Export (0) Print
Expand All

Implementing the OR Operator in Natively Compiled Stored Procedures

SQL Server 2014

OR operators are not supported in query predicates in natively compiled stored procedures. Because NOT operators are also not supported in query predicates in natively compiled stored procedures, the effects of OR operators cannot be simulated through the use of equivalent logical operators alone. However, the effects of an OR operator may be simulated with memory-optimized table variables.

If you have an OR operator in a WHERE clause, you can use the following approach to simulate its behavior:

  1. Create a memory-optimized table variable with the appropriate schema. This requires a predefined memory-optimized table type.

  2. Starting with the top level OR operator, separate the WHERE clause into two parts according to the predicates joined by the OR operator. If you have more than one OR operator in a WHERE clause, you may need to do this more than once. Repeat this step until no OR operators remain. For example, if you have the following predicate:

    pred1 OR (pred2 AND (pred3 OR pred4)) OR (pred5 AND pred6)
    

    After this step, you should have the following predicates:

    pred1
    pred5 AND pred6
    pred2 AND pred3
    pred2 AND pred4
    
  3. Execute a query with each of the two parts found in Step 2 as the predicate. Insert the result for each query into the memory-optimized table variable created in Step 1.

  4. If necessary, remove duplicates from the memory-optimized table variable.

  5. Use the content of the memory-optimized table variable as the result from the query.

The following sample uses tables from the AdventureWorks2012 database that were updated for In-Memory OLTP. To download the files for this sample, goto AdventureWorks Databases – 2012, 2008R2 and 2008. To apply In-Memory OLTP code sample to AdventureWorks2012, go to SQL Server 2014 In-Memory OLTP Sample.

Add the following stored procedure to the database. We will convert this stored procedure to use native compilation.

CREATE PROCEDURE Sales.usp_fuzzySearchSalesOrderDetail_ondisk
  @SalesOrderId int = 0, @SalesOrderDetailId int = 0, 
  @CarrierTrackingNumber nvarchar(25) = N'', @ProductId int = 0, 
  @minUnitPrice money = 0, @maxUnitPrice money = 0
AS BEGIN
  SELECT SalesOrderId, SalesOrderDetailId, ModifiedDate
  FROM Sales.SalesOrderDetail_ondisk s
  WHERE  s.SalesOrderId = @SalesOrderId
      OR s.SalesOrderDetailId = @SalesOrderDetailId
      OR s.CarrierTrackingNumber = @CarrierTrackingNumber
      OR s.ProductID = @ProductId
      OR (s.UnitPrice > @minUnitPrice AND s.UnitPrice < @maxUnitPrice)
END
GO

After conversion, the table and stored procedure schema is as follows,

CREATE TYPE Sales.fuzzySearchSalesOrderDetailType AS TABLE
(
  SalesOrderId int not null,
  SalesOrderDetailId int not null,
  ModifiedDate datetime2(7) not null
  INDEX ix_fuzzySearchSalesOrderDetailType NONCLUSTERED (SalesOrderId, SalesOrderDetailId)
) WITH (MEMORY_OPTIMIZED = ON)
GO

CREATE TYPE Sales.fuzzySearchSalesOrderDetailTempType AS TABLE
(
  SalesOrderId int not null,
  SalesOrderDetailId int not null,
  recordcount int not null
  INDEX ix_fuzzySearchSalesOrderDetailTempType NONCLUSTERED (SalesOrderId, SalesOrderDetailId)
) WITH (MEMORY_OPTIMIZED = ON)
GO

CREATE PROCEDURE Sales.usp_fuzzySearchSalesOrderDetail_inmem
  @SalesOrderId int = 0, @SalesOrderDetailId int = 0, 
  @CarrierTrackingNumber nvarchar(25) = N'', @ProductId int = 0, 
  @minUnitPrice money = 0, @maxUnitPrice money = 0
WITH NATIVE_COMPILATION, SCHEMABINDING, EXECUTE AS OWNER
AS BEGIN ATOMIC WITH (TRANSACTION ISOLATION LEVEL = SNAPSHOT, LANGUAGE = N'ENGLISH')
  
