提高sys.dm_db_index_physical_stats的性能

在维护工作期间，我正在尝试获取碎片索引的列表。但是查询非常慢，需要30多分钟才能执行。我认为这是由于对sys.dm_db_index_physical_stats进行的远程扫描。

有什么方法可以加快以下查询的速度：

SELECT
    OBJECT_NAME(i.OBJECT_ID) AS TableName,
    i.name AS TableIndexName
FROM
    sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, 'DETAILED') phystat 
    INNER JOIN sys.indexes i 
        ON i.OBJECT_ID = phystat.OBJECT_ID AND i.index_id = phystat.index_id 
WHERE 
    phystat.avg_fragmentation_in_percent > 20 
    AND OBJECT_NAME(i.OBJECT_ID) IS NOT NULL
ORDER BY phystat.avg_fragmentation_in_percent DESC

我不是DBA，可能在上面的查询中犯了一个明显的错误，或者也许有些索引或统计信息会有所帮助？也许只是数据库的大小（大约20Gb，大约140个表）。

我问的原因是，我们在夜间只有一个很小的维护窗口，这在大多数时候都占用了。

'DETAILED'意味着对索引（或堆）中的每个页面进行全面扫描。对每个表和每个辅助索引执行此操作，结果意味着您正在端到端进行完整的数据库扫描，而效率却不是很高（例如，其读取速度不及备份读取的速度快）。时间取决于：

• 您的数据库有多大
• 您的IO子系统读取整个数据库有多快
• 并发负载争夺IO吞吐量

基本上，如果您只有一根稻草（您的IO吞吐量），那么要花30分钟才能喝一桶（您的数据库大小）。购买更快的IO，减少数据量或使用SAMPLED扫描。

话虽这么说... 20Gb很小。30分钟内读取20Gb是很多时间。就是你IO子系统是慢？您是否在7200 RPM消费者1TB驱动器上进行了部署？

除了@Remus建议使用SAMPLED扫描外，我不知道在您的维护窗口启动之前无法启动此查询。为什么不使用结果预填充表？如果您在维护窗口之前约15-20分钟启动此查询（例如，采样扫描需要10分钟），然后将结果填充到表中，则维护窗口启动后即可立即使用数据，并且同时，基础数据实际上并不会发生太大变化。如果您避免对原始查询进行排序和过滤，那么它也应该更快地完成，例如

CREATE TABLE dbo.IndexStats
(
  TableName SYSNAME,
  IndexName SYSNAME,
  Frag DECIMAL(5,2)
);
CREATE INDEX x ON dbo.IndexStats(Frag);

然后，在您的第一个夜间工作（该工作在维护窗口之前开始）中：

TRUNCATE TABLE dbo.IndexStats;

INSERT dbo.IndexStats
SELECT 
  OBJECT_NAME(i.[object_id]),
  i.name,
  s.avg_fragmentation_in_percent
FROM
  sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, 'SAMPLED') AS s 
INNER JOIN sys.indexes AS i 
ON i.[object_id] = s.[object_id]
AND i.index_id = s.index_id;

DELETE dbo.IndexStats WHERE Frag < 20
  OR TableName IS NULL;

现在，您实际的碎片整理脚本已经具有立即进行所需的所有信息。（您甚至可以将这些作业链接在一起，或者使用强制上述操作等待维护窗口的开始时间WAITFOR TIME。）

您也可以考虑一起玩，LIMITED看看票价如何。

免责声明：这些脚本已在SQL Server 2005/2008上进行了测试。但是，此代码和信息按“现状”提供，没有任何形式的明示或暗示保证，包括但不限于暗示保证或适销性和/或对特定目的的适用性。与往常一样，在尝试部署到生产环境之前，请在测试环境中进行测试。 现在那已经不合时宜了...

我在处理索引DMV时遇到的问题之一是它们无法关联。意思是，您不能对它们使用“交叉/外部应用”，以限制您要对其执行扫描的索引。为了解决这个问题，我将用于物理和操作索引DMV的包装函数部署到我的主数据库中：

物理：

ALTER FUNCTION [dbo].[tfn_IndexPhysicalStats_select]
(
    @DatabaseID SMALLINT = 0,
    @ObjectID INT = 0,
    @IndexID INT = -1,
    @PartitionNumber INT = 0,
    @Mode NVARCHAR(20) = NULL
)
RETURNS @IndexPhysicalStats TABLE
(
    database_id SMALLINT NOT NULL,
    object_id INT NOT NULL,
    index_id INT NOT NULL,
    partition_number INT NOT NULL,
    index_type_desc NVARCHAR(60) NULL,
    alloc_unit_type_desc NVARCHAR(60) NULL,
    index_depth TINYINT NOT NULL,
    index_level TINYINT NOT NULL,
    avg_fragmentation_in_percent FLOAT NULL,
    fragment_count BIGINT NULL,
    avg_fragment_size_in_pages FLOAT NULL,
    page_count BIGINT NOT NULL,
    avg_page_space_used_in_percent FLOAT NULL,
    record_count BIGINT NULL,
    ghost_record_count BIGINT NULL,
    version_ghost_record_count BIGINT NULL,
    min_record_size_in_bytes INT NULL,
    max_record_size_in_bytes INT NULL,
    avg_record_size_in_bytes FLOAT NULL,
    forwarded_record_count BIGINT NULL
)
AS
BEGIN

    INSERT INTO @IndexPhysicalStats
    (
        database_id,
        object_id,
        index_id,
        partition_number,
        index_type_desc,
        alloc_unit_type_desc,
        index_depth,
        index_level,
        avg_fragmentation_in_percent,
        fragment_count,
        avg_fragment_size_in_pages,
        page_count,
        avg_page_space_used_in_percent,
        record_count,
        ghost_record_count,
        version_ghost_record_count,
        min_record_size_in_bytes,
        max_record_size_in_bytes,
        avg_record_size_in_bytes,
        forwarded_record_count
    )
    SELECT
        ddips.database_id,
        ddips.object_id,
        ddips.index_id,
        ddips.partition_number,
        ddips.index_type_desc,
        ddips.alloc_unit_type_desc,
        ddips.index_depth,
        ddips.index_level,
        ddips.avg_fragmentation_in_percent,
        ddips.fragment_count,
        ddips.avg_fragment_size_in_pages,
        ddips.page_count,
        ddips.avg_page_space_used_in_percent,
        ddips.record_count,
        ddips.ghost_record_count,
        ddips.version_ghost_record_count,
        ddips.min_record_size_in_bytes,
        ddips.max_record_size_in_bytes,
        ddips.avg_record_size_in_bytes,
        ddips.forwarded_record_count
    FROM sys.dm_db_index_physical_stats
    (
        @DatabaseID,
        @ObjectID,
        @IndexID,
        @PartitionNumber,
        @Mode
    ) AS ddips;

    RETURN;
END

操作：

ALTER FUNCTION [dbo].[tfn_IndexOperationalStats_select]
(
    @DatabaseID SMALLINT = 0,
    @TableID INT = 0,
    @IndexID INT = -1,
    @PartitionNumber INT = 0
)
RETURNS @IndexOperationalStats TABLE
(
    database_id SMALLINT NOT NULL,
    object_id INT NOT NULL,
    index_id INT NOT NULL,
    partition_number INT NOT NULL,
    leaf_insert_count BIGINT NULL,
    leaf_delete_count BIGINT NULL,
    leaf_update_count BIGINT NULL,
    leaf_ghost_count BIGINT NULL,
    nonleaf_insert_count BIGINT NULL,
    nonleaf_delete_count BIGINT NULL,
    nonleaf_update_count BIGINT NULL,
    leaf_allocation_count BIGINT NULL,
    nonleaf_allocation_count BIGINT NULL,
    leaf_page_merge_count BIGINT NULL,
    nonleaf_page_merge_count BIGINT NULL,
    range_scan_count BIGINT NULL,
    singleton_lookup_count BIGINT NULL,
    forwarded_fetch_count BIGINT NULL,
    lob_fetch_in_pages BIGINT NULL,
    lob_fetch_in_bytes BIGINT NULL,
    lob_orphan_create_count BIGINT NULL,
    lob_orphan_insert_count BIGINT NULL,
    row_overflow_fetch_in_pages BIGINT NULL,
    row_overflow_fetch_in_bytes BIGINT NULL,
    column_value_push_off_row_count BIGINT NULL,
    column_value_pull_in_row_count BIGINT NULL,
    row_lock_count BIGINT NULL,
    row_lock_wait_count BIGINT NULL,
    row_lock_wait_in_ms BIGINT NULL,
    page_lock_count BIGINT NULL,
    page_lock_wait_count BIGINT NULL,
    page_lock_wait_in_ms BIGINT NULL,
    index_lock_promotion_attempt_count BIGINT NULL,
    index_lock_promotion_count BIGINT NULL,
    page_latch_wait_count BIGINT NULL,
    page_latch_wait_in_ms BIGINT NULL,
    page_io_latch_wait_count BIGINT NULL,
    page_io_latch_wait_in_ms BIGINT NULL
    PRIMARY KEY CLUSTERED
    (
        database_id ASC,
        object_id ASC,
        index_id ASC,
        partition_number ASC
    )
)
AS
BEGIN
    INSERT INTO @IndexOperationalStats
    (
        database_id,
        object_id,
        index_id,
        partition_number,
        leaf_insert_count,
        leaf_delete_count,
        leaf_update_count,
        leaf_ghost_count,
        nonleaf_insert_count,
        nonleaf_delete_count,
        nonleaf_update_count,
        leaf_allocation_count,
        nonleaf_allocation_count,
        leaf_page_merge_count,
        nonleaf_page_merge_count,
        range_scan_count,
        singleton_lookup_count,
        forwarded_fetch_count,
        lob_fetch_in_pages,
        lob_fetch_in_bytes,
        lob_orphan_create_count,
        lob_orphan_insert_count,
        row_overflow_fetch_in_pages,
        row_overflow_fetch_in_bytes,
        column_value_push_off_row_count,
        column_value_pull_in_row_count,
        row_lock_count,
        row_lock_wait_count,
        row_lock_wait_in_ms,
        page_lock_count,
        page_lock_wait_count,
        page_lock_wait_in_ms,
        index_lock_promotion_attempt_count,
        index_lock_promotion_count,
        page_latch_wait_count,
        page_latch_wait_in_ms,
        page_io_latch_wait_count,
        page_io_latch_wait_in_ms
    )
    SELECT
        ddios.database_id,
        ddios.object_id,
        ddios.index_id,
        ddios.partition_number,
        ddios.leaf_insert_count,
        ddios.leaf_delete_count,
        ddios.leaf_update_count,
        ddios.leaf_ghost_count,
        ddios.nonleaf_insert_count,
        ddios.nonleaf_delete_count,
        ddios.nonleaf_update_count,
        ddios.leaf_allocation_count,
        ddios.nonleaf_allocation_count,
        ddios.leaf_page_merge_count,
        ddios.nonleaf_page_merge_count,
        ddios.range_scan_count,
        ddios.singleton_lookup_count,
        ddios.forwarded_fetch_count,
        ddios.lob_fetch_in_pages,
        ddios.lob_fetch_in_bytes,
        ddios.lob_orphan_create_count,
        ddios.lob_orphan_insert_count,
        ddios.row_overflow_fetch_in_pages,
        ddios.row_overflow_fetch_in_bytes,
        ddios.column_value_push_off_row_count,
        ddios.column_value_pull_in_row_count,
        ddios.row_lock_count,
        ddios.row_lock_wait_count,
        ddios.row_lock_wait_in_ms,
        ddios.page_lock_count,
        ddios.page_lock_wait_count,
        ddios.page_lock_wait_in_ms,
        ddios.index_lock_promotion_attempt_count,
        ddios.index_lock_promotion_count,
        ddios.page_latch_wait_count,
        ddios.page_latch_wait_in_ms,
        ddios.page_io_latch_wait_count,
        ddios.page_io_latch_wait_in_ms
    FROM sys.dm_db_index_operational_stats
    (
        @DatabaseID,
        @TableID,
        @IndexID,
        @PartitionNumber
    ) AS ddios;

    RETURN;
END

然后，我以下列方式在索引维护作业中引用此函数：

DECLARE 
    @DDL NVARCHAR(MAX);

DECLARE ddl_cursor CURSOR
FOR
    SELECT
        CONVERT(NVARCHAR(MAX), DDL.DDL) AS DDL
    FROM
    (
        SELECT
            MasterIndexes.SchemaName,
            MasterIndexes.TableName,
            MasterIndexes.IndexName,
            MasterIndexes.DatabaseID,
            MasterIndexes.ObjectID,
            MasterIndexes.IndexID,
            MasterIndexes.PartitionNumber,
            MasterIndexes.type_desc,
            MasterIndexes.is_unique,
            MasterIndexes.is_primary_key,
            MasterIndexes.is_unique_constraint,
            MasterIndexes.fill_factor,
            MasterIndexes.allow_row_locks,
            MasterIndexes.allow_page_locks,
            MasterIndexes.UpdateStatisticsIndicator,
            1 AS SortInTempDB,
            CASE
                WHEN CONVERT(VARCHAR(100), SERVERPROPERTY('edition')) LIKE 'Enterprise Edition%' THEN 1
                ELSE 0
            END AS OnlineIndicator,
            CASE
                WHEN 
                    ips.avg_fragmentation_in_percent BETWEEN CONVERT(FLOAT, 10) AND CONVERT(FLOAT, 30)
                    AND ips.page_count >= 100
                THEN
                    1
                ELSE
                    0
            END AS ReorganizationIndicator,
            CASE
                WHEN
                (
                    ips.avg_fragmentation_in_percent >= 30
                    AND ips.page_count >= 100
                )
                OR
                (
                    ips.avg_fragmentation_in_percent BETWEEN CONVERT(FLOAT, 10) AND CONVERT(FLOAT, 30)
                    AND ips.page_count < 100
                )
                THEN 
                    1
                ELSE
                    0
            END AS RebuildIndicator
        FROM
        (
            SELECT
                s.name AS SchemaName,
                t.name AS TableName,
                ix.name AS IndexName,
                DB_ID() AS DatabaseID,
                ddps.object_id AS ObjectID,
                ddps.index_id AS IndexID,
                ddps.partition_number AS PartitionNumber,
                ix.type_desc,
                ix.is_unique,   
                ix.is_primary_key,
                ix.is_unique_constraint,
                ix.fill_factor, 
                ix.allow_row_locks,
                ix.allow_page_locks,
                1 AS UpdateStatisticsIndicator  
            FROM sys.schemas AS s

                INNER JOIN sys.tables AS t
                    ON s.schema_id = t.schema_id

                    INNER JOIN sys.indexes AS ix
                        ON t.object_id = ix.object_id

                        INNER JOIN sys.dm_db_partition_stats AS ddps
                            ON ix.object_id = ddps.object_id
                            AND ix.index_id = ddps.index_id

                CROSS APPLY master.dbo.tfn_IndexOperationalStats_select
                (
                    DB_ID(),
                    t.object_id,
                    ix.index_id,
                    NULL
                ) AS ios

            WHERE
                CASE
                    WHEN ddps.row_count = 0 THEN 0
                    ELSE
                    (
                        (
                            CONVERT
                            (
                                FLOAT,
                                (
                                    ios.nonleaf_insert_count + 
                                    ios.nonleaf_update_count + 
                                    ios.leaf_insert_count + 
                                    ios.leaf_update_count
                                )
                            ) /
                            CONVERT
                            (
                                FLOAT,
                                ddps.row_count
                            )
                        ) * 100.0
                    ) 
                END >= 10.0
            AND t.is_ms_shipped = 0
            AND t.name NOT LIKE 'MSmerge%'
            AND ix.index_id > 0
        ) AS MasterIndexes

            CROSS APPLY master.dbo.tfn_IndexPhysicalStats_select
            (
                MasterIndexes.DatabaseID,
                MasterIndexes.ObjectID,
                MasterIndexes.IndexID,
                MasterIndexes.PartitionNumber,
                'SAMPLED'
            ) AS ips
    ) AS MasterIndexList    

