这个查询如何使用窗口函数为每个键返回多个结果？

Posted 2023-03-21

【中文标题】这个查询如何使用窗口函数为每个键返回多个结果？

【英文标题】：How is this query using a window function returning multiple results per key?

【发布时间】：2021-06-05 16:38:24

【问题描述】：

我写了以下查询

SELECT
    data.id,
    LAST_VALUE(data.access_type) OVER (
        PARTITION BY data.id
        ORDER BY CAST(FROM_UNIXTIME(ts) AS TIMESTAMP)
    ) AS access_type,
    LAST_VALUE(CAST(FROM_UNIXTIME(ts) AS TIMESTAMP)) OVER (
        PARTITION BY data.id
        ORDER BY CAST(FROM_UNIXTIME(ts) AS TIMESTAMP)
    ) AS access_type_timestamp
FROM
    table

data 是一个结构体。

我希望每个id 返回一行，最新的access_type 和ts 对应id。但是，它有时仍会为每个 id 返回多行。

我做错了什么？

【问题讨论】：

窗口函数应用于表格的每一行，不影响返回的行数。使用SELECT DISTINCT ....

【参考方案1】：

使用SELECT DISTINCT。我建议FIRST_VALUE():

SELECT DISTINCT
    data.id,
    FIRST_VALUE(data.access_type) OVER (
        PARTITION BY data.id
        ORDER BY CAST(FROM_UNIXTIME(ts) AS TIMESTAMP) DESC
    ) AS access_type,
    FIRST_VALUE(CAST(FROM_UNIXTIME(ts) AS TIMESTAMP)) OVER (
        PARTITION BY data.id
        ORDER BY CAST(FROM_UNIXTIME(ts) AS TIMESTAMP) DESC
    ) AS access_type_timestamp
FROM table;

窗口函数不会减少行数。

另外，我会假设ts 的排序方式与时间戳相同，因此可以简化。另外，第二个就是MAX()：

SELECT DISTINCT data.id,
       FIRST_VALUE(data.access_type) OVER (PARTITION BY data.id
                                           ORDER BY ts DESC
                                          ) AS access_type,
       MAX(CAST(FROM_UNIXTIME(ts) AS TIMESTAMP)) OVER (PARTITION BY data.id) AS access_type_timestamp
FROM table;

如果你使用LAST_VALUE()，你需要一个窗口子句：

LAST_VALUE(data.access_type) OVER (PARTITION BY data.id
                                   ORDER BY ts
                                   ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING
                                 ) AS access_type

带有ORDER BY 的默认窗口子句是RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW——这有一个令人讨厌的习惯，导致LAST_VALUE() 只返回当前行中的值。