窗口帧

Snowflake 中的 Window Functions

Jake Roach

Field Data Engineer

窗口帧

显示由所有记录组成的窗口的输出。

窗口帧

显示使用 member_id 字段对记录分区构建窗口的输出。

动态窗口帧

显示"滑动"窗口帧的输出。

寻找累计计算

SELECT
    ...

    AVG(<1>) OVER(
        PARTITION BY ...
        ORDER BY <2>

        -- 窗口：从第一行
        -- 到当前行


        ROWS BETWEEN UNBOUNDED PRECEDING 
            AND CURRENT ROW

    )
...

ROWS BETWEEN 允许创建动态窗口帧

UNBOUNDED PRECEEDING AND CURRENT ROW
序列首行至当前行
"累计"计算，非滚动
CURRENT ROW AND UNBOUNDED FOLLOWING

<1>：参与计算的字段

<2>：确定顺序并构建窗口帧

燃烧卡路里的累计总和

SELECT
    member_id AS m_id, 
    calories_burned AS cb,

    -- 累计总和！
    SUM(calories_burned) OVER(
        PARTITION BY member_id
        ORDER BY workout_date


        ROWS BETWEEN UNBOUNDED PRECEDING
            AND CURRENT ROW


    ) AS running_total

FROM fitness.workouts;


         m_id  |    cb   |  running_total
       ------- | ------- | --------------
        m_192  |   105   |      105 
        m_192  |   156   |      261 
        m_192  |    69   |      330 
        m_192  |   102   |      432  

        m_74   |   374   |      374  
        m_74   |   396   |      770  
        m_74   |   504   |     1274  

        m_233  |    51   |       51  
        m_233  |    81   |      132

CURRENT ROW AND UNBOUNDED FOLLOWING

SELECT
    member_id AS m_id, 
    calories_burned AS cb,

    SUM(calories_burned) OVER(
        PARTITION BY member_id
        ORDER BY workout_date

        -- 窗口：当前行到最后一行

        ROWS BETWEEN CURRENT ROW
            AND UNBOUNDED FOLLOWING


    ) AS left_to_burn

FROM fitness.workouts;


         m_id  |    cb   |  left_to_burn
       ------- | ------- | --------------
        m_192  |   105   |      432 
        m_192  |   156   |      327 
        m_192  |    69   |      171 
        m_192  |   102   |      102  

        m_74   |   374   |     1274  
        m_74   |   396   |      900 
        m_74   |   504   |      504  

        m_233  |    51   |      132  
        m_233  |    81   |       81

卡路里消耗趋势

SELECT
    member_id,
    workout_date,
    calories_burned,

    AVG(calories_burned) OVER(   -- 燃烧卡路里的累计平均
        PARTITION BY member_id

        -- 按 workout_date 建窗口：从首条到当前行 
        ORDER BY workout_date
        ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW

    ) AS avg_calories_burned

FROM FITNESS.workouts;

卡路里消耗趋势


            member_id  |  workout_date  |  calories_burned  |  average_calories_burned
           ----------- | -------------- | ----------------- | -------------------------
              m_192    |   2025-01-01   |        105        |          105.0
              m_192    |   2025-01-03   |        156        |          130.5
              m_192    |   2025-01-04   |         69        |          110.0
              m_192    |   2025-01-10   |        102        |          108.0

               m_74    |   2025-02-10   |        374        |          374.0
               m_74    |   2025-02-13   |        396        |          385.0
               m_74    |   2025-02-14   |        504        |          426.7

              m_233    |   2025-03-05   |         51        |           51.0
              m_233    |   2025-03-12   |         81        |           66.0

Passons à la pratique !

Snowflake 中的 Window Functions