sql-server - FileTable 分隔字符串拆分

Question

编辑 - 为了清楚起见，我会将目标句放在顶部。测试和我的问题是是否有办法在不使用临时表的情况下获得与临时表相同的性能。

我觉得这应该是一个简单的问题，但我被困住了。我正在 SQL2014 中试验 FileTables。我知道一些可行的替代方案，但目标是确定从文件表中提取文本子字符串的可行性。

该测试有 35,000 个文本文件，其中一行文本如下，每个文件平均有 100 个字节的非 unicode 文本。

Aaa|Bbb|Ccc|Ddd|Eee|Fff|Ggg

所需的输出是每个文件的一行，并将分隔字符串分成七列。

我找到了一个快速的字符串解析器函数，但与 varchar 列相比，在文件流上运行对性能有显着影响。

此查询需要 18 秒才能运行。我试图让从文件流到 varchar 的转换只执行一次，但我认为调用 UDF 可能会导致它发生在每一行（文件）。

Create View vAddresses As
Select file_type, Convert(Varchar(8000),file_stream) TextData /* Into #Temp */ From InputFiles Where file_type = 'adr'
Go
Select  --TextData,
        dbo.udf_StringSplit(TextData, 1, '|'), dbo.udf_StringSplit(TextData, 2, '|'), dbo.udf_StringSplit(TextData, 3, '|'),
        dbo.udf_StringSplit(TextData, 4, '|'), dbo.udf_StringSplit(TextData, 5, '|'), dbo.udf_StringSplit(TextData, 6, '|'),
        dbo.udf_StringSplit(TextData, 7, '|')--, TextData
    From vAddresses

我已经尝试将其作为视图、cte 和子查询。唯一似乎有帮助的是创建一个临时表。创建临时表需要 1 秒，查询需要 1 秒。因此，对于 35k 行，总查询时间为 2 秒，而总查询时间为 18 秒。

Drop Table #Temp
(Select file_type, Convert(Varchar(8000),file_stream) TextData Into #Temp From HumanaInputFiles Where file_type = 'adr')
Select  --TextData,
        dbo.udf_StringSplit(TextData, 1, '|'), dbo.udf_StringSplit(TextData, 2, '|'), dbo.udf_StringSplit(TextData, 3, '|'),
        dbo.udf_StringSplit(TextData, 4, '|'), dbo.udf_StringSplit(TextData, 5, '|'), dbo.udf_StringSplit(TextData, 6, '|'),
        dbo.udf_StringSplit(TextData, 7, '|')--, TextData
    From #Temp

我已经阅读了很多关于文件表和临时表与单个查询性能主题的文章和博客，但我似乎无法弄清楚。它可能与 sargable 或统计有关？非常感谢任何建议。

这是 UDF，我在 MSDN 博客/论坛上找到了它，它是迄今为止我发现的表现最好的。

ALTER FUNCTION [dbo].[udf_StringSplit](
 @TEXT      varchar(8000)
,@COLUMN    tinyint
,@SEPARATOR char(1)
)RETURNS varchar(8000)
AS
  BEGIN
       DECLARE @POS_START  int = 1
       DECLARE @POS_END    int = CHARINDEX(@SEPARATOR, @TEXT, @POS_START)

       WHILE (@COLUMN >1 AND @POS_END> 0)
         BEGIN
             SET @POS_START = @POS_END + 1
             SET @POS_END = CHARINDEX(@SEPARATOR, @TEXT, @POS_START)
             SET @COLUMN = @COLUMN - 1
         END 

