您好，我在 SQL Server 数据库中有一个仪表值表，其中包含一个包含以下列的表：
Timestamp, meterID, rawValue
我正在尝试使用查询和谷歌图表绘制用水率图表，问题是我需要根据每 15 到 30 分钟更新一次的原始水表值计算用水率。
我想运行一个返回特定水表值的查询。
MeterID, Timestamp, (rawValue-previousRawValue)/(timestamp difference in seconds)
非常感谢任何帮助。
【问题讨论】：
标签： sql-server coalesce rate
编辑 1：我修改了索引定义以消除查找运算符 => 更少的逻辑读取。
编辑 2： 我添加了基于 quirky update 方法的第二个解决方案。请阅读 Jeff Moden 撰写的这篇文章 (Solving the Running Total and Ordinal Rank Problems)。
可以使用 SQL Server 2005/2008 测试第一个解决方案：
--Create test table
CREATE TABLE dbo.MeterValues
(
ID INT IDENTITY(1,1) PRIMARY KEY
,[Timestamp] DATETIME NOT NULL
,MeterID INT NOT NULL
,RawValue INT NOT NULL
);
CREATE UNIQUE INDEX IUN_MeterValues_MeterID_Timestamp
--SQL Server 2008
ON dbo.MeterValues (MeterID, [Timestamp])
INCLUDE (RawValue)
--SQL Server 2005
--ON dbo.MeterValues (MeterID, [Timestamp],RawValue)
--DROP INDEX dbo.MeterValues.IUN_MeterValues_MeterID_Timestamp
--Insert some values
INSERT dbo.MeterValues ([Timestamp], MeterID, RawValue)
SELECT '2011-01-01T00:00:00', 1, 100
UNION ALL
SELECT '2011-01-01T00:00:15', 1, 105
UNION ALL
SELECT '2011-01-01T00:00:30', 1, 102
UNION ALL
SELECT '2011-01-01T00:00:45', 1, 108
UNION ALL
SELECT '2011-01-01T00:01:00', 1, 109
UNION ALL
SELECT '2011-01-01T00:00:00', 2, 1000
UNION ALL
SELECT '2011-01-01T00:00:15', 2, 900
UNION ALL
SELECT '2011-01-01T00:00:30', 2, 1105
UNION ALL
SELECT '2011-01-01T00:00:45', 2, 1050
UNION ALL
SELECT '2011-01-01T00:01:00', 2, 910;
--Check test data
SELECT *
FROM dbo.MeterValues mv
ORDER BY mv.MeterID, mv.ID DESC;
--Solution
WITH ValuesWithRowNumber
AS
(
SELECT mv.MeterID
,mv.RawValue
,mv.[Timestamp]
,ROW_NUMBER() OVER(PARTITION BY mv.MeterID ORDER BY mv.[Timestamp] ASC) RowNum
FROM dbo.MeterValues mv
)
SELECT crt.MeterID
,crt.[Timestamp] AS CrtTimestamp
,prev.[Timestamp] AS PrevTimestamp
,crt.RawValue AS CrtRawValue
,prev.RawValue AS PrevRawValue
,(crt.RawValue - prev.RawValue)*1.00/DATEDIFF(SECOND, prev.[Timestamp], crt.[Timestamp]) Diff
,STR((crt.RawValue - prev.RawValue)*1.00/DATEDIFF(SECOND, prev.[Timestamp], crt.[Timestamp])*100, 10, 2)+'%' [Percent]
FROM ValuesWithRowNumber crt --crt=current
LEFT JOIN ValuesWithRowNumber prev ON crt.MeterID = prev.MeterID --prev=previous
AND crt.RowNum - 1 = prev.RowNum
ORDER BY crt.MeterID, crt.[Timestamp] DESC;
--By, by
DROP TABLE dbo.MeterValues;
结果：
MeterID CrtTimestamp PrevTimestamp CrtRawValue PrevRawValue Diff Percent
----------- ----------------------- ----------------------- ----------- ------------ --------------------------------------- -----------
1 2011-01-01 00:01:00.000 2011-01-01 00:00:45.000 109 108 0.0666666666666 6.67%
1 2011-01-01 00:00:45.000 2011-01-01 00:00:30.000 108 102 0.4000000000000 40.00%
1 2011-01-01 00:00:30.000 2011-01-01 00:00:15.000 102 105 -0.2000000000000 -20.00%
1 2011-01-01 00:00:15.000 2011-01-01 00:00:00.000 105 100 0.3333333333333 33.33%
1 2011-01-01 00:00:00.000 NULL 100 NULL NULL NULL
2 2011-01-01 00:01:00.000 2011-01-01 00:00:45.000 910 1050 -9.3333333333333 -933.33%
2 2011-01-01 00:00:45.000 2011-01-01 00:00:30.000 1050 1105 -3.6666666666666 -366.67%
2 2011-01-01 00:00:30.000 2011-01-01 00:00:15.000 1105 900 13.6666666666666 1366.67%
2 2011-01-01 00:00:15.000 2011-01-01 00:00:00.000 900 1000 -6.6666666666666 -666.67%
2 2011-01-01 00:00:00.000 NULL 1000 NULL NULL NULL
第二种解决方案可以/应该使用 SQL 2000/2005/2008（请阅读 Jeff Moden 文章中的“规则”部分）：
--Create test table
CREATE TABLE dbo.MeterValues
(
MeterID INT NOT NULL
,[Timestamp] DATETIME NOT NULL
,RawValue INT NOT NULL
,Diff NUMERIC(10,3) NULL
,PRIMARY KEY CLUSTERED(MeterID,[Timestamp])
);
--Insert some values
INSERT dbo.MeterValues ([Timestamp], MeterID, RawValue)
SELECT '2011-01-01T00:00:00', 1, 100
UNION ALL
SELECT '2011-01-01T00:00:15', 1, 105
UNION ALL
SELECT '2011-01-01T00:00:30', 1, 102
UNION ALL
SELECT '2011-01-01T00:00:45', 1, 108
UNION ALL
SELECT '2011-01-01T00:01:00', 1, 109
UNION ALL
SELECT '2011-01-01T00:00:00', 2, 1000
UNION ALL
SELECT '2011-01-01T00:00:15', 2, 900
UNION ALL
SELECT '2011-01-01T00:00:30', 2, 1105
UNION ALL
SELECT '2011-01-01T00:00:45', 2, 1050
UNION ALL
SELECT '2011-01-01T00:01:00', 2, 910;
--Check test data
SELECT *
FROM dbo.MeterValues mv
ORDER BY mv.MeterID, mv.[Timestamp];
DECLARE @OldRawValue INT
,@Diff NUMERIC(10,3)
,@OldMeterID INT
,@OldTimestamp DATETIME;
PRINT '*****Star*****'
--Calculations
UPDATE dbo.MeterValues WITH(TABLOCKX)
SET @Diff = CASE WHEN @OldMeterID = MeterID THEN (RawValue - @OldRawValue)*1.00/DATEDIFF(SECOND,@OldTimeStamp,[TimeStamp]) END
,Diff = @Diff
,@OldRawValue = RawValue
,@OldMeterID = MeterID
,@OldTimestamp = [Timestamp]
OPTION(MAXDOP 1);
--Results
SELECT *
FROM dbo.MeterValues mv
ORDER BY mv.MeterID, mv.[Timestamp];
PRINT '*****Stop*****'
--By, by
DROP TABLE dbo.MeterValues;
结果：
MeterID Timestamp RawValue Diff
----------- ----------------------- ----------- ---------------------------------------
1 2011-01-01 00:01:00.000 109 0.067
1 2011-01-01 00:00:45.000 108 0.400
1 2011-01-01 00:00:30.000 102 -0.200
1 2011-01-01 00:00:15.000 105 0.333
1 2011-01-01 00:00:00.000 100 NULL
2 2011-01-01 00:01:00.000 910 -9.333
2 2011-01-01 00:00:45.000 1050 -3.667
2 2011-01-01 00:00:30.000 1105 13.667
2 2011-01-01 00:00:15.000 900 -6.667
2 2011-01-01 00:00:00.000 1000 NULL
【讨论】：
Try this
Select a.metered,a.timestamp,
(a.rawValue-b.rawValue)/(a.timestamp-b.timestamp)
From meters A
Join (selec top 2 rawValue,Timestamp
From meters where metered = @meter
order by timestamp DESC) b
On b.timestamp <> a.timestamp and a.meterId=B.meterId
在中间查询的时间戳中添加了 DESC。这将导致返回最近的两个时间戳，然后 JOIN 会从 A 中“过滤掉”与当前行匹配的那个
【讨论】：
我对我的查询和@Bogdan 的查询进行了一些小改动，以使它们尽可能相似，然后进行比较。 Bogdan 修改后的查询在这篇文章的底部。
在同一个查询中堆叠在一起，根据 SQL Server 查询执行计划，我的是 53% 的查询成本，而博格丹的是 47%。
对于博格丹帖子中提供的数据集：
我每隔 15 秒为meterID 1 和 2 添加值，最多 5 分钟，总共 42 条记录，然后使用 SQL Server Profiler 重新运行查询。
我的查询在读取方面胜出，Bogdan 在 CPU 和持续时间方面仍然胜出。
CPU SCANS READS DURATION
--------------------------------------
Mine 47 22 313 249ms
Bogdan's 16 22 972 15ms
--------------------------------------
我做了一些假设，比如您的 MeterID 是一个 INT。根据需要进行更改。
我还假设，由于您要对特定仪表 ID 运行查询，因此它将作为参数传递给存储过程。
这应该适用于 SQL Server 2005 及更高版本。
我做了一些可能会分散实际解决方案的事情。核心逻辑确实在 WHILE 循环中。
CREATE PROCEDURE [dbo].[GetMeterResults]
@MeterID INT
AS
BEGIN
-- create a temp table to store results
CREATE TABLE #tempResults
(
MeterID INT,
[Timestamp] DATETIME,
Result FLOAT
)
DECLARE
@Timestamp DATETIME,
@RawValue INT,
@LastTimestamp DATETIME,
@LastRawValue INT,
@FirstRun BIT = 1
DECLARE cr CURSOR FAST_FORWARD FOR
SELECT
[Timestamp],
RawValue
FROM
YourTable
WHERE
MeterID = @MeterID
ORDER BY
[Timestamp]
OPEN cr
FETCH NEXT FROM cr INTO @Timestamp, @RawValue
WHILE (@@FETCH_STATUS = 0)
BEGIN
IF (@FirstRun = 1)
BEGIN -- the first run
SELECT @FirstRun = 0 -- flip the bit for all future runs
END
ELSE -- all subsequent runs after the first
BEGIN
INSERT INTO
#tempResults
SELECT
@MeterID,
@Timestamp,
(@RawValue - @LastRawValue) * 1.00 / DATEDIFF(s, @LastTimestamp, @Timestamp)
END
-- save the current values for comparison on the next run
SELECT
@LastTimestamp = @Timestamp,
@LastRawValue = @RawValue
FETCH NEXT FROM cr INTO @Timestamp, @RawValue
END
CLOSE CR
DEALLOCATE CR
-- return the result set
SELECT
*
FROM
#tempResults
-- clean up the temp table
DROP TABLE #tempResults
END
GO
Bogdan 修改后的查询，按 MeterID 过滤，与我上面的查询进行比较：
DECLARE @MeterID INT = 1;
WITH ValuesWithRowNumber
AS
(
SELECT mv.MeterID
,mv.RawValue
,mv.[Timestamp]
,ROW_NUMBER() OVER(PARTITION BY mv.MeterID ORDER BY mv.[Timestamp] ASC) RowNum
FROM dbo.MeterValues mv
WHERE mv.MeterID = @MeterID
)
SELECT crt.MeterID
,crt.[Timestamp] AS CrtTimestamp
,prev.[Timestamp] AS PrevTimestamp
,crt.RawValue AS CrtRawValue
,prev.RawValue AS PrevRawValue
,(crt.RawValue - prev.RawValue)*1.00/DATEDIFF(SECOND, prev.[Timestamp], crt.[Timestamp]) Diff
FROM ValuesWithRowNumber crt --crt=current
JOIN ValuesWithRowNumber prev ON crt.RowNum - 1 = prev.RowNum
ORDER BY crt.[Timestamp];
【讨论】：
FORWARD_ONLY FAST_FORWARD READ_ONLY ? FAST_FORWARD = FORWARD_ONLY + READ_ONLY
IUN_MeterValues_MeterID_Timestamp 索引，所以Bookmark Lookup 运算符将被淘汰。这意味着更少的逻辑读取。此外，我还添加了另一种基于“古怪更新”方法的解决方案。