HiveQL 查询不返回任何结果，也没有错误

Question

我在 Ubuntu 14.0 上运行 Apache Hadoop 2.6.0，我在 Hive 0.13.0 中创建了一个表：

CREATE TABLE IF NOT EXISTS recipes_hive.cuisine (
ID INT COMMENT 'Cuisine ID.', 
name STRING COMMENT 'Cusine name - primary key.', 
area STRING COMMENT 'Name of the area of origin - foreign key.', 
scope STRING COMMENT 'Either country or area.') 
COMMENT 'Table containing cuisines data.'
ROW FORMAT DELIMITED FIELDS TERMINATED BY ','
STORED AS TEXTFILE;

我用数据填充它：

LOAD DATA LOCAL INPATH 'path_to_file/CUISINE.csv'
OVERWRITE INTO TABLE recipes_hive.cuisine;

我的数据库有几个这样的表，它们都是用相同的过程创建和填充的。运行简单查询时，例如：

SELECT * FROM cuisine

或者甚至在 WHERE 子句中的某些条件下，我得到了预期的结果，但运行更复杂的查询我得到了蹲下。例如：

SELECT cuisine.name, SUM(IF (ingredient.category = "fruit",1,2))/count(*) AS      PERC 
FROM cuisine JOIN recipe ON recipe.cuisine = cuisine.name JOIN part_of ON part_of.id_recipe = recipe.id JOIN ingredient ON ingredient.name = part_of.ingredient 
GROUP BY cuisine.name 
ORDER BY PERC DESC

，或者：

SELECT ingredient.id, ingredient.name 
FROM cuisine JOIN recipe ON recipe.cuisine = cuisine.name JOIN part_of ON part_of.id_recipe = recipe.id JOIN ingredient ON ingredient.name = part_of.ingredient 
WHERE ingredient.id IN (
SELECT ingredient.id 
FROM cuisine c JOIN recipe ON recipe.cuisine = c.name JOIN part_of ON part_of.id_recipe = recipe.id JOIN ingredient ON ingredient.name = part_of.ingredient 
WHERE c.name = "Pakistan") AND cuisine.name = "Bangladesh"

第一个示例计算某个百分比，第二个示例检查相互元素。

MapReduce 和 Hadoop 被正确调用并且它们没有返回错误。输出结束于：

Execution completed successfully
MapredLocal task succeeded
OK
Time taken: 122.119 seconds

我已经检查了网络，人们也遇到了与我类似的问题。我检查了：

Hive Table returning empty result set on all queries

Simple Hive query is empty

但未能解决我的问题。数据实际上位于 HDFS 中，如前所述，它适用于简单查询。

所以要么我的 Hive 实例有问题，要么我的查询写得不正确。

任何帮助将不胜感激。 最好的问候。

Answer 1

如果我们有包含 ID = {1,2,3} 的 Cuisine 表和包含 ID = {5,6,7} 的 Recipe 表，那么即使这些表不是空的，我们仍然没有返回任何行做 INNER JOIN Cuisine.ID = Recipe.ID （因为 ID 在 2 个表中不同） 你能检查一下没有这样的情况吗？

SELECT count(1)
FROM cuisine c JOIN recipe ON recipe.cuisine = c.name WHERE c.name = "Pakistan";

--- must return > 0 

select count(1) from recipe as recipe
JOIN part_of ON part_of.id_recipe = recipe.id ;

--- must return > 0 

select count(1) from part_of as part_of
JOIN ingredient ON ingredient.name = part_of.ingredient ;

--- must return > 0 

因此，当所有 count(*) 非零时，内部查询返回一行。现在测试外部选择：

SELECT ingredient.id, ingredient.name 
FROM cuisine JOIN recipe ON recipe.cuisine = cuisine.name JOIN part_of ON part_of.id_recipe = recipe.id JOIN ingredient ON ingredient.name = part_of.ingredient 
WHERE ingredient.id = <inner query result> and cuisine.name = "Bangladesh";

Answer 2

您确定生成的联接不是空的吗？因为，您已经实现了内连接，即使一个表缺少记录，整个结果集也是 0。尝试添加一个带有“IS NULL”的左连接，以验证所有表都对结果集有贡献。如果所有子表在连接后各自的列中都有非空值，则查询是好的。