我们有一个生产数据库，可以提前数年管理 100 多个分支机构的人员预订，精度达到分钟级。
该系统的一部分是突出显示差距的报告，即比较分行营业时间和员工预订情况，以查看是否有任何分行营业但没有人预订。
它还会同时检查重叠、重复预订等，基本上需要分钟级的精度。
我们这样做的方法是使用整数计数表将开放时间和预订的开始和结束时间扩展为分钟:
--===== Create and populate the Tally table on the fly
SELECT TOP 16777216
IDENTITY(INT,1,1) AS N
INTO dbo.Tally
FROM Master.dbo.SysColumns sc1,
Master.dbo.SysColumns sc2,
Master.dbo.SysColumns sc3
--===== Add a Primary Key to maximize performance
ALTER TABLE dbo.Tally
ADD CONSTRAINT PK_Tally_N
PRIMARY KEY CLUSTERED (N) WITH FILLFACTOR = 100
我们利用这个静态索引理货表来扩展开放时间和预订，如下所示:
SELECT [BranchID] ,
[DayOfWeek] ,
DATEADD(MINUTE, N - 1, StartTime)
FROM OpeningHours
LEFT OUTER JOIN tally ON tally.N BETWEEN 0
AND DATEDIFF(MINUTE, OpeningHours.StartTime, OpeningHours.EndTime) + 1
问题是，一旦我们有了 13,000,000 分钟的“开放分钟数”和“预订分钟数”，我们就需要加入结果以查看涵盖的内容:
SELECT OpenDatesAndMinutes.[Date] ,
OpenDatesAndMinutes.[Time] ,
OpenDatesAndMinutes.[BranchID] ,
ISNULL(BookedMinutes.BookingCount, 0) AS BookingCount
FROM OpenDatesAndMinutes
LEFT OUTER JOIN BookedMinutes ON OpenDatesAndMinutes.BranchID = BookedMinutes.BranchID
AND OpenDatesAndMinutes.[Date] = BookedMinutes.[Date]
AND OpenDatesAndMinutes.[Time] = BookedMinutes.[Time]
正如您所想象的那样，加入存储在 CTE 表中的 13,000,000 行的分支、日期和时间需要 AGES - 运行一周并不算太糟糕，大约 10 秒，但如果我们运行 6 个月(13,000,000 分钟) ) 膨胀到 25 分钟+
一旦我们将开放分钟数加入预订分钟数，我们就会将数据分组并呈现给用户:
CrossTabPrep ( [Date], [Time], [BranchID], [BookingCount], [Grp] )
AS ( SELECT [Date] ,
[Time] ,
[BranchID] ,
[BookingCount] ,
DATEPART(HOUR, Time) * 60 + DATEPART(MINUTE, Time) - ROW_NUMBER() OVER ( PARTITION BY [BranchID], Date, [BookingCount] ORDER BY Time ) AS [Grp]
FROM PreRender
),
FinalRender ( [BranchID], [Date], [Start Time], [End Time], [Duration], [EntryCount], [EntryColour] )
AS ( SELECT [BranchID] ,
[Date] ,
MIN([Time]) AS [Start Time] ,
MAX([Time]) AS [End Time] ,
ISNULL(DATEDIFF(MINUTE, MIN([Time]), MAX([Time])), 0) AS Duration ,
[BookingCount] AS EntryCount ,
CASE WHEN [BookingCount] = 0 THEN 'Red'
WHEN [BookingCount] = 1 THEN 'Green'
ELSE 'Yellow'
END AS EntryColour
FROM CrossTabPrep
GROUP BY [BranchID] ,
[Date] ,
[BookingCount] ,
[Grp]
)
很简单，我的方法是否有效？有什么办法可以改进这种方法，同时保持分钟级的准确性？在处理像这样的大量 CTE 表时，将这些数据转储到索引临时表并加入它们会有什么好处吗？
我正在考虑的另一件事是替换大连接使用的 DATE & TIME(0) 数据类型，如果我将它们转换为整数会更有效吗？
以下是完整的 CTE，以防有帮助:
WITH OpeningHours ( [BranchID], [DayOfWeek], [StartTime], [EndTime] )
AS ( SELECT BranchID ,
DayOfWeek ,
CONVERT(TIME(0), AM_open) ,
CONVERT(TIME(0), AM_close)
FROM db_BranchDetails.dbo.tbl_ShopOpeningTimes (NOLOCK)
INNER JOIN @tbl_Days Filter_Days ON db_BranchDetails.dbo.tbl_ShopOpeningTimes.DayOfWeek = Filter_Days.DayNumber
WHERE CONVERT(TIME(0), AM_open) <> CONVERT(TIME(0), '00:00:00')
UNION ALL
SELECT BranchID ,
DayOfWeek ,
CONVERT(TIME(0), PM_open) ,
CONVERT(TIME(0), PM_close)
FROM db_BranchDetails.dbo.tbl_ShopOpeningTimes (NOLOCK)
INNER JOIN @tbl_Days Filter_Days ON db_BranchDetails.dbo.tbl_ShopOpeningTimes.DayOfWeek = Filter_Days.DayNumber
WHERE CONVERT(TIME(0), PM_open) <> CONVERT(TIME(0), '00:00:00')
UNION ALL
SELECT BranchID ,
