Actualmente estoy trabajando en mi camino a través delMicrosoft SQL Server 2008 - Desarrollo de bases de datos (Examen MCTS 70-433) proceso de dar un título. En uno de los capítulos anteriores sobreCombinando conjuntos de datosMe encontré con elEXCEPT (yINTERSECT) comandos. Un ejemplo muestra cómo usarEXCEPT para obtener todos los valores de una tabla que no tiene un valor relacionado en una segunda tabla como esta:

SELECT EmployeeKey FROM DimEmployee
EXCEPT
SELECT EmployeeKey FROM FactResellerSales

losEXCEPT el comando era nuevo para mí, pero con lo que sabía antes de hoy, todavía resolvería fácilmente el problema usando unLEFT JOIN y verificaIS NULL en la restricción de unión de la siguiente manera:

SELECT DISTINCT DimEmployee.EmployeeKey FROM DimEmployee
LEFT JOIN FactResellerSales ON FactResellerSales.EmployeeKey = DimEmployee.EmployeeKey
WHERE FactResellerSales.EmployeeKey IS NULL

Ahora comencé a preguntarme cuál de estos tiene el mejor rendimiento. Traté de estudiar los planes de ejecución de consultas, pero no soy demasiado bueno para leerlos, por lo que no me hizo más sabio. Para la consulta usandoEXCEPT, el plan se ve así:

|--Nested Loops(Left Anti Semi Join, OUTER REFERENCES:([AdventureWorksDW2008].[dbo].[DimEmployee].[EmployeeKey], [Expr1006]) WITH UNORDERED PREFETCH)
    |--Index Scan(OBJECT:([AdventureWorksDW2008].[dbo].[DimEmployee].[IX_DimEmployee_SalesTerritoryKey]))
    |--Top(TOP EXPRESSION:((1)))
        |--Index Seek(OBJECT:([AdventureWorksDW2008].[dbo].[FactResellerSales].[IX_FactResellerSales_EmployeeKey]), SEEK:([AdventureWorksDW2008].[dbo].[FactResellerSales].[EmployeeKey]=[AdventureWorksDW2008].[dbo].[DimEmployee].[EmployeeKey]) ORDERED FORWARD)

Y para el que usa elLEFT JOIN se parece a esto:

|--Stream Aggregate(GROUP BY:([AdventureWorksDW2008].[dbo].[DimEmployee].[EmployeeKey]))
    |--Filter(WHERE:([AdventureWorksDW2008].[dbo].[FactResellerSales].[EmployeeKey] IS NULL))
        |--Merge Join(Left Outer Join, MERGE:([AdventureWorksDW2008].[dbo].[DimEmployee].[EmployeeKey])=([AdventureWorksDW2008].[dbo].[FactResellerSales].[EmployeeKey]), RESIDUAL:([AdventureWorksDW2008].[dbo].[FactResellerSales].[EmployeeKey]=[AdventureWorksDW2008].[dbo].[DimEmployee].[EmployeeKey]))
            |--Clustered Index Scan(OBJECT:([AdventureWorksDW2008].[dbo].[DimEmployee].[PK_DimEmployee_EmployeeKey]), ORDERED FORWARD)
            |--Index Scan(OBJECT:([AdventureWorksDW2008].[dbo].[FactResellerSales].[IX_FactResellerSales_EmployeeKey]), ORDERED FORWARD)

Las tablas utilizadas en la consulta son delAdventureWorksDW2008 base de datos de ejemplo, así que a continuación también incluyo los scripts de creación para las dos tablas en caso de que sea necesario para responder la pregunta correctamente:

USE [AdventureWorksDW2008]
GO

/****** Object:  Table [dbo].[DimEmployee]    Script Date: 11/22/2010 20:30:20 ******/
SET ANSI_NULLS ON
GO

SET QUOTED_IDENTIFIER ON
GO

CREATE TABLE [dbo].[DimEmployee](
    [EmployeeKey] [int] IDENTITY(1,1) NOT NULL,
    [ParentEmployeeKey] [int] NULL,
    [EmployeeNationalIDAlternateKey] [nvarchar](15) NULL,
    [ParentEmployeeNationalIDAlternateKey] [nvarchar](15) NULL,
    [SalesTerritoryKey] [int] NULL,
    [FirstName] [nvarchar](50) NOT NULL,
    [LastName] [nvarchar](50) NOT NULL,
    [MiddleName] [nvarchar](50) NULL,
    [NameStyle] [bit] NOT NULL,
    [Title] [nvarchar](50) NULL,
    [HireDate] [date] NULL,
    [BirthDate] [date] NULL,
    [LoginID] [nvarchar](256) NULL,
    [EmailAddress] [nvarchar](50) NULL,
    [Phone] [nvarchar](25) NULL,
    [MaritalStatus] [nchar](1) NULL,
    [EmergencyContactName] [nvarchar](50) NULL,
    [EmergencyContactPhone] [nvarchar](25) NULL,
    [SalariedFlag] [bit] NULL,
    [Gender] [nchar](1) NULL,
    [PayFrequency] [tinyint] NULL,
    [BaseRate] [money] NULL,
    [VacationHours] [smallint] NULL,
    [SickLeaveHours] [smallint] NULL,
    [CurrentFlag] [bit] NOT NULL,
    [SalesPersonFlag] [bit] NOT NULL,
    [DepartmentName] [nvarchar](50) NULL,
    [StartDate] [date] NULL,
    [EndDate] [date] NULL,
    [Status] [nvarchar](50) NULL,
 CONSTRAINT [PK_DimEmployee_EmployeeKey] PRIMARY KEY CLUSTERED 
(
    [EmployeeKey] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]

