        | {"age":25,"gender":"female","city":"哈尔滨"}  |
| 6        | 马达         | {"age":17,"gender":"male","city":"北京"}     |
| 7        | 赵小雪        | {"age":23,"gender":"female","city":"杭州"}   |
| 8        | 薛文泉        | {"age":26,"gender":"male","city":"上海"}     |
| 9        | 丁建         | {"age":29,"gender":"male","city":"北京"}     |
+----------+------------+--------------------------------------------+--+

4.1 查询出每个人的id,名字,居住地址

0: jdbc:hive2://node01:10000> select id,name,info.city from dept;
INFO  : Compiling command(queryId=hadoop_20191115224040_796610ca-d24a-48ac-a8aa-d1eaee1ddce9): select id,name,info.city from dept
INFO  : Semantic Analysis Completed
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:id, type:int, comment:null), FieldSchema(name:name, type:string, comment:null), FieldSchema(name:city, type:string, comment:null)], properties:null)
INFO  : Completed compiling command(queryId=hadoop_20191115224040_796610ca-d24a-48ac-a8aa-d1eaee1ddce9); Time taken: 0.081 seconds
INFO  : Concurrency mode is disabled, not creating a lock manager
INFO  : Executing command(queryId=hadoop_20191115224040_796610ca-d24a-48ac-a8aa-d1eaee1ddce9): select id,name,info.city from dept
INFO  : Completed executing command(queryId=hadoop_20191115224040_796610ca-d24a-48ac-a8aa-d1eaee1ddce9); Time taken: 0.001 seconds
INFO  : OK
+-----+-------+-------+--+
| id  | name  | city  |
+-----+-------+-------+--+
| 1   | 张三    | 北京    |
| 2   | 李四    | 上海    |
| 3   | 杨朝来   | 深圳?   |
| 4   | 蒋平    | 成都    |
| 5   | 唐灿华   | 哈尔滨   |
| 6   | 马达    | 北京    |
| 7   | 赵小雪   | 杭州    |
| 8   | 薛文泉   | 上海    |
| 9   | 丁建    | 北京    |
+-----+-------+-------+--+

以上为hive练习部分，记录一下。