       IF @COLUMN > 1  SET @POS_START = LEN(@TEXT) + 1
       IF @POS_END = 0 SET @POS_END = LEN(@TEXT) + 1 

       RETURN SUBSTRING (@TEXT, @POS_START, @POS_END - @POS_START)
  END

这是临时表的执行计划。

<?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="12.0.4100.1" xmlns="http://schemas.microsoft.com/sqlserver/2004/07/showplan">
  <BatchSequence>
    <Batch>
      <Statements>
        <StmtSimple StatementCompId="1" StatementEstRows="17486" StatementId="1" StatementOptmLevel="TRIVIAL" CardinalityEstimationModelVersion="120" StatementSubTreeCost="0.166487" StatementText="Select --TextData,&#xD;&#xA;       dbo.udf_StringSplit(TextData, 1, '|'), dbo.udf_StringSplit(TextData, 2, '|'), dbo.udf_StringSplit(TextData, 3, '|'),&#xD;&#xA;      dbo.udf_StringSplit(TextData, 4, '|'), dbo.udf_StringSplit(TextData, 5, '|'), dbo.udf_StringSplit(TextData, 6, '|'),&#xD;&#xA;      dbo.udf_StringSplit(TextData, 7, '|')--, TextData&#xD;&#xA; From #Temp" StatementType="SELECT" QueryHash="0xC4D6F0215D332F3D" QueryPlanHash="0xC50CFAF9494B5DBE" 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="0" NonParallelPlanReason="CouldNotGenerateValidParallelPlan" CachedPlanSize="24" CompileTime="1" CompileCPU="1" CompileMemory="168">
            <MemoryGrantInfo SerialRequiredMemory="0" SerialDesiredMemory="0" />
            <OptimizerHardwareDependentProperties EstimatedAvailableMemoryGrant="838735" EstimatedPagesCached="419367" EstimatedAvailableDegreeOfParallelism="4" />
            <RelOp AvgRowSize="28023" EstimateCPU="0.0017486" EstimateIO="0" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="17486" LogicalOp="Compute Scalar" NodeId="0" Parallel="false" PhysicalOp="Compute Scalar" EstimatedTotalSubtreeCost="0.166487">
              <OutputList>
                <ColumnReference Column="Expr1003" />
                <ColumnReference Column="Expr1004" />
                <ColumnReference Column="Expr1005" />
                <ColumnReference Column="Expr1006" />
                <ColumnReference Column="Expr1007" />
                <ColumnReference Column="Expr1008" />
                <ColumnReference Column="Expr1009" />
              </OutputList>
              <RunTimeInformation>
                <RunTimeCountersPerThread Thread="0" ActualRows="17486" ActualEndOfScans="1" ActualExecutions="1" />
              </RunTimeInformation>
              <ComputeScalar>
                <DefinedValues>
                  <DefinedValue>
                    <ColumnReference Column="Expr1003" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([tempdb].[dbo].[#Temp].[TextData],(1),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(1)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1004" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([tempdb].[dbo].[#Temp].[TextData],(2),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(2)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1005" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([tempdb].[dbo].[#Temp].[TextData],(3),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(3)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1006" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([tempdb].[dbo].[#Temp].[TextData],(4),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(4)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1007" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([tempdb].[dbo].[#Temp].[TextData],(5),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(5)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1008" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([tempdb].[dbo].[#Temp].[TextData],(6),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(6)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1009" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([tempdb].[dbo].[#Temp].[TextData],(7),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(7)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                </DefinedValues>
                <RelOp AvgRowSize="4011" EstimateCPU="0.0193131" EstimateIO="0.145426" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="17486" LogicalOp="Table Scan" NodeId="1" Parallel="false" PhysicalOp="Table Scan" EstimatedTotalSubtreeCost="0.164739" TableCardinality="17486">
                  <OutputList>
                    <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                  </OutputList>
                  <RunTimeInformation>
                    <RunTimeCountersPerThread Thread="0" ActualRows="17486" ActualEndOfScans="1" ActualExecutions="1" />
                  </RunTimeInformation>
                  <TableScan Ordered="false" ForcedIndex="false" ForceScan="false" NoExpandHint="false" Storage="RowStore">
                    <DefinedValues>
                      <DefinedValue>
                        <ColumnReference Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" Column="TextData" />
                      </DefinedValue>
                    </DefinedValues>
                    <Object Database="[tempdb]" Schema="[dbo]" Table="[#Temp]" IndexKind="Heap" Storage="RowStore" />
                  </TableScan>
                </RelOp>
              </ComputeScalar>
            </RelOp>
          </QueryPlan>
        </StmtSimple>
      </Statements>
    </Batch>
  </BatchSequence>
</ShowPlanXML>

