Filtered Indexes Work-Around
Filed under: Performance & Tuning, SQL 2008, SQL Tips, Syndication
Recently, I needed to create a stored procedure that queried a rather large table. The table has a filtered index on a date column, and it covers the query. However, the Query Optimizer was not using the index, which was increasing the execution time (not to mention IO!) by at least 10x. This wasn’t the first time I’ve had the Optimizer fail to use a filtered index. Normally when this happens, I use a table hint to force the filtered index — after I verify that it is indeed faster, of course. However, since this was a stored procedure, I was receiving the following error message whenever I tried to execute the proc:
Query processor could not produce a query plan because of the hints defined in this query. Resubmit the query without specifying any hints and without using SET FORCEPLAN.
SQL Server would not allow me to execute the stored procedure using the filtered index hint. If I removed the hint, it executed, but it used a different, non-covering and far more expensive index. For those of you not familiar with this issue, allow me to illustrate the problem.
First, create a table to play with and populate it with some bogus data:
CREATE TABLE dbo.filteredIndexTest ( myID INT IDENTITY(1,3) , myDate SMALLDATETIME , myData CHAR(100) CONSTRAINT PK_filteredIndexTest PRIMARY KEY CLUSTERED(myID) ); SET NOCOUNT ON; DECLARE @DATE SMALLDATETIME = '2010-01-01'; WHILE @DATE < '2010-02-01' BEGIN INSERT INTO dbo.filteredIndexTest ( myDate , myData ) SELECT @DATE , 'Date: ' + CONVERT(VARCHAR(20), @DATE, 102); SET @DATE = DATEADD(MINUTE, 1, @DATE); END; SELECT COUNT(*) FROM dbo.filteredIndexTest;
It looks like this will generate 44,640 rows of test data… plenty enough for our purposes. Now, let’s create our filtered index and write a query that will use it:
CREATE NONCLUSTERED INDEX IX_filteredIndexTest_1 ON dbo.filteredIndexTest(myDate) Include (myData) WHERE myDate >= '2010-01-27'; SELECT DISTINCT myData FROM dbo.filteredIndexTest WHERE myDate >= '2010-01-28';
If you look at the execution plan for this query, you’ll notice that the Optimizer is using the filtered index. Perfect! Now let’s parameterize it.
DECLARE @myDate1 SMALLDATETIME = '2010-01-28'; SELECT DISTINCT myData FROM dbo.filteredIndexTest WHERE myDate >= @myDate1;
Uh oh. Looking at the execution plan, we see that SQL Server is no longer using the filtered index. Instead, it’s scanning the clustered index! Why is this? There’s actually a good explanation for it. The reason is that I could, in theory, pass a date to my parameter that fell outside of the filtered date range. If that’s the case, then SQL Server could not utilize the filtered index. Personally, I think it’s a bug and SQL Server should identify whether or not a filtered index could be used based on the actual value submitted, but… that’s a whole other blog post. 
So what can we do? Well, dynamic SQL may be able to help us out in this case. Let’s give it a go. First, let’s try parameterized dynamic SQL.
DECLARE @mySQL1 NVARCHAR(2000) , @myParam NVARCHAR(2000) = '@p_myDate2 smalldatetime' , @myDate2 SMALLDATETIME = '2010-01-28'; SET @mySQL1 = 'Select Distinct myData From dbo.filteredIndexTest Where myDate >= @p_myDate2'; EXECUTE SP_EXECUTESQL @mySQL1, @myParam, @p_myDate2 = @myDate2;
Looking at the execution plan, we see we’re still scanning on the clustered index. This is because the parameterized dynamic SQL resolves to be the exact same query as the one above it. Let’s try unparameterized SQL instead:
DECLARE @mySQL2 NVARCHAR(2000) , @myDate3 SMALLDATETIME = '2010-01-28'; SET @mySQL2 = 'Select Distinct myData From dbo.filteredIndexTest Where myDate >= ''' + CAST(@myDate3 AS VARCHAR(20)) + ''''; EXECUTE SP_EXECUTESQL @mySQL2; -- Drop Table dbo.filteredIndexTest;
Voila! We have a seek on our filtered index. Why? Because the statement resolves to be identical to our first query, where we hard-coded the date value in the WHERE clause.
Now, I want to stress this fact: you should always, ALWAYS use parameterized dynamic SQL whenever possible. Not only is it safer, but it’s also faster, because it can reuse cached plans. But sometimes you just cannot accomplish the same tasks with it. This is one of those times. If you do end up needing to use unparameterized dynamic SQL as a work-around, please make sure you’re validating your input, especially if you’re interfacing with any sort of external source.
There’s an even easier work-around for this problem that Dave (http://www.crappycoding.com) shared with me: recompile.
Adding “Option (Recompile)” to the end of your statements will force the Optimizer to re-evaluate which index will best meet the needs of your query every time the statement is executed. More importantly, it evaluates the plan based on the actual values passed to the parameter… just like in our hard-coded and dynamic SQL examples. Let’s see it in action:
DECLARE @myDate4 SMALLDATETIME = '2010-01-28'; SELECT DISTINCT myData FROM dbo.filteredIndexTest WHERE myDate >= @myDate4 OPTION (RECOMPILE); DECLARE @myDate5 SMALLDATETIME = '2010-01-20'; SELECT DISTINCT myData FROM dbo.filteredIndexTest WHERE myDate >= @myDate5 OPTION (RECOMPILE);
If we look at the execution plans for the 2 queries above, we see that the first query seeks on the filtered index, and the second query scans on the clustered index. This is because the second query cannot be satisfied with the filtered index because we initially limited our index to dates greater than or equal to 1/27/2010.
