Question

我在SQL Server 2012 / SQL Server 2016上。

Xquery功能＆＃39;修改＆＃39;只会一次更改一行。所以我逐个元素地遍历我的XML数据。这很慢。当只有一个不同的值要更改时，我会转换XML并替换值。但是在大多数情况下（行）有多个不同的值。

我希望my_xml的结构从（stronly simplified）数据变得清晰：元素组（a和b）中有多个标签。

--My data

if object_id ('tempdb..#t1') is not null drop table #t1

create table #t1 (id int not null identity (1,1) unique clustered, my_xml xml)
insert #t1 (my_xml) values 
     ('<root><a label=''name A'' /> <b /> </root>'),        --A
     ('<root><a label=''name A'' /> <b label=''name B'' /> </root>'),   --AB
     ('<root><a label=''name A'' /> <b label=''name B'' /> <a label=''name A'' /> <b label=''name B'' /><a label=''name A'' /> <b label=''name B'' /></root>'), --ABABAB
     ('<root><a label=''name B'' /> <b label=''name A'' /> <a label=''name B'' /> <a label=''name B'' /> <a label=''name B'' /> </root>') --BABBB

select * from #t1

我想将标签的值更改为原始值的SHA-Hash。

我查询了一些关于my_xml结构的信息，以最小化要修改的行：

--My helper-table    

if object_id ('tempdb..#t2') is not null drop table #t2

select 
     id, 
     max(my_xml.value('count(/root/*[self::a or self::b]/@label)', 'int')) max_label,
     max(my_xml.value('count(distinct-values(/root/*[self::a or self::b]/@label))', 'int')) max_label_distinct
into #t2
from #t1
group by id

select * from #t2

然后我修改XML（只使用带有多个不同标签的行，因为只需要它们，我需要遍历XML）

 --My query:

 set nocount on
 declare 
     @label         char(64), 
     @id                int,
     @n_max_label   smallint

 declare test cursor for 

     select 
         id, 
         max_label 
     from #t2
     where max_label_distinct > 1

 open test
 fetch next from test into @id, @n_max_label
 while (@@FETCH_STATUS <> -1)
 begin

     while @n_max_label > 0 
     begin

         select @label = isnull(convert(char(64), HASHBYTES('SHA2_256', my_xml.value('((/root/*[self::a or self::b]/@label)[sql:variable("@n_max_label")])[1]', 'char(64)')), 1), 'x') 
         from #t1
         where id = @id

         update #t1
         set my_xml.modify('replace value of ((/root/*[self::a or self::b]/@label)[sql:variable("@n_max_label")])[1] with sql:variable("@label")')
         where id = @id

         set @n_max_label = @n_max_label - 1

     end

     fetch next from test into @id, @n_max_label
 end
 close test
 deallocate test

最终结果：

 select * from #t1

我想以10到100倍的速度（就速度而言）改进解决方案，因为现在我需要大约2个月来计算结果 - 长期操作大约600 GB的数据。（卸载和使用其他工具是没有选择的）

Answer 1

您发现.modify()会一次更改一个值。这很慢，需要程序化（CURSOR）逻辑。

您没有仔细查看原始XML，因此这可能指向错误的方向，但您可以尝试这样做：

您的新帮助程序表

WITH cte AS
(
    select 
          id 
         ,A.AllLabels.value('local-name(..)','nvarchar(max)') AS ElementName
         ,A.AllLabels.value('local-name(.)','nvarchar(max)') AS LabelName
         ,A.AllLabels.value('.','nvarchar(max)') AS LabelValue
         ,HASHBYTES('SHA2_256', A.AllLabels.value('.','nvarchar(max)')) AS LabelHash
    from #t1
     --use "/root/" instead of "//" if always below <root>
    cross apply my_xml.nodes('//*[local-name()="a" or local-name()="b"]/@label') AS A(AllLabels)
)
SELECT id
      ,ElementName
      ,LabelName
      ,LabelValue
      ,LabelHash
      ,(SELECT LabelHash FOR XML PATH(''),TYPE).value('.','nvarchar(max)') AS HashBase64
      ,CONCAT(' ',LabelName,'="',LabelValue,'"' ) AS ReplaceThis
      ,CONCAT(' ',LabelName,'="',(SELECT LabelHash FOR XML PATH(''),TYPE).value('.','nvarchar(max)'),'"' ) AS WithThat
INTO #t2
FROM cte
GROUP BY id,ElementName,LabelName,LabelValue,LabelHash;

SELECT * FROM #t2;

结果

+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| id | ElementName | LabelName | LabelValue | LabelHash                                                          | HashBase64                                   | ReplaceThis    | WithThat                                             |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| 1  | a           | label     | name A     | 0x6F5A56BD0B5E367787D4D3F798F5467C17FB798C0849ECF88373A7346EB405CB | b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs= | label="name A" | label="b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs=" |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| 2  | a           | label     | name A     | 0x6F5A56BD0B5E367787D4D3F798F5467C17FB798C0849ECF88373A7346EB405CB | b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs= | label="name A" | label="b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs=" |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| 3  | a           | label     | name A     | 0x6F5A56BD0B5E367787D4D3F798F5467C17FB798C0849ECF88373A7346EB405CB | b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs= | label="name A" | label="b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs=" |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| 4  | b           | label     | name A     | 0x6F5A56BD0B5E367787D4D3F798F5467C17FB798C0849ECF88373A7346EB405CB | b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs= | label="name A" | label="b1pWvQteNneH1NP3mPVGfBf7eYwISez4g3OnNG60Bcs=" |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| 2  | b           | label     | name B     | 0xC8144356A508FA516CC263E0839B7D22811A6F8333851F5319E778D94FE5B0C3 | yBRDVqUI+lFswmPgg5t9IoEab4MzhR9TGed42U/lsMM= | label="name B" | label="yBRDVqUI+lFswmPgg5t9IoEab4MzhR9TGed42U/lsMM=" |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| 3  | b           | label     | name B     | 0xC8144356A508FA516CC263E0839B7D22811A6F8333851F5319E778D94FE5B0C3 | yBRDVqUI+lFswmPgg5t9IoEab4MzhR9TGed42U/lsMM= | label="name B" | label="yBRDVqUI+lFswmPgg5t9IoEab4MzhR9TGed42U/lsMM=" |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+
| 4  | a           | label     | name B     | 0xC8144356A508FA516CC263E0839B7D22811A6F8333851F5319E778D94FE5B0C3 | yBRDVqUI+lFswmPgg5t9IoEab4MzhR9TGed42U/lsMM= | label="name B" | label="yBRDVqUI+lFswmPgg5t9IoEab4MzhR9TGed42U/lsMM=" |
+----+-------------+-----------+------------+--------------------------------------------------------------------+----------------------------------------------+----------------+------------------------------------------------------+

您可以使用此代码使用列REPLACE和ReplaceThis在字符串基础上尝试WithThat，也可以在.modify <中使用CURSOR / p>

要记住的一些事情：

在XML中，二进制值应为Base64而不是十六进制字符串（通常为）
您可以使用 quirky update UDF （例如provided here）一次性执行多次替换，根据需要进行修改
如果您不必查看包含@label的元素，则可以在帮助程序表中省略此列以减少行数。
有数以千计的数据我建议在帮助表上使用索引
您必须知道，空格可能会导致此问题（label = "blah"与字符串级别的label="blah"不同）
与单引号和双引号相同
如果你选择.modify()，你可以动态地在CURSOR内建立一个声明并使用EXEC(@YourStatement)，但我怀疑这会很快......

如何加速XQuery修改？

1 个答案: