SQL Server表查询带分页性能调优，了解当前解决方案

Question

如标题所述，我开始对表查询进行性能调整，该查询使用由使用 Linq To SQL 作为 ORM 的遗留程序生成的分页。

我发现这个资源强烈建议在分页之前对表进行排序： https: //rimdev.io/optimizing-linq-sql-skip-take/

所以我遵循了提供的建议并尝试了巨大的差异。我很清楚这与 row_number 的计算方式有一定关系，但我不清楚到底发生了什么以及为什么两个查询之间有如此大的差异。

原始慢查询（〜7K元素的数据集，需要〜3s）：

SELECT [t7].[ID], [t7].[ID_BRAND], [t7].[CODE], [t7].[CODFOR], [t7].[COD_ALT01], [t7].[COD_ALT02], [t7].[COD_ALT03], [t7].[ID_UOM], [t7].[IS_ACTIVE], [t7].[_ATTRIBUTES] AS [_ATTRIBUTES], [t7].[_DOCUMENTS] AS [_DOCUMENTS], [t7].[_SEO] AS [_SEO], [t7].[_TRANSLATIONS] AS [_TRANSLATIONS], [t7].[_TAGS] AS [_TAGS], [t7].[_NOTES] AS [_NOTES], [t7].[_METADATA] AS [_METADATA], [t7].[IS_B2B], [t7].[IS_B2C], [t7].[IS_PROMO], [t7].[IS_NEWS], [t7].[CAN_BE_RETURNED], [t7].[IS_SHIPPABLE], [t7].[HAS_SHIPPING_COSTS], [t7].[IS_PURCHEASABLE], [t7].[test], [t7].[ID2], [t7].[CODE2], [t7].[BUSINESS_NAME], [t7].[NAME], [t7].[PHONE_01], [t7].[PHONE_02], [t7].[PHONE_03], [t7].[FAX_01], [t7].[FAX_02], [t7].[COUNTRY_01], [t7].[CITY_01], [t7].[ADDRESS_01], [t7].[COUNTRY_02], [t7].[CITY_02], [t7].[ADDRESS_02], [t7].[EMAIL_01], [t7].[EMAIL_02], [t7].[PEC], [t7].[SITE_01], [t7].[SITE_02], [t7].[SITE_03], [t7].[SITE_04], [t7].[VAT_NUMBER], [t7].[SORT], [t7].[GROUPID_01], [t7].[IS_GROUPLEADER_01], [t7].[GROUPID_02], [t7].[IS_GROUPLEADER_02],[t7].[IS_ACTIVE2], [t7].[[_DOCUMENTS]]2] AS [_DOCUMENTS2], [t7].[[_SEO]]2] AS [_SEO2], [t7].[[_METADATA]]2] AS [_METADATA2], [t7].[test2], [t7].[ID3], [t7].[CODE3], [t7].[[_TRANSLATIONS]]2] AS [_TRANSLATIONS2], [t7].[[_METADATA]]3] AS [_METADATA3], [t7].[test3], [t7].[ID4], [t7].[ID_LINE], [t7].[ID_GROUP], [t7].[ID_CLASS], [t7].[ID_FAM], [t7].[ID_ARTICLE]
FROM (
    SELECT ROW_NUMBER() OVER (ORDER BY [t0].[ID], [t0].[ID_BRAND], [t0].[CODE], [t0].[CODFOR], [t0].[COD_ALT01], [t0].[COD_ALT02], [t0].[COD_ALT03], [t0].[ID_UOM], [t0].[IS_ACTIVE], [t0].[_ATTRIBUTES], [t0].[_DOCUMENTS], [t0].[_SEO], [t0].[_TRANSLATIONS], [t0].[_TAGS], [t0].[_NOTES], [t0].[_METADATA], [t0].[IS_B2B], [t0].[IS_B2C], [t0].[IS_PROMO], [t0].[IS_NEWS], [t0].[CAN_BE_RETURNED], [t0].[IS_SHIPPABLE], [t0].[HAS_SHIPPING_COSTS], [t0].[IS_PURCHEASABLE], [t2].[test], [t2].[ID], [t2].[CODE], [t2].[BUSINESS_NAME], [t2].[NAME], [t2].[PHONE_01], [t2].[PHONE_02], [t2].[PHONE_03], [t2].[FAX_01], [t2].[FAX_02], [t2].[COUNTRY_01], [t2].[CITY_01], [t2].[ADDRESS_01], [t2].[COUNTRY_02], [t2].[CITY_02], [t2].[ADDRESS_02], [t2].[EMAIL_01], [t2].[EMAIL_02], [t2].[PEC], [t2].[SITE_01], [t2].[SITE_02], [t2].[SITE_03], [t2].[SITE_04], [t2].[VAT_NUMBER], [t2].[SORT], [t2].[GROUPID_01], [t2].[IS_GROUPLEADER_01], [t2].[GROUPID_02], [t2].[IS_GROUPLEADER_02], [t2].[IS_ACTIVE], [t2].[_DOCUMENTS], [t2].[_SEO], [t2].[_METADATA], [t4].[test], [t4].[ID], [t4].[CODE], [t4].[_TRANSLATIONS], [t4].[_METADATA], [t6].[test], [t6].[ID], [t6].[ID_LINE], [t6].