计算布尔列更改之间的间隔答案

【问题标题】：Calculate interval between boolean column change计算布尔列更改之间的间隔
【发布时间】：2020-01-31 21:28:30
【问题描述】：

我有一张天气测量表，这里是它的简化版本：

"station_id","measured_at","rainy"
-------------------------------------------------------------------------
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:35:35.27+00",FALSE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:36:33.976+00",FALSE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:37:33.864+00",FALSE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:38:34.767+00",TRUE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:39:36.076+00",TRUE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:40:29.776+00",FALSE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:41:35.579+00",FALSE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:42:34.274+00",TRUE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:43:23.842+00",TRUE
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485","2020-01-31 18:44:35.08+00",FALSE

我需要计算有多少时间是晴天。我知道我需要使用窗口功能，但我被卡住了。因为我需要在某个范围内计算这个间隔，例如最后一天。我已经可以查询到这个阶段了。

SELECT 
    prev.station_id,
    prev.rainy,
    prev.measured_at AS started_at,
    COALESCE(LEAD(prev.measured_at) OVER (ORDER BY prev.measured_at ASC), NOW()) AS ended_at,
    (COALESCE(LEAD(prev.measured_at) OVER (ORDER BY prev.measured_at ASC), NOW()) - prev.measured_at) AS diff
FROM (
    SELECT
        m.station_id,
        m.measured_at,
        m.rainy,
        COALESCE(LEAD(m.rainy) OVER (ORDER BY m.measured_at ASC), m.rainy) AS prev_rainy
    FROM
        z_measurements m
    WHERE m.measured_at >= '2020-01-30T00:00:00.000Z'
    ORDER BY m.measured_at ASC
) prev
WHERE prev.rainy IS DISTINCT FROM prev.prev_rainy
ORDER BY prev.measured_at ASC;

此查询结果为：

"station_id","rainy","started_at","ended_at","diff"
---------------------------------------------------
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485",FALSE,"2020-01-31 18:37:33.864","2020-01-31 18:39:36.076+00","00:02:02.212"
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485",TRUE,"2020-01-31 18:39:36.076","2020-01-31 18:41:35.579+00","00:01:59.503"
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485",FALSE,"2020-01-31 18:41:35.579","2020-01-31 18:43:23.842+00","00:01:48.263"
"b6b53561-dab5-4b9a-8d28-a7de1e4d1485",TRUE,"2020-01-31 18:43:23.842","2020-01-31 21:18:04.89333+00","02:34:41.05133"

但如果这个结果的第一条车道是rainy = False，那么start_date 应该与查询2020-01-30T00:00:00.000Z 中的一条相同（我假设如果第一个结果是晴天，那么它一直都是晴天在我的范围开始和第一次记录的测量之间）并且还缺少最后一行，因为最后测量的状态是晴天。所以最后一行的started_at 应该是2020-01-31 18:44:35.08+00 和end_at 应该是NOW()。

有人可以帮帮我吗？

我使用的是 postgresql 12.1。

【问题讨论】：

标签： sql postgresql intervals window-functions gaps-and-islands

【解决方案1】：

SELECT t.station_id, t.rainy, t.started_at, t.ended_at, t.ended_at - t.started_at AS diff
FROM (
    SELECT
        prev.station_id,
        prev.rainy,
        CASE
            WHEN LAG(prev.measured_at) OVER measured_at_by_station_id IS NULL THEN '2020-01-30T00:00:00.000Z'
            ELSE prev.measured_at
        END AS started_at,
        LEAD(prev.measured_at, 1, NOW()) OVER measured_at_by_station_id AS ended_at
    FROM (
        SELECT
            m.station_id,
            m.measured_at,
            m.rainy,
            LAG(m.rainy, 1, NOT(m.rainy)) OVER (PARTITION BY m.station_id ORDER BY m.measured_at ASC) AS prev_rainy
        FROM z_measurements m
        WHERE m.measured_at >= '2020-01-30T00:00:00.000Z'
        ORDER BY m.station_id ASC, m.measured_at ASC
    ) prev
    WHERE prev.rainy IS DISTINCT FROM prev.prev_rainy
    WINDOW measured_at_by_station_id AS (PARTITION BY prev.station_id ORDER BY prev.measured_at ASC)
) t
ORDER BY t.station_id ASC, t.started_at ASC

【讨论】：

【解决方案2】：

这是一个间隙和孤岛问题，您希望将一系列相邻记录组合在一起。

这是在两个不同分区上使用row_number() 解决它的一种方法：排名之间的差异为您提供每条记录所属的组，您可以使用它来聚合结果集。

select
    station_id,
    rainy,
    min(measured_at) started_at,
    max(measured_at) ended_at,
    max(measured_at) - min(measured_at) diff
from (
    select 
        t.*,
        row_number() over(partition by station_id order by measured_at) rn1,
        row_number() over(partition by station_id, rainy order by measured_at) rn2
    from mytable t
) t
group by station_id, rainy, rn1 - rn2
order by station_id, started_at

【讨论】：