这是视图的计划。

<?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="12.0.4100.1" xmlns="http://schemas.microsoft.com/sqlserver/2004/07/showplan">
  <BatchSequence>
    <Batch>
      <Statements>
        <StmtSimple StatementCompId="1" StatementEstRows="17486" StatementId="1" StatementOptmLevel="FULL" StatementOptmEarlyAbortReason="GoodEnoughPlanFound" CardinalityEstimationModelVersion="120" StatementSubTreeCost="0.905265" StatementText="Select    --TextData,&#xD;&#xA;       dbo.udf_StringSplit(TextData, 1, '|'), dbo.udf_StringSplit(TextData, 2, '|'), dbo.udf_StringSplit(TextData, 3, '|'),&#xD;&#xA;      dbo.udf_StringSplit(TextData, 4, '|'), dbo.udf_StringSplit(TextData, 5, '|'), dbo.udf_StringSplit(TextData, 6, '|'),&#xD;&#xA;      dbo.udf_StringSplit(TextData, 7, '|')--, TextData&#xD;&#xA; From vAddresses" StatementType="SELECT" QueryHash="0xB4F8A0B288802C4E" QueryPlanHash="0x28DA02D774B1AF53" 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="0" NonParallelPlanReason="CouldNotGenerateValidParallelPlan" CachedPlanSize="32" CompileTime="3" CompileCPU="3" CompileMemory="520">
            <Warnings>
              <PlanAffectingConvert ConvertIssue="Cardinality Estimate" Expression="CONVERT(varchar(8000),[DmProd01].[dbo].[HumanaInputFiles].[file_stream],0)" />
            </Warnings>
            <MemoryGrantInfo SerialRequiredMemory="0" SerialDesiredMemory="0" />
            <OptimizerHardwareDependentProperties EstimatedAvailableMemoryGrant="838735" EstimatedPagesCached="419367" EstimatedAvailableDegreeOfParallelism="4" />
            <RelOp AvgRowSize="28023" EstimateCPU="0.0017486" EstimateIO="0" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="17486" LogicalOp="Compute Scalar" NodeId="0" Parallel="false" PhysicalOp="Compute Scalar" EstimatedTotalSubtreeCost="0.905265">
              <OutputList>
                <ColumnReference Column="Expr1004" />
                <ColumnReference Column="Expr1005" />
                <ColumnReference Column="Expr1006" />
                <ColumnReference Column="Expr1007" />
                <ColumnReference Column="Expr1008" />
                <ColumnReference Column="Expr1009" />
                <ColumnReference Column="Expr1010" />
              </OutputList>
              <RunTimeInformation>
                <RunTimeCountersPerThread Thread="0" ActualRows="17486" ActualEndOfScans="1" ActualExecutions="1" />
              </RunTimeInformation>
              <ComputeScalar>
                <DefinedValues>
                  <DefinedValue>
                    <ColumnReference Column="Expr1004" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([Expr1011],(1),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Column="Expr1011" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(1)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1005" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([Expr1011],(2),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Column="Expr1011" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(2)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1006" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([Expr1011],(3),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Column="Expr1011" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(3)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1007" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([Expr1011],(4),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Column="Expr1011" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(4)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1008" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([Expr1011],(5),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Column="Expr1011" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(5)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1009" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([Expr1011],(6),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Column="Expr1011" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(6)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                  <DefinedValue>
                    <ColumnReference Column="Expr1010" />
                    <ScalarOperator ScalarString="[DmProd01].[dbo].[udf_StringSplit]([Expr1011],(7),'|')">
                      <UserDefinedFunction FunctionName="[DmProd01].[dbo].[udf_StringSplit]">
                        <ScalarOperator>
                          <Identifier>
                            <ColumnReference Column="Expr1011" />
                          </Identifier>
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="(7)" />
                        </ScalarOperator>
                        <ScalarOperator>
                          <Const ConstValue="'|'" />
                        </ScalarOperator>
                      </UserDefinedFunction>
                    </ScalarOperator>
                  </DefinedValue>
                </DefinedValues>
                <RelOp AvgRowSize="4019" EstimateCPU="0.0034972" EstimateIO="0" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="17486" LogicalOp="Compute Scalar" NodeId="1" Parallel="false" PhysicalOp="Compute Scalar" EstimatedTotalSubtreeCost="0.88673">
                  <OutputList>
                    <ColumnReference Column="Expr1011" />
                  </OutputList>
                  <ComputeScalar>
                    <DefinedValues>
                      <DefinedValue>
                        <ColumnReference Column="Expr1011" />
                        <ScalarOperator ScalarString="CONVERT(varchar(8000),[DmProd01].[dbo].[HumanaInputFiles].[file_stream],0)">
                          <Convert DataType="varchar" Length="8000" Style="0" Implicit="false">
                            <ScalarOperator>
                              <Identifier>
                                <ColumnReference Database="[DmProd01]" Schema="[dbo]" Table="[HumanaInputFiles]" Column="file_stream" />
                              </Identifier>
                            </ScalarOperator>
                          </Convert>
                        </ScalarOperator>
                      </DefinedValue>
                    </DefinedValues>
                    <RelOp AvgRowSize="4043" EstimateCPU="0.0386262" EstimateIO="0.844606" EstimateRebinds="0" EstimateRewinds="0" EstimatedExecutionMode="Row" EstimateRows="17486" LogicalOp="Table Scan" NodeId="2" Parallel="false" PhysicalOp="Table Scan" EstimatedTotalSubtreeCost="0.883233" TableCardinality="34972">
                      <OutputList>
                        <ColumnReference Database="[DmProd01]" Schema="[dbo]" Table="[HumanaInputFiles]" Column="file_stream" />
                      </OutputList>
                      <RunTimeInformation>
                        <RunTimeCountersPerThread Thread="0" ActualRows="17486" ActualEndOfScans="1" ActualExecutions="1" />
                      </RunTimeInformation>
                      <TableScan Ordered="false" ForcedIndex="false" ForceScan="false" NoExpandHint="false" Storage="RowStore">
                        <DefinedValues>
                          <DefinedValue>
                            <ColumnReference Database="[DmProd01]" Schema="[dbo]" Table="[HumanaInputFiles]" Column="file_stream" />
                          </DefinedValue>
                        </DefinedValues>
                        <Object Database="[DmProd01]" Schema="[dbo]" Table="[HumanaInputFiles]" IndexKind="Heap" Storage="RowStore" />
                        <Predicate>
                          <ScalarOperator ScalarString="[DmProd01].[dbo].[HumanaInputFiles].[file_type]=N'adr'">
                            <Compare CompareOp="EQ">
                              <ScalarOperator>
                                <Identifier>
                                  <ColumnReference Database="[DmProd01]" Schema="[dbo]" Table="[HumanaInputFiles]" Column="file_type" ComputedColumn="true" />
                                </Identifier>
                              </ScalarOperator>
                              <ScalarOperator>
                                <Const ConstValue="N'adr'" />
                              </ScalarOperator>
                            </Compare>
                          </ScalarOperator>
                        </Predicate>
                      </TableScan>
                    </RelOp>
                  </ComputeScalar>
                </RelOp>
              </ComputeScalar>
            </RelOp>
          </QueryPlan>
        </StmtSimple>
      </Statements>
    </Batch>
  </BatchSequence>
</ShowPlanXML>

