SQL：按日期报告数据答案

【问题标题】：SQL: Reporting on data by dateSQL：按日期报告数据
【发布时间】：2011-06-20 21:06:43
【问题描述】：

我有一张表，我正试图从中生成报告。它基本上是记录什么时候发生故障（发生故障），然后得到修复的日志。

表架构和一些示例数据如下。举例说明：
下降时插入 1 行
恢复时插入 1 行。

我想做的是报告各个方面，例如：
给定日/周/月的停机时间
在给定的一天/一周/一个月内下降的次数。

理想情况下，可以轻松导出到 Excel 或类似的图形。

我无法提出任何类型的查询来获取此信息。

我有这个例子：

SELECT [Name], datepart(day,[Inserted]), count([SystemDown])     
FROM [DownTimeLog]
WHERE [SystemDown]=1
GROUP BY [Name],datepart(day,[Inserted])

这给了我系统每天出现故障的次数，这是一个很好的起点。

但我试图想出一种方法来显示它的总时间，但我正在画一个空白。例如，有些天可能是 0，有时它可能会下降几次，因此尝试将 2 个对应行之间的时间差相加是很困难的。

CREATE TABLE [dbo].[DownTimeLog]
(
    [ID] [int] IDENTITY(1,1) NOT NULL,
    [Name] VARCHAR(30) NOT NULL,
    [SystemDown] BIT NOT NULL,
    [Inserted] DATETIME NOT NULL
)

INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 14 2011  1:49:58:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 14 2011  2:49:58:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 15 2011  1:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 15 2011  2:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 15 2011  4:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 15 2011  5:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 17 2011  1:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 17 2011  3:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 18 2011 10:00:00:000AM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 18 2011 11:00:00:000AM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 18 2011  1:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 18 2011  3:30:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 18 2011  4:00:00:000PM')
INSERT INTO [DownTimeLog] ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 18 2011  8:00:00:000PM')

因此，例如使用上面的数据，我想拉回类似的数据：

系统1 | 2011 年 6 月 14 日 | 1 小时 | 1 次出现
系统1 | 2011 年 6 月 15 日 | 2 小时 | 2次出现的
系统1 | 2011 年 6 月 16 日 | 0 小时 | 0 次出现的
系统1 | 2011 年 6 月 17 日 | 2 小时 | 1 次出现
系统1 | 2011 年 6 月 18 日 | 7.5 小时 | 3次出现的

我希望有人能给我一个方法来做我想做的事情。

-- 编辑：

感谢大家的一些很棒的答案。帮了我一吨。我认为我的 sql 是很漂亮的字符串，但从未听说过交叉申请 - 我想我需要回到学校！

干杯！

【问题讨论】：

您使用的是什么版本的 SQL Server？像this self-join solution 这样的东西可能对你有用。
如果系统从晚上 11 点到凌晨 1 点停机，是每天 1 小时，还是第一天 2 小时？
如果系统在某一天完全没有宕机，还需要显示0小时的记录吗？

标签： sql sql-server tsql

【解决方案1】：

以下将产生您的确切输出，包括缺失的天数。请注意，此代码基于 Mikael Eriksson 的回答。

CREATE TABLE #DownTimeLog
(
    [ID] [int] IDENTITY(1,1) NOT NULL,
    [Name] VARCHAR(30) NOT NULL,
    [SystemDown] BIT NOT NULL,
    [Inserted] DATETIME NOT NULL
)

INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 14 2011  1:49:58:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 14 2011  2:49:58:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 15 2011  1:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 15 2011  2:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 15 2011  4:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 15 2011  5:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 17 2011  1:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 17 2011  3:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 18 2011 10:00:00:000AM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 18 2011 11:00:00:000AM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 18 2011  1:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 18 2011  3:30:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',1,'Jun 18 2011  4:00:00:000PM')
INSERT INTO #DownTimeLog ([Name],[SystemDown],[Inserted])VALUES('System1',0,'Jun 18 2011  8:00:00:000PM')

CREATE TABLE #DownTimeLogModified (Name nvarchar(512), [Date] nvarchar(512), DownTime int,  Occurrences int)

INSERT INTO #DownTimeLogModified (Name, [Date], DownTime,  Occurrences)
select D1.Name,
       CONVERT(VARCHAR(12), dateadd(d, datediff(d, 0, D1.Inserted), 0), 107) as [Date],
       sum(datediff(mi, D1.Inserted, D2.Inserted)) as DownTime,
       count(*) as Occurrences
from #DownTimeLog as D1
  cross apply ( select top 1 Name,
                             Inserted
                from #DownTimeLog
                where SystemDown = 0 and
                      Name = D1.Name and
                      Inserted > D1.Inserted 
                order by Inserted             
              ) as D2
where D1.SystemDown = 1
group by D1.Name, dateadd(d, datediff(d, 0, D1.Inserted), 0)

DECLARE @startdate datetime,@enddate datetime
select @startdate = MIN(Inserted) FROM #DownTimeLog
select @enddate = MAX(Inserted) FROM #DownTimeLog

;WITH DateIntervalsCTE AS
        (
        SELECT 1 i,@startdate AS Date
        UNION ALL
        SELECT i + 1, DATEADD(day, i, @startdate )
        FROM DateIntervalsCTE 
        WHERE DATEADD(day, i, @startdate ) <= @enddate
        )

SELECT CASE WHEN a.Name is null THEN 'No System downtime' ELSE a.Name END as Name,CONVERT(VARCHAR(12), b.Date, 107) AS Date,CASE WHEN a.DownTime is null THEN 0 ELSE a.DownTime END AS DownTime,CASE WHEN a.Occurrences is null THEN 0 ELSE a.Occurrences END AS Occurrences 
FROM DateIntervalsCTE b
LEFT JOIN #DownTimeLogModified a ON a.Date = CONVERT(VARCHAR(12), b.Date, 107)

DROP TABLE #DownTimeLog
DROP TABLE #DownTimeLogModified

输出

Name                           Date                    DownTime    Occurrences
------------------------------ ----------------------- ----------- -----------
System1                        Jun 14, 2011                60           1
System1                        Jun 15, 2011               120           2
No system downtime             Jun 16, 2011                 0           0
System1                        Jun 17, 2011               120           1
System1                        Jun 18, 2011               450           3

【讨论】：

【解决方案2】：

这样的事情可能对你有用。

select D1.Name,
       dateadd(d, datediff(d, 0, D1.Inserted), 0) as [Date],
       sum(datediff(mi, D1.Inserted, D2.Inserted)) as DownTime,
       count(*) as Occurrences
from DownTimeLog as D1
  cross apply ( select top 1 Name,
                             Inserted
                from DownTimeLog
                where Name = D1.Name and
                      Inserted > D1.Inserted 
                order by Inserted             
              ) as D2
where D1.SystemDown = 1
group by D1.Name, dateadd(d, datediff(d, 0, D1.Inserted), 0)

结果：

Name                           Date                    DownTime    Occurrences
------------------------------ ----------------------- ----------- -----------
System1                        2011-06-14 00:00:00.000 60          1
System1                        2011-06-15 00:00:00.000 120         2
System1                        2011-06-17 00:00:00.000 120         1
System1                        2011-06-18 00:00:00.000 450         3

您不会连续几天出现 0 次。如果您在午夜有停机时间，则所有时间都将计入停机事件当天。停机时间以分钟为单位。

【讨论】：

【解决方案3】：

我认为，如果您将它们分成两个伪表，您将能够做更多事情。试试这个：

SELECT
    *
FROM
    ( SELECT
        *
      FROM
        [DownTimeLog] AS down
      WHERE
        [SystemDown] = 1
    ) down
    INNER JOIN ( SELECT
                    *
                 FROM
                    [DownTimeLog]
                 WHERE
                    SystemDown = 0
               ) up
    ON down.id = ( up.id - 1 ) and down.name = up.name
    ORDER BY down.[Inserted], up.inserted

然后，您可以对 down 记录与随后的 up 记录进行各种计算。

编辑：当然，@Dems 指出这假设 ID 是连续的。如果您更愿意使用日期作为过滤器，请执行以下操作：

SELECT
    *
FROM
    ( SELECT
        *
      FROM
        [DownTimeLog] AS down
      WHERE
        [SystemDown] = 1
    ) down
    CROSS APPLY ( SELECT TOP 1
                    *
                 FROM
                    [DownTimeLog] up
                 WHERE
                    SystemDown = 0

                    AND up.INSERTED > down.INSERTED
                    AND up.NAME = down.name
                    ORDER BY up.inserted 
               ) up
    ORDER BY down.[Inserted], up.inserted

下面是一个更详细的例子：

SELECT
        [Down_Id]
    ,   [Down_Name]
    ,   [Down_SystemDown]
    ,   [Down_Inserted]
    ,   [Up_Id]
    ,   [Up_Name]
    ,   [Up_SystemDown]
    ,   [Up_Inserted]
    ,   CAST(DATEDIFF(mi,Down_Inserted,Up_Inserted) AS DECIMAL)/60  AS Hours_Down
    ,   DATEDIFF(mi,Down_Inserted,Up_Inserted)  AS Minutes_Down
FROM
    ( SELECT
        [ID] AS Down_Id
      , [Name] AS Down_Name
      , [SystemDown] AS Down_SystemDown
      , [Inserted] AS Down_Inserted
      FROM
        [DownTimeLog] AS down
      WHERE
        [SystemDown] = 1
    ) down
    CROSS APPLY ( SELECT TOP 1
                    [ID] AS Up_Id
                  , [Name] AS Up_Name
                  , [SystemDown] AS Up_SystemDown
                  , [Inserted] AS Up_Inserted
                  FROM
                    [DownTimeLog] up
                  WHERE
                    SystemDown = 0
                    AND up.[Inserted] > down.Down_Inserted
                    AND up.NAME = down.Down_name
                  ORDER BY
                    up.Inserted
                ) up
ORDER BY
    down.[Down_Inserted]
,   up.up_inserted

【讨论】：

您已经假设这里只有一个系统插入到表中，因此所有记录都是顺序的。 System 字段的存在导致我做出相反的假设。
@Dems，你说得很好。一开始我忘了加入name。此外，根据系统的实际设置方式，使用inserted 列可能比依赖顺序 ID 更好。

【解决方案4】：

WITH
  enhanced_log
AS
(
SELECT
  [start].Name                                              AS [Name],
  DATEADD(DAY, DATEDIFF(DAY, 0, [start].Inserted), 0)       AS [Date],
  [start].Inserted                                          AS [Start],
  [finish].Inserted                                         AS [Finish]
FROM
  DownTimeLog  AS [start]
OUTER APPLY
  (SELECT TOP 1 * FROM DownTimeLog WHERE name = [start].name AND Inserted > [start].Inserted ORDER BY Inserted ASC) AS [finish]
WHERE
  [start].SystemDown = 1

UNION ALL

SELECT
  [start].Name,
  [Calendar].date,
  [Calendar].date,
  [finish].Inserted
FROM
  calendar
CROSS JOIN
  system
CROSS APPLY
  (SELECT TOP 1 * FROM DownTimeLog WHERE name = [system].name AND Inserted < [calendar].date ORDER BY Inserted ASC) AS [start]
OUTER APPLY
  (SELECT TOP 1 * FROM DownTimeLog WHERE name = [system].name AND Inserted > [start].Inserted ORDER BY Inserted ASC) AS [finish]
WHERE
  [start].SystemDown = 1
)

SELECT
  Name,
  Date,
  SUM(DATEDIFF(MINUTE, Start, CASE WHEN Finish < Date + 1 THEN Finish ELSE Date + 1 END)) AS Duration,
  COUNT(*) AS Instances
FROM
  enhanced_log
WHERE
  Start <> CASE WHEN Finish < Date + 1 THEN Finish ELSE Date + 1 END
GROUP BY
  Name,
  Date

【讨论】：