在 postgres 窗口框架中定义相对百分比框架限制

Question

我有一个查询，试图查找一组服务地点的平均能耗，这些服务地点在特定月份内每个服务地点的能耗差异在 ±5% 之内。我知道 range 和 rows 函数无法定义特定的数字范围，而是使用前面和后面的子句，但是我如何编写这个查询来满足我的需要？

WITH energy_consumption_cte AS (
    SELECT
        sl.id,
        sl.sqft,
        SUM(CASE WHEN dl.log_label = 'energy use' THEN dl.value END) AS monthly_consumption
    FROM
        serv_location sl
    INNER JOIN
        device d ON d.service_loc_id = sl.id
    INNER JOIN
        device_log dl ON dl.device_id = d.id
    WHERE
        dl."timestamp" >= '2022-08-01 00:01:00.000'::timestamp
        AND dl."timestamp" < '2022-09-01 00:01:00.000'::timestamp
        AND sl.sqft IS NOT NULL
    GROUP BY
        sl.id, sl.sqft
)

SELECT
    id,
    sqft,
    monthly_consumption,
    (monthly_consumption / AVG(monthly_consumption) OVER w) * 100 
    AS average_monthly_for_sqft_range
FROM
    energy_consumption_cte
window w as (PARTITION BY sqft ORDER BY sqft range between 
    ROUND(sqft * 0.95) and ROUND(sqft * 1.05))
ORDER BY
    sqft ASC;

Answer 1

窗框

range

模式可以定义特定范围。要使用包含与当前记录相比向上/向下 5% 的记录值的聚合函数，您应该想到

range between 0.05*sqft preceding and 0.05*sqft following

。问题是，你不能在那里使用变量。

在 offset

PRECEDING

和 offset

FOLLOWING

框架选项中，offset 必须是不包含任何变量、聚合函数或窗口函数的表达式。

如果您尝试引用

sqft

列/字段，您将得到：

ERROR:  argument of RANGE must not contain variables
LINE 27:              RANGE BETWEEN sqft*0.05 PRECEDING
                                    ^

您可以通过首先使用

lag()

或

lead()

计算行与行之间的百分比变化来解决该限制，然后将其收集到步进

sum()

中，以便它保持

sqft

的初始顺序。一旦你有了这个，你可以在框架子句中使用常量偏移来达到 5% 来回。 db<>fiddle 的演示：

WITH energy_consumption_cte AS (
    SELECT sl.id,
           sl.sqft,
           SUM(dl.value)filter(where dl.log_label='energy use') AS monthly_consumption
    FROM serv_location sl
    INNER JOIN device d ON d.service_loc_id = sl.id
    INNER JOIN device_log dl ON dl.device_id = d.id
    WHERE dl."timestamp" >= '2022-08-01 00:01:00.000'::timestamp
      AND dl."timestamp" < '2022-09-01 00:01:00.000'::timestamp
      AND sl.sqft IS NOT NULL
    GROUP BY sl.id, sl.sqft 
),square_footage_steps_cte AS (
    SELECT *, 1-lag(sqft,1,sqft)over(order by sqft)/sqft::numeric AS sqft_step
    FROM energy_consumption_cte
    ORDER BY sqft ASC
),square_footage_steps_cumulative_cte AS (
    SELECT *, sum(sqft_step)over(order by sqft) AS sqft_step_sum
    FROM square_footage_steps_cte
    ORDER BY sqft ASC)
SELECT id,
       sqft,
       monthly_consumption,
       AVG(monthly_consumption) OVER w AS average_monthly_for_sqft_range,
       array_agg(id)over w as rows_in_frame
FROM square_footage_steps_cumulative_cte 
WINDOW w as (ORDER BY sqft_step_sum
             RANGE BETWEEN 0.05 PRECEDING
                   AND     0.05 FOLLOWING)
ORDER BY sqft ASC;

id	平方英尺	每月_消费	平方英尺范围的平均每月	帧内行数
1	100	400	700.0000000000000000	{1,2}
2	105	1000	1000.0000000000000000	{1,2,3}
3	110	1600	1600.0000000000000000	{2,3,4}
4	115	2200	1900.0000000000000000	{3,4}
5	200	2800	3100.0000000000000000	{5,6}
6	210	3400	3100.0000000000000000	{5,6}