编辑：对此进行了不同的搜索，发现答案是使用 top 和 order by。这使它缩短到 4 秒。似乎有点做作，仍然没有解释查看查询计划如何帮助解决这个问题，所以我自己不会回答这个问题，而是让它保持打开状态。

Answer 1

MSDN 中有建议不要在您的情况下使用 FileTable：

FileTable 功能建立在 SQL Server FILESTREAM 技术之上。

对于小对象，FileStream 的性能不好。Filestream 旨在处理大约 1MB 或更多的文件，但您只有 100 个字节（https://docs.microsoft.com/en-us/sql/relational-databases/blob/filestream-sql-server）：

何时使用 FILESTREAM 在 SQL Server 中，BLOB 可以是在表中存储数据的标准 varbinary(max) 数据，也可以是在文件系统中存储数据的 FILESTREAM varbinary(max) 对象。数据的大小和用途决定了您应该使用数据库存储还是文件系统存储。如果满足以下条件，则应考虑使用 FILESTREAM：

• 正在存储的对象平均大于 1 MB。
• 快速读取访问很重要。
• 您正在开发将中间层用于应用程序逻辑的应用程序。

对于较小的对象，将 varbinary(max) BLOB 存储在数据库中通常可以提供更好的流式传输性能。

您可以模拟持久列 - 创建另一个表并使用触发器填充它。在这种情况下，您可以从这两种情况下获得奖金

PS 你可以使用内联 TVF + union all 来实现@TT 的评论

Answer 2

编辑 - 更快的方法

而不是调用 Parse/Split 函数 7 次，也许这个 TVF 可能更有效。以下将在0.773 秒内处理35,000 条唯一记录。

例子

-- Create Some Sample/UNIQUE Data
Select N,TextData =concat(N,TextData )
Into #Temp
From  (values ('Aaa|Bbb|Ccc|Ddd|Eee|Fff|Ggg') ) A (TextData )
Cross Apply (Select Top 35000 N=Row_Number() Over (Order By (Select NULL)) From master..spt_values n1,master..spt_values n2) B

