[2019.03.15] FI NuGet.Jobs DEV into MASTER

src/Stats.Warehouse/Programmability/Stored Procedures/dbo.DownloadReportLast6Weeks.sql

@@ -4,54 +4,22 @@ AS
 BEGIN
 	SET NOCOUNT ON;
 
+	DECLARE @6WeeksAgo DATE = CAST(DATEADD(WEEK, -6, @ReportGenerationTime) AS DATE)
 	DECLARE @MinDate DATE
-	DECLARE @MinWeekOfYear INT
-	DECLARE @MinYear INT
 
-	SELECT	@MinWeekOfYear = [WeekOfYear],
-			@MinYear = [Year]
-	FROM	[dbo].[Dimension_Date] (NOLOCK)
-	WHERE	[Date] = CAST(DATEADD(day, -42, @ReportGenerationTime) AS DATE)
-
-	SELECT	@MinDate = MIN([Date])
-	FROM	[dbo].[Dimension_Date] (NOLOCK)
-	WHERE	[WeekOfYear] = @MinWeekOfYear
-		AND	[Year] = @MinYear
+	SELECT @MinDate = MIN(d1.Date)
+	FROM [dbo].View_Fixed_Week_Dimension_Date d1
+	JOIN [dbo].View_Fixed_Week_Dimension_Date d2 ON d2.[Date] = @6WeeksAgo AND d2.[WeekOfYear] = d1.[WeekOfYear] AND d2.[Year] = d1.[Year]
+	WHERE d1.[Date] >= DATEADD(WEEK, -1, @6WeeksAgo) AND d1.[Date] <= @6WeeksAgo
 
 	DECLARE @Cursor DATETIME = (SELECT ISNULL(MAX([Position]), @ReportGenerationTime) FROM [dbo].[Cursors] (NOLOCK) WHERE [Name] = 'GetDirtyPackageId')
-	DECLARE @MaxDate DATE = DATEADD(DAY, 42, @MinDate);
+	DECLARE @MaxDate DATE = DATEADD(WEEK, 6, @MinDate);
 
-	WITH WeekLookup AS 
-	(
-		-- If we just take all the rows between @MinDate and @MaxDate from Dimension_Date table
-		-- we might end up the days from the week that contains 1st of January to have two
-		-- different week numbers:
-		-- 12/30/2018 -> 53rd of 2018
-		-- 12/31/2018 -> 53rd of 2018
-		-- 1/1/2019 -> 1st of 2019
-		-- 1/2/2019 -> 1st of 2019
-		-- etc.
-		-- which results in the wrong grouping when group by [WeekOfYear] and [Year] is done:
-		-- the single week gets split into two portions, one for the previous year and another for
-		-- the new one with aggregations calculated separately for each of the portions.
-		-- This CTE works around the issue by making sure that all days of the week map to
-		-- the same [WeekOfYear] and [Year], specifically to that of the first day of that week.
-		SELECT d.[Id], dd.[WeekOfYear], dd.[Year]
-		FROM [dbo].[Dimension_Date] AS d WITH(NOLOCK)
-		CROSS APPLY
-		(
-			SELECT TOP(1) [WeekOfYear], [Year]
-			FROM [dbo].[Dimension_Date] AS d2 WITH(NOLOCK)
-			WHERE d2.[Date] <= d.[Date] AND d2.[DayOfWeek] = 1
-			ORDER BY d2.[Date] DESC
-		) AS dd
-		WHERE d.[Date] >= @MinDate AND d.[Date] < @MaxDate AND d.[Date] <= @Cursor
-	)
 	SELECT	D.[Year],
 			D.[WeekOfYear],
 			SUM(ISNULL(Facts.[DownloadCount], 0)) AS [Downloads]
 	FROM	[dbo].[Fact_Download] AS Facts (NOLOCK)
-	INNER JOIN WeekLookup AS D ON D.Id = Facts.Dimension_Date_Id
+	INNER JOIN [dbo].View_Fixed_Week_Dimension_Date AS D ON D.Id = Facts.Dimension_Date_Id AND D.[Date] >= @MinDate AND D.[Date] < @MaxDate
 	GROUP BY D.[Year], D.[WeekOfYear]
 	ORDER BY [Year], [WeekOfYear]
 END

src/Stats.Warehouse/Stats.Warehouse.sqlproj

@@ -138,6 +138,7 @@
     <Build Include="Programmability\Stored Procedures\dbo.CheckLogFileHasToolStatistics.sql" />
     <Build Include="Programmability\Stored Procedures\dbo.DownloadReportRecentCommunityPopularity.sql" />
     <Build Include="Programmability\Stored Procedures\dbo.DownloadReportRecentCommunityPopularityDetail.sql" />
+    <Build Include="Views\dbo.View_Fixed_Week_Dimension_Date.sql" />
   </ItemGroup>
   <ItemGroup>
     <None Include="StaticCodeAnalysis.SuppressMessages.xml" />

src/Stats.Warehouse/Views/dbo.View_Fixed_Week_Dimension_Date.sql

@@ -0,0 +1,27 @@
+CREATE VIEW [dbo].View_Fixed_Week_Dimension_Date
+	WITH SCHEMABINDING
+AS
+	-- [Dimension_Date] table have [WeekOfYear] and [Year] columns that are set to the
+	-- number of the week in a year and a year respectively.
+	-- This might lead to a situation around a new year day that within single week,
+	-- there are days that have different values of [WeekOfYear] and [Year]:
+	-- 12/30/2018 -> 53rd of 2018 [Sunday]
+	-- 12/31/2018 -> 53rd of 2018 [Monday]
+	-- 1/1/2019 -> 1st of 2019 [Tuesday]
+	-- 1/2/2019 -> 1st of 2019 [Wednesday]
+	-- etc.
+	-- which results in the wrong grouping when group by [WeekOfYear] and [Year] is done:
+	-- the single week gets split into two portions, one for the previous year and another for
+	-- the new one with aggregations calculated separately for each of the portions.
+	-- This view works around the issue by making sure that all days of the week map to
+	-- the same [WeekOfYear] and [Year], specifically to that of the first day of that week.
+	SELECT d.[Id], d.[Date], dd.WeekOfYear, dd.[Year]
+	FROM [dbo].Dimension_Date AS d WITH(NOLOCK)
+	CROSS APPLY
+	(
+		SELECT TOP(1) [WeekOfYear], [Year]
+		FROM [dbo].[Dimension_Date] AS d2 WITH(NOLOCK)
+		WHERE d2.[Date] IS NOT NULL AND d.[Date] IS NOT NULL AND d2.[DayOfWeek] IS NOT NULL
+			AND ISNULL(d2.[Date], '1980-01-01') <= ISNULL(d.[Date], '1980-01-01') AND ISNULL(d2.[DayOfWeek], 0) = 1
+		ORDER BY d2.[Date] DESC
+	) AS dd