Hive 的隐式连接总是内部连接吗?

dat*_*ess 4 hadoop hive join hiveql

Hive的连接文档鼓励使用隐式连接,即

SELECT * 
FROM table1 t1, table2 t2, table3 t3 
WHERE t1.id = t2.id AND t2.id = t3.id AND t1.zipcode = '02535';
Run Code Online (Sandbox Code Playgroud)

这是否相当于

SELECT t1.*, t2.*, t3.*
FROM table1 t1
INNER JOIN table2 t2 ON
  t1.id = t2.id 
INNER JOIN table3 t3 ON
  t2.id = t3.id
WHERE t1.zipcode = '02535'
Run Code Online (Sandbox Code Playgroud)

,或者上面会返回额外的记录?

lef*_*oin 7

不总是。您的查询是等效的。但没有WHERE t1.id = t2.id AND t2.id = t3.id它就会CROSS JOIN

更新:

这是一个有趣的问题,我决定添加一些演示。让我们创建两个表:

A(c1 int, c2 string)B(c1 int, c2 string)

加载数据:

insert into table A
 select 1, 'row one' union all
 select 2, 'row two';

insert into table B
 select 1, 'row one' union all
 select 3, 'row three';
Run Code Online (Sandbox Code Playgroud)

检查数据:

hive> select * from A;
OK
1       row one
2       row two
Time taken: 1.29 seconds, Fetched: 2 row(s)
hive> select * from B;
OK
1       row one
3       row three
Time taken: 0.091 seconds, Fetched: 2 row(s)
Run Code Online (Sandbox Code Playgroud)

检查交叉连接(隐式连接而不where转换为交叉):

hive> select a.c1, a.c2, b.c1, b.c2 from a,b;
Warning: Map Join MAPJOIN[14][bigTable=a] in task 'Stage-3:MAPRED' is a cross product
Warning: Map Join MAPJOIN[22][bigTable=b] in task 'Stage-4:MAPRED' is a cross product
Warning: Shuffle Join JOIN[4][tables = [a, b]] in Stage 'Stage-1:MAPRED' is a cross product

OK
1       row one 1       row one
2       row two 1       row one
1       row one 3       row three
2       row two 3       row three
Time taken: 54.804 seconds, Fetched: 4 row(s)
Run Code Online (Sandbox Code Playgroud)

检查内连接(隐式连接where作为 INNER 工作):

hive> select a.c1, a.c2, b.c1, b.c2 from a,b where a.c1=b.c1;
OK
1       row one 1       row one
Time taken: 38.413 seconds, Fetched: 1 row(s)
Run Code Online (Sandbox Code Playgroud)

尝试通过添加OR b.c1 is null到 where来执行左连接:

hive> select a.c1, a.c2, b.c1, b.c2 from a,b where (a.c1=b.c1) OR (b.c1 is null);
OK
1       row one 1       row one
Time taken: 57.317 seconds, Fetched: 1 row(s)
Run Code Online (Sandbox Code Playgroud)

如您所见,我们再次获得了内连接。or b.c1 is null被忽略

现在left join没有whereandON子句(转换为 CROSS):

 select a.c1, a.c2, b.c1, b.c2 from a left join b;
OK
1       row one 1       row one
1       row one 3       row three
2       row two 1       row one
2       row two 3       row three
Time taken: 37.104 seconds, Fetched: 4 row(s)
Run Code Online (Sandbox Code Playgroud)

正如你所看到的,我们再次交叉。

尝试使用带where子句和不带子句的左连接ON(用作内部):

select a.c1, a.c2, b.c1, b.c2 from a left join b where a.c1=b.c1;
OK
1       row one 1       row one
Time taken: 40.617 seconds, Fetched: 1 row(s)
Run Code Online (Sandbox Code Playgroud)

我们得到了 INNER 加入

尝试 left join withwhere子句,而 without ON+ 尝试允许空值:

 select a.c1, a.c2, b.c1, b.c2 from a left join b where a.c1=b.c1 or b.c1 is null;
OK
1       row one 1       row one
Time taken: 53.873 seconds, Fetched: 1 row(s)
Run Code Online (Sandbox Code Playgroud)

再次获得了内在。或被b.c1 is null忽略。

左连接on子句:

hive> select a.c1, a.c2, b.c1, b.c2 from a left join b on a.c1=b.c1;
OK
1       row one 1       row one
2       row two NULL    NULL
Time taken: 48.626 seconds, Fetched: 2 row(s)
Run Code Online (Sandbox Code Playgroud)

是的,这是真正的左连接。

左连接on+ where(得到 INNER):

hive> select a.c1, a.c2, b.c1, b.c2 from a left join b on a.c1=b.c1 where a.c1=b.c1;
OK
1       row one 1       row one
Time taken: 49.54 seconds, Fetched: 1 row(s)
Run Code Online (Sandbox Code Playgroud)

我们得到 INNER 是因为 WHERE 不允许 NULLS。

左连接 where + 允许空值:

hive> select a.c1, a.c2, b.c1, b.c2 from a left join b on a.c1=b.c1 where a.c1=b.c1 or b.c1 is null;
OK
1       row one 1       row one
2       row two NULL    NULL
Time taken: 55.951 seconds, Fetched: 2 row(s)
Run Code Online (Sandbox Code Playgroud)

是的,它是左连接。

结论:

  1. 如果没有 WHERE 子句,则隐式连接用作 INNNER(带 where)或 CROSS。
  2. 如果没有 ON 和 WHERE,左连接可以作为 CROSS 工作,如果 WHERE 子句不允许右表为空,也可以作为 INNER 工作。
  3. 最好使用 ANSI 语法,因为它是不言自明的,并且很容易理解您期望它的工作方式。作为 INNER 或 CROSS 工作的隐式连接或左连接很难理解并且很容易出错。