Select B.*
 From  #Temp A
 Cross Apply (
                Select Pos1=max(case when RetSeq=1 then RetVal end)
                      ,Pos2=max(case when RetSeq=2 then RetVal end)
                      ,Pos3=max(case when RetSeq=3 then RetVal end)
                      ,Pos4=max(case when RetSeq=4 then RetVal end)
                      ,Pos5=max(case when RetSeq=5 then RetVal end)
                      ,Pos6=max(case when RetSeq=6 then RetVal end)
                      ,Pos7=max(case when RetSeq=7 then RetVal end)
                 From [dbo].[udf-Str-Parse-8K](A.TextData,'|') B1
             ) B

有兴趣的UDF

CREATE FUNCTION [dbo].[udf-Str-Parse-8K] (@String varchar(max),@Delimiter varchar(25))
Returns Table 
As
Return (  
    with   cte1(N)   As (Select 1 From (Values(1),(1),(1),(1),(1),(1),(1),(1),(1),(1)) N(N)),
           cte2(N)   As (Select Top (IsNull(DataLength(@String),0)) Row_Number() over (Order By (Select NULL)) From (Select N=1 From cte1 a,cte1 b,cte1 c,cte1 d) A ),
           cte3(N)   As (Select 1 Union All Select t.N+DataLength(@Delimiter) From cte2 t Where Substring(@String,t.N,DataLength(@Delimiter)) = @Delimiter),
           cte4(N,L) As (Select S.N,IsNull(NullIf(CharIndex(@Delimiter,@String,s.N),0)-S.N,8000) From cte3 S)

    Select RetSeq = Row_Number() over (Order By A.N)
          ,RetVal = LTrim(RTrim(Substring(@String, A.N, A.L)))
    From   cte4 A
);
--Orginal Source http://www.sqlservercentral.com/articles/Tally+Table/72993/

只是为了帮助可视化

作为独立的 TVF 包括可以应用于条件聚合的项目序列。

Select * from [dbo].[udf-Str-Parse-8K]('Aaa|Bbb|Ccc|Ddd|Eee|Fff|Ggg','|')

退货

RetSeq  RetVal
1       Aaa
2       Bbb
3       Ccc
4       Ddd
5       Eee
6       Fff
7       Ggg

Answer 3

由于您有固定/预定数量的列要拆分，因此您实际上根本不需要使用字符串拆分功能。

以下内容能够在相当贫乏的开发服务器上在约 2 秒内咀嚼 100​​K 行。

注意...此解决方案结合了Adam Machanic 的 MakeParallel 函数来强制执行并行执行计划。

IF OBJECT_ID('tempdb..#TestData', 'U') IS NOT NULL 
DROP TABLE #TestData;

WITH 
    cte_n1 (n) AS (SELECT 1 FROM (VALUES (1),(1),(1),(1),(1),(1),(1),(1),(1),(1)) n (n)), 
    cte_n2 (n) AS (SELECT 1 FROM cte_n1 a CROSS JOIN cte_n1 b),
    cte_n3 (n) AS (SELECT 1 FROM cte_n2 a CROSS JOIN cte_n2 b),
    cte_Tally (n) AS (
        SELECT TOP 100000
            ROW_NUMBER() OVER (ORDER BY (SELECT NULL))
        FROM
            cte_n3 a CROSS JOIN cte_n3 b
        )
SELECT 
    ID = ISNULL(CAST(t.n AS INT), 0),
    FilePath = CAST(fp.FilePath AS VARCHAR(1000))
    INTO #TestData
FROM
    cte_Tally t
    CROSS APPLY ( VALUES (CONCAT(
                                ABS(CHECKSUM(NEWID())) % 9999 + 1000, '|',
                                ABS(CHECKSUM(NEWID())) % 9999 + 1000, '|',
                                ABS(CHECKSUM(NEWID())) % 9999 + 1000, '|',
                                ABS(CHECKSUM(NEWID())) % 9999 + 1000, '|',
                                ABS(CHECKSUM(NEWID())) % 9999 + 1000, '|',
                                ABS(CHECKSUM(NEWID())) % 9999 + 1000, '|',
                                ABS(CHECKSUM(NEWID())) % 9999 + 1000
                                )
                            ) ) fp (FilePath);