DayOfWeek ,
CONVERT(TIME(0), EVE_open) ,
CONVERT(TIME(0), EVE_close)
FROM db_BranchDetails.dbo.tbl_ShopOpeningTimes (NOLOCK)
INNER JOIN @tbl_Days Filter_Days ON db_BranchDetails.dbo.tbl_ShopOpeningTimes.DayOfWeek = Filter_Days.DayNumber
WHERE CONVERT(TIME(0), EVE_open) <> CONVERT(TIME(0), '00:00:00')
),
DateRange ( [Date], [DayOfWeek] )
AS ( SELECT CONVERT(DATE, DATEADD(DAY, N - 1, @StartDate)) ,
DATEPART(WEEKDAY, DATEADD(DAY, N - 1, @StartDate))
FROM tally (NOLOCK)
WHERE N <= DATEDIFF(DAY, @StartDate, @EndDate) + 1
),
OpenMinutes ( [BranchID], [DayOfWeek], [Time] )
AS ( SELECT [BranchID] ,
[DayOfWeek] ,
DATEADD(MINUTE, N - 1, StartTime)
FROM OpeningHours
LEFT OUTER JOIN tally ON tally.N BETWEEN 0
AND DATEDIFF(MINUTE, OpeningHours.StartTime, OpeningHours.EndTime) + 1
),
OpenDatesAndMinutes ( [Date], [Time], [BranchID] )
AS ( SELECT DateRange.[Date] ,
OpenMinutes.[Time] ,
OpenMinutes.BranchID
FROM DateRange
LEFT OUTER JOIN OpenMinutes ON DateRange.DayOfWeek = OpenMinutes.DayOfWeek
WHERE OpenMinutes.BranchID IS NOT NULL
),
WhiteListEmployees ( [DET_NUMBERA] )
AS ( SELECT DET_NUMBERA
FROM [dbo].[tbl_ChrisCache_WhiteList]
WHERE [TimeSheetV2_SecurityContext] = @TimeSheetV2_SecurityContext
),
BookedMinutesByRole ( [Date], [Time], [BranchID], BookingCount )
AS ( SELECT [BookingDate] ,
DATEADD(MINUTE, N - 1, StartTime) ,
BranchID ,
COUNT(BookingID) AS Bookings
FROM tbl_Booking (NOLOCK)
INNER JOIN tbl_BookingReason (NOLOCK) ON dbo.tbl_BookingReason.ReasonID = dbo.tbl_Booking.ReasonID
INNER JOIN tbl_ChrisCache (NOLOCK) ON dbo.tbl_Booking.DET_NUMBERA = dbo.tbl_ChrisCache.DET_NUMBERA
INNER JOIN @ValidPosCodes AS Filter_PostCodes ON dbo.tbl_ChrisCache.POS_NUMBERA = Filter_PostCodes.POSCODE
LEFT OUTER JOIN tally (NOLOCK) ON tally.N BETWEEN 0
AND DATEDIFF(MINUTE, tbl_Booking.StartTime, tbl_Booking.EndTime) + 1
WHERE ( Void = 0 )
AND tbl_BookingReason.CoverRequired = 0 --#### Only use bookings that dont require cover
AND tbl_booking.BranchID <> '023' --#### Branch 23 will always have messy data
AND ( dbo.tbl_Booking.BookingDate BETWEEN @StartDate
AND @EndDate )
GROUP BY [BookingDate] ,
BranchID ,
DATEADD(MINUTE, N - 1, StartTime)
),
BookedMinutesByWhiteList ( [Date], [Time], [BranchID], BookingCount )
AS ( SELECT [BookingDate] ,
DATEADD(MINUTE, N - 1, StartTime) ,
BranchID ,
COUNT(BookingID) AS Bookings
FROM tbl_Booking(NOLOCK)
INNER JOIN tbl_BookingReason (NOLOCK) ON dbo.tbl_BookingReason.ReasonID = dbo.tbl_Booking.ReasonID
INNER JOIN tbl_ChrisCache (NOLOCK) ON dbo.tbl_Booking.DET_NUMBERA = dbo.tbl_ChrisCache.DET_NUMBERA
INNER JOIN WhiteListEmployees Filter_WhiteList ON dbo.tbl_Booking.DET_NUMBERA = Filter_WhiteList.DET_NUMBERA
LEFT OUTER JOIN tally (NOLOCK) ON tally.N BETWEEN 0
AND DATEDIFF(MINUTE, tbl_Booking.StartTime, tbl_Booking.EndTime) + 1
WHERE ( Void = 0 )
AND tbl_BookingReason.CoverRequired = 0 --#### Only use bookings that dont require cover
AND tbl_booking.BranchID <> '023' --#### Branch 23 will always have messy data
AND ( dbo.tbl_Booking.BookingDate BETWEEN @StartDate
