SQL ：在分区上使用 LEAD/LAG and/or ROW_NUMBER 查找 2 个事件之间的持续时间

Question

我有一个 table，其中包含按车辆和时间划分的一系列事件。车辆在这段时间内连接或断开连接。

我有兴趣计算连接事件 (connected = 1) 和随后的断开连接事件 (connected = 0) 之间的持续时间

我想在分区上使用 SQL 的 LEAD and LAG。我想知道如何对我的数据进行分区以实现此目标。

当然VehicleId是第一候选。第二个计算域是什么？

示例数据：

CREATE TABLE #Events (VehicleId int, connected bit, Time DateTime);

INSERT INTO #Events (VehicleId, connected, Time)
VALUES(25931, 0, '2020-10-13 16:02:10.117'),
(25931, 1, '2020-10-13 11:32:39.213'),
(25931, 1, '2020-10-13 10:04:29.470'),
(25925, 1, '2020-10-13 07:41:31.637'),
(25925, 1, '2020-10-13 06:06:22.600'),
(25931, 1, '2020-10-13 05:23:19.433'),
(25927, 1, '2020-10-13 01:01:36.460'),
(25931, 0, '2020-10-13 17:55:10.380'),
(25931, 0, '2020-10-13 12:14:10.837'),
(25931, 0, '2020-10-13 10:53:54.527'),
(25925, 0, '2020-10-13 09:06:52.063'),
(25931, 0, '2020-10-13 08:32:45.230'),
(25925, 0, '2020-10-13 06:42:37.627'),
(25925, 0, '2020-10-13 05:12:08.070'),
(25927, 0, '2020-10-13 04:42:23.887'),
(25927, 0, '2020-10-13 00:56:36.090')

SELECT * FROM #Events ORDER BY Time ASC

DROP TABLE #Events

SELECT查询结果：

VehicleId   connected   Time
25927   0   2020-10-13 00:56:36.090
25927   1   2020-10-13 01:01:36.460
25927   0   2020-10-13 04:42:23.887
25925   0   2020-10-13 05:12:08.070
25931   1   2020-10-13 05:23:19.433
25925   1   2020-10-13 06:06:22.600
25925   0   2020-10-13 06:42:37.627
25925   1   2020-10-13 07:41:31.637
25931   0   2020-10-13 08:32:45.230
25925   0   2020-10-13 09:06:52.063
25931   1   2020-10-13 10:04:29.470
25931   0   2020-10-13 10:53:54.527
25931   1   2020-10-13 11:32:39.213
25931   0   2020-10-13 12:14:10.837
25931   0   2020-10-13 16:02:10.117
25931   0   2020-10-13 17:55:10.380

编辑：我期待的结果集类似于

VehicleId, Duration (min)
25927, 221

以下两个事件：

25927   1   2020-10-13 01:01:36.460
25927   0   2020-10-13 04:42:23.887

每个 vehicleid 和一对 connected/disconnected.

依此类推

提前致谢。

编辑 2：根据评论 FIRST_VALUE/LAST_VALUE 不适合。问题已更新。

Answer 1

看起来您想要对“过渡”行的时间差求和，其中车辆从状态断开连接到连接状态。如果是这样，您可以使用 lag():

select vehicleid, sum(datediff(minute, lag_time, time)) sum_diff
from (
    select e.*, 
        lag(connected) over(partition by vehicleid order by time) lag_connected,
        lag(time)      over(partition by vehicleid order by time) lag_time
    from #events e
) e
where connected = 0 and lag_connected = 1
group by vehicleid

对于您的示例数据，this returns:

vehicleid | sum_diff
--------: | -------:
    25925 |      121
    25927 |      221
    25931 |      280

Answer 2

假设1总是后跟[=12=]那辆车，那么你可以使用 LEAD 获取下一次，然后 SUM DATEDIFF:

WITH CTE AS(
    SELECT VehicleId,
           connected,
           [Time], --Time is a data type, and doesn't have a date portion, I would suggest using a different name
           LEAD(Time) OVER (PARTITION BY VehicleID ORDER BY [Time]) AS NextTime
    FROM #Events E
    WHERE VehicleID = 25927)
SELECT VehicleID,
       SUM(DATEDIFF(MINUTE,[Time],NextTime)) AS Duration
FROM CTE
WHERE Connected = 1
GROUP BY VehicleId;

Answer 3

此解决方案不使用 FIRST_VALUE/LAST_VALUE。

;
WITH Ranked AS
(
    SELECT 
        *,
        DateRowNumber = ROW_NUMBER()OVER(ORDER BY Time)
    FROM  #Events            
)
,Joined AS
(
    SELECT 
        *,
        JoiningId = CASE WHEN connected=1 THEN LEAD(DateRowNumber) OVER(PARTITION BY VehicleId ORDER BY Time) ELSE NULL END  
    FROM Ranked 
)
SELECT 
    J.VehicleId, 
    J.Time, 
    R.Time, 
    DifferencetInSeconds = DATEDIFF(SECOND,J.Time, R.Time) 
FROM 
    Joined J
    INNER JOIN Ranked R ON r.DateRowNumber = J.JoiningId
ORDER BY 
    J.VehicleId, J.Time

    SELECT * FROM #Events ORDER BY VehicleID, Time


VehicleId   Time                    Time                    DifferencetInSeconds
----------- ----------------------- ----------------------- --------------------
25925       2020-10-13 06:06:22.600 2020-10-13 06:42:37.627 2175
25925       2020-10-13 07:41:31.637 2020-10-13 09:06:52.063 5121
25927       2020-10-13 01:01:36.460 2020-10-13 04:42:23.887 13247
25931       2020-10-13 05:23:19.433 2020-10-13 08:32:45.230 11366
25931       2020-10-13 10:04:29.470 2020-10-13 10:53:54.527 2965
25931       2020-10-13 11:32:39.213 2020-10-13 12:14:10.837 2491

SQL ：在分区上使用 LEAD/LAG and/or ROW_NUMBER 查找 2 个事件之间的持续时间

SQL : Finding duration between 2 events using LEAD/LAG and/or ROW_NUMBER over a partition

tsql

datetime

window-functions

gaps-and-islands

sql-server-2012