ALTER TABLE #TestData ADD PRIMARY KEY CLUSTERED (ID);

--=============================================================================
DECLARE     -- Dump values into variables to eliminate display rendering from execution time.
    @id INT,
    @Col_1 VARCHAR(5),
    @Col_2 VARCHAR(5),
    @Col_3 VARCHAR(5),
    @Col_4 VARCHAR(5),
    @Col_5 VARCHAR(5),
    @Col_6 VARCHAR(5),
    @Col_7 VARCHAR(5);

SELECT 
    @ID = td.ID,
    @Col_1 = SUBSTRING(td.FilePath, 1, ABS(d1.DelimLocation - 1)),
    @Col_2 = SUBSTRING(td.FilePath, d1.DelimLocation + 1, ABS(d2.DelimLocation - d1.DelimLocation - 1)),
    @Col_3 = SUBSTRING(td.FilePath, d2.DelimLocation + 1, ABS(d3.DelimLocation - d2.DelimLocation - 1)),
    @Col_4 = SUBSTRING(td.FilePath, d3.DelimLocation + 1, ABS(d4.DelimLocation - d3.DelimLocation - 1)),
    @Col_5 = SUBSTRING(td.FilePath, d4.DelimLocation + 1, ABS(d5.DelimLocation - d4.DelimLocation - 1)),
    @Col_6 = SUBSTRING(td.FilePath, d5.DelimLocation + 1, ABS(d6.DelimLocation - d5.DelimLocation - 1)),
    @Col_7 = SUBSTRING(td.FilePath, d6.DelimLocation + 1, 1000)
FROM
    #TestData td
    CROSS APPLY ( VALUES (LEN(td.FilePath) - LEN(REPLACE(td.FilePath, '|', ''))) ) dc (DelimiterCount)
    CROSS APPLY ( VALUES (IIF(dc.DelimiterCount < 1, 1000,  CHARINDEX('|', td.FilePath, 1))) ) d1 (DelimLocation)
    CROSS APPLY ( VALUES (IIF(dc.DelimiterCount < 2, 1000,  CHARINDEX('|', td.FilePath, d1.DelimLocation + 1))) ) d2 (DelimLocation)
    CROSS APPLY ( VALUES (IIF(dc.DelimiterCount < 3, 1000,  CHARINDEX('|', td.FilePath, d2.DelimLocation + 1))) ) d3 (DelimLocation)
    CROSS APPLY ( VALUES (IIF(dc.DelimiterCount < 4, 1000,  CHARINDEX('|', td.FilePath, d3.DelimLocation + 1))) ) d4 (DelimLocation)
    CROSS APPLY ( VALUES (IIF(dc.DelimiterCount < 5, 1000,  CHARINDEX('|', td.FilePath, d4.DelimLocation + 1))) ) d5 (DelimLocation)
    CROSS APPLY ( VALUES (IIF(dc.DelimiterCount < 6, 1000,  CHARINDEX('|', td.FilePath, d5.DelimLocation + 1))) ) d6 (DelimLocation)
    CROSS APPLY dbo.MakeParallel() mp;  -- Forces a parallel execution plan.
                                        -- http://dataeducation.com/next-level-parallel-plan-forcing-an-alternative-to-8649/

当然，MakeParallel 函数也可以与 Splitter 函数结合使用。在这种情况下，使用Jeff Moden 的 DelimitedSplit8K 函数。

--=============================================================================
DECLARE     -- Dump values into variables to eliminate display rendering from execution time.
    @id INT,
    @Col_1 VARCHAR(5),
    @Col_2 VARCHAR(5),
    @Col_3 VARCHAR(5),
    @Col_4 VARCHAR(5),
    @Col_5 VARCHAR(5),
    @Col_6 VARCHAR(5),
    @Col_7 VARCHAR(5);

