Hive-将层次结构表展平到各个级别

时间:2019-02-10 06:23:40

标签: hadoop hive hiveql impala

我的层次结构表的父子关系最大级别为15。我需要找出每个父节点的所有子节点级别。 我已经尝试了递归查询,但是在Hive和Impala中不起作用。 请提出一些查询来解决此问题。

+---------------+----------------+
|parent_node_id | child_node_id  |
+---------------+----------------+
|  C1           |    C11         |
+---------------+----------------+
|  C11          |    C12         |
+---------------+----------------+
|  C12          |    123         |
+---------------+----------------+
|  C12          |    456         |
+---------------+----------------+
|  P1           |    C1          |
+---------------+----------------+
|  P1           |    C2          | 
+---------------+----------------+
|  P2           |    C3          | 
+---------------+----------------+


+-------+----------+
| node  | balance  |
+-------+----------+
|  123  |  100     |
+-------+----------+
|  456  |  150     |
+-------+----------+
|  C2   |  400     |
+-------+----------+
|  C3   |  200     |
+-------+----------+

输出数据:

+---------------+-------------+---------+----------+----------+
|Parent_node_id |   Level 1   | Level 2 | Level 2  | Level3   | 
+---------------+-------------+---------+----------+----------+
|  P1           |    C1       | C11     |  C12     |  123     |
+---------------+-------------+---------+----------+----------+
|  P1           |    C1       | C11     |  C12     |  456     |
+---------------+-------------+---------+----------+----------+
|  P1           |    C2       | NULL    |  NULL    |  NULL    |
+---------------+-------------+---------+----------+----------+
|  P2           |    C3       | NULL    |  NULL    |  NULL    |
+---------------+-------------+---------+----------+----------+


+-------+----------+
| node  | balance  |
+-------+----------+
|  123  |  100     |
+-------+----------+
|  456  |  150     |
+-------+----------+
|  C12  |  250     |
+-------+----------+
|  C11  |  250     |
+-------+----------+
|  C1   |  250     |
+-------+----------+
|  P1   |  650     |
+-------+----------+
|  C2   |  200     |
+-------+----------+
|  P2   |  200     |
+-------+----------+

1 个答案:

答案 0 :(得分:1)

由于您知道可以在层次结构中建立的最大关系,

下面是查询-

hive> select * from hier;
OK
parent child child_desc
C1      C11     Party
C1      C111    Placement
C111    C1111   XM
C12     C121    Brand
C2      C12     Equity
XN12    C1      Private
XN12    C2      Public
Time taken: 0.123 seconds, Fetched: 7 row(s)

查询-

with lvl1 as
(
select p1.parent, p1.child child1, p1.child_desc child_desc1, p2.child child2, p2.child_desc child_desc2
from hier p1 left outer join hier p2
on p1.child = p2.parent
)
, lvl2 as
(
select lvl1.parent, lvl1.child1, lvl1.child_desc1, lvl1.child2 child2, lvl1.child_desc2, p3.parent p2, 
p3.child as child3, p3.child_desc child_desc3
from lvl1 left outer join hier p3 on nvl(lvl1.child2,'A') = nvl(p3.parent,'A')
)
, lvl3 as
(
select lvl2.parent, lvl2.child1 child1, lvl2.child_desc1, lvl2.child2 child2, lvl2.child_desc2, 
lvl2.child3 hild3, lvl2.child_desc3, p3.child as child4, p3.child as child_desc4
from lvl2 left outer join hier p3 on nvl(lvl2.child3,'A') = nvl(p3.parent,'A')
)
select * from lvl3;

C1      C11     Party   NULL    NULL    NULL    NULL    NULL    NULL
C1      C111    Placement       C1111   XM      NULL    NULL    NULL    NULL
C111    C1111   XM      NULL    NULL    NULL    NULL    NULL    NULL
C12     C121    Brand   NULL    NULL    NULL    NULL    NULL    NULL
C2      C12     Equity  C121    Brand   NULL    NULL    NULL    NULL
XN12    C1      Private C11     Party   NULL    NULL    NULL    NULL
XN12    C1      Private C111    Placement       C1111   XM      NULL    NULL
XN12    C2      Public  C12     Equity  C121    Brand   NULL    NULL

以上查询适用于3级,您可以将其复制到15级。

我已经对上述查询进行了3个级别的测试。

hive> select * from hier;
OK
parent child
C1      C11
C11     C12
C12     123
P1      C1
P2      C2

o / p-

C1      C11     C12     123     NULL
C11     C12     123     NULL    NULL
C12     123     NULL    NULL    NULL
P1      C1      C11     C12     123
P2      C2      NULL    NULL    NULL

如果您要查询n级并且不想对查询进行硬编码,则将需要一些脚本(工作量大)来创建查询或多次迭代同一查询。 / p>

希望这会有所帮助