AND @EndDate )
GROUP BY [BookingDate] ,
BranchID ,
DATEADD(MINUTE, N - 1, StartTime)
),
BookedMinutes ( [Date], [Time], [BranchID], BookingCount )
AS ( SELECT [Date] ,
[Time] ,
[BranchID] ,
BookingCount
FROM BookedMinutesByRole
UNION
SELECT [Date] ,
[Time] ,
[BranchID] ,
BookingCount
FROM BookedMinutesByWhiteList
),
PreRender ( [Date], [Time], [BranchID], [BookingCount] )
AS ( SELECT OpenDatesAndMinutes.[Date] ,
OpenDatesAndMinutes.[Time] ,
OpenDatesAndMinutes.[BranchID] ,
ISNULL(BookedMinutes.BookingCount, 0) AS BookingCount
FROM OpenDatesAndMinutes
LEFT OUTER JOIN BookedMinutes ON OpenDatesAndMinutes.BranchID = BookedMinutes.BranchID
AND OpenDatesAndMinutes.[Date] = BookedMinutes.[Date]
AND OpenDatesAndMinutes.[Time] = BookedMinutes.[Time]
),
CrossTabPrep ( [Date], [Time], [BranchID], [BookingCount], [Grp] )
AS ( SELECT [Date] ,
[Time] ,
[BranchID] ,
[BookingCount] ,
DATEPART(HOUR, Time) * 60 + DATEPART(MINUTE, Time) - ROW_NUMBER() OVER ( PARTITION BY [BranchID], Date, [BookingCount] ORDER BY Time ) AS [Grp]
FROM PreRender
),
DeletedBranches ( [BranchID] )
AS ( SELECT [ShopNo]
FROM [dbo].[vw_BranchList]
WHERE [Branch_Deleted] = 1
),
FinalRender ( [BranchID], [Date], [Start Time], [End Time], [Duration], [EntryCount], [EntryColour] )
AS ( SELECT [BranchID] ,
[Date] ,
MIN([Time]) AS [Start Time] ,
MAX([Time]) AS [End Time] ,
ISNULL(DATEDIFF(MINUTE, MIN([Time]), MAX([Time])), 0) AS Duration ,
--dbo.format_timeV2(ISNULL(DATEDIFF(SECOND, MIN([Time]), MAX([Time])), 0)) AS DurationF ,
[BookingCount] AS EntryCount ,
CASE WHEN [BookingCount] = 0 THEN 'Red'
WHEN [BookingCount] = 1 THEN 'Green'
ELSE 'Yellow'
END AS EntryColour
FROM CrossTabPrep
GROUP BY [BranchID] ,
[Date] ,
[BookingCount] ,
[Grp]
)
SELECT [BranchID] ,
CONVERT(VARCHAR(10), DATEADD(DAY, 7, CONVERT(DATETIME, CONVERT(VARCHAR(10), DATEADD(day, -1 - ( DATEPART(dw, [Date]) + @@DATEFIRST - 2 ) % 7, [Date]), 103) + ' 23:59:59', 103)), 103) AS WeekEnding ,
[Date] ,
[Start Time] ,
[End Time] ,
[Duration] ,
CONVERT(VARCHAR, ( [Duration] * 60 ) / 3600) + 'h ' + CONVERT(VARCHAR, ROUND(( ( CONVERT(FLOAT, ( ( [Duration] * 60 ) % 3600 )) ) / 3600 ) * 60, 0)) + 'm' AS [DurationF] ,
[EntryCount] ,
[EntryColour] ,
CASE WHEN [EntryCount] = 0 THEN 'Red'
WHEN [EntryCount] >= 1 THEN 'Green'
END AS DurationColour ,
CASE WHEN [EntryCount] = 0 THEN 'This period of open-time isnt covered'
WHEN [EntryCount] >= 1 THEN 'This period of open-time is covered by ' + CONVERT(VARCHAR, [EntryCount]) + ' booking(s)'
END AS [DurationComment]
FROM FinalRender
WHERE FinalRender.BranchID NOT IN ( SELECT [BranchID]
FROM DeletedBranches )
最佳答案
这很有趣，因为你在最后用你的问题回答了你自己的问题。您应该尝试所有方法，但总结一下:
- 实现 CTE 以获得更好的性能。您永远不知道 SQL Server 何时会多次评估 CTE
- 您可以针对临时表构建 indexex。
- 我不确定您是如何从
[DayOfWeek],DATEADD(MINUTE, N - 1, StartTime)跳到加入[Date],[Time]另一方面，但这里有两列没有意义。使用单个datetime或代表一个纪元的seconds的 bigint。 UnixTimestamp 在这里工作得很好。
关于sql - 连接海量 CTE 表(13,000,000 行以上)性能问题，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/13293359/