HIVE原生和复合类型的数据加载和使用

原生类型

原生类型包括TINYINT,SMALLINT,INT,BIGINT,BOOLEAN,FLOAT,DOUBLE,STRING,BINARY (Hive 0.8.0以上才可用),TIMESTAMP (Hive 0.8.0以上才可用)，这些数据加载很容易，只要设置好列分隔符，按照列分隔符输出到文件就可以了。

假设有这么一张用户登陆表

CREATE TABLE login (
  uid  BIGINT,
  ip  STRING
)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
STORED AS TEXTFILE;

这表示登陆表ip字段和uid字段以分隔符','隔开。

输出hive表对应的数据

# printf "%s,%s\n" 3105007001 192.168.1.1 >> login.txt
# printf "%s,%s\n" 3105007002 192.168.1.2 >> login.txt

login.txt的内容:

# cat login.txt                                                                                                                        
3105007001,192.168.1.1
3105007002,192.168.1.2

加载数据到hive表

LOAD DATA LOCAL INPATH '/home/hadoop/login.txt' OVERWRITE INTO TABLE login PARTITION (dt='20130101'); 

查看数据

select uid,ip from login where dt='20130101';
3105007001    192.168.1.1
3105007002    192.168.1.2

 

array

假设登陆表是

CREATE TABLE login_array (
  ip  STRING,
  uid  array<BIGINT>
)
PARTITIONED BY (dt STRING)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
COLLECTION ITEMS TERMINATED BY '|'
STORED AS TEXTFILE;

这表示登陆表每个ip有多个用户登陆，ip和uid字段之间使用','隔开，而uid数组之间的元素以'|'隔开。

 

输出hive表对应的数据

# printf "%s,%s|%s|%s\n" 192.168.1.1 3105007010 3105007011 3105007012 >> login_array.txt
# printf "%s,%s|%s|%s\n" 192.168.1.2 3105007020 3105007021 3105007022 >> login_array.txt

login_array.txt的内容:

cat login_array.txt                                                                                                                    
192.168.1.1,3105007010|3105007011|3105007012
192.168.1.2,3105007020|3105007021|3105007022

 

加载数据到hive表

LOAD DATA LOCAL INPATH '/home/hadoop/login_array.txt' OVERWRITE INTO TABLE login_array PARTITION (dt='20130101'); 

 

查看数据

select ip,uid from login_array where dt='20130101';
192.168.1.1    [3105007010,3105007011,3105007012]
192.168.1.2    [3105007020,3105007021,3105007022]

使用数组

select ip,uid[0] from login_array where dt='20130101'; --使用下标访问数组

select ip,size(uid) from login_array where dt='20130101'; #查看数组长度

select ip from login_array where dt='20130101'  where array_contains(uid,'3105007011');#数组查找

更多操作参考 https://cwiki.apache.org/confluence/display/Hive/LanguageManual+UDF#LanguageManualUDF-CollectionFunctions

 

map

假设登陆表是

CREATE TABLE login_map (
  ip  STRING,
  uid  STRING,
  gameinfo map<string,bigint>
)
PARTITIONED BY (dt STRING)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
COLLECTION ITEMS TERMINATED BY '|'
MAP KEYS TERMINATED BY ':'
STORED AS TEXTFILE;

这表示登陆表每个用户都会有游戏信息，而用户的游戏信息有多个，key是游戏名，value是游戏的积分。map中的key和value以'':"分隔，map的元素以'|'分隔。 

 

输出hive表对应的数据

# printf "%s,%s,%s:%s|%s:%s|%s:%s\n" 192.168.1.1  3105007010 wow 10 cf 1 qqgame 2  >> login_map.txt
# printf "%s,%s,%s:%s|%s:%s|%s:%s\n" 192.168.1.2  3105007012 wow 20 cf 21 qqgame 22  >> login_map.txt

 

login_map.txt的内容:

# cat login_map.txt
192.168.1.1,3105007010,wow:10|cf:1|qqgame:2
192.168.1.2,3105007012,wow:20|cf:21|qqgame:22

 

 

加载数据到hive表

LOAD DATA LOCAL INPATH '/home/hadoop/login_map.txt' OVERWRITE INTO TABLE login_map PARTITION (dt='20130101'); 

 

查看数据

select ip,uid,gameinfo from login_map where dt='20130101';
192.168.1.1    3105007010    {"wow":10,"cf":1,"qqgame":2}
192.168.1.2    3105007012    {"wow":20,"cf":21,"qqgame":22}

 

使用map

select ip,uid,gameinfo['wow'] from login_map where dt='20130101'; --使用下标访问map

select ip,uid,size(gameinfo) from login_map where dt='20130101'; #查看map长度

select ip,uid from login_map where dt='20130101'  where array_contains(map_keys(gameinfo),'wow');#查看map的key，找出有玩wow游戏的记录

更多操作参考 https://cwiki.apache.org/confluence/display/Hive/LanguageManual+UDF#LanguageManualUDF-CollectionFunctions

