Hive SELECT 语句来创建一个 STRUCTS 数组

Question

我在 Hive 中选择结构数组时遇到问题。

我的源表如下所示：

+-------------+--+
|    field    |
+-------------+--+
| id          |
| fieldid     |
| fieldlabel  |
| fieldtype   |
| answer_id   |
| unitname    |
+-------------+--+

这是调查数据，其中id是调查id，中间的四个字段是响应数据，unitname是调查所属的业务单位。

我需要为每个调查 ID 的所有答案创建一个结构数组。我认为这会起作用，但它不起作用：

select id, 
array( 
    named_struct(
        "field_id",
        fieldid,
        "field_label",
        fieldlabel,
        "field_type",
        fieldtype,
        "answer_id",
        answer_id,)) as answers,
unitname
from new_answers;

返回的是每个调查答案 (field_id) 作为该答案的一个结构的数组，如下所示：

id | answers | unitname
1 | [{"field_id":175877,"field_label":"Comment","field_type":"COMMENT","answer_id":8990947803}] | Location1
2 | [{"field_id":47824,"field_label":"Language","field_type":"MULTIPLE_CHOICE","answer_id":8990950069}] | Location2
2 | [{"field_id":48187,"field_label":"Language Type","field_type":"MULTIPLE_CHOICE","answer_id":8990950070}] | Location2
2 | [{"field_id":47829,"field_label":"Trans #","field_type":"TEXT","answer_id":8990950071}] | Location2

但我需要了解的是：

id | answers | unitname    
1 | [{"field_id":175877,"field_label":"Comment","field_type":"COMMENT","answer_id":8990947803}] | Location1
2 | [{"field_id":47824,"field_label":"Language","field_type":"MULTIPLE_CHOICE","answer_id":8990950069},
   {"field_id":48187,"field_label":"Language Type","field_type":"MULTIPLE_CHOICE","answer_id":8990950070},
   {"field_id":47829,"field_label":"Trans #","field_type":"TEXT","answer_id":8990950071}] | Location2

我已经搜索并搜索过，但我找到的所有答案似乎都与使用 INSERT INTO ....VALUES() 查询有关。我已经有了一个表结构；我只是无法让 ARRAY 以它应该的方式进行 ARRAY。

非常感激任何的帮助。

出于复制目的，如果需要：

CREATE TABLE `new_answers`( 
`id` bigint,
`fieldid` bigint,
`fieldlabel` string,
`fieldtype` string,
`answer_id` bigint,
`unitname` string)

INSERT INTO new_answers VALUES
(1,175877,"Comment","COMMENT",8990947803,"Location1"),
(2,47824,"Language","MULTIPLE_CHOICE",8990950069,"Location2"),
(2,48187,"Language Type","MULTIPLE_CHOICE",8990950070,"Location2"),
(2,47829,"Trans #","TEXT",8990950071,"Location2");

Answer 1

您似乎正在寻找的功能是将结构收集到一个数组中。Hive 带有两个用于将事物收集到数组中的函数：collect_set 和 collect_list。但是，这些函数只能用于创建基本类型的数组。

砖房项目的 jar ( https://github.com/klout/brickhouse/wiki/Downloads ) 提供了许多功能，包括收集复杂类型的能力。

add jar hdfs://path/to/your/jars/brickhouse-0.6.0.jar

然后您可以collect使用您喜欢的任何名称添加函数：

create temporary function collect_struct as 'brickhouse.udf.collect.CollectUDAF';

以下查询：

select id
     , collect_struct( 
         named_struct(
           "field_id", fieldid,
           "field_label", fieldlabel,
           "field_type", fieldtype,
           "answer_id", answer_id)) as answers
     , unitname
  from new_answers
 group by id, unitname
;

提供以下结果：

id  answers unitname
1   [{"field_id":175877,"field_label":"Comment","field_type":"COMMENT","answer_id":8990947803}] Location1
2   [{"field_id":47824,"field_label":"Language","field_type":"MULTIPLE_CHOICE","answer_id":8990950069},{"field_id":48187,"field_label":"Language Type","field_type":"MULTIPLE_CHOICE","answer_id":8990950070},{"field_id":47829,"field_label":"Trans #","field_type":"TEXT","answer_id":8990950071}]    Location2