SQL Server - 存储过程突然变慢

Question

我昨天编写了一个存储过程，通常在一秒钟内完成。今天，大约需要 18 秒。我昨天也遇到了这个问题，似乎可以通过删除并重新创建存储过程来解决。如今，这个伎俩似乎不起作用了。 :(

有趣的是，如果我复制存储过程的主体并将其作为简单查询执行，它会很快完成。事实似乎是它是一个存储过程，导致速度减慢......！

有谁知道问题可能是什么我已经搜索过答案，但他们通常建议通过查询分析器运行它，但我没有 - 我现在使用 SQL Server 2008 Express

如下；

ALTER PROCEDURE [dbo].[spGetPOIs]
    @lat1 float,
    @lon1 float,
    @lat2 float,
    @lon2 float,
    @minLOD tinyint, 
    @maxLOD tinyint,
    @exact bit
AS
BEGIN
    -- Create the query rectangle as a polygon
    DECLARE @bounds geography;
    SET @bounds = dbo.fnGetRectangleGeographyFromLatLons(@lat1, @lon1, @lat2, @lon2);

    -- Perform the selection
    if (@exact = 0)
    BEGIN
        SELECT [ID], [Name], [Type], [Data], [MinLOD], [MaxLOD], [Location].[Lat] AS [Latitude], [Location].[Long] AS [Longitude], [SourceID]
        FROM [POIs]
        WHERE
            NOT ((@maxLOD  [MaxLOD])) AND
            (@bounds.Filter([Location]) = 1)
    END
    ELSE
    BEGIN
        SELECT [ID], [Name], [Type], [Data], [MinLOD], [MaxLOD], [Location].[Lat] AS [Latitude], [Location].[Long] AS [Longitude], [SourceID]
        FROM [POIs]
        WHERE
            NOT ((@maxLOD  [MaxLOD])) AND
            (@bounds.STIntersects([Location]) = 1)
    END

END

存储过程 表具有 MinLOD、MaxLOD 索引和 Location 空间索引。

Answer 1

啊，难道查询计划很糟糕吗？

SP 的编译/查询计划在首次使用时确定 - 取决于参数。因此，第一次调用的参数（当不存在计划时）确定查询计划。有一次我从缓存中删除，生成了新计划。

下次运行缓慢时，可能会使用查询分析器进行调用并获取选定的计划 - 并检查它的外观。

如果是这样 - 添加一个选项，在每次调用时重新编译 SP（使用重新编译）。

Answer 2

parameter sniffing 谷歌一下。 try this, which will "remap" the input parameters to local variables to prevent SQL Server from trying to guess the query plan based on parameters:

ALTER PROCEDURE [dbo].[spGetPOIs]
    @lat1 float,
    @lon1 float,
    @lat2 float,
    @lon2 float,
    @minLOD tinyint, 
    @maxLOD tinyint,
    @exact bit
AS
BEGIN
DECLARE @X_lat1 float,
    @X_lon1 float,
    @X_lat2 float,
    @X_lon2 float,
    @X_minLOD tinyint, 
    @X_maxLOD tinyint,
    @X_exact bit



    -- Create the query rectangle as a polygon
    DECLARE @bounds geography;
    SET @bounds = dbo.fnGetRectangleGeographyFromLatLons(@X_lat1, @X_lon1, @lX_at2, @X_lon2);

    -- Perform the selection
    if (@exact = 0)
    BEGIN
        SELECT [ID], [Name], [Type], [Data], [MinLOD], [MaxLOD], [Location].[Lat] AS [Latitude], [Location].[Long] AS [Longitude], [SourceID]
        FROM [POIs]
        WHERE
            NOT ((@X_maxLOD  [MaxLOD])) AND
            (@bounds.Filter([Location]) = 1)
    END
    ELSE
    BEGIN
        SELECT [ID], [Name], [Type], [Data], [MinLOD], [MaxLOD], [Location].[Lat] AS [Latitude], [Location].[Long] AS [Longitude], [SourceID]
        FROM [POIs]
        WHERE
            NOT ((@X_maxLOD  [MaxLOD])) AND
            (@bounds.STIntersects([Location]) = 1)
    END