[ID_GROUP], [t6].[ID_CLASS], [t6].[ID_FAM], [t6].[ID_ARTICLE]) AS [ROW_NUMBER], [t0].[ID], [t0].[ID_BRAND], [t0].[CODE], [t0].[CODFOR], [t0].[COD_ALT01], [t0].[COD_ALT02], [t0].[COD_ALT03], [t0].[ID_UOM], [t0].[IS_ACTIVE], [t0].[_ATTRIBUTES], [t0].[_DOCUMENTS], [t0].[_SEO], [t0].[_TRANSLATIONS], [t0].[_TAGS], [t0].[_NOTES], [t0].[_METADATA], [t0].[IS_B2B], [t0].[IS_B2C], [t0].[IS_PROMO], [t0].[IS_NEWS], [t0].[CAN_BE_RETURNED], [t0].[IS_SHIPPABLE], [t0].[HAS_SHIPPING_COSTS], [t0].[IS_PURCHEASABLE], [t2].[test], [t2].[ID] AS [ID2], [t2].[CODE] AS [CODE2], [t2].[BUSINESS_NAME], [t2].[NAME], [t2].[PHONE_01], [t2].[PHONE_02], [t2].[PHONE_03], [t2].[FAX_01], [t2].[FAX_02], [t2].[COUNTRY_01], [t2].[CITY_01], [t2].[ADDRESS_01], [t2].[COUNTRY_02], [t2].[CITY_02], [t2].[ADDRESS_02], [t2].[EMAIL_01], [t2].[EMAIL_02], [t2].[PEC], [t2].[SITE_01], [t2].[SITE_02], [t2].[SITE_03], [t2].[SITE_04], [t2].[VAT_NUMBER], [t2].[SORT], [t2].[GROUPID_01], [t2].[IS_GROUPLEADER_01], [t2].[GROUPID_02], [t2].[IS_GROUPLEADER_02], [t2].[IS_ACTIVE] AS [IS_ACTIVE2], [t2].[_DOCUMENTS] AS [[_DOCUMENTS]]2], [t2].[_SEO] AS [[_SEO]]2], [t2].[_METADATA] AS [[_METADATA]]2], [t4].[test] AS [test2], [t4].[ID] AS [ID3], [t4].[CODE] AS [CODE3], [t4].[_TRANSLATIONS] AS [[_TRANSLATIONS]]2], [t4].[_METADATA] AS [[_METADATA]]3], [t6].[test] AS [test3], [t6].[ID] AS [ID4], [t6].[ID_LINE], [t6].[ID_GROUP], [t6].[ID_CLASS], [t6].[ID_FAM], [t6].[ID_ARTICLE]
    FROM [dbo].[tbl_ana_Articles] AS [t0]
    LEFT OUTER JOIN (
        SELECT 1 AS [test], [t1].[ID], [t1].[CODE], [t1].[BUSINESS_NAME], [t1].[NAME], [t1].[PHONE_01], [t1].[PHONE_02], [t1].[PHONE_03], [t1].[FAX_01], [t1].[FAX_02], [t1].[COUNTRY_01], [t1].[CITY_01], [t1].[ADDRESS_01], [t1].[COUNTRY_02], [t1].[CITY_02], [t1].[ADDRESS_02], [t1].[EMAIL_01], [t1].[EMAIL_02], [t1].[PEC], [t1].[SITE_01], [t1].[SITE_02], [t1].[SITE_03], [t1].[SITE_04], [t1].[VAT_NUMBER], [t1].[SORT], [t1].[GROUPID_01], [t1].[IS_GROUPLEADER_01], [t1].[GROUPID_02], [t1].[IS_GROUPLEADER_02], [t1].[IS_ACTIVE], [t1].[_DOCUMENTS], [t1].[_SEO], [t1].[_METADATA]
        FROM [dbo].[tbl_ana_Brands] AS [t1]
        ) AS [t2] ON [t2].[ID] = [t0].[ID_BRAND]
    LEFT OUTER JOIN (
        SELECT 1 AS [test], [t3].[ID], [t3].[CODE], [t3].[_TRANSLATIONS], [t3].[_METADATA]
        FROM [dbo].[tbl_ana_UoMs] AS [t3]
        ) AS [t4] ON [t4].[ID] = [t0].[ID_UOM]
    LEFT OUTER JOIN (
        SELECT 1 AS [test], [t5].[ID], [t5].[ID_LINE], [t5].[ID_GROUP], [t5].[ID_CLASS], [t5].[ID_FAM], [t5].[ID_ARTICLE]
        FROM [dbo].[tbl_src_ArticlesCategories] AS [t5]
        ) AS [t6] ON [t6].[ID_ARTICLE] = [t0].[ID]
    WHERE (
        (CASE 
            WHEN 1 = 1 THEN CONVERT(Int,[t0].[IS_ACTIVE])
            ELSE 0
         END)) = 1
    ) AS [t7]
WHERE [t7].[ROW_NUMBER]  BETWEEN 7272 + 1 AND 7284
ORDER BY [t7].[ROW_NUMBER]

这里慢速查询执行计划：https://www.brentozar.com/pastetheplan/ ?id=Sk-rLnY3F

修改后的快速查询（约 7K 元素的数据集，需要约 0 秒）：

SELECT [t7].[ID], [t7].[ID_BRAND], [t7].[CODE], [t7].[CODFOR], [t7].[COD_ALT01], [t7].[COD_ALT02], [t7].[COD_ALT03], [t7].[ID_UOM], [t7].[IS_ACTIVE], [t7].[_ATTRIBUTES] AS [_ATTRIBUTES], [t7].[_DOCUMENTS] AS [_DOCUMENTS], [t7].[_SEO] AS [_SEO], [t7].[_TRANSLATIONS] AS [_TRANSLATIONS], [t7].[_TAGS] AS [_TAGS], [t7].[_NOTES] AS [_NOTES], [t7].[_METADATA] AS [_METADATA], [t7].[IS_B2B], [t7].[IS_B2C], [t7].[IS_PROMO], [t7].[IS_NEWS], [t7].[CAN_BE_RETURNED], [t7].[IS_SHIPPABLE], [t7].[HAS_SHIPPING_COSTS], [t7].[IS_PURCHEASABLE], [t7].[test], [t7].[ID2], [t7].[CODE2], [t7].[BUSINESS_NAME], [t7].[NAME], [t7].[PHONE_01], [t7].[PHONE_02], [t7].[PHONE_03], [t7].[FAX_01], [t7].[FAX_02], [t7].[COUNTRY_01], [t7].[CITY_01], [t7].[ADDRESS_01], [t7].[COUNTRY_02], [t7].[CITY_02], [t7].[ADDRESS_02], [t7].[EMAIL_01], [t7].[EMAIL_02], [t7].[PEC], [t7].[SITE_01], [t7].[SITE_02], [t7].[SITE_03], [t7].[SITE_04], [t7].[VAT_NUMBER], [t7].[SORT], [t7].[GROUPID_01], [t7].[IS_GROUPLEADER_01], [t7].[GROUPID_02], [t7].[IS_GROUPLEADER_02],[t7].[IS_ACTIVE2], [t7].[[_DOCUMENTS]]2] AS [_DOCUMENTS2], [t7].[[_SEO]]2] AS [_SEO2], [t7].[[_METADATA]]2] AS [_METADATA2], [t7].[test2], [t7].[ID3], [t7].[CODE3], [t7].[[_TRANSLATIONS]]2] AS [_TRANSLATIONS2], [t7].[[_METADATA]]3] AS [_METADATA3], [t7].[test3], [t7].[ID4], [t7].[ID_LINE], [t7].[ID_GROUP], [t7].[ID_CLASS], [t7].[ID_FAM], [t7].[ID_ARTICLE]
FROM (
    SELECT ROW_NUMBER() OVER (ORDER BY [t0].[ID]) AS [ROW_NUMBER], [t0].[ID], [t0].[ID_BRAND], [t0].[CODE], [t0].[CODFOR], [t0].[COD_ALT01], [t0].[COD_ALT02], [t0].[COD_ALT03], [t0].[ID_UOM], [t0].[IS_ACTIVE], [t0].[_ATTRIBUTES], [t0].[_DOCUMENTS], [t0].[_SEO], [t0].[_TRANSLATIONS], [t0].[_TAGS], [t0].[_NOTES], [t0].[_METADATA], [t0].[IS_B2B], [t0].[IS_B2C], [t0].[IS_PROMO], [t0].[IS_NEWS], [t0].[CAN_BE_RETURNED], [t0].[IS_SHIPPABLE], [t0].[HAS_SHIPPING_COSTS], [t0].[IS_PURCHEASABLE], [t2].[test], [t2].[ID] AS [ID2], [t2].[CODE] AS [CODE2], [t2].[BUSINESS_NAME], [t2].[NAME], [t2].[PHONE_01], [t2].[PHONE_02], [t2].[PHONE_03], [t2].[FAX_01], [t2].[FAX_02], [t2].