There are, of course, trade-offs associated with each approach, so use whichever one best meets your needs. Do you have another work-around for this issue? If so, please let me know.
Update:
Alex Kuznetsov (http://www.simple-talk.com/author/alex-kuznetsov/) shared this method too:
DECLARE @myDate1 SMALLDATETIME = '2010-01-28'; SELECT DISTINCT myData FROM dbo.filteredIndexTest WHERE myDate = @myDate1 AND myDate >= '2010-01-27';
Like the other examples, this will result in an index seek on the filtered index. Basically, by explicitly declaring the start date of your filter, you’re letting the Optimizer know that the filtered index can satisfy the request, regardless of the parameter value passed. Thanks for the tip, Alex!
Index Defrag Script Updates – Beta Testers Needed
Filed under: Miscellaneous, Performance & Tuning, SQL 2008, Syndication, T-SQL Scripts
Update: Wow! I’ve received a ton of responses to my request for beta testers. Thank you all! The SQL Community is really amazing. I’ll hopefully have the new version online in just a few days.
Over the last few months, I’ve received many great comments and suggestions regarding my Index Defrag Script v3.0. I’ve just recently had time to implement most of these suggestions, plus some other things that I thought would be useful.
Here’s some of what you can look forward to shortly:
- Probably the single most requested feature, the new version of the script allows you to set a time limit for index defrags.
- There’s now a static table for managing the status of index defrags. This way, when your time limit is reached, you can pick up where you left off the next day, without the need to rescan indexes.
- There’s now an option to prioritize defrags by range scan counts, fragmentation level, or page counts.
- For those using partitioning, there is now an option to exclude the right-most populated partition from defrags (in theory, the one you’re writing to in a sliding-window scenario).
- Options such as page count limits and SORT_IN_TEMPDB are now parameterized.
- I’ve enhanced error logging.
- … and more!
Right now, I’m looking for a few folks who are willing to beta test the script. If you’re interested, please send me an e-mail at michelle at sqlfool dot com with the editions of SQL Server you can test this on (i.e. 2005 Standard, 2008 Enterprise, etc.).
Thank you!
Partitioning Tricks
Filed under: Miscellaneous, Performance & Tuning, SQL 2008, SQL Tips, Syndication
For those of you who are using partitioning, or who are considering using partitioning, allow me to share some tips with you.
Easy Partition Staging Tables
Switching partitions (or more specifically, hobts) in and out of a partitioned table requires the use of a staging table. The staging table has very specific requirements: it must be completely identical to the partitioned table, including indexing structures, and it must have a check constraint that limits data to the partitioning range. Thanks to my co-worker Jeff, I’ve recently started using the SQL Server Partition Management tool on CodePlex. I haven’t used the automatic partition switching feature — frankly, using any sort of data modification tool in a production environment makes me nervous — but I’ve been using the scripting option to create staging tables in my development environment, which I then copy to production for use. It’s nothing you can’t do yourself, but it does make the whole process easy and painless, plus it saves you from annoying typos. But be careful when using this tool to just create the table and check constraints automatically, because you may need to…
Add Check Constraints After Loading Data
Most of the time, I add the check constraint when I create the staging table, then I load data and perform the partition switch. However, for some reason, I was receiving the following error:
.Net SqlClient Data Provider: Msg 4972, Level 16, State 1, Line 1
ALTER TABLE SWITCH statement failed. Check constraints or partition function of source table ‘myStagingTable’ allows values that are not allowed by check constraints or partition function on target table ‘myDestinationTable’.
This drove me crazy. I confirmed my check constraints were correct, that I had the correct partition number, and that all schema and indexes matched identically. After about 30 minutes of this, I decided to drop and recreate the constraint. For some reason, it fixed the issue. Repeat tests produced the same results: the check constraint needed to be added *after* data was loaded. This error is occurring on a SQL Server 2008 SP1 box; to be honest, I’m not sure what’s causing the error, so if you know, please leave me a comment. But I figured I’d share so that anyone else running into this issue can hopefully save some time and headache.
Replicating Into Partitioned and Non-Partitioned Tables
Recently, we needed to replicate a non-partitioned table to two different destinations. We wanted to use partitioning for Server A, which has 2008 Enterprise; Server B, which is on 2005 Standard, could not take advantage of partitioning. The solution was really easy: create a pre-snapshot and post-snapshot script for the publication, then modify to handle each server group differently. Using pseudo-code, it looked something like this:
/* Identify which servers get the partitioned version */ IF @@SERVERNAME In ('yourServerNameList') BEGIN /* Create your partitioning scheme if necessary */ IF Not Exists(SELECT * FROM sys.partition_schemes WHERE name = 'InsertPartitionScheme') CREATE PARTITION SCHEME InsertPartitionScheme AS PARTITION InsertPartitionFunction ALL TO ([PRIMARY]); /* Create your partitioning function if necessary */ IF Not Exists(SELECT * FROM sys.partition_functions WHERE name = 'InsertPartitionFunction') CREATE PARTITION FUNCTION InsertPartitionFunction (SMALLDATETIME) AS RANGE RIGHT FOR VALUES ('insertValues'); /* Create a partitioned version of your table */ CREATE TABLE [dbo].[yourTableName] ( [yourTableSchema] ) ON InsertPartitionScheme([partitioningKey]); END ELSE BEGIN /* Create a non-partitioned version of your table */ CREATE TABLE [dbo].[yourTableName] ( [yourTableSchema] ) ON [PRIMARY]; END
You could also use an edition check instead of a server name check, if you prefer. The post-snapshot script basically looked the same, except you create partitioned indexes instead.
Compress Old Partitions
Did you know you can set different compression levels for individual partitions? It’s true! I’ve just completed doing this on our largest partitioned table. Here’s how:
/* Apply compression to your partitioned table */ ALTER TABLE dbo.yourTableName Rebuild Partition = All WITH ( Data_Compression = Page ON Partitions(1 TO 9) , Data_Compression = ROW ON Partitions(10 TO 11) , Data_Compression = NONE ON Partitions(12) ); /* Apply compression to your partitioned index */ ALTER INDEX YourPartitionedIndex ON dbo.yourTableName Rebuild Partition = All WITH ( Data_Compression = Page ON Partitions(1 TO 9) , Data_Compression = ROW ON Partitions(10 TO 11) , Data_Compression = NONE ON Partitions(12) ); /* Apply compression to your unpartitioned index */ ALTER INDEX YourUnpartitionedIndex ON dbo.yourTableName Rebuild WITH (Data_Compression = ROW);
A couple of things to note. In all of our proof-of-concept testing, we found that compression significantly reduced query execution time, reads (IO), and storage. However, CPU was also increased significantly. The results were more dramatic, both good and bad, with page compression versus row compression. Still, for our older partitions, which aren’t queried regularly, it made sense to turn on page compression. The newer partitions receive row compression, and the newest partitions, which are still queried very regularly by routine processes, were left completely uncompressed. This seems to strike a nice balance in our environment, but of course, results will vary depending on how you use your data.
Something to be aware of is that compressing your clustered index does *not* compress your non-clustered indexes; those are separate operations. Lastly, for those who are curious, it took us about 1 minute to apply row compression and about 7 minutes to apply page compression to partitions averaging 30 million rows.
Looking for more information on table partitioning? Check out my overview of partitioning, my example code, and my article on indexing on partitioned tables.
Undocumented Function in SQL 2008
If you’ve been following my blog for a little while, you’ll know that I’m a fan of SQL Server internals. There’s a lot that can be learned or better understood by rolling up your sleeves and getting into the nitty-gritty of data pages (i.e. see my post on Overhead in Non-Unique Clustered Indexes). So imagine how happy I was when my co-worker Jeff shared an undocumented function with me today that retrieves the file number, page number, and slot number of a single record. Very cool! Well, at least to me. So now let’s see how you can use it.
The fn_physLocCracker function can be called in the following way:
SELECT TOP 100 plc.*, soh.SalesOrderID FROM Sales.SalesOrderHeader AS soh Cross Apply sys.fn_physLocCracker (%%physloc%%) AS plc;
Results (just a sample):
file_id page_id slot_id SalesOrderID ----------- ----------- ----------- ------------ 1 14032 0 43659 1 14032 1 43660 1 14032 2 43661 1 14032 3 43662 1 14032 4 43663
If you look at the sp_helptext for sys.fn_physLocCracker, %%physloc%% is apparently a virtual column that contains information on where the record is stored. In fact, you can even append %%physloc%% to your column list if you want to see how the information is stored. But for our purposes, we now have a file number, page number, and slot number. What do we do with it?
Well, you can use the investigation proc I wrote to retrieve the actual data page:
EXECUTE dba_viewPageData_sp @databaseName = 'AdventureWorks' , @fileNumber = 1 , @pageNumber = 14032;
Results (just a sample):
Slot 0 Column 1 Offset 0x4 Length 4 Length (physical) 4 SalesOrderID = 43659 Slot 0 Column 2 Offset 0x8 Length 1 Length (physical) 1 RevisionNumber = 1 Slot 0 Column 3 Offset 0x9 Length 8 Length (physical) 8 OrderDate = 2001-07-01 00:00:00.000 Slot 0 Column 4 Offset 0x11 Length 8 Length (physical) 8 DueDate = 2001-07-13 00:00:00.000
Neat, huh? So why would you use it to look up the data page and file number when you can just pass the table name and index name to my proc and retrieve data pages? Well, my investigation proc will retrieve data pages for any index type — the fn_physLocCracker function will only retrieve data for the clustered index — but it will not retrieve the data page for a specific record. So just something to be aware of.
That’s all for now. Back to the #24HoursOfPASS!
Index Defrag Script, v3.0
Filed under: Performance & Tuning, SQL 2008, Syndication, T-SQL Scripts
I’ve just completed the latest version of my index defrag script! Here’s a brief list of the updates:
- Fixed a bug with the LOB logic. In the previous version, after a LOB was encountered, all subsequent indexes would be reorganized.
