未能将窗口调用分解为组。至少 1 个组必须仅依赖于输入列答案

【问题标题】：Failed to breakup Windowing invocations into Groups. At least 1 group must only depend on input columns未能将窗口调用分解为组。至少 1 个组必须仅依赖于输入列
【发布时间】：2021-04-22 07:52:34
【问题描述】：

我有一个预订酒店的数据集。 date_in 的格式为“yyyy-MM-dd”。我需要按月选择前 10 家访问量最大的酒店。

SELECT top_visits.date_ci, top_visits.hotel_id, top_visits.count_visits
FROM (
   SELECT  date_ci, hotel_id, COUNT(id) AS count_visits, 
   RANK() OVER (
   PARTITION BY date_ci, hotel_id ORDER BY COUNT(id) DESC) as rank
   FROM ( 
      SELECT id, hotel_id, SUBSTRING(my_tab.date_in, 1, 7) as date_ci 
      FROM my_database.my_tab) x
   ) top_visits
GROUP BY date_ci, hotel_id HAVING rank <= 10;

我收到以下错误：

错误：编译语句时出错：FAILED: SemanticException 未能将窗口调用分解为组。至少1组必须只依赖于输入列。还要检查循环依赖关系。潜在错误： org.apache.hadoop.hive.ql.parse.SemanticException：第 4:13 行表达式不在 GROUP BY 键 'hotel_id' 中

【问题讨论】：

PARTITION BY date_ci, hotel_id ORDER BY COUNT(id) DESC) as rank -> PARTITION BY date_ci ORDER BY COUNT(id) DESC) as rank

标签： sql hive hiveql window-functions aggregation

【解决方案1】：

将 COUNT(id) 聚合移动到子查询中，添加 group by：

SELECT top_visits.date_ci, top_visits.hotel_id, top_visits.count_visits
FROM (
      SELECT date_ci, hotel_id, count_visits, 
             RANK() OVER (PARTITION BY date_ci, hotel_id ORDER BY count_visits DESC) as rank
        FROM ( 
              SELECT hotel_id, SUBSTRING(my_tab.date_in, 1, 7) as date_ci,
                     COUNT(id) AS count_visits 
                FROM my_database.my_tab
              GROUP BY hotel_id, SUBSTRING(my_tab.date_in, 1, 7)
             ) x
     ) top_visits
WHERE rank <= 10;

【讨论】：

【解决方案2】：

我会建议这样的事情：

SELECT ymh.*
FROM (SELECT YEAR(date_in) as yyyy, MONTH(date_in) as mm, hotel_id, 
             COUNT(*) AS count_visits, 
             ROW_NUMBER() OVER (PARTITION BY YEAR(date_in), MONTH(date_in), COUNT(*) DESC) as seqnum
      FROM my_database.my_tab
      GROUP BY YEAR(date_in), MONTH(date_in)
     ) ymh
WHERE seqnum <= 10;

也就是说，有一个聚合和一个窗口函数调用来枚举值。

【讨论】：