Sql server 如何使用SQL Server 2016中另一个表中的值替换长字符串中的模式子字符串

我在SQL Server 2016中遇到了替换函数的问题，我无法用任何模式替换。问题是：

我有一个大字符串，里面有两个查询，它们之间用@@@模式分隔，如下所示：

我有个问题@@你叫什么名字@@@你多大了@@@我需要 更多答案。@@您好@。如果您愿意，请回答

现在，@@@中的文本是一个查询。此查询的答案保存在另一个数据库表中

我想用表值替换文本中的查询。我们可以从表中运行游标来获取值，然后替换为特定位置。假设我们在表格（答案）中有如下查询值，我们可以按照查询顺序获取数据：

• 你叫什么名字
• 你多大年纪->29岁
• 你好吗->我很好

最后的答案是：我有一个问题Doe29我需要更多的答案。我很好。如果你想回答的话

现在，我发现在SQL Server 2016中替换两个查询之间的每个查询都有困难

---

是否有人可以帮助我了解在哪一个SQL Server函数上可以获得按模式替换的功能，并使用表中的原始值替换查询？

必须存在使用SQL游标的解决方案，但此问题不需要该解决方案。下面的解决方案使用两个（递归CTE）和一些。完整的解决方案似乎令人望而生畏，因此我将其背后的方法分为三个步骤

样本数据

declare @question table
(
    question nvarchar(200) primary key clustered
);

insert into @question (question) values
('I have a question. @@What is your name@@@@what is your age@@I need more answers.@@How are you@@.Answer if you want.');

declare @answers table
(
    key_ nvarchar(30) primary key clustered,
    value_ nvarchar(30)
);

insert into @answers (key_, value_) values
('What is your name', 'Doe'),
('what is your age', '29'),
('How are you', 'I am fine');

-- 1. Sample Data
SET NOCOUNT ON;
IF OBJECT_ID('dbo.question','U') IS NOT NULL DROP TABLE dbo.question;
IF OBJECT_ID('dbo.answers','U')  IS NOT NULL DROP TABLE dbo.answers;
CREATE TABLE dbo.question
(
  qID      int  PRIMARY KEY CLUSTERED,
  question nvarchar(200) NOT NULL
);
CREATE TABLE dbo.answers
(
  aID    int          NOT NULL INDEX nc__answers__aID,
  key_   nvarchar(30) NOT NULL,
  value_ nvarchar(30) NOT NULL
);
insert into dbo.question (qID,question) values
(1,'I have a question. @@What is your name@@@@what is your age@@I need more answers.'+
   '@@@@@@@@@@@@@@@@@@@How are you@@@@@@@@@@@@@@@@@@@@@@@@@@@.Answer if you want.'),
(2,'I have a question too.@@@@@@@@@@@@Your Name is?@@@how young are you?@@Answer This-'+
   '@@@@@@@@@@@You good?@@@@@@@@@@@@@@@How is Bill@@@@@@@@Answer if you must!.');
insert into dbo.answers (aID,key_, value_) values
(1,'What is your name','Doe'),(1,'what is your age','29'),(1,'How are you','I am fine'),
(2,'Your Name is?','Fred'),(2,'how young are you?','38'),(2,'Answer This-','What?'),
(2,'You good?','Yeah!'),(2,'How is Bill','Sleepy');

-- 2. Add Many New rows for performance testing
DECLARE @multiplier BIGINT = 10000;

INSERT dbo.question
SELECT     ROW_NUMBER() OVER (ORDER BY (SELECT NULL))+2, q.question
FROM       dbo.question                    AS q
CROSS JOIN core.rangeAB(1,@multiplier,1,1) AS r;

INSERT     dbo.answers
SELECT     a.aID%2+((a.aID+r.RN)*2), a.key_, a.value_
FROM       dbo.answers                     AS a
CROSS JOIN core.rangeAB(1,@multiplier,1,0) AS r;