        CROSS APPLY
        (
            SELECT          
                'ALTER INDEX ' + 
                MasterIndexList.IndexName + 
                ' ON ' + 
                MasterIndexList.SchemaName + '.' + MasterIndexList.TableName +
                ' REBUILD WITH(' + 
                'FILLFACTOR = ' + 
                    CASE
                        WHEN MasterIndexList.fill_factor = 0 THEN '100'
                        ELSE CONVERT(VARCHAR(3), MasterIndexList.fill_factor)
                    END + ', ' +
                'SORT_IN_TEMPDB = ' + 
                    CASE
                        WHEN MasterIndexList.SortInTempDB = 1 THEN 'ON'
                        ELSE 'OFF'
                    END + ', ' +
                'ONLINE = ' + 
                    CASE
                        WHEN MasterIndexList.OnlineIndicator = 1 THEN 'ON'
                        ELSE 'OFF'
                    END + ', ' + 
                'ALLOW_ROW_LOCKS = ' + 
                    CASE
                        WHEN MasterIndexList.[allow_row_locks] = 1 THEN 'ON'
                        ELSE 'OFF'
                    END + ', ' + 
                'ALLOW_PAGE_LOCKS = ' + 
                    CASE
                        WHEN MasterIndexList.[allow_page_locks] = 1 THEN 'ON'
                        ELSE 'OFF'
                    END + ');' AS [DDL],

                1 AS DDLOrdinal

            WHERE MasterIndexList.RebuildIndicator = 1

            UNION ALL
            SELECT          
                'ALTER INDEX ' + 
                MasterIndexList.IndexName + 
                ' ON ' + 
                MasterIndexList.SchemaName + '.' + MasterIndexList.TableName +
                ' REORGANIZE;' AS [DDL],

                2 AS DDLOrdinal

            WHERE MasterIndexList.ReorganizationIndicator = 1

            UNION ALL
            SELECT
                'UPDATE STATISTICS ' + 
                MasterIndexList.SchemaName + '.' + MasterIndexList.TableName + ' ' + 
                MasterIndexList.IndexName + ' ' + 
                'WITH FULLSCAN;' AS [DDL],

                3 AS DDLOrdinal

            WHERE MasterIndexList.UpdateStatisticsIndicator = 1
            AND MasterIndexList.RebuildIndicator = 0
            AND STATS_DATE(MasterIndexList.ObjectID, MasterIndexList.IndexID) <= DATEADD(hh, -20, GETDATE())
        ) AS [DDL]

    ORDER BY
        ObjectID ASC,
        IndexID ASC,
        DDLOrdinal ASC;

OPEN ddl_cursor;

FETCH NEXT FROM ddl_cursor
INTO @DDL;

WHILE @@FETCH_STATUS = 0
BEGIN

    EXECUTE sys.sp_executesql 
        @stmt = @DDL;

    FETCH NEXT FROM ddl_cursor
    INTO @DDL;
END

CLOSE ddl_cursor;
DEALLOCATE ddl_cursor;
GO

与往常一样，您的里程可能会有所不同，但是可以随时使用/更改这些脚本来满足您的需求。

祝你有个好的一天，

马特

下面的编解码器在〜185 GB数据库上运行良好。

DECLARE @dbid int

SET @dbid = DB_ID()

select o.name as ObjectName, 
       O.id as ObjectID,
       I.name as IndexName,
       I.index_id as IndexID,
       I.type,
       I.type_Desc,
       ps.avg_fragmentation_in_percent 

from sysobjects O with (nolock)
inner join sys.indexes i with (nolock) ON O.id = i.object_id 
inner join sys.dm_db_index_physical_stats (@dbid,null,null,null,'LIMITED') ps on ps.database_id = @dbid
                                                                             and ps.object_id = O.id 
                                                                             and ps.index_id = I.index_id
where xtype = 'U'
and LEFT(o.name,2) <> 'MS'
and ps.avg_fragmentation_in_percent > 10
order by o.name 

我注意到，通过进行更新统计信息，我收集有关碎片索引的信息的时间从3个半小时缩短到5分钟。使用Ola Hallengren更新统计信息表更新统计信息，应该这样做。