  DECLARE @retValue Sales.fuzzySearchSalesOrderDetailType

  INSERT INTO @retValue (SalesOrderId, SalesOrderDetailId, ModifiedDate)
  SELECT SalesOrderId, SalesOrderDetailId, ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  WHERE s.SalesOrderId = @SalesOrderId
  
  INSERT INTO @retValue (SalesOrderId, SalesOrderDetailId, ModifiedDate)
  SELECT SalesOrderId, SalesOrderDetailId, ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  WHERE s.SalesOrderDetailId = @SalesOrderDetailId
  
  INSERT INTO @retValue (SalesOrderId, SalesOrderDetailId, ModifiedDate)
  SELECT SalesOrderId, SalesOrderDetailId, ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  WHERE s.CarrierTrackingNumber COLLATE Latin1_General_BIN2 = @CarrierTrackingNumber COLLATE Latin1_General_BIN2 
  
  INSERT INTO @retValue (SalesOrderId, SalesOrderDetailId, ModifiedDate)
  SELECT SalesOrderId, SalesOrderDetailId, ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  WHERE s.ProductID = @ProductId
  
  INSERT INTO @retValue (SalesOrderId, SalesOrderDetailId, ModifiedDate)
  SELECT SalesOrderId, SalesOrderDetailId, ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  WHERE (s.UnitPrice > @minUnitPrice AND s.UnitPrice < @maxUnitPrice)

  -- After the above statements, there will be duplicates inside @retValue
  -- Delete the duplicates from @retValue
  DECLARE @duplicates Sales.fuzzySearchSalesOrderDetailTempType

  INSERT INTO @duplicates (SalesOrderId, SalesOrderDetailId, recordcount) 
  SELECT SalesOrderId, SalesOrderDetailId, COUNT(*) AS recordCount
  FROM @retValue
  GROUP BY SalesOrderId, SalesOrderDetailId
  
  -- Now we have one row per pair
  -- clear and rebuild the result set
  DELETE FROM @retValue

  INSERT INTO @retValue
  SELECT s.SalesOrderId, s.SalesOrderDetailId, s.ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  JOIN @duplicates d ON s.SalesOrderId = d.SalesOrderId AND s.SalesOrderDetailId = d.SalesOrderDetailId
  
  -- After this every pair of (SalesOrderId, SalesOrderDetailId) in @retValue should be unique.
  SELECT SalesorderId, SalesOrderDetailId, ModifiedDate FROM @retValue
END
GO

If you have an OR operator in a JOIN condition of a SELECT statement, you may use the following approach to simulate its behavior. If you have more than one OR operator in a JOIN condition or you have multiple JOIN condition with OR operators, you may need to do this more than once.

If you have OUTER JOIN conditions, you may combine this workaround with the workaround for OUTER JOIN conditions.

  1. Create a memory-optimized table variable with the appropriate schema. This requires a predefined memory-optimized table type.

  2. Separate the predicate in the JOIN condition into two parts according to the predicates joined by the OR operator. If you have multiple JOIN conditions, you may need to do this for each JOIN condition and then create a set of combinations of the resulting fragments. For example, if you have three JOIN conditions with one OR operator in each JOIN condition, you may have 2x2x2=8 predicates.

  3. For each predicate produced by Step 2, create a query that will insert its result into the memory-optimized table variable created in Step 1.

  4. If necessary, remove duplicates from the memory-optimized table variable.

  5. Use the content of the memory-optimized table variable as the result from the query.

The following sample uses tables from the AdventureWorks2012 database that were updated for In-Memory OLTP. To download the files for this sample, goto AdventureWorks Databases – 2012, 2008R2 and 2008. To apply In-Memory OLTP code sample to AdventureWorks2012, go to SQL Server 2014 In-Memory OLTP Sample.

Add the following stored procedure to the database. We will convert this stored procedure to use native compilation. This sample uses INNER JOIN conditions.

CREATE PROCEDURE Sales.usp_fuzzySearchSalesSpecialOffers_ondisk
  @SpecialOfferId int
AS BEGIN
  
  SELECT s.SalesOrderId, s.SalesOrderDetailId, s.SpecialOfferID, s.ModifiedDate
  FROM Sales.SalesOrderDetail_ondisk s
  JOIN Sales.SpecialOffer_onDisk offer 
    ON s.SpecialOfferID = offer.SpecialOfferID 
    OR s.ProductID IN (SELECT ProductId FROM Sales.SpecialOfferProduct sop WHERE sop.SpecialOfferID = @SpecialOfferId)
END

After conversion, the table and stored procedure schema is as follows,

CREATE TYPE Sales.fuzzySearchSalesSpecialOffers_Type AS TABLE
(
  SalesOrderId int not null,
  SalesOrderDetailId int not null,
  SpecialOfferId int not null,
  ModifiedDate datetime2(7) not null
  INDEX ix_fuzzySearchSalesSpecialOffers_Type NONCLUSTERED (SalesOrderId, SalesOrderDetailId)
) WITH (MEMORY_OPTIMIZED = ON)
GO

CREATE TYPE Sales.fuzzySearchSalesSpecialOffers_TempType AS TABLE
(
  SalesOrderId int not null,
  SalesOrderDetailId int not null,
  SpecialOfferId int not null,
  recordcount int null
  INDEX ix_fuzzySearchSalesSpecialOffers_TempType NONCLUSTERED (SalesOrderId, SalesOrderDetailId)
) WITH (MEMORY_OPTIMIZED = ON)
GO

CREATE PROCEDURE Sales.usp_fuzzySearchSalesSpecialOffers_inmem
  @SpecialOfferId int
WITH NATIVE_COMPILATION, SCHEMABINDING, EXECUTE AS OWNER
AS BEGIN ATOMIC WITH (TRANSACTION ISOLATION LEVEL = SNAPSHOT, LANGUAGE = N'ENGLISH')

  DECLARE @retValue Sales.FuzzySearchSalesSpecialOffers_Type
  
  -- Find all special offers matching the conditions

  INSERT INTO @retValue (SalesOrderId, SalesOrderDetailId, SpecialOfferid, ModifiedDate)
  SELECT s.SalesOrderId, s.SalesOrderDetailId, s.SpecialOfferID, s.ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  JOIN Sales.SpecialOffer_inmem offer 
    ON s.SpecialOfferID = offer.SpecialOfferID 
  
  
  INSERT INTO @retValue (SalesOrderId, SalesOrderDetailId, SpecialOfferid, ModifiedDate)
  SELECT s.SalesOrderId, s.SalesOrderDetailId, s.SpecialOfferID, s.ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  JOIN Sales.SpecialOfferProduct_inmem sop 
    ON sop.SpecialOfferId = @SpecialOfferId AND s.ProductID = sop.ProductId
  
  -- Now we need to remove the duplicates from @matchingSpecialOffers
  DECLARE @duplicates Sales.fuzzySearchSalesSpecialOffers_TempType

  INSERT INTO @duplicates (SalesOrderId, SalesOrderDetailId, SpecialOfferid, recordcount)
  SELECT SalesOrderId, SalesOrderDetailId, SpecialOfferId, COUNT(*) 
  FROM @retValue
  GROUP BY SalesOrderId, SalesOrderDetailId, SpecialOfferId
  
  -- now there should be no duplicates within @duplicate
  -- use @duplicate for join.
  SELECT s.SalesOrderId, s.SalesOrderDetailId, s.SpecialOfferID, s.ModifiedDate
  FROM Sales.SalesOrderDetail_inmem s
  JOIN @duplicates offer 
    ON    s.SalesOrderId = offer.SalesOrderId 
      AND s.SalesOrderDetailId = offer.SalesOrderDetailID 
      AND s.SpecialOfferId = offer.SpecialOfferId
END
GO

If you have more than one OR operator in the WHERE clause or JOIN condition, the number of queries you need to execute to simulate the behavior may increase exponentially. This may slow down query performance and may increase memory usage due to the need to use memory-optimized table variables.

Was this page helpful?
(1500 characters remaining)
Thank you for your feedback

Community Additions

ADD
Show:
© 2014 Microsoft