完整解决方案

with cte_position as
(
    select  charindex('@@', q.question, 0) as 'position',
            1 as 'count_',
            convert(nvarchar(30), null) as 'key_',
            q.question
    from @question q
        union all
    select  charindex('@@', cp.question, cp.position+2),
            cp.count_+1,
            convert(nvarchar(30), case
                when (cp.count_+1) % 2 = 0
                then substring(cp.question, cp.position+2, charindex('@@', cp.question, cp.position+2)-cp.position-2)
                else null
            end),
            cp.question
    from cte_position cp
    where charindex('@@', cp.question, cp.position+2) > 0
),
cte_replace as
(
    select  cp.count_,
            cp.key_,
            replace(cp.question, '@@'+cp.key_+'@@', a.value_) as 'newquestion'
    from cte_position cp
    join @answers a
        on a.key_ = cp.key_
    where cp.key_ is not null
      and not exists (  select top 1 'x'
                        from cte_position cp2
                        where cp2.key_ is not null
                          and cp2.count_ < cp.count_ )
        union all
    select  cp.count_,
            cp.key_,
            replace(cr.newquestion, '@@'+cp.key_+'@@', a.value_)
    from cte_replace cr
    join cte_position cp
        on cp.count_ = cr.count_ + 2
    join @answers a
        on a.key_ = cp.key_
)
select top 1 cr.newquestion
from cte_replace cr
order by cr.count_ desc;

第2部分-用值替换键

重复CTE的结构

CTE\u replace

：

仅选择具有要替换的键的中间记录=

cp.key\uu.不为空

以计数最低的第一条记录为例（因此计数较低的记录必须不存在）

选择计数器

选择要替换的键

加入

@answer

执行实际操作，并将结果重命名为

newquestion

对于递归部分：

采用之前的记录（上述步骤1至5）

通过加入

cte_positions

并使用大于2的

count_

对记录进行过滤（我们在步骤1中仅选择偶数行），获得下一个键

加入

@answer

执行实际的更换操作

当在

@answer

中找不到更多要替换的

键时，递归停止

中间结果

第3部分-选择最终结果

newquestion
-----------------------------------------------------------------------------
I have a question. Doe29I need more answers.I am fine.Answer if you want.

-- 3. Performance Test
BEGIN
  PRINT CHAR(10)+'Start Test'+CHAR(10)+REPLICATE('-',90);
    IF OBJECT_ID('tempdb..#t1') IS NOT NULL DROP TABLE #t1;
    IF OBJECT_ID('tempdb..#t2') IS NOT NULL DROP TABLE #t2;
    IF OBJECT_ID('tempdb..#t3') IS NOT NULL DROP TABLE #t3;
    IF OBJECT_ID('tempdb..#t4') IS NOT NULL DROP TABLE #t4;
  
  PRINT CHAR(10)+'delimitedSplit8K serial'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t1
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY dbo.delimitedSplit8K(q.question,'@')                     AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID  = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (MAXDOP 1);
  
  PRINT CHAR(10)+'delimitedSplit8K parallel'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t2
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY dbo.delimitedSplit8K(q.question,'@')                     AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID  = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (QUERYTRACEON 8649);
  
  PRINT CHAR(10)+'patExtract8K Serial'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t3
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY samd.patExtract8K(q.question,'@')                        AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (MAXDOP 1);
  
  PRINT CHAR(10)+'patExtract8K Parallel'+CHAR(10)+REPLICATE('-',90);
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t4
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY samd.patExtract8K(q.question,'@')                        AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (QUERYTRACEON 8649);
END

最简单的部分：从最后一个计数最高的结果集中选择记录，以便在按计数递减排序时第一个（

top 1

）记录（

cr.count\uu desc

）

结果

newquestion
-----------------------------------------------------------------------------
I have a question. Doe29I need more answers.I am fine.Answer if you want.

-- 3. Performance Test
BEGIN
  PRINT CHAR(10)+'Start Test'+CHAR(10)+REPLICATE('-',90);
    IF OBJECT_ID('tempdb..#t1') IS NOT NULL DROP TABLE #t1;
    IF OBJECT_ID('tempdb..#t2') IS NOT NULL DROP TABLE #t2;
    IF OBJECT_ID('tempdb..#t3') IS NOT NULL DROP TABLE #t3;
    IF OBJECT_ID('tempdb..#t4') IS NOT NULL DROP TABLE #t4;
  
  PRINT CHAR(10)+'delimitedSplit8K serial'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t1
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY dbo.delimitedSplit8K(q.question,'@')                     AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID  = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (MAXDOP 1);
  
  PRINT CHAR(10)+'delimitedSplit8K parallel'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t2
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY dbo.delimitedSplit8K(q.question,'@')                     AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID  = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (QUERYTRACEON 8649);
  
