强制索引假脱机

Question

我知道出于性能原因应该避免它，但我试图展示一个条件，它显示为演示如何确保它不会出现。

但是，我最终得到了缺少索引警告，但优化器选择不创建临时索引。

我正在使用的查询是

SELECT 
    z.a
FROM dbo.t5 AS z WITH(INDEX(0))
WHERE 
    EXISTS 
    (
        SELECT y.a 
        FROM dbo.t4 AS y
        WHERE y.a = z.a
    )
OPTION (MAXDOP 1);

表模式是：

CREATE TABLE dbo.t4
(
    a   integer NULL,
    b   varchar(1000) NULL,
    p   varchar(100) NULL
);

CREATE TABLE dbo.t5
(
    a   integer NULL,
    b   varchar(1000) NULL
);

CREATE UNIQUE CLUSTERED INDEX c1 
ON dbo.t5 (a);

两个表都有 10,000 行，您可以使用以下方法进行模拟：

UPDATE STATISTICS dbo.t4 
WITH 
    ROWCOUNT = 10000, 
    PAGECOUNT = 1000;

UPDATE STATISTICS dbo.t5 
WITH 
    ROWCOUNT = 10000,
    PAGECOUNT = 1000;

查询计划是：

<?xml version="1.0" encoding="utf-16"?>
<ShowPlanXML xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema" Version="1.2" Build="11.0.2218.0" xmlns="http://schemas.microsoft.com/sqlserver/2004/07/showplan">
  <BatchSequence>
    <Batch>
      <Statements>
        <StmtSimple StatementCompId="1" StatementEstRows="5532.16" StatementId="1" StatementOptmLevel="FULL" StatementOptmEarlyAbortReason="GoodEnoughPlanFound" StatementSubTreeCost="0.407384" StatementText="select a from t5  z WITH(INDEX(0))  where exists (select a from t4 where a=z.a )" StatementType="SELECT" QueryHash="0x1B882FCEA34AEAF4" QueryPlanHash="0x1B276DC04B718F7C" RetrievedFromCache="true">
          <StatementSetOptions ANSI_NULLS="true" ANSI_PADDING="true" ANSI_WARNINGS="true" ARITHABORT="true" CONCAT_NULL_YIELDS_NULL="true" NUMERIC_ROUNDABORT="false" QUOTED_IDENTIFIER="true" />
          <QueryPlan DegreeOfParallelism="1" MemoryGrant="2912" CachedPlanSize="32" CompileTime="10" CompileCPU="10" CompileMemory="296">
            <MissingIndexes>
              <MissingIndexGroup Impact="82.4536">
                <MissingIndex Database="[planoper]" Schema="[dbo]" Table="[t4]">
                  <ColumnGroup Usage="EQUALITY">
                    <Column Name="[a]" ColumnId="1" />
                  </ColumnGroup>
                </MissingIndex>
              </MissingIndexGroup>
            </MissingIndexes>
            <MemoryGrantInfo SerialRequiredMemory="1024" SerialDesiredMemory="2912" RequiredMemory="1024" DesiredMemory="2912" RequestedMemory="2912" GrantWaitTime="0" GrantedMemory="2912" MaxUsedMemory="896" />
            <OptimizerHardwareDependentProperties EstimatedAvailableMemoryGrant="104846" EstimatedPagesCached="11834" EstimatedAvailableDegreeOfParallelism="2" />
            <RelOp AvgRowSize="11" EstimateCPU="0.228447" EstimateIO="0" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="5532.16" LogicalOp="Left Semi Join" NodeId="0" Parallel="false" PhysicalOp="Hash Match" EstimatedTotalSubtreeCost="0.407384">
              <OutputList>
                <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t5]" Alias="[z]" Column="a" />
              </OutputList>
              <MemoryFractions Input="1" Output="1" />
              <RunTimeInformation>
                <RunTimeCountersPerThread Thread="0" ActualRows="10000" ActualEndOfScans="1" ActualExecutions="1" />
              </RunTimeInformation>
              <Hash>
                <DefinedValues />
                <HashKeysBuild>
                  <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t5]" Alias="[z]" Column="a" />
                </HashKeysBuild>
                <HashKeysProbe>
                  <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t4]" Column="a" />
                </HashKeysProbe>
                <ProbeResidual>
                  <ScalarOperator ScalarString="[planoper].[dbo].[t4].[a]=[planoper].[dbo].[t5].[a] as [z].[a]">
                    <Compare CompareOp="EQ">
                      <ScalarOperator>
                        <Identifier>
                          <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t4]" Column="a" />
                        </Identifier>
                      </ScalarOperator>
                      <ScalarOperator>
                        <Identifier>
                          <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t5]" Alias="[z]" Column="a" />
                        </Identifier>
                      </ScalarOperator>
                    </Compare>
                  </ScalarOperator>
                </ProbeResidual>
                <RelOp AvgRowSize="11" EstimateCPU="0.0110785" EstimateIO="0.0565368" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="10000" LogicalOp="Clustered Index Scan" NodeId="1" Parallel="false" PhysicalOp="Clustered Index Scan" EstimatedTotalSubtreeCost="0.0676153" TableCardinality="10000">
                  <OutputList>
                    <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t5]" Alias="[z]" Column="a" />
                  </OutputList>
                  <RunTimeInformation>
                    <RunTimeCountersPerThread Thread="0" ActualRows="10000" ActualEndOfScans="1" ActualExecutions="1" />
                  </RunTimeInformation>
                  <IndexScan Ordered="false" ForcedIndex="true" ForceScan="false" NoExpandHint="false">
                    <DefinedValues>
                      <DefinedValue>
                        <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t5]" Alias="[z]" Column="a" />
                      </DefinedValue>
                    </DefinedValues>
                    <Object Database="[planoper]" Schema="[dbo]" Table="[t5]" Index="[c1]" Alias="[z]" IndexKind="Clustered" />
                  </IndexScan>
                </RelOp>
                <RelOp AvgRowSize="11" EstimateCPU="0.011157" EstimateIO="0.100162" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="10000" LogicalOp="Table Scan" NodeId="2" Parallel="false" PhysicalOp="Table Scan" EstimatedTotalSubtreeCost="0.111319" TableCardinality="10000">
                  <OutputList>
                    <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t4]" Column="a" />
                  </OutputList>
                  <RunTimeInformation>
                    <RunTimeCountersPerThread Thread="0" ActualRows="10000" ActualEndOfScans="1" ActualExecutions="1" />
                  </RunTimeInformation>
                  <TableScan Ordered="false" ForcedIndex="false" ForceScan="false" NoExpandHint="false">
                    <DefinedValues>
                      <DefinedValue>
                        <ColumnReference Database="[planoper]" Schema="[dbo]" Table="[t4]" Column="a" />
                      </DefinedValue>
                    </DefinedValues>
                    <Object Database="[planoper]" Schema="[dbo]" Table="[t4]" IndexKind="Heap" />
                  </TableScan>
                </RelOp>
              </Hash>
            </RelOp>
          </QueryPlan>
        </StmtSimple>
      </Statements>
    </Batch>
  </BatchSequence>
</ShowPlanXML>