 

struct

假设登陆表是

CREATE TABLE login_struct (
  ip  STRING,
  user  struct<uid:bigint,name:string>
)
PARTITIONED BY (dt STRING)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
COLLECTION ITEMS TERMINATED BY '|'
MAP KEYS TERMINATED BY ':'
STORED AS TEXTFILE;

user是一个struct,分别包含用户uid和用户名。

 

输出hive表对应的数据

printf "%s,%s|%s|\n" 192.168.1.1  3105007010 blue  >> login_struct.txt
printf "%s,%s|%s|\n" 192.168.1.2  3105007012 ggjucheng  >> login_struct.txt

 

 

login_struct.txt的内容:

# cat login_struct.txt
192.168.1.1,3105007010,wow:10|cf:1|qqgame:2
192.168.1.2,3105007012,wow:20|cf:21|qqgame:22

 

 

加载数据到hive表

LOAD DATA LOCAL INPATH '/home/hadoop/login_struct.txt' OVERWRITE INTO TABLE login_struct PARTITION (dt='20130101'); 

 

查看数据

select ip,user from login_struct where dt='20130101';
192.168.1.1    {"uid":3105007010,"name":"blue"}
192.168.1.2    {"uid":3105007012,"name":"ggjucheng"}

 

使用struct

select ip,user.uid,user.name from login_map where dt='20130101'; 

 

union

用的比较少，暂时不讲

 

嵌套复合类型

之前讲的array,map,struct这几种复合类型，里面的元素都是原生类型，如果元素是复合类型，那该怎么加载数据呢。

假设登陆表是

CREATE TABLE login_game_complex (
  ip STRING,
  uid STRING,
  gameinfo map<bigint,struct<name:string,score:bigint,level:string>> ) 
PARTITIONED BY (dt STRING) 
ROW FORMAT DELIMITED 
STORED AS TEXTFILE;

这表示登陆表每个用户都会有游戏信息，而用户的游戏信息有多个，key是游戏id，value是一个struct，包含游戏的名字，积分，等级。

这种复杂类型的入库格式很麻烦，而且复合嵌套层次很多时，要生成的正确的格式也比较复杂，很容易出错。这里稍微提下，在嵌套层次多的情况下，分隔符会会随着复合类型嵌套层次的递增，分隔符默认会以\0,\1,\2....变化。

这里不介绍从shell下生成文件load data入库，感兴趣的同学，可以看看hive的源代码的org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe的serialize方法。

这里介绍使用另一种数据操作方式:insert，先把一个简单的表的数据，加载load到hive，再使用insert插入数据到一个嵌套复杂类型的表。

 

创建简单的表

CREATE TABLE login_game_simple (
  ip STRING,
  uid STRING,
  gameid bigint,
  gamename string,
  gamescore bigint,
  gamelevel string 
) 
PARTITIONED BY (dt STRING) 
ROW FORMAT DELIMITED 
FIELDS TERMINATED BY ','
STORED AS TEXTFILE;

生成login_game_simple.txt的内容:

192.168.1.0,3105007010,1,wow,100,v1
192.168.1.0,3105007010,2,cf,100,v2
192.168.1.0,3105007010,3,qqgame,100,v3
192.168.1.2,3105007011,1,wow,101,v1
192.168.1.2,3105007011,3,qqgame,101,v3
192.168.1.2,3105007012,1,wow,102,v1
192.168.1.2,3105007012,2,cf,102,v2
192.168.1.2,3105007012,3,qqgame,102,v3

load data到hive后，再生成复杂的gameinfo map结构，插入到表login_game_complex

INSERT OVERWRITE TABLE login_game_complex PARTITION (dt='20130101')  
select ip,uid,map(gameid, named_struct('name',gamename,'score',gamescore,'level',gamelevel) ) FROM login_game_simple  where dt='20130101' ;

 

查询数据

select ip,uid,gameinfo from login_game_complex where dt='20130101';
192.168.1.0    3105007010    {1:{"name":"wow","score":100,"level":"v1"}}
192.168.1.0    3105007010    {2:{"name":"cf","score":100,"level":"v2"}}
192.168.1.0    3105007010    {3:{"name":"qqgame","score":100,"level":"v3"}}
192.168.1.2    3105007011    {1:{"name":"wow","score":101,"level":"v1"}}
192.168.1.2    3105007011    {3:{"name":"qqgame","score":101,"level":"v3"}}
192.168.1.2    3105007012    {1:{"name":"wow","score":102,"level":"v1"}}
192.168.1.2    3105007012    {2:{"name":"cf","score":102,"level":"v2"}}
192.168.1.2    3105007012    {3:{"name":"qqgame","score":102,"level":"v3"}}

这里只是演示了嵌套复杂类型的入库方式，所以这里只是例子。真正要完美入库，还是需要写一个自定义函数，根据ip和uid做group by，然后把gameinfo合并起来。hive没有这样的自定义函数，篇幅着想，不引进复杂的自定义函数编写。