  PRINT CHAR(10)+'patExtract8K Serial'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t3
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY samd.patExtract8K(q.question,'@')                        AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (MAXDOP 1);
  
  PRINT CHAR(10)+'patExtract8K Parallel'+CHAR(10)+REPLICATE('-',90);
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t4
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY samd.patExtract8K(q.question,'@')                        AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (QUERYTRACEON 8649);
END

---

更新20200725

不需要游标、循环或递归CTE。游标或递归解决方案几乎肯定是缓慢且过于庞大的。一个基于集合的解决方案，利用或使用是最好的选择

注意更新的样本数据（包括更多@）：

每个解决方案的结果：

FinalAnswer
--------------------------------------------------------------------------
I have a question. Doe29I need more answers.I am fine.Answer if you want.

delimitedSplit8K serial
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 7641 ms,  elapsed time = 7643 ms.

delimitedSplit8K parallel
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 12454 ms,  elapsed time = 2118 ms.

patExtract8K Serial
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 4172 ms,  elapsed time = 4216 ms.

patExtract8K Parallel
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 7453 ms,  elapsed time = 1343 ms.

与方法相比，提取器方法的优点是该方法“跳过”分隔符（@）。比较执行计划：

DSPLIT8K执行计划

patExtract8K执行计划

我在示例数据中添加了一些额外的分隔符，以显示它如何使用提取器提取6个必需项，并开始只向前处理6行（而不是55行）

注意，在本例中，我们只处理一条记录。为了更好地理解如何对多行执行此操作，您需要有更好的示例数据

最终编辑20200726（性能测试）：

FinalAnswer
--------------------------------------------------------------------------
I have a question. Doe29I need more answers.I am fine.Answer if you want.

delimitedSplit8K serial
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 7641 ms,  elapsed time = 7643 ms.

delimitedSplit8K parallel
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 12454 ms,  elapsed time = 2118 ms.

patExtract8K Serial
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 4172 ms,  elapsed time = 4216 ms.

patExtract8K Parallel
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 7453 ms,  elapsed time = 1343 ms.

基于此线程中的讨论。我包括了两个不同的拆分器，可以用于此目的，并利用串行和并行执行计划对它们进行了测试

样本数据

declare @question table
(
    question nvarchar(200) primary key clustered
);

insert into @question (question) values
('I have a question. @@What is your name@@@@what is your age@@I need more answers.@@How are you@@.Answer if you want.');

declare @answers table
(
    key_ nvarchar(30) primary key clustered,
    value_ nvarchar(30)
);

insert into @answers (key_, value_) values
('What is your name', 'Doe'),
('what is your age', '29'),
('How are you', 'I am fine');

-- 1. Sample Data
SET NOCOUNT ON;
IF OBJECT_ID('dbo.question','U') IS NOT NULL DROP TABLE dbo.question;
IF OBJECT_ID('dbo.answers','U')  IS NOT NULL DROP TABLE dbo.answers;
CREATE TABLE dbo.question
(
  qID      int  PRIMARY KEY CLUSTERED,
  question nvarchar(200) NOT NULL
);
CREATE TABLE dbo.answers
(
  aID    int          NOT NULL INDEX nc__answers__aID,
  key_   nvarchar(30) NOT NULL,
  value_ nvarchar(30) NOT NULL
);
insert into dbo.question (qID,question) values
(1,'I have a question. @@What is your name@@@@what is your age@@I need more answers.'+
   '@@@@@@@@@@@@@@@@@@@How are you@@@@@@@@@@@@@@@@@@@@@@@@@@@.Answer if you want.'),
(2,'I have a question too.@@@@@@@@@@@@Your Name is?@@@how young are you?@@Answer This-'+
   '@@@@@@@@@@@You good?@@@@@@@@@@@@@@@How is Bill@@@@@@@@Answer if you must!.');
insert into dbo.answers (aID,key_, value_) values
(1,'What is your name','Doe'),(1,'what is your age','29'),(1,'How are you','I am fine'),
(2,'Your Name is?','Fred'),(2,'how young are you?','38'),(2,'Answer This-','What?'),
(2,'You good?','Yeah!'),(2,'How is Bill','Sleepy');

-- 2. Add Many New rows for performance testing
DECLARE @multiplier BIGINT = 10000;

INSERT dbo.question
SELECT     ROW_NUMBER() OVER (ORDER BY (SELECT NULL))+2, q.question
FROM       dbo.question                    AS q
CROSS JOIN core.rangeAB(1,@multiplier,1,1) AS r;

INSERT     dbo.answers
SELECT     a.aID%2+((a.aID+r.RN)*2), a.key_, a.value_
FROM       dbo.answers                     AS a
CROSS JOIN core.rangeAB(1,@multiplier,1,0) AS r;

性能测试

newquestion
-----------------------------------------------------------------------------
I have a question. Doe29I need more answers.I am fine.Answer if you want.

