728x90
728x170

Select  A.[object_id]
   , OBJECT_NAME(A.[object_id]) AS Table_Name
   , A.Index_ID
   , A.[Name] As Index_Name
   , CAST(
     Case When A.type = 1 AND is_unique = 1 Then 'Create Unique Clustered Index '
       When A.type = 1 AND is_unique = 0 Then 'Create Clustered Index '
       When A.type = 2 AND is_unique = 1 Then 'Create Unique NonClustered Index '
       When A.type = 2 AND is_unique = 0 Then 'Create NonClustered Index '
     End
     + quotename(A.[Name]) + ' On ' + quotename(S.name) + '.' + quotename(OBJECT_NAME(A.[object_id])) + ' ('
     + Stuff(
       (
        Select
          ',[' + COL_NAME(A.[object_id],C.column_id)
          + Case When C.is_descending_key = 1 Then '] Desc' Else '] Asc' End
        From sys.index_columns C WITH (NOLOCK)
        Where A.[Object_ID] = C.object_id
          And A.Index_ID = C.Index_ID
          And C.is_included_column = 0
        Order by C.key_Ordinal Asc
        For XML Path('')
       )
       ,1,1,'') + ') '

     + CASE WHEN A.type = 1 THEN ''
       ELSE Coalesce('Include ('
         + Stuff(
           (
            Select
              ',' + QuoteName(COL_NAME(A.[object_id],C.column_id))
            From sys.index_columns C WITH (NOLOCK)
            Where A.[Object_ID] = C.object_id
              And A.Index_ID = C.Index_ID
              And C.is_included_column = 1
            Order by C.index_column_id Asc
            For XML Path('')
           )
           ,1,1,'') + ') '
      ,'') End
     + Case When A.has_filter = 1 Then 'Where ' + A.filter_definition Else '' End
     + ' With (Drop_Existing = OFF, SORT_IN_TEMPDB = ON'
     --when the same index exists you'd better to set the Drop_Existing = ON
     --SORT_IN_TEMPDB = ON is recommended but based on your own environment.
     + ', Fillfactor = ' + Cast(Case When fill_factor = 0 Then 100 Else fill_factor End As varchar(3))
     + Case When A.[is_padded] = 1 Then ', PAD_INDEX = ON' Else ', PAD_INDEX = OFF' END
     + Case When D.[no_recompute] = 1 Then ', STATISTICS_NORECOMPUTE  = ON' Else ', STATISTICS_NORECOMPUTE  = OFF' End                 
     + Case When A.[ignore_dup_key] = 1 Then ', IGNORE_DUP_KEY = ON' Else ', IGNORE_DUP_KEY = OFF' End
     + Case When A.[ALLOW_ROW_LOCKS] = 1 Then ', ALLOW_ROW_LOCKS = ON' Else ', ALLOW_ROW_LOCKS = OFF' END
     + Case When A.[ALLOW_PAGE_LOCKS] = 1 Then ', ALLOW_PAGE_LOCKS = ON' Else ', ALLOW_PAGE_LOCKS = OFF' End 
     + Case When P.[data_compression] = 0 Then ', DATA_COMPRESSION = NONE'
       When P.[data_compression] = 1 Then ', DATA_COMPRESSION = ROW'
       Else ', DATA_COMPRESSION = PAGE' End                
     + ') On '
     + Case when C.type = 'FG' THEN quotename(C.name)
       ELSE quotename(C.name) + '(' + F.Partition_Column + ')' END + ';' --if it uses partition scheme then need partition column
   As nvarchar(Max)) As Index_Create_Statement
   , C.name AS FileGroupName
   , 'DROP INDEX ' + quotename(A.[Name]) + ' On ' + quotename(S.name) + '.' + quotename(OBJECT_NAME(A.[object_id])) + ';' AS Index_Drop_Statement
From  SYS.Indexes A WITH (NOLOCK)
    INNER JOIN
   sys.objects B WITH (NOLOCK)
    ON A.object_id = B.object_id
    INNER JOIN
   SYS.schemas S
    ON B.schema_id = S.schema_id           
    INNER JOIN
   SYS.data_spaces C WITH (NOLOCK)
    ON A.data_space_id =   C.data_space_id  
    INNER JOIN
   SYS.stats D WITH (NOLOCK)
    ON A.object_id = D.object_id
     AND A.index_id = D.stats_id 
    Inner Join
   --The below code is to find out what data compression type was used by the index. If an index is not partitioned, it is easy as only one data compression
   --type can be used. If the index is partitioned, then each partition can be configued to use the different data compression. This is hard to generalize,
   --for simplicity, I just use the data compression type used most for the index partitions for all partitions. You can later rebuild the index partition to
   --the appropriate data compression type you want to use
   (
    select object_id, index_id, Data_Compression, ROW_NUMBER() Over(Partition By object_id, index_id Order by COUNT(*) Desc) As Main_Compression
    From sys.partitions WITH (NOLOCK)
    Group BY object_id, index_id, Data_Compression
   ) P
    ON A.object_id = P.object_id
     AND A.index_id = P.index_id
     AND P.Main_Compression = 1
    Outer APPLY
   (
    SELECT COL_NAME(A.object_id, E.column_id) AS Partition_Column
    From sys.index_columns E WITH (NOLOCK)
    WHERE E.object_id = A.object_id
      AND E.index_id = A.index_id
      AND E.partition_ordinal = 1
   ) F    
Where  A.type IN (1,2) --clustered and nonclustered
   AND B.Type != 'S'
   AND is_primary_key = 0 --this is not for primary key constraint
   AND OBJECT_NAME(A.[object_id]) not like 'queue_messages_%'
   AND OBJECT_NAME(A.[object_id]) not like 'filestream_tombstone_%'
   AND OBJECT_NAME(A.[object_id]) not like 'sys%' --if you have index start with sys then remove it
   
OPTION (Recompile);

 

 

http://www.msbicoe.com/post/2012/12/11/Script-out-Indexes-and-Index-Option-Settings-for-a-SQL-Server-Database.aspx

728x90
그리드형
Posted by kjun.kr
,