[COUNTRY_01], [t2].[CITY_01], [t2].[ADDRESS_01], [t2].[COUNTRY_02], [t2].[CITY_02], [t2].[ADDRESS_02], [t2].[EMAIL_01], [t2].[EMAIL_02], [t2].[PEC], [t2].[SITE_01], [t2].[SITE_02], [t2].[SITE_03], [t2].[SITE_04], [t2].[VAT_NUMBER], [t2].[SORT], [t2].[GROUPID_01], [t2].[IS_GROUPLEADER_01], [t2].[GROUPID_02], [t2].[IS_GROUPLEADER_02], [t2].[IS_ACTIVE] AS [IS_ACTIVE2], [t2].[_DOCUMENTS] AS [[_DOCUMENTS]]2], [t2].[_SEO] AS [[_SEO]]2], [t2].[_METADATA] AS [[_METADATA]]2], [t4].[test] AS [test2], [t4].[ID] AS [ID3], [t4].[CODE] AS [CODE3], [t4].[_TRANSLATIONS] AS [[_TRANSLATIONS]]2], [t4].[_METADATA] AS [[_METADATA]]3], [t6].[test] AS [test3], [t6].[ID] AS [ID4], [t6].[ID_LINE], [t6].[ID_GROUP], [t6].[ID_CLASS], [t6].[ID_FAM], [t6].[ID_ARTICLE]
    FROM [dbo].[tbl_ana_Articles] AS [t0]
    LEFT OUTER JOIN (
        SELECT 1 AS [test], [t1].[ID], [t1].[CODE], [t1].[BUSINESS_NAME], [t1].[NAME], [t1].[PHONE_01], [t1].[PHONE_02], [t1].[PHONE_03], [t1].[FAX_01], [t1].[FAX_02], [t1].[COUNTRY_01], [t1].[CITY_01], [t1].[ADDRESS_01], [t1].[COUNTRY_02], [t1].[CITY_02], [t1].[ADDRESS_02], [t1].[EMAIL_01], [t1].[EMAIL_02], [t1].[PEC], [t1].[SITE_01], [t1].[SITE_02], [t1].[SITE_03], [t1].[SITE_04], [t1].[VAT_NUMBER], [t1].[SORT], [t1].[GROUPID_01], [t1].[IS_GROUPLEADER_01], [t1].[GROUPID_02], [t1].[IS_GROUPLEADER_02], [t1].[IS_ACTIVE], [t1].[_DOCUMENTS], [t1].[_SEO], [t1].[_METADATA]
        FROM [dbo].[tbl_ana_Brands] AS [t1]
        ) AS [t2] ON [t2].[ID] = [t0].[ID_BRAND]
    LEFT OUTER JOIN (
        SELECT 1 AS [test], [t3].[ID], [t3].[CODE], [t3].[_TRANSLATIONS], [t3].[_METADATA]
        FROM [dbo].[tbl_ana_UoMs] AS [t3]
        ) AS [t4] ON [t4].[ID] = [t0].[ID_UOM]
    LEFT OUTER JOIN (
        SELECT 1 AS [test], [t5].[ID], [t5].[ID_LINE], [t5].[ID_GROUP], [t5].[ID_CLASS], [t5].[ID_FAM], [t5].[ID_ARTICLE]
        FROM [dbo].[tbl_src_ArticlesCategories] AS [t5]
        ) AS [t6] ON [t6].[ID_ARTICLE] = [t0].[ID]
    WHERE (
        (CASE 
            WHEN 1 = 1 THEN CONVERT(Int,[t0].[IS_ACTIVE])
            ELSE 0
         END)) = 1
    ) AS [t7]
WHERE [t7].[ROW_NUMBER] BETWEEN 7272 + 1 AND 7284
ORDER BY [t7].[ROW_NUMBER]