- Added support for stat rebuilds after the index defrag is complete (@rebuildStats)
- Added an exclusion list table (dba_indexDefragExclusion) to support index scheduling
- Modified logging to show which defrags are “in progress”; added columns to dba_indexDefragLog
- Added support for the defrag of the model and msdb databases
- Added @scanMode as a configurable parameter
So what can this index defrag script do? Well, for starters, you can:
- Schedule it to run with the default settings; it works “right out of the box” with no additional configuration necessary
- Run this one script from a centralized database for all databases on a server
- Run this script for a specific database or table
- Configure custom threshold limits and the point at which a rebuild should be performed (instead of a reorganize)
- Defrag individual partitions
- Log its actions and the duration of the defrag
- Run in “commands only” mode (@executeSQL = 0, @printCommands = 1)
- Customize performance parameters such as @maxDopRestriction and @defragDelay to minimize impact on the server
- Schedule specific indexes to only be defragged on weekends, or every other day
To use this last option, you need to add a record to the dba_indexDefragExclusion table. I think all of the columns are pretty self-explanatory except the [exclusionMask] column. The way this works is each day of the week is assigned a value:
1=Sunday, 2=Monday, 4=Tuesday, 8=Wednesday, 16=Thursday, 32=Friday, 64=Saturday
Take a SUM of the values for the days that you want excluded. So if you want an index to only be defragged on weekends, you would add up Monday through Friday (2+4+8+16+32) and use a value of 62 for the exclusionMask column. For a little more information on how this works, check out my blog post on Bitwise Operations.
Please note: if you don’t insert any records into the dba_indexDefragExclusion table, by default all indexes will be defragged every run-time if they exceed the specified thresholds. This is normal behavior and may be perfectly fine in your environment. However, if the dba_indexDefragExclusion table does not exist, the script will fail.
I try to document each parameter within the code, so check the comments section in the script for a full list of parameters and what they do.
Special thanks to everyone who helped beta test this script!
Without further ado, the script:
/* Drop Table Scripts: Drop Table dbo.dba_indexDefragLog; Drop Table dbo.dba_indexDefragExclusion; */ IF Not Exists(SELECT [OBJECT_ID] FROM sys.tables WHERE [name] In (N'dba_indexDefragLog', 'dba_indexDefragExclusion')) BEGIN CREATE TABLE dbo.dba_indexDefragLog ( indexDefrag_id INT IDENTITY(1,1) Not Null , databaseID INT Not Null , databaseName NVARCHAR(128) Not Null , objectID INT Not Null , objectName NVARCHAR(128) Not Null , indexID INT Not Null , indexName NVARCHAR(128) Not Null , partitionNumber SMALLINT Not Null , fragmentation FLOAT Not Null , page_count INT Not Null , dateTimeStart DATETIME Not Null , dateTimeEnd DATETIME Null , durationSeconds INT Null CONSTRAINT PK_indexDefragLog PRIMARY KEY CLUSTERED (indexDefrag_id) ); PRINT 'dba_indexDefragLog Table Created'; CREATE TABLE dbo.dba_indexDefragExclusion ( databaseID INT Not Null , databaseName NVARCHAR(128) Not Null , objectID INT Not Null , objectName NVARCHAR(128) Not Null , indexID INT Not Null , indexName NVARCHAR(128) Not Null , exclusionMask INT Not Null /* 1=Sunday, 2=Monday, 4=Tuesday, 8=Wednesday, 16=Thursday, 32=Friday, 64=Saturday */ CONSTRAINT PK_indexDefragExclusion PRIMARY KEY CLUSTERED (databaseID, objectID, indexID) ); PRINT 'dba_indexDefragExclusion Table Created'; END ELSE RAISERROR('One or more tables already exist. Please drop or rename before proceeding.', 16, 0); IF OBJECTPROPERTY(OBJECT_ID('dbo.dba_indexDefrag_sp'), N'IsProcedure') = 1 BEGIN DROP PROCEDURE dbo.dba_indexDefrag_sp; PRINT 'Procedure dba_indexDefrag_sp dropped'; END; Go CREATE PROCEDURE dbo.dba_indexDefrag_sp /* Declare Parameters */ @minFragmentation FLOAT = 5.0 /* in percent, will not defrag if fragmentation less than specified */ , @rebuildThreshold FLOAT = 30.0 /* in percent, greater than @rebuildThreshold will result in rebuild instead of reorg */ , @executeSQL BIT = 1 /* 1 = execute; 0 = print command only */ , @DATABASE VARCHAR(128) = Null /* Option to specify a database name; null will return all */ , @tableName VARCHAR(4000) = Null -- databaseName.schema.tableName /* Option to specify a table name; null will return all */ , @scanMode VARCHAR(10) = N'LIMITED' /* Options are LIMITED, SAMPLED, and DETAILED */ , @onlineRebuild BIT = 1 /* 1 = online rebuild; 0 = offline rebuild; only in Enterprise */ , @maxDopRestriction TINYINT = Null /* Option to restrict the number of processors for the operation; only in Enterprise */ , @printCommands BIT = 0 /* 1 = print commands; 0 = do not print commands */ , @printFragmentation BIT = 0 /* 1 = print fragmentation prior to defrag; 0 = do not print */ , @defragDelay CHAR(8) = '00:00:05' /* time to wait