SQL Server 的 DMV 建议的缺失索引数量限制的解决方法

Question

在生产中，每个 SQL Server 实例都有超过 250 个数据库（我们称之为“OrgDb”）。我目前正在进行的项目旨在将 SQL Server 的 DMV 报告的所有缺失索引发送到遥测，以便对发送到这些 orgDb 的查询的执行情况进行一些后期分析，并可能进行一些优化。

听起来很简单，对吧？然而，问题是DMV 可以在单个 SQL Server 中报告的缺失索引数量的最大限制为 500，而我们预计每个 OrgDb 大约有 20 个缺失索引（总共约 5000 个）。

任何人都可以想出解决此限制的方法吗？我最初想到的一个解决方案是删除 DMV 缺少的索引表：

sys.dm_db_missing_index_details
sys.dm_db_missing_index_groups
sys.dm_db_missing_index_group

每次更新后，结果这些表无法修改：

Error:Ad hoc updates to system catalogs are not allowed.

Answer 1

您无法重置 DMV，但是您可以通过在 DMV 中提到的表上创建一个小的过滤索引然后立即删除该索引来解决此限制并从 DMV 中删除行。

例如：

CREATE INDEX IX_temp
ON dbo.SomeTable(SomeKey)
WHERE SomeKey IS NULL;

DROP INDEX dbo.SomeTable.IX_temp;

我创建了一个脚本来自动化这个过程。

IF OBJECT_ID('dbo.RemoveMissingIndexSuggestions') IS NOT NULL
DROP PROCEDURE RemoveMissingIndexSuggestions;
GO
CREATE PROCEDURE dbo.RemoveMissingIndexSuggestions
(
    @Database SYSNAME = NULL --optional, if NULL, clear all suggestions
                             --if specified, only clear suggestions for that database
    , @Table SYSNAME = NULL --if not NULL, only clear suggestions for the specified table 
)
AS
BEGIN
    /*
        Max Vernon, 2016-04-08
        Inspired by work by Joe Sack and Glenn Berry at
        http://www.sqlskills.com/blogs/joe/clearing-missing-index-suggestions-for-a-single-table/

        Creates one index for each table that is mentioned in sys.dm_db_missing_index_details
        then promply drops that index.  The index is created with a WHERE clause that is likely 
        to eliminate all or almost all rows, and therefore will be created quite quickly.
    */
    SET NOCOUNT ON;
    DECLARE @ObjectName SYSNAME;
    DECLARE @DatabaseName SYSNAME;
    DECLARE @CreateStmt NVARCHAR(MAX);
    DECLARE @DropStmt NVARCHAR(MAX);
    DECLARE @stmt NVARCHAR(MAX);
    DECLARE @msg NVARCHAR(2000);
    DECLARE @vars NVARCHAR(1000);
    DECLARE @Uniquifier NVARCHAR(48);
    SET @vars = '@stmt NVARCHAR(MAX)';
    DECLARE cur CURSOR LOCAL FORWARD_ONLY STATIC
    FOR
    WITH cte AS
    (
        SELECT ObjectName = d.name + '.' + s.name + '.' + o.name
            , DatabaseName = d.name
            , CreateStmt = N'CREATE INDEX [IX_temp] 
ON ' + QUOTENAME(s.name) + N'.' + QUOTENAME(o.name) + N'(' + mid.equality_columns + N') 
WHERE ' 
              + (
                SELECT TOP(1) cols.ColName FROM (
                    SELECT TOP(1) ColName = QUOTENAME(c.name) + N' IS NULL'
                    FROM sys.columns c 
                        INNER JOIN sys.key_constraints kc ON c.object_id = kc.parent_object_id 
                    WHERE c.object_id = o.object_id 
                        AND kc.type_desc = N'PRIMARY_KEY_CONSTRAINT'
                    UNION ALL
                    SELECT TOP(1) QUOTENAME(c.name) + N' = -2147483648'
                    FROM sys.columns c
                        INNER JOIN sys.types ty ON c.system_type_id = ty.system_type_id
                    WHERE ty.name IN 
                        (
                              N'bigint'
                            , N'binary'
                            , N'hierarchyid'
                            , N'int'
                            , N'uniqueidentifier'
                            , N'varbinary'
                        )
                    ) cols
                ) 
                + ';'
                , DropStmt = N'DROP INDEX ' + QUOTENAME(s.name) + N'.' + QUOTENAME(o.name) + '.[IX_temp];'
                , rn = ROW_NUMBER() OVER (PARTITION BY mid.object_id ORDER BY mid.index_handle)
        FROM sys.dm_db_missing_index_details mid
            INNER JOIN sys.objects o ON mid.object_id = o.object_id
            INNER JOIN sys.schemas s ON o.schema_id = s.schema_id
            INNER JOIN sys.databases d ON mid.database_id = d.database_id
        WHERE o.name NOT LIKE '#%' -- ignore temp tables
            AND (d.name = @Database OR @Database IS NULL)
            AND (o.name = @Table OR @Table IS NULL)
    )
    SELECT cte.ObjectName
        , cte.DatabaseName
        , cte.CreateStmt
        , cte.DropStmt
    FROM cte
    WHERE rn = 1;
    OPEN cur;
    FETCH NEXT FROM cur INTO @ObjectName, @DatabaseName, @CreateStmt, @DropStmt;
    WHILE @@FETCH_STATUS = 0
    BEGIN
        SET @msg = 'Flushing ' + @ObjectName + ' indexes.

