如何使用嵌套循环加速查询
Posted
技术标签:
【中文标题】如何使用嵌套循环加速查询【英文标题】：How to speed up a query with nested loops 【发布时间】：2014-01-16 17:38:03 【问题描述】：我有这个查询有效，但速度很慢
SELECT
ID_NODE,
-- this case slows down the query!!!
CASE WHEN (EXISTS (SELECT MV.ID_CHILD FROM MYVIEW MV INNER JOIN MYTABLE1 MT1 ON MT1.ID_NODE = MV.ID_CHILD WHERE MV.ID_PARENT = CA.ID_NODE AND ID_FATHER IS NOT NULL)) THEN 'Y' ELSE 'N' END AS HAVE_CHILDREN,
OTHER_FIELDS
FROM
MYTABLE2
更新：在第一个答案之后，我意识到我的示例并不完美，所以我对其进行了 2 处更改（CA 到 MT1，并写了 MT1.ID_FATHER 而不是 ID_FATHER )
SELECT
ID_NODE,
-- this case slows down the query!!!
CASE WHEN (EXISTS (SELECT MV.ID_CHILD FROM MYVIEW MV INNER JOIN MYTABLE1 MT1 ON MT1.ID_NODE = MV.ID_CHILD WHERE MV.ID_PARENT = MT2.ID_NODE AND MT1.ID_FATHER IS NOT NULL)) THEN 'Y' ELSE 'N' END AS HAVE_CHILDREN,
OTHER_FIELDS
FROM
MYTABLE2
更新结束
基本上我想要一个'y'/'n' 关于“这个节点有孩子吗？”的结果
在执行计划中，我只看到一个警告：
嵌套循环（内连接））43%
您能否建议对查询进行改进？
作为 exterme 解决方案，我可以将 HAVE_CHILDREN 值作为新字段存储在表中，但我不喜欢这样，因为它是“通往 bug 的高速公路”。
赏金注意事项：
我在这里发布原始表、视图（使用 CREATE 语句制作）和查询以帮助提供回复：
--This is MYTABLE1
CREATE TABLE [dbo].[MAN_PRG_OPERAZIONI](
[ID_PROG_OPERAZIONE] [int] NOT NULL,
[ID_CESPITE] [int] NOT NULL,
[ID_TIPO_OPERAZIONE] [int] NOT NULL,
[SEQUENZA] [int] NOT NULL,
[ID_RESPONSABILE] [int] NULL,
[DATA_SCADENZA] [datetime] NULL,
[DATA_ULTIMA] [datetime] NULL,
[ID_TIPO_FREQUENZA] [int] NOT NULL,
[FREQUENZA] [int] NOT NULL,
[NOTIFICA_SCADENZA] [nchar](1) NOT NULL,
[COSTO_FISSO] [numeric](19, 4) NOT NULL,
[NOTE] [nvarchar](max) NULL,
[ID_CONTO_FORNITORE] [int] NULL,
[ID_ESECUTORE] [int] NULL,
[GIORNI_INTERVENTO_PREVISTI] [int] NOT NULL,
[RIPETIZIONE] [nchar](1) NOT NULL,
[RIPETIZIONE_CONTINUA] [nchar](1) NOT NULL,
[RIPETI_FINO_A] [datetime] NULL,
[SOSPESO] [nchar](1) NOT NULL,
[ORE_LAVORO_PREVISTE] [decimal](8, 2) NOT NULL,
[DESCR_TITOLO_OPERAZIONE] [nvarchar](100) NOT NULL,
[ID_TEMPLATE] [int] NULL,
[TEMPLATE] [nvarchar](25) NULL,
[ID_PARENT_TEMPLATE_REMOTE] [int] NULL,
[ATTIVO] [nchar](1) NOT NULL,
[ID_PARENT_TEMPLATE] [int] NULL,
[NOTIFY_RESPONSIBLE] [nchar](1) NULL,
[NOTIFY_EXECUTOR] [nchar](1) NULL,
[NOTIFY_OTHERS] [nvarchar](200) NULL,
[NOTIFY_INADVANCE] [nchar](1) NULL,
[NOTIFY_ADVANCE_DAYS] [int] NULL,
[NOTIFY_ONEXECUTION] [nchar](1) NULL,
[NOTIFY_ONCLOSE] [nchar](1) NULL,
[ID_UTENTE_INS] [int] NULL,
[DATA_INS] [datetime] NULL,
[ID_UTENTE_ULT_MOD] [int] NULL,
[DATA_ULTIMA_MOD] [datetime] NULL,
[STATO_CKL] [int] NOT NULL,
[TAGAPPSYNC] [nchar](1) NOT NULL,
[IS_FATHER] [nchar](1) NOT NULL,
[ID_FATHER] [int] NULL,
[NOTIFY_DELAYS] [nchar](1) NULL,
[NOTIFY_DELAYS_DAYS] [int] NULL,
[NOTIFY_INS_USER] [nchar](1) NULL,
CONSTRAINT [PK_MAN_PRG_OPERAZIONI] PRIMARY KEY CLUSTERED
(
[ID_PROG_OPERAZIONE] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON, FILLFACTOR = 90) ON [PRIMARY]
) ON [PRIMARY]
--This is MYTABLE2
CREATE TABLE [dbo].[CES_ANAGRAFICA](
[ID_CESPITE] [int] NOT NULL,
[ID_CESPITE_PADRE] [int] NULL,
[COD_CESPITE] [nvarchar](50) NOT NULL,
[DESCR_CESPITE] [nvarchar](120) NOT NULL,
[IMMATERIALE] [nchar](1) NOT NULL,
[DATA_ACQUISTO] [datetime] NULL,
[DATA_ENTRATA_FUNZIONE] [datetime] NULL,
[DATA_DISMISSIONE] [datetime] NULL,
[BENE_USATO] [nchar](1) NOT NULL,
[ID_UBICAZIONE] [int] NULL,
[NRO_IDENTIFICAZIONE] [nvarchar](50) NULL,
[MARCA] [nvarchar](50) NULL,
[MODELLO] [nvarchar](50) NULL,
[MARCATURA_CE] [nchar](1) NULL,
[ANNO_COSTRUZIONE] [int] NULL,
[MATRICOLA_COSTRUTTORE] [nvarchar](50) NULL,
[COSTRUTTORE] [nvarchar](80) NULL,
[ID_CONTO_FORNITORE] [int] NULL,
[NOTE] [nvarchar](max) NULL,
[ID_TIPO_CESPITE] [int] NOT NULL,
[ID_STATO_CESPITE] [int] NULL,
[ID_CONTO_PROPRIETA] [int] NULL,
[ID_RESPONSABILE] [int] NULL,
[DATA_SCAD_GARANZIA] [datetime] NULL,
[CAMPO_MISURA] [nvarchar](80) NULL,
[CRITERI_ACC] [nvarchar](80) NULL,
[RISOLUZIONE] [nvarchar](80) NULL,
[ID_USO_STRUMENTO] [int] NULL,
[ID_REFERENTE] [int] NULL,
[FOTO] [varbinary](max) NULL,
[PROF_ID] [int] NOT NULL,
[ID_TEMPLATE] [int] NULL,
[TEMPLATE] [nvarchar](25) NULL,
[ID_PARENT_TEMPLATE_REMOTE] [int] NULL,
[ID_PARENT_TEMPLATE] [int] NULL,
[ISLOCKED] [nchar](1) NULL,
[TAGAPPSYNC] [nchar](1) NOT NULL,
[TAGAPPDOCSYNC] [nchar](1) NOT NULL,
CONSTRAINT [PK_CES_ANAGRAFICA] PRIMARY KEY CLUSTERED
(
[ID_CESPITE] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON, FILLFACTOR = 90) ON [PRIMARY]
) ON [PRIMARY]
--This is MYVIEW
CREATE VIEW [dbo].[V_CESPITE_TREE] AS
--BEGIN
WITH q AS
(
SELECT ID_CESPITE , ID_CESPITE AS ID_CESPITE_ANCESTOR
FROM CES_ANAGRAFICA c
JOIN CES_TIPI_CESPITE ctc ON ctc.ID_TIPI_INFRSTR = c.ID_TIPO_CESPITE
UNION ALL
SELECT c.ID_CESPITE, q.ID_CESPITE_ANCESTOR
FROM q
JOIN CES_ANAGRAFICA c
ON c.ID_CESPITE_PADRE = q.ID_CESPITE
JOIN CES_TIPI_CESPITE ctc ON ctc.ID_TIPI_INFRSTR = c.ID_TIPO_CESPITE
)
select ID_CESPITE AS ID_CHILD, ID_CESPITE_ANCESTOR AS ID_PARENT from q
GO
-- So my original query was this:
SELECT
CA.ID_CESPITE,CASE WHEN (EXISTS (SELECT VCA.ID_CHILD FROM V_CESPITE_TREE VCA INNER JOIN MAN_PRG_OPERAZIONI MPO ON MPO.ID_CESPITE = VCA.ID_CHILD WHERE VCA.ID_PARENT = CA.ID_CESPITE AND ID_FATHER IS NOT NULL)) THEN 'Y' ELSE 'N' END AS HAVE_CHILD_PRG,
CA.ID_CESPITE_PADRE
<Other Fields>
FROM
CES_ANAGRAFICA CA LEFT OUTER JOIN
CES_PERMESSI CP ON ((CA.ID_CESPITE = CP.ID_CESPITE)) INNER JOIN CES_TIPI_CESPITE CTCS ON CA.ID_TIPO_CESPITE = CTCS.ID_TIPI_INFRSTR LEFT OUTER JOIN
V_UTENTI_DIPENDENTI VUD ON CA.ID_RESPONSABILE = VUD.ID_DIPENDENTE
【问题讨论】：
带前缀CA的引用表是什么？ 嵌套查询与mytable2 有何关系？
您能否更具体地说明您希望在解决方案中看到什么？
你能说出每张表有多少行吗？
我认为您缺少一些引用的表。指示性行数确实也很有帮助。我假设 MT2 是 MYTABLE2 的别名？此外，这必须是单查询解决方案还是我们可以使用中间临时表和/或表变量？
【参考方案1】：
SELECT
CA.ID_CESPITE,
CASE WHEN child.[Count] > 0 THEN 'Y' ELSE 'N' END AS HAVE_CHILD_PRG,
CA.ID_CESPITE_PADRE
FROM
CES_ANAGRAFICA CA
LEFT OUTER JOIN CES_PERMESSI CP
ON ((CA.ID_CESPITE = CP.ID_CESPITE))
INNER JOIN CES_TIPI_CESPITE CTCS
ON CA.ID_TIPO_CESPITE = CTCS.ID_TIPI_INFRSTR
LEFT OUTER JOIN V_UTENTI_DIPENDENTI VUD
ON CA.ID_RESPONSABILE = VUD.ID_DIPENDENTE
CROSS APPLY (
SELECT [Count] = COUNT(*)
FROM V_CESPITE_TREE VCA
JOIN MAN_PRG_OPERAZIONI MPO
ON MPO.ID_CESPITE = VCA.ID_CHILD
WHERE VCA.ID_PARENT = CA.ID_CESPITE
AND MPO.ID_FATHER is not null
) child
这应该会稍微好一些，因为它不必执行聚合函数。
SELECT
CA.ID_CESPITE,
CASE WHEN child.[Exists] = 1 THEN 'Y' ELSE 'N' END AS HAVE_CHILD_PRG,
CA.ID_CESPITE_PADRE
FROM
CES_ANAGRAFICA CA
LEFT OUTER JOIN CES_PERMESSI CP
ON ((CA.ID_CESPITE = CP.ID_CESPITE))
INNER JOIN CES_TIPI_CESPITE CTCS
ON CA.ID_TIPO_CESPITE = CTCS.ID_TIPI_INFRSTR
LEFT OUTER JOIN V_UTENTI_DIPENDENTI VUD
ON CA.ID_RESPONSABILE = VUD.ID_DIPENDENTE
OUTER APPLY (
SELECT TOP (1) 1 [Exists]
FROM V_CESPITE_TREE VCA
JOIN MAN_PRG_OPERAZIONI MPO
ON MPO.ID_CESPITE = VCA.ID_CHILD
WHERE VCA.ID_PARENT = CA.ID_CESPITE
AND MPO.ID_FATHER is not null
) child
【讨论】：
也许它会起作用，CROSS JOIN 和 CROSS APPLY 让我害怕进一步阅读......随着数据量的增长，计算量将成倍增长。