between defrag commands */ , @debugMode BIT = 0 /* display some useful comments to help determine if/where issues occur */ , @rebuildStats BIT = 1 /* option to rebuild stats after completed index defrags */ AS /********************************************************************************* Name: dba_indexDefrag_sp Author: Michelle Ufford, http://sqlfool.com Purpose: Defrags all indexes for the current database Notes: CAUTION: TRANSACTION LOG SIZE SHOULD BE MONITORED CLOSELY WHEN DEFRAGMENTING. @minFragmentation defaulted to 10%, will not defrag if fragmentation is less than that @rebuildThreshold defaulted to 30% as recommended by Microsoft in BOL; greater than 30% will result in rebuild instead @executeSQL 1 = execute the SQL generated by this proc; 0 = print command only @database Optional, specify specific database name to defrag; If not specified, all non-system databases will be defragged. @tableName Specify if you only want to defrag indexes for a specific table, format = databaseName.schema.tableName; if not specified, all tables will be defragged. @scanMode Specifies which scan mode to use to determine fragmentation levels. Options are: LIMITED - scans the parent level; quickest mode, recommended for most cases. SAMPLED - samples 1% of all data pages; if less than 10k pages, performs a DETAILED scan. DETAILED - scans all data pages. Use great care with this mode, as it can cause performance issues. @onlineRebuild 1 = online rebuild; 0 = offline rebuild @maxDopRestriction Option to specify a processor limit for index rebuilds @printCommands 1 = print commands to screen; 0 = do not print commands @printFragmentation 1 = print fragmentation to screen; 0 = do not print fragmentation @defragDelay Time to wait between defrag commands; gives the server a little time to catch up @debugMode 1 = display debug comments; helps with troubleshooting 0 = do not display debug comments @rebuildStats Affects only statistics that need to be rebuilt 1 = rebuild stats 0 = do not rebuild stats Called by: SQL Agent Job or DBA Date Initials Version Description ---------------------------------------------------------------------------- 2007-12-18 MFU 1.0 Initial Release 2008-10-17 MFU 1.1 Added @defragDelay, CIX_temp_indexDefragList 2008-11-17 MFU 1.2 Added page_count to log table , added @printFragmentation option 2009-03-17 MFU 2.0 Provided support for centralized execution , consolidated Enterprise & Standard versions , added @debugMode, @maxDopRestriction , modified LOB and partition logic 2009-06-18 MFU 3.0 Fixed bug in LOB logic, added @scanMode option , added support for stat rebuilds (@rebuildStats) , support model and msdb defrag , added columns to the dba_indexDefragLog table , modified logging to show "in progress" defrags , added defrag exclusion list (scheduling) ********************************************************************************* Exec dbo.dba_indexDefrag_sp @executeSQL = 0 , @printCommands = 1 , @debugMode = 1 , @printFragmentation = 1; *********************************************************************************/ SET NOCOUNT ON; SET XACT_Abort 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; BEGIN IF @debugMode = 1 RAISERROR('Undusting the cogs and starting up...', 0, 42) WITH NoWait; /* Declare our variables */ DECLARE @objectID INT , @databaseID INT , @databaseName NVARCHAR(128) , @indexID INT , @partitionCount BIGINT , @schemaName NVARCHAR(128) , @objectName NVARCHAR(128) , @indexName NVARCHAR(128) , @partitionNumber SMALLINT , @fragmentation FLOAT , @pageCount INT , @sqlCommand NVARCHAR(4000) , @rebuildCommand NVARCHAR(200) , @dateTimeStart DATETIME , @dateTimeEnd DATETIME , @containsLOB BIT , @editionCheck BIT , @debugMessage VARCHAR(128) , @updateSQL NVARCHAR(4000) , @partitionSQL NVARCHAR(4000) , @partitionSQL_Param NVARCHAR(1000) , @LOB_SQL NVARCHAR(4000) , @LOB_SQL_Param NVARCHAR(1000) , @rebuildStatsID INT , @rebuildStatsSQL NVARCHAR(1000) , @indexDefrag_id INT; /* Create our temporary tables */ CREATE TABLE #indexDefragList ( databaseID INT , databaseName NVARCHAR(128) , objectID INT , indexID INT , partitionNumber SMALLINT , fragmentation FLOAT , page_count INT , defragStatus BIT , schemaName NVARCHAR(128) Null , objectName NVARCHAR(128) Null , indexName NVARCHAR(128) Null ); CREATE TABLE #databaseList ( databaseID INT , databaseName VARCHAR(128) , scanStatus BIT , statsStatus BIT ); CREATE TABLE #processor ( [INDEX] INT , Name VARCHAR(128) , Internal_Value INT , Character_Value INT ); IF @debugMode = 1 RAISERROR('Beginning validation...', 0, 42) WITH NoWait; /* Just a little validation... */ IF @minFragmentation Not Between 0.00 And 100.0 SET @minFragmentation = 10.0; IF @rebuildThreshold Not Between 0.00 And 100.0 SET @rebuildThreshold = 30.0; IF @defragDelay Not Like '00:[0-5][0-9]:[0-5][0-9]' SET @defragDelay = '00:00:05'; IF @scanMode Not In ('LIMITED', 'SAMPLED', 'DETAILED') SET @scanMode = 'LIMITED'; /* Make sure we're not exceeding the number of processors we have available */ INSERT INTO #processor EXECUTE XP_MSVER 'ProcessorCount'; IF @maxDopRestriction IS Not Null And @maxDopRestriction > (SELECT Internal_Value FROM #processor) SELECT @maxDopRestriction = Internal_Value FROM #processor; /* Check our server version; 1804890536 = Enterprise, 610778273 = Enterprise Evaluation, -2117995310 = Developer */ IF (SELECT SERVERPROPERTY('EditionID')) In (1804890536, 610778273, -2117995310) SET @editionCheck = 1 -- supports online rebuilds ELSE SET @editionCheck = 0; -- does not support online rebuilds IF @debugMode = 1 RAISERROR('Grabbing a list of our databases...', 0, 42) WITH NoWait; /* Retrieve the list of databases to investigate */ INSERT INTO #databaseList SELECT database_id , name , 0 -- not scanned yet for fragmentation , 0 -- statistics not yet updated FROM sys.databases WHERE name = IsNull(@DATABASE, name) And [name] Not In ('master', 'tempdb')-- exclude system databases And [STATE] = 0; -- state must be ONLINE IF @debugMode = 1 RAISERROR('Looping through our list of databases and checking for fragmentation...', 0, 42) WITH NoWait; /* Loop through our list of databases */ WHILE (SELECT COUNT(*) FROM #databaseList WHERE scanStatus = 0) > 0 BEGIN SELECT TOP 1 @databaseID = databaseID FROM #databaseList WHERE scanStatus = 0; SELECT @debugMessage = ' working on ' + DB_NAME(@databaseID) + '...'; IF @debugMode = 1 RAISERROR(@debugMessage, 0, 42) WITH NoWait; /* Determine which indexes to defrag using our user-defined parameters */ INSERT INTO #indexDefragList SELECT database_id AS databaseID , QUOTENAME(DB_NAME(database_id)) AS 'databaseName' , [OBJECT_ID] AS objectID , index_id AS indexID , partition_number AS partitionNumber , avg_fragmentation_in_percent AS fragmentation , page_count , 0 AS 'defragStatus' /* 0 = unprocessed, 1 = processed */ , Null AS 'schemaName' , Null AS 'objectName' , Null AS 'indexName' FROM sys.dm_db_index_physical_stats (@databaseID, OBJECT_ID(@tableName), Null , Null, @scanMode) WHERE avg_fragmentation_in_percent >= @minFragmentation And index_id > 0 -- ignore heaps And page_count > 8 -- ignore objects with less than 1 extent And index_level = 0 -- leaf-level nodes only, supports @scanMode OPTION (MaxDop 2); /* Keep track of which databases have already been scanned */ UPDATE #databaseList SET scanStatus = 1 WHERE databaseID = @databaseID; END CREATE CLUSTERED INDEX CIX_temp_indexDefragList ON #indexDefragList(databaseID, objectID, indexID, partitionNumber); /* Delete any indexes from our to-do that are also in our exclusion list for today */ DELETE idl FROM #indexDefragList AS idl Join dbo.dba_indexDefragExclusion AS ide ON idl.databaseID = ide.databaseID And idl.objectID = ide.objectID And idl.indexID = ide.indexID WHERE exclusionMask & POWER(2, DATEPART(weekday, GETDATE())-1) > 0; SELECT @debugMessage = 'Looping through our list... there''s ' + CAST(COUNT(*) AS VARCHAR(10)) + ' indexes to defrag!' FROM #indexDefragList; IF @debugMode = 1 RAISERROR(@debugMessage, 0, 42) WITH NoWait; /* Begin our loop for defragging */ WHILE (SELECT COUNT(*) FROM #indexDefragList WHERE defragStatus = 0) > 0 BEGIN IF @debugMode = 1 RAISERROR(' Picking an index to beat into shape...', 0, 42) WITH NoWait; /* Grab the most fragmented index first to defrag */ SELECT TOP 1 @objectID = objectID , @indexID = indexID , @databaseID = databaseID , @databaseName = databaseName , @fragmentation = fragmentation , @partitionNumber = partitionNumber , @pageCount = page_count FROM #indexDefragList WHERE defragStatus = 0 ORDER BY fragmentation DESC; IF @debugMode = 1 RAISERROR(' Looking up the specifics for our index...', 0, 42) WITH NoWait; /* Look up index information */ SELECT @updateSQL = N'Update idl Set schemaName = QuoteName(s.name) , objectName = QuoteName(o.name) , indexName = QuoteName(i.name) From #indexDefragList As idl Inner Join ' + @databaseName + '.sys.objects As o On idl.objectID = o.object_id Inner Join ' + @databaseName + '.sys.indexes As i On o.object_id = i.object_id Inner Join ' + @databaseName + '.sys.schemas As s On o.schema_id = s.schema_id Where o.object_id = ' + CAST(@objectID AS VARCHAR(10)) + ' And i.index_id = ' + CAST(@indexID AS VARCHAR(10)) + ' And i.type > 0 And idl.databaseID = ' + CAST(@databaseID AS VARCHAR(10)); EXECUTE SP_EXECUTESQL @updateSQL; /* Grab our object names */ SELECT @objectName = objectName , @schemaName = schemaName , @indexName = indexName FROM #indexDefragList WHERE objectID = @objectID And indexID = @indexID And databaseID = @databaseID; IF @debugMode = 1 RAISERROR(' Grabbing the partition count...', 0, 42) WITH NoWait; /* Determine if the index is partitioned */ SELECT @partitionSQL = 'Select @partitionCount_OUT = Count(*) From ' + @databaseName + '.sys.partitions Where object_id = ' + CAST(@objectID AS VARCHAR(10)) + ' And index_id = ' + CAST(@indexID AS VARCHAR(10)) + ';' , @partitionSQL_Param = '@partitionCount_OUT int OutPut'; EXECUTE SP_EXECUTESQL @partitionSQL, @partitionSQL_Param, @partitionCount_OUT = @partitionCount OUTPUT; IF @debugMode = 1 RAISERROR(' Seeing if there''s any LOBs to be handled...', 0, 42) WITH NoWait; /* Determine if the table contains LOBs */ SELECT @LOB_SQL = ' Select @containsLOB_OUT = Count(*) From ' + @databaseName + '.sys.columns With (NoLock) Where [object_id] = ' + CAST(@objectID AS VARCHAR(10)) + ' And (system_type_id In (34, 35, 99) Or max_length = -1);' /* system_type_id --> 34 = image, 35 = text, 99 = ntext max_length = -1 --> varbinary(max), varchar(max), nvarchar(max), xml */ , @LOB_SQL_Param = '@containsLOB_OUT int OutPut'; EXECUTE SP_EXECUTESQL @LOB_SQL, @LOB_SQL_Param, @containsLOB_OUT = @containsLOB OUTPUT; IF @debugMode = 1 RAISERROR(' Building our SQL statements...', 0, 42) WITH NoWait; /* If there's not a lot of fragmentation, or if we have a LOB, we should reorganize */ IF @fragmentation < @rebuildThreshold Or @containsLOB >= 1 Or @partitionCount > 1 BEGIN SET @sqlCommand = N'Alter Index ' + @indexName + N' On ' + @databaseName + N'.' + @schemaName + N'.' + @objectName + N' ReOrganize'; /* If our index is partitioned, we should always reorganize */ IF @partitionCount > 1 SET @sqlCommand = @sqlCommand + N' Partition = ' + CAST(@partitionNumber AS NVARCHAR(10)); END; /* If the index is heavily fragmented and doesn't contain any partitions or LOB's, rebuild it */ IF @fragmentation >= @rebuildThreshold And IsNull(@containsLOB, 0) != 1 And @partitionCount <= 1 BEGIN /* Set online rebuild options; requires Enterprise Edition */ IF @onlineRebuild = 1 And @editionCheck = 1 SET @rebuildCommand = N' Rebuild With (Online = On'; ELSE SET @rebuildCommand = N' Rebuild With (Online = Off'; /* Set processor restriction options; requires Enterprise Edition */ IF @maxDopRestriction IS Not Null And @editionCheck = 1 SET @rebuildCommand = @rebuildCommand + N', MaxDop = ' + CAST(@maxDopRestriction AS VARCHAR(2)) + N')'; ELSE SET @rebuildCommand = @rebuildCommand + N')'; SET @sqlCommand = N'Alter Index ' + @indexName + N' On ' + @databaseName + N'.' + @schemaName + N'.' + @objectName + @rebuildCommand; END; /* Are we executing the SQL? If so, do it */ IF @executeSQL = 1 BEGIN IF @debugMode = 1 RAISERROR(' Executing SQL statements...', 0, 42) WITH NoWait; /* Grab the time for logging purposes */ SET @dateTimeStart = GETDATE(); /* Log our actions */ INSERT INTO dbo.dba_indexDefragLog ( databaseID , databaseName , objectID , objectName , indexID , indexName , partitionNumber , fragmentation , page_count , dateTimeStart ) SELECT @databaseID , @databaseName , @objectID , @objectName , @indexID , @indexName , @partitionNumber , @fragmentation , @pageCount , @dateTimeStart; SET @indexDefrag_id = SCOPE_IDENTITY(); /* Execute our defrag! */ EXECUTE SP_EXECUTESQL @sqlCommand; SET @dateTimeEnd = GETDATE(); /* Update our log with our completion time */ UPDATE dbo.dba_indexDefragLog SET dateTimeEnd = @dateTimeEnd , durationSeconds = DATEDIFF(SECOND, @dateTimeStart, @dateTimeEnd) WHERE indexDefrag_id = @indexDefrag_id; /* Just a little breather for the server */ WAITFOR Delay @defragDelay; /* Print if specified to do so */ IF @printCommands = 1 PRINT N'Executed: ' + @sqlCommand; END ELSE /* Looks like we're not executing, just printing the commands */ BEGIN IF @debugMode = 1 RAISERROR(' Printing SQL statements...', 0, 42) WITH NoWait; IF @printCommands = 1 PRINT IsNull(@sqlCommand, 'error!'); END IF @debugMode = 1 RAISERROR(' Updating our index defrag status...', 0, 42) WITH NoWait; /* Update our index defrag list so we know we've finished with that index */ UPDATE #indexDefragList SET defragStatus = 1 WHERE databaseID = @databaseID And objectID = @objectID And indexID = @indexID And partitionNumber = @partitionNumber; END /* Do we want to output our fragmentation results? */ IF @printFragmentation = 1 BEGIN IF @debugMode = 1 RAISERROR(' Displaying fragmentation results...', 0, 42) WITH NoWait; SELECT databaseID , databaseName , objectID , objectName , indexID , indexName , fragmentation , page_count FROM #indexDefragList; END; /* Do we want to rebuild stats? */ IF @rebuildStats = 1 BEGIN WHILE Exists(SELECT TOP 1 * FROM #databaseList WHERE statsStatus = 0) BEGIN /* Build our SQL statement to update stats */ SELECT TOP 1 @rebuildStatsSQL = 'Use [' + databaseName + ']; ' + 'Execute sp_updatestats;' , @rebuildStatsID = databaseID FROM #databaseList WHERE statsStatus = 0; SET @debugMessage = 'Rebuilding Statistics: ' + @rebuildStatsSQL; IF @debugMode = 1 RAISERROR(@debugMessage, 0, 42) WITH NoWait; /* Execute our stats update! */ EXECUTE SP_EXECUTESQL @rebuildStatsSQL; /* Keep track of which databases have been updated */ UPDATE #databaseList SET statsStatus = 1 WHERE databaseID = @rebuildStatsID; END; END; /* When everything is said and done, make sure to get rid of our temp table */ DROP TABLE #indexDefragList; DROP TABLE #databaseList; DROP TABLE #processor; IF @debugMode = 1 RAISERROR('DONE! Thank you for taking care of your indexes! :)', 0, 42) WITH NoWait; SET NOCOUNT OFF; RETURN 0 END Go SET Quoted_Identifier OFF SET ANSI_Nulls ON Go
Webcast Tomorrow!
Filed under: PASS, Presentations, SQL 2008, SQL Tips, Syndication
I’m excited to be doing a webcast tomorrow with the infamous illustrious Brent Ozar for Quest’s Pain-of-the-Week. The title is “Getting Started with SQL Server Management Studio,” and as you’ve probably gathered, it’s pretty entry-level stuff. If you read my blog, then chances are you don’t need to watch this webcast. But if you know anyone who’s trying to learn SQL Server or is trying to make the upgrade from 2000 to 2005/2008, this may be a good webcast for them.
I’ve also got a few other speaking engagements coming up:
June 2nd: Cedar Valley .NET User Group
I’ll be reprising my Iowa Code Camp presentation on “SQL Server for the .NET Developer” for CVINETA. This presentation focuses on what you need to know about good table design, indexing strategies, and fragmentation… you know, what you wish every .NET developer knew about SQL Server.
June 11th: PoTW: Time-Saving SQL Server Management Studio Tips & Tricks
I’ll also be doing this webcast with @BrentO as a follow-up to our webcast tomorrow. It will focus on how to save time and improve your sanity by using some neat little tricks in SSMS 2008.
BOL 2008 Update Released
If you haven’t heard, Microsoft released an update to Books Online for SQL Server 2008 yesterday. You can find the download here:
Overhead in Non-Unique Clustered Indexes
Filed under: Internals, Performance & Tuning, SQL 2008, Syndication
I’ve received a couple of questions regarding my article, Performance Considerations of Data Types, and the overhead associated with non-unique clustered indexes. I started to respond via e-mail, but my response was so long I decided to turn it into a blog post instead. 
I should start by clarifying that non-unique clustered indexes do not necessarily consume more space and overhead; it depends on the data stored. If you have duplicate clustered key values, the first instance of the value will be handled as though it were unique. Any subsequent values, however, will incur overhead to manage the uniquifier that SQL Server adds to maintain row uniqueness. This same overhead is also incurred in non-clustered indexes, too, adding to the overall expense of this approach.
I think it helps to actually look at the data, so let’s walk through a few different common scenarios. We’ll create a table with a unique clustered index, a table with a non-unique clustered index but no duplicates, and a table with duplicate key values.
Also, a little warning that I started to write this in SQL Server 2008, and since I’m on a 2008 kick, I decided to leave it that way. You can modify this pretty easily to work in 2005, if necessary.
USE sandbox; Go /* Unique, clustered index, no duplicate values */ CREATE TABLE dbo.uniqueClustered ( myDate DATE Not Null , myNumber INT Not Null , myColumn CHAR(995) Not Null ); CREATE UNIQUE CLUSTERED INDEX CIX_uniqueClustered ON dbo.uniqueClustered(myDate); /* Non-unique clustered index, but no duplicate values */ CREATE TABLE dbo.nonUniqueNoDups ( myDate DATE Not Null , myNumber INT Not Null , myColumn CHAR(995) Not Null ); CREATE CLUSTERED INDEX CIX_nonUniqueNoDups ON dbo.nonUniqueNoDups(myDate); /* Non-unique clustered index, duplicate values */ CREATE TABLE dbo.nonUniqueDuplicates ( myDate DATE Not Null , myNumber INT Not Null , myColumn CHAR(995) Not Null ); CREATE CLUSTERED INDEX CIX_nonUniqueDuplicates ON dbo.nonUniqueDuplicates(myDate);
I’m going to use the date data type in 2008 for my clustered index key. To ensure uniqueness for the first two tables, I’ll iterate through a few years’ worth of dates. This is typical of what you may see in a data mart, where you’d have one record with an aggregation of each day’s data. For the table with duplicate values, I’m going to insert the same date for each row.
/* Populate some test data */ SET NOCOUNT ON; DECLARE @myDate DATE = '1990-01-01' , @myNumber INT = 1; WHILE @myDate < '2010-01-01' BEGIN INSERT INTO dbo.uniqueClustered SELECT @myDate, @myNumber, 'data'; INSERT INTO dbo.nonUniqueNoDups SELECT @myDate, @myNumber, 'data'; INSERT INTO dbo.nonUniqueDuplicates SELECT '2009-01-01', @myNumber, 'data'; SELECT @myDate = DATEADD(DAY, 1, @myDate) , @myNumber += 1; END;
After running the above script, each table should have 7,305 records. This is obviously pretty small for a table, but it’ll serve our purposes. Now let’s take a look at the size of our tables:
/* Look at the details of our indexes */ /* Unique, clustered index, no duplicate values */ SELECT 'unique' AS 'type', page_