';
        RAISERROR (@msg, 0, 1) WITH NOWAIT;
        SET @stmt = 'EXEC ' + QUOTENAME(@DatabaseName) + '.sys.sp_executesql @stmt;'
        SET @Uniquifier = CONVERT(NVARCHAR(48), NEWID(), 0);
        SET @CreateStmt = REPLACE(@CreateStmt, '[IX_Temp]', '[IX_Temp_' + @Uniquifier + ']')
        SET @DropStmt = REPLACE(@DropStmt, '[IX_Temp]', '[IX_Temp_' + @Uniquifier + ']')
        SET @CreateStmt = 'SET ANSI_NULLS ON;
SET ANSI_PADDING ON;
SET ANSI_WARNINGS ON;
SET ARITHABORT ON;
SET CONCAT_NULL_YIELDS_NULL ON;
SET NUMERIC_ROUNDABORT OFF;
SET QUOTED_IDENTIFIER ON;
' + @CreateStmt + '

';
        SET @DropStmt = 'SET ANSI_NULLS ON;
SET ANSI_PADDING ON;
SET ANSI_WARNINGS ON;
SET ARITHABORT ON;
SET CONCAT_NULL_YIELDS_NULL ON;
SET NUMERIC_ROUNDABORT OFF;
SET QUOTED_IDENTIFIER ON;
' + @DropStmt + '

';
        RAISERROR (@CreateStmt, 0, 1) WITH NOWAIT;
        RAISERROR (@DropStmt, 0, 1) WITH NOWAIT;
        EXEC sp_executesql @stmt, @vars, @stmt = @CreateStmt;
        EXEC sp_executesql @stmt, @vars, @stmt = @DropStmt;
        FETCH NEXT FROM cur INTO @ObjectName, @DatabaseName, @CreateStmt, @DropStmt;
    END
    CLOSE cur;
    DEALLOCATE cur;
END
GO

这可以在@Database参数设置为数据库名称的情况下执行以消除仅与该数据库相关的推荐索引，或者不使用参数以消除所有推荐索引。通过将表的名称传递到@Table参数中，可以选择性地将其限制为单个表的建议。

EXEC dbo.RemoveMissingIndexSuggestions @Database = 'tempdb', @Table = 'SomeTable';

它最多为每个表创建一个索引。索引具有唯一的名称，并使用单个列构造，优先使用表的PRIMARY KEY，如果有的话。如果表没有主键或至少以下类型的列之一，则此过程将丢失表：

bigint
binary
hierarchyid
int
uniqueidentifier
varbinary

我在SQLServerScience上写了一篇关于这个问题的简短博客文章