Question

我有以下查询，我在我的数据库服务器上运行，但运行大约需要 30秒，我无法理解为什么会这样。

SELECT *
FROM [dbo].[PackageInstance] AS packInst
  INNER JOIN [dbo].[PackageDefinition] AS packageDef 
    ON packInst.[PackageDefinitionID] = packageDef.[PackageDefinitionID]
  LEFT OUTER JOIN [dbo].[PackageInstanceContextDef] AS contextDef 
   ON packInst.[PackageInstanceID] = contextDef.[PackageInstanceID]

这产生了以下执行计划，对我来说看起来很好....所以我无法理解为什么需要这么多时间来执行结果数据只有100,000条记录（这应该是一个步骤park for SQL Server）。

SQL Server Execution Plan

任何想法可能导致这么长的执行时间？

我查看了Profiler中的查询，看看其中的统计信息及其位置如下：

CPU - 4711
Reads - 744453
Writes - 9
Duration - 26329

以下是表格定义：

CREATE TABLE [dbo].[PackageDefinition](
    [PackageDefinitionID] [int] IDENTITY(1,1) NOT NULL,
    [ts] [timestamp] NOT NULL,
    [ProgramID] [int] NULL,
    [VendorID] [int] NULL,
    [PackageExecutionTypeID] [int] NULL,
    [PackageDefinitionStatusID] [int] NOT NULL,
    [IsInternal] [bit] NOT NULL,
    [Name] [dbo].[D_Name] NOT NULL,
    [Description] [dbo].[D_Description] NOT NULL,
    [CreatedDate] [datetime] NOT NULL,
    [PublishedDate] [datetime] NULL,
    [OwnerUserGuid] [uniqueidentifier] NOT NULL,
    [ProcessDefinitionMainID] [int] NULL,
    [KeyInfoHtml] [nvarchar](max) NULL,
    [DescriptionHtml] [nvarchar](max) NULL,
    [WhatToExpectHtml] [nvarchar](max) NULL,
    [BestPracticesHtml] [nvarchar](max) NULL,
    [RecommendedJourneysHtml] [nvarchar](max) NULL,
    [RequiresSLAAgreement] [bit] NOT NULL,
    [SLAFileAssetID] [int] NULL,
    [ImageDataID] [int] NULL,
    [VideoHtml] [nvarchar](max) NULL,
    [VideoAssetID] [int] NULL,
    [UseMapCosts] [bit] NOT NULL,
    [CostMin] [money] NOT NULL,
    [CostMax] [money] NOT NULL,
    [LandingPageVisitCount] [int] NOT NULL,
    [IsDeleted] [dbo].[D_IsDeleted] NOT NULL,
    [CreatedByUserGuid] [uniqueidentifier] NOT NULL,
    [OrderHtml] [nvarchar](max) NULL,
 CONSTRAINT [PK_PackageDefinition] PRIMARY KEY CLUSTERED 
(
    [PackageDefinitionID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]

CREATE TABLE [dbo].[PackageInstance](
    [PackageInstanceID] [int] IDENTITY(1,1) NOT NULL,
    [ts] [timestamp] NOT NULL,
    [PackageDefinitionID] [int] NOT NULL,
    [PackageStatusID] [int] NOT NULL,
    [Name] [dbo].[D_Description] NOT NULL,
    [CampaignID] [int] NULL,
    [MarketingPlanID] [int] NULL,
    [CountryID] [int] NULL,
    [DateEntered] [datetime] NULL,
    [DateExecuted] [datetime] NULL,
    [ProcessID] [int] NULL,
    [OrderedByUserGuid] [uniqueidentifier] NULL,
    [RequestedByUserGuid] [uniqueidentifier] NULL,
    [SLAEndDate] [datetime] NULL,
 CONSTRAINT [PK_PackageInstance] PRIMARY KEY CLUSTERED 
(
    [PackageInstanceID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]

CREATE TABLE [dbo].[PackageInstanceContextDef](
    [PackageInstanceContextDefID] [int] IDENTITY(1,1) NOT NULL,
    [ts] [timestamp] NOT NULL,
    [PackageInstanceID] [int] NOT NULL,
    [ContextObjectDefID] [int] NOT NULL,
    [EnteredFieldValue] [varchar](max) NULL,
    [SelectedListValueID] [int] NULL,
    [AssetIdsString] [nvarchar](max) NULL,
    [SelectedListValueIdsString] [nvarchar](max) NULL,
    [ContextObjectFieldName] [nvarchar](30) NOT NULL,
 CONSTRAINT [PK_PackageInstanceContextDef] PRIMARY KEY CLUSTERED 
(
    [PackageInstanceContextDefID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]

Answer 1

删除*

中的SELECT *

它将始终扫描，因为您要求所有列。你有聚集索引吗？

Answer 2

答案结果证明是@MartinSmith的建议。因为PackageDefinition表包含大约8个NVARCHAR（MAX）列，所以当创建结果连接并且超过100k行时，这导致varchar（max）值被重复读取并且它们存在于行外页面。因此，大量的逻辑读取。

感谢大家的支持，只需要让实体框架产生我想要的查询。

Answer 3

如果添加以下索引会发生什么......

CREATE NONCLUSTERED INDEX ix ON  PackageDefinition(PackageDefinitionID)

...并尝试以下方法来减少进入排序的数据宽度？

SELECT packInst.*,
       packageDef2.*,
       contextDef.*
FROM   [dbo].[PackageInstance] AS packInst
       INNER MERGE JOIN [dbo].[PackageDefinition] AS packageDef
         ON packInst.[PackageDefinitionID] = packageDef.[PackageDefinitionID]
       LEFT OUTER MERGE JOIN [dbo].[PackageInstanceContextDef] AS contextDef
        ON packInst.[PackageInstanceID] = contextDef.[PackageInstanceID]
       INNER MERGE JOIN [dbo].[PackageDefinition] AS packageDef2
        ON packageDef.[PackageDefinitionID] = packageDef2.[PackageDefinitionID]

当然*不应该被用作，即使你需要所有列，你绝对不需要两次JOIN的相同列，但这只是为了保持语义你的原始查询。

为什么这次加入需要这么长时间？

3 个答案: