让实体框架与 SQL Server 地理空间索引一起工作

Question

我们有一个相当简单的表，我们正在尝试使用 Entity Framework 查询它，它看起来像这样：

CREATE TABLE [dbo].[EFBlkFireRisks](
    [EFBlkFireRiskId] [int] IDENTITY(1,1) NOT NULL,
    [EFStateId] [int] NOT NULL,
    [Lat] [decimal](18, 8) NOT NULL,
    [Lon] [decimal](18, 8) NOT NULL,
    [AdjustedPremiumMultiplier] [decimal](18, 4) NOT NULL DEFAULT ((0)),
    [GeoLocation] [geography] NULL,
 CONSTRAINT [PK_dbo.EFBlkFireRisks] PRIMARY KEY CLUSTERED 
(
    [EFBlkFireRiskId] ASC
)

它在 GeoLocation 列上有一个地理空间索引，如下所示：

CREATE SPATIAL INDEX [IX_EFBlkFireRisks_Spatial] ON [dbo].[EFBlkFireRisks]
(
    [GeoLocation]
)USING  GEOGRAPHY_AUTO_GRID

我们正在尝试在该表中找到最接近给定纬度/经度点的条目。

C# 查询如下所示：

var geoPoint = SqlSchemaFunctions.GetDbGeography(lat, lon);
var query = ctx.BulkFireRisks
    .Where(x => x.EFStateId == stateId && x.GeoLocation.Distance(geoPoint) != null)
    .OrderBy(x => x.GeoLocation.Distance(geoPoint));
var bfr = query.First();

查询生成的 SQL 基本上是这样的：

declare @p3 sys.geography set @p3=convert(sys.geography,0xE6100000010C8FC2F5285CFF434052B81E85EB515AC0)
declare @p4 sys.geography set @p4=convert(sys.geography,0xE6100000010C8FC2F5285CFF434052B81E85EB515AC0)
exec sp_executesql N'SELECT TOP (1) 
    [Project1].[EFBlkFireRiskId] AS [EFBlkFireRiskId], 
    [Project1].[EFStateId] AS [EFStateId], 
    [Project1].[GeoId] AS [GeoId], 
    [Project1].[Lat] AS [Lat], 
    [Project1].[Lon] AS [Lon], 
    [Project1].[GeoLocation] AS [GeoLocation], 
    [Project1].[AdjustedPremiumMultiplier] AS [AdjustedPremiumMultiplier],
    [Project1].C1 as Distance
    FROM ( SELECT 
        [Extent1].[GeoLocation].STDistance(@p__linq__1) AS [C1], 
        [Extent1].[EFBlkFireRiskId] AS [EFBlkFireRiskId], 
        [Extent1].[EFStateId] AS [EFStateId], 
        [Extent1].[GeoId] AS [GeoId], 
        [Extent1].[Lat] AS [Lat], 
        [Extent1].[Lon] AS [Lon], 
        [Extent1].[GeoLocation] AS [GeoLocation], 
        [Extent1].[AdjustedPremiumMultiplier] AS [AdjustedPremiumMultiplier]
        FROM [dbo].[EFBlkFireRisks] AS [Extent1]
        WHERE (11 = [Extent1].[EFStateId]) AND ([Extent1].[GeoLocation].STDistance(@p__linq__0) IS NOT NULL)
    )  AS [Project1]
    ORDER BY [Project1].[C1] ASC',N'@p__linq__0 [geography],@p__linq__1 [geography]',@p__linq__0=@p3,@p__linq__1=@p4

并且该查询不使用索引 - 将其置于循环中并运行 10 次大约需要 30 秒（无论我是通过 SQL 手动执行它，还是通过 C# 中的 EF 运行它） .