SELECT 
    @ID = td.ID,
    @Col_1 = MAX(CASE WHEN sc.ItemNumber = 1 THEN sc.Item END),
    @Col_2 = MAX(CASE WHEN sc.ItemNumber = 2 THEN sc.Item END),
    @Col_3 = MAX(CASE WHEN sc.ItemNumber = 3 THEN sc.Item END),
    @Col_4 = MAX(CASE WHEN sc.ItemNumber = 4 THEN sc.Item END),
    @Col_5 = MAX(CASE WHEN sc.ItemNumber = 5 THEN sc.Item END),
    @Col_6 = MAX(CASE WHEN sc.ItemNumber = 6 THEN sc.Item END),
    @Col_7 = MAX(CASE WHEN sc.ItemNumber = 7 THEN sc.Item END)
FROM
    #TestData td
    CROSS APPLY dbo.DelimitedSplit8K(td.FilePath, '|') sc
    CROSS APPLY dbo.MakeParallel() mp       -- Forces a parallel execution plan.
                                            -- http://dataeducation.com/next-level-parallel-plan-forcing-an-alternative-to-8649/
GROUP BY
    td.ID;

HTH，杰森

Answer 4

我已经玩了一点，除了编写 CLR 程序集（不是非常快）之外，最快的解决方案似乎是以下某种类型的变体：

CREATE FUNCTION [UDF_Split] (
    @InputStr NVARCHAR(Max),
    @Delimiter NVARCHAR(255)
)
RETURNS TABLE
AS RETURN(

   WITH lv0 AS (SELECT 0 g UNION ALL SELECT 0)
    ,lv1 AS (SELECT 0 g FROM lv0 a CROSS JOIN lv0 b) -- 4
    ,lv2 AS (SELECT 0 g FROM lv1 a CROSS JOIN lv1 b) -- 16
    ,lv3 AS (SELECT 0 g FROM lv2 a CROSS JOIN lv2 b) -- 256
    ,lv4 AS (SELECT 0 g FROM lv3 a CROSS JOIN lv3 b) -- 65,536
    ,lv5 AS (SELECT 0 g FROM lv4 a CROSS JOIN lv4 b) -- 4,294,967,296
    ,Tally (n) AS (SELECT ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) FROM lv5)

 SELECT SUBSTRING(@InputStr, N, CHARINDEX(@Delimiter, @InputStr + @Delimiter, N) - N) AS TextLine
 FROM   Tally
 WHERE  N BETWEEN 1 AND LEN(@InputStr) + LEN(@InputStr)
        AND SUBSTRING(@Delimiter + @InputStr, N, LEN(@Delimiter)) = @Delimiter);

不过，从 MS SQL 2016 开始，我们得到了自己的拆分函数： Sql Server 2016 STRING_SPLIT (string, separator)

Answer 5

[1] 我已经多次使用以下解决方案，即 (1.1) 将源字符串转换为 XML，然后 (1.2) 每次都使用以下value方法从 XMl 中提取：

USE tempdb
GO
IF OBJECT_ID('dbo.SourceTable') IS NOT NULL
    DROP TABLE dbo.SourceTable
GO
CREATE TABLE dbo.SourceTable (
    ID      INT IDENTITY PRIMARY KEY,
    Col1    VARCHAR(100) NOT NULL
);
INSERT  dbo.SourceTable (Col1) VALUES ('Aaa|Bbb|Ccc|Ddd|Eee|Fff|Ggg')
INSERT  dbo.SourceTable (Col1) VALUES ('hhh|iii|JJJ|kkk')

SELECT  b.ID, c.XmlCol.value('.', 'VARCHAR(100)') AS ItemVal--, ROW_NUMBER() OVER(PARTITION BY b.ID ORDER BY c.XmlCol) AS RowNum
FROM (
    SELECT  a.ID, CONVERT(XML, '<root><i>' + REPLACE(a.Col1, '|', '</i><i>') + '</i></root>') AS Col1AsXML
    FROM    dbo.SourceTable a
) b OUTER APPLY b.Col1AsXML.nodes('root/i') c(XmlCol)
--OPTION(FORCE ORDER)

[2] 如果您提供 DDL 和 DML 语句，我会进行一些性能测试。

[3] 如果您能提供以下详细信息，将会有所帮助

1. 每个字符串有最大数量的项目吗？
2. 如果所有项目的长度相同（例如 3 个字符）？