GO

ALTER TABLE [dbo].[DimEmployee]  WITH CHECK ADD  CONSTRAINT [FK_DimEmployee_DimEmployee] FOREIGN KEY([ParentEmployeeKey])
REFERENCES [dbo].[DimEmployee] ([EmployeeKey])
GO

ALTER TABLE [dbo].[DimEmployee] CHECK CONSTRAINT [FK_DimEmployee_DimEmployee]
GO

ALTER TABLE [dbo].[DimEmployee]  WITH CHECK ADD  CONSTRAINT [FK_DimEmployee_DimSalesTerritory] FOREIGN KEY([SalesTerritoryKey])
REFERENCES [dbo].[DimSalesTerritory] ([SalesTerritoryKey])
GO

ALTER TABLE [dbo].[DimEmployee] CHECK CONSTRAINT [FK_DimEmployee_DimSalesTerritory]
GO

Y la segunda mesa:

USE [AdventureWorksDW2008]
GO

/****** Object:  Table [dbo].[FactResellerSales]    Script Date: 11/22/2010 20:30:47 ******/
SET ANSI_NULLS ON
GO

SET QUOTED_IDENTIFIER ON
GO

CREATE TABLE [dbo].[FactResellerSales](
    [ProductKey] [int] NOT NULL,
    [OrderDateKey] [int] NOT NULL,
    [DueDateKey] [int] NOT NULL,
    [ShipDateKey] [int] NOT NULL,
    [ResellerKey] [int] NOT NULL,
    [EmployeeKey] [int] NOT NULL,
    [PromotionKey] [int] NOT NULL,
    [CurrencyKey] [int] NOT NULL,
    [SalesTerritoryKey] [int] NOT NULL,
    [SalesOrderNumber] [nvarchar](20) NOT NULL,
    [SalesOrderLineNumber] [tinyint] NOT NULL,
    [RevisionNumber] [tinyint] NULL,
    [OrderQuantity] [smallint] NULL,
    [UnitPrice] [money] NULL,
    [ExtendedAmount] [money] NULL,
    [UnitPriceDiscountPct] [float] NULL,
    [DiscountAmount] [float] NULL,
    [ProductStandardCost] [money] NULL,
    [TotalProductCost] [money] NULL,
    [SalesAmount] [money] NULL,
    [TaxAmt] [money] NULL,
    [Freight] [money] NULL,
    [CarrierTrackingNumber] [nvarchar](25) NULL,
    [CustomerPONumber] [nvarchar](25) NULL,
 CONSTRAINT [PK_FactResellerSales_SalesOrderNumber_SalesOrderLineNumber] PRIMARY KEY CLUSTERED 
(
    [SalesOrderNumber] ASC,
    [SalesOrderLineNumber] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]

GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimCurrency] FOREIGN KEY([CurrencyKey])
REFERENCES [dbo].[DimCurrency] ([CurrencyKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimCurrency]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimDate] FOREIGN KEY([OrderDateKey])
REFERENCES [dbo].[DimDate] ([DateKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimDate]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimDate1] FOREIGN KEY([DueDateKey])
REFERENCES [dbo].[DimDate] ([DateKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimDate1]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimDate2] FOREIGN KEY([ShipDateKey])
REFERENCES [dbo].[DimDate] ([DateKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimDate2]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimEmployee] FOREIGN KEY([EmployeeKey])
REFERENCES [dbo].[DimEmployee] ([EmployeeKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimEmployee]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimProduct] FOREIGN KEY([ProductKey])
REFERENCES [dbo].[DimProduct] ([ProductKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimProduct]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimPromotion] FOREIGN KEY([PromotionKey])
REFERENCES [dbo].[DimPromotion] ([PromotionKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimPromotion]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimReseller] FOREIGN KEY([ResellerKey])
REFERENCES [dbo].[DimReseller] ([ResellerKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimReseller]
GO

ALTER TABLE [dbo].[FactResellerSales]  WITH CHECK ADD  CONSTRAINT [FK_FactResellerSales_DimSalesTerritory] FOREIGN KEY([SalesTerritoryKey])
REFERENCES [dbo].[DimSalesTerritory] ([SalesTerritoryKey])
GO

ALTER TABLE [dbo].[FactResellerSales] CHECK CONSTRAINT [FK_FactResellerSales_DimSalesTerritory]
GO

Esta pregunta está usando un ejemplo muy específico, pero también estoy interesado en información general sobre cuándo / si sería apropiado usarEXCEPT en vez de unaLEFT JOIN y verifiqueIS NULL.

También noté que si intentaba ejecutar la primera consulta conINTERCEPT, el equivalente en la segunda consulta sería usar un estándarJOIN ySELECT DISTINCT DimEmployee.EmployeeKey (y noWHERE cláusula en absoluto). Sin embargo, en este caso, el plan de ejecución fue exactamente el mismo en los dos casos.

Actualizar
Actualización menor a la segunda consulta (ver historial de revisiones), lo que resultó en un plan de consulta un poco más complejo para esa. Supongo que un plan de consulta más grande indica una consulta menos óptima, pero todavía me gustaría que esta pregunta sea respondida.