但是，可以删除sp_executesql 位，然后直接运行查询，如下所示：

declare @p3 sys.geography set @p3=convert(sys.geography,0xE6100000010C8FC2F5285CFF434052B81E85EB515AC0)
SELECT TOP (1) 
    [Project1].[EFBlkFireRiskId] AS [EFBlkFireRiskId], 
    [Project1].[EFStateId] AS [EFStateId], 
    [Project1].[GeoId] AS [GeoId], 
    [Project1].[Lat] AS [Lat], 
    [Project1].[Lon] AS [Lon], 
    [Project1].[GeoLocation] AS [GeoLocation], 
    [Project1].[AdjustedPremiumMultiplier] AS [AdjustedPremiumMultiplier],
    [Project1].C1 as Distance
    FROM ( SELECT 
        [Extent1].[GeoLocation].STDistance(@p3) AS [C1], 
        [Extent1].[EFBlkFireRiskId] AS [EFBlkFireRiskId], 
        [Extent1].[EFStateId] AS [EFStateId], 
        [Extent1].[GeoId] AS [GeoId], 
        [Extent1].[Lat] AS [Lat], 
        [Extent1].[Lon] AS [Lon], 
        [Extent1].[GeoLocation] AS [GeoLocation], 
        [Extent1].[AdjustedPremiumMultiplier] AS [AdjustedPremiumMultiplier]
        FROM [dbo].[EFBlkFireRisks] AS [Extent1]
        WHERE (11 = [Extent1].[EFStateId]) AND ([Extent1].[GeoLocation].STDistance(@p3) IS NOT NULL)
    )  AS [Project1]
    ORDER BY [Project1].[C1] asc

而那个查询运行的时间不是 30 秒，而是 30 毫秒。

我在挠头。知道为什么这两个查询具有如此完全不同的执行模式吗？知道如何让 EF 使用这个特定的地理空间索引吗？我做错了什么？

编辑 6/18 - 如果我尝试向较慢的 sp_executesql 查询添加索引提示，我会收到以下错误消息：

查询处理器无法为带有空间索引提示的查询生成查询计划。原因：空间索引不支持谓词中提供的比较器。

同样的提示在另一个查询中有效 - 或者至少不会出错。

慢查询的查询计划如下：

换句话说，它没有获取地理空间索引。快速版本的查询计划正在获取索引，并且看起来像您期望的那样：

这是在 SQL Server 2014 上运行的：

(Microsoft SQL Server 2014 - 12.0.2000.8 (X64) 2014 年 2 月 20 日 20:04:26 版权所有 (c) 微软公司 Windows NT 6.3（内部版本 9600：）上的 Express 版（64 位）

Answer 1

好的，我想通了。

事实证明这两个查询并不完全相同：sp_executesql 版本有 两个 参数，而快速版本只有一个：这是关键。这个查询（只有一个参数）像你期望的那样执行：

declare @p3 sys.geography
set @p3=convert(sys.geography,0xE6100000010C8FC2F5285CFF434052B81E85EB515AC0)
exec sp_executesql N'SELECT TOP (1) 
    [Project1].[EFBlkFireRiskId] AS [EFBlkFireRiskId], 
    [Project1].[EFStateId] AS [EFStateId], 
    [Project1].[GeoId] AS [GeoId], 
    [Project1].[Lat] AS [Lat], 
    [Project1].[Lon] AS [Lon], 
    [Project1].[GeoLocation] AS [GeoLocation], 
    [Project1].[AdjustedPremiumMultiplier] AS [AdjustedPremiumMultiplier]
    FROM ( SELECT 
        [Extent1].[EFBlkFireRiskId] AS [EFBlkFireRiskId], 
        [Extent1].[EFStateId] AS [EFStateId], 
        [Extent1].[GeoId] AS [GeoId], 
        [Extent1].[Lat] AS [Lat], 
        [Extent1].[Lon] AS [Lon], 
        [Extent1].[GeoLocation] AS [GeoLocation], 
        [Extent1].[AdjustedPremiumMultiplier] AS [AdjustedPremiumMultiplier], 
        [Extent1].[GeoLocation].STDistance(@p__linq__0) AS [C1]
        FROM [dbo].[EFBlkFireRisks] AS [Extent1]
        WHERE (11 = [Extent1].[EFStateId]) AND ([Extent1].[GeoLocation].STDistance(@p__linq__0) IS NOT NULL)
    )  AS [Project1]
    ORDER BY [Project1].[C1] ASC',N'@p__linq__0 [geography]',@p__linq__0=@p3

在 LINQ 中获取该查询的方法是这样的：

var geoPoint = DbGeography.PointFromText(string.Format("POINT({0} {1})", lon, lat), 4326);
var q2 = (from b in ctx.BulkFireRisks
          let distance = b.GeoLocation.Distance(geoPoint)
          where b.EFStateId == stateId && distance != null
          orderby distance
          select b);
var bfr = q2.First();

关键是只需定义任何sys.geography 变量一次。