END

Answer 3

我有一个类似的问题，它与索引有关。
重建它们可以帮助 SP 再次快速运行。

我在此处找到了解决方案

USE master;
GO

CREATE PROC DatabaseReIndex(@Database VARCHAR(100)) AS 
BEGIN
  DECLARE @DbID SMALLINT=DB_ID(@Database)--Get Database ID
  IF EXISTS(SELECT * FROM tempdb.sys.objects WHERE name='Indexes') 
  BEGIN --Delete Temp Table if exists, then create
    DROP TABLE TempDb.dbo.Indexes
  END

CREATE TABLE TempDb.dbo.Indexes(IndexTempID INT IDENTITY(1,1),SchemaName NVARCHAR(128),TableName NVARCHAR(128),IndexName NVARCHAR(128),IndexFrag FLOAT)
EXEC ('USE '+@Database+';
INSERT INTO TempDb.dbo.Indexes(TableName,SchemaName,IndexName,IndexFrag)
SELECT OBJECT_NAME(ind.OBJECT_ID) AS TableName,sch.name,ind.name IndexName,indexstats.avg_fragmentation_in_percent
FROM sys.dm_db_index_physical_stats('+@DbID+', NULL, NULL, NULL, NULL) indexstats
INNER JOIN sys.indexes ind ON ind.object_id = indexstats.object_id AND ind.index_id = indexstats.index_id
INNER JOIN sys.objects obj on obj.object_id=indexstats.object_id
INNER JOIN sys.schemas as sch ON sch.schema_id = obj.schema_id
WHERE indexstats.avg_fragmentation_in_percent > 10 AND indexstats.index_type_desc<>''HEAP''
ORDER BY indexstats.avg_fragmentation_in_percent DESC')--Get index data and fragmentation, set the percentage as high or low as you need

DECLARE @IndexTempID BIGINT=0,@SchemaName NVARCHAR(128),@TableName NVARCHAR(128),@IndexName NVARCHAR(128),@IndexFrag FLOAT
SELECT * FROM TempDb.dbo.Indexes --View your results, comment out if not needed...

-- Loop through the indexes
WHILE @IndexTempID IS NOT NULL 
  BEGIN
    SELECT @SchemaName=SchemaName,@TableName=TableName,@IndexName=IndexName,@IndexFrag=IndexFrag FROM TempDb.dbo.Indexes WHERE IndexTempID=@IndexTempID
    IF @IndexName IS NOT NULL AND @SchemaName IS NOT NULL AND @TableName IS NOT NULL 
    BEGIN
      IF @IndexFrag<30. 
      BEGIN --Low fragmentation can use re-organise, set at 30 as per most articles
        PRINT 'USE '+@Database+'; ALTER INDEX ' + @IndexName + N' ON ' + @SchemaName + N'.' + @TableName + N' REORGANIZE'
        EXEC('USE '+@Database+'; ALTER INDEX ' + @IndexName + N' ON ' + @SchemaName + N'.' + @TableName + N' REORGANIZE')
      END
    ELSE 
      BEGIN --High fragmentation needs re-build
        PRINT 'USE '+@Database+'; ALTER INDEX ' + @IndexName + N' ON ' + @SchemaName + N'.' + @TableName + N' REBUILD'
        EXEC('USE '+@Database+'; ALTER INDEX ' + @IndexName + N' ON ' + @SchemaName + N'.' + @TableName + N' REBUILD')
      END
    END

    SET @IndexTempID=(SELECT MIN(IndexTempID) FROM TempDb.dbo.Indexes WHERE IndexTempID>@IndexTempID)
  END
END

DROP TABLE TempDb.dbo.Indexes

GO