它甚至告诉我创建这个索引：

USE [planoper];
GO
CREATE NONCLUSTERED INDEX [<Name of Missing Index, sysname,>]
ON [dbo].[t4] ([a]);

Answer 1

让索引假脱机自然出现的一种方法是使用稍微不同的语法来表达需求：

SELECT DISTINCT 
    z.a
FROM dbo.t5 AS z
JOIN dbo.t4 AS y ON
    y.a >= z.a AND y.a <= z.a
OPTION (LOOP JOIN, MAXDOP 1, FORCE ORDER);

这会产生一个执行计划，如：

将等式重写为一对等价的不等式鼓励使用索引假脱机，尽管假脱机谓词并不完全是您所追求的，但语义最终是相同的。

另一种方式是滥用TOP：

SELECT 
    z.a
FROM dbo.t5 AS z
WHERE 
    EXISTS 
    (
        SELECT TOP ((SELECT 100)) PERCENT y.a 
        FROM dbo.t4 AS y
        WHERE y.a = z.a
    );

使用原始语法没有简单的方法来引入所需的索引假脱机；然而，这并不是说这是不可能的。由于您只需要在演示中使用它，并且不会在生产系统附近的任何地方使用它，我将向您展示另一种方式：

SELECT 
    z.a
FROM dbo.t5 AS z WITH(INDEX(0))
WHERE 
    EXISTS 
    (
        SELECT y.a 
        FROM dbo.t4 AS y
        WHERE y.a = z.a
    )
OPTION (MAXDOP 1, LOOP JOIN, QUERYTRACEON 9114);

执行计划是：

索引假脱机谓词是需要的：

SELECT DISTINCT 
    z.a
FROM dbo.t5 AS z
JOIN dbo.t4 AS y ON
    y.a >= z.a AND y.a <= z.a
OPTION (LOOP JOIN, MAXDOP 1, FORCE ORDER);

您将无法在USE PLAN提示中使用此计划，因为优化器通常不会考虑它。

进一步阅读：

• 嵌套循环连接和性能假脱机
• CROSS APPLY 产生外连接