-- 3. Performance Test
BEGIN
  PRINT CHAR(10)+'Start Test'+CHAR(10)+REPLICATE('-',90);
    IF OBJECT_ID('tempdb..#t1') IS NOT NULL DROP TABLE #t1;
    IF OBJECT_ID('tempdb..#t2') IS NOT NULL DROP TABLE #t2;
    IF OBJECT_ID('tempdb..#t3') IS NOT NULL DROP TABLE #t3;
    IF OBJECT_ID('tempdb..#t4') IS NOT NULL DROP TABLE #t4;
  
  PRINT CHAR(10)+'delimitedSplit8K serial'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t1
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY dbo.delimitedSplit8K(q.question,'@')                     AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID  = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (MAXDOP 1);
  
  PRINT CHAR(10)+'delimitedSplit8K parallel'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t2
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY dbo.delimitedSplit8K(q.question,'@')                     AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID  = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (QUERYTRACEON 8649);
  
  PRINT CHAR(10)+'patExtract8K Serial'+CHAR(10)+REPLICATE('-',90);
  
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t3
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY samd.patExtract8K(q.question,'@')                        AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (MAXDOP 1);
  
  PRINT CHAR(10)+'patExtract8K Parallel'+CHAR(10)+REPLICATE('-',90);
  WITH q2 AS (SELECT q.qID FROM dbo.question AS q GROUP BY q.qID)
  SELECT
    q2.qID,
    FinalAnswer = fa.FinalAnswer
  INTO   #t4
  FROM   q2
  CROSS APPLY
  (
    SELECT      ISNULL(a.value_,split.item)
    FROM        (SELECT q.* FROM dbo.question AS q WHERE q.qID = q2.qID) AS q
    CROSS APPLY samd.patExtract8K(q.question,'@')                        AS split
    LEFT JOIN   (SELECT a.* FROM dbo.answers  AS a WHERE a.aID = q2.qID) AS a
      ON        a.aID = q.qID
      AND       a.key_ = split.item
    ORDER BY    split.ItemNumber
    FOR XML PATH('')
  ) AS fa(FinalAnswer)
  OPTION (QUERYTRACEON 8649);
END

结果：

FinalAnswer
--------------------------------------------------------------------------
I have a question. Doe29I need more answers.I am fine.Answer if you want.

delimitedSplit8K serial
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 7641 ms,  elapsed time = 7643 ms.

delimitedSplit8K parallel
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 12454 ms,  elapsed time = 2118 ms.

patExtract8K Serial
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 4172 ms,  elapsed time = 4216 ms.

patExtract8K Parallel
------------------------------------------------------------------------------------------
 SQL Server Execution Times: CPU time = 7453 ms,  elapsed time = 1343 ms.

---

为此，需要一个基于集合的解决方案。Patextract在这里工作是因为有超长的分隔符。我尝试了下面的递归CTE解决方案，但无法使其工作；部分原因是，逻辑需要活在一个比我有时间做的工作更多的函数中。尽管如此，对于（大多数情况下。）

准备充分的答案，做得好。然而，递归会很慢，而且对于这种类型的事情来说有点复杂。性能完全取决于数据集的大小和复杂性以及可用的系统资源。如果没有更多的样本数据，很难验证。我没有为我的解决方案考虑预定义函数。@ AalbBistin对单行样本数据做了一些快速测试。如果我像您一样使用

主键（解决方案更新）定义表变量，那么我的查询计划的成本（00547064）大约是
delimitedSplit8K
解决方案（00253554）的两倍。所以，如果函数是一个选项，那么您的解决方案将是首选方法。说得清楚，我喜欢您的解决方案，在某些方面，我喜欢递归CTES。我在上面添加了一个性能测试线束。我将不得不把你的逻辑转换成一个比计划的更耗时的函数。我怀疑rCTE解决方案会更慢-它肯定会产生更多IO。