这里是快速查询执行计划：https://www.brentozar.com/pastetheplan/ ?id=B10l82K2Y

注意：所有查询代码都是由 ORM 自动生成的

这两个查询看起来非常相似，我不清楚是什么显着地提高了性能。我真的很感谢对 SQL Server 有如此大帮助的提示，以便我可以更好地理解如何在将来调整 ORM。

Answer 1

线轴

\n

这两个查询之间的主要区别是Eager Index Spool的存在。

\n

来自文章：

\n

\n

急切索引假脱机可以发生在嵌套循环连接的内侧，\n通过为循环创建\n更合适的索引来减少需要在那里完成的工作量。

\n

一切听起来都很好，但是存在一些问题：

\n

\n
• 索引是单线程创建的
\n
• 数据加载到spool的方式效率很低
\n
• 查询完成后，假脱机将被释放，It\xe2\x80\x99 将在以后的执行中一遍又一遍地构建
\n
• \xe2\x80\x99s 任何地方都没有丢失线轴的索引请求
\n

\n

\n

但就您而言，由于线轴非常小，因此它对您有利。

\n

慢行者

\n

在慢速查询中，您有一个嵌套循环连接，tbl_src_ArticlesCategories该连接执行约 7k 次，但没有有用的索引，因此每次执行都会扫描整个表。

\n

扫描：

\n

\n

细节：

\n

\n

总而言之，您最终总共读取了约 5300 万行，因为您扫描了约 736k 行约 7k 次。

\n

快速戳

\n

在快速计划中，您会得到以下结果：

\n

扫描和假脱机（带查找）：

\n

\n

细节

\n

\n

该计划的优化器决定为您创建一个好的索引，因此它有一个更合适的结构来定位 中的匹配行ID_ARTICLE。

\n

您执行了大约 7k 次搜索，考虑到这种情况，这效率要高得多。

\n

均衡器

\n

通过添加此索引，您可能会从两个查询中获得相同的性能：

\n

CREATE /*UNIQUE?*/ INDEX spool_b_gone\n    ON [dbo].[tbl_src_ArticlesCategories]\n(\n    [ID_ARTICLE]\n)\nINCLUDE\n(\n    [ID_LINE],\n    [ID_GROUP], \n    [ID_CLASS], \n    [ID_FAM]\n);\n

\n

尽管有时优化器是愚蠢的，即使您拥有正确的索引，也可能会决定假脱机。

\n

差异？

\n

我看到的直接区别是，在较慢的执行中，您会在不同的表中生成更多列的行号：

\n

\n

但我的时间有点短，所以可能还有其他因素影响在线轴/无线轴之间的选择。

\n