这适用于 4500 条记录（考虑到这对我来说是现实生活的上限）。即使你应得的，我也不会奖励你，因为 David Khaykin 帮助了我很多。谢谢。
@user193655 你能告诉我第二个查询是否更好吗？【参考方案2】：
使用 CTE ;它会加载一次你的内部连接，然后缓存它。
注意我不知道 CA.ID_NODE 来自哪里，因为你没有解释。 您的内部查询也连接到 MyTable1，但您没有关联 MyTable2 和子查询。
鉴于您提供的内容，伪代码应该是这样的： （如果您使用相关信息更新您的问题，我将更新此答案以反映它）。
更新：
这是基于您的架构更新和一些示例数据的更新版本。我确认这不再导致重复。真正的问题是您检查它并确保它提高性能，也就是降低执行时间。
; with hasChildCte(ID_CESPITE, ID_PARENT)
As (
SELECT VCA.ID_CHILD,
vca.ID_PARENT
FROM V_CESPITE_TREE VCA
INNER JOIN MAN_PRG_OPERAZIONI MPO
ON MPO.ID_CESPITE = VCA.ID_CHILD
WHERE ID_FATHER
IS NOT NULL
)
Select
CA.ID_CESPITE,
Case
When Exists (
Select ID_PARENT
From hasChildCte cte
Where cte.ID_PARENT = ca.ID_CESPITE
) Then 'Y'
Else 'N'
End As HAVE_CHILDREN,
CA.ID_CESPITE_PADRE
From CES_ANAGRAFICA CA
另请注意，如果您没有在所有连接列上都有索引，那么将它们放入其中是一个不错的举措。这将有助于进一步加快查询速度，尤其是在您处理大量数据。
更新 2
关于 CTE 执行不止一次的评论让我思考，显然由 SQL Server 决定是否缓存 CTE，而不是总是缓存。在许多情况下，CTE 只会执行一次，但其他时候它类似于 SQL Server 中的视图并且不会被缓存。
因此，我修改了代码以改用table variable。不过，我没有足够的测试数据来查看哪个性能更好或更快。
试试这个，看看它是否会产生更快的查询执行时间。还要注意，无论您选择哪种重构和性能改进方法，最好使用您在 JOIN 中使用的列上的索引正确设置您的数据库。这会显着增加查询执行时间，但需要更新索引的插入成本。
更新的非 CTE 代码，使用表变量代替：
Declare @HasChildren table (ID_CESPITE int, ID_PARENT int)
Insert into @HasChildren
Select VCA.ID_CHILD,
vca.ID_PARENT
From V_CESPITE_TREE VCA
Inner Join MAN_PRG_OPERAZIONI MPO
On MPO.ID_CESPITE = VCA.ID_CHILD
Where ID_FATHER
Is Not Null
Select
CA.ID_CESPITE,
Case
When Exists (
Select ID_PARENT
From @HasChildren c
Where c.ID_PARENT = ca.ID_CESPITE
) Then 'Y'
Else 'N'
End As HAVE_CHILDREN,
CA.ID_CESPITE_PADRE
From CES_ANAGRAFICA CA
【讨论】：
感谢您的回复，正如您所指出的，由于我的示例不完美，您的建议不起作用。我对其进行了修改，您会在原始版本的下方找到修改后的版本。 @user193655 我更新了答案以反映您的新信息，试试吧。如果仍然难以使其工作，则需要发布 MyTable1、MyTable2 和 MyView 的架构 再次感谢，这一次我至少可以尝试一下，无论如何我仍然看到不同之处（您的查询中的记录比我原来的多 50 条）。我将发布原始表结构（所以从现在开始，我指的是原始名称而不是“模拟的名称”）。我还将开始赏金以奖励您应得的声誉。不幸的是，我是 CTE 的新手，所以我不能说太多. 太棒了！查询结果现在真的是一样的。 +500 给你！ 太棒了，只是好奇为您节省了多少执行时间？【参考方案3】：对于优化器来说，下面这样的事情可能更容易处理。
SELECT DISTINCT
ID_NODE,
CASE WHEN MTSub.ID_CHILD is null then 'N' else 'Y' END,
OTHERFIELDS
FROM MYTABLE2
LEFT JOIN
(SELECT MV.ID_CHILD FROM MYVIEW MV INNER JOIN MYTABLE1 MT1 ON MT1.ID_NODE = MV.ID_CHILD WHERE MV.ID_PARENT = MT2.ID_NODE AND MT1.ID_FATHER IS NOT NULL) MTSub
【讨论】：
您缺少 LEFT JOIN 的连接谓词以上是关于如何使用嵌套循环加速查询的主要内容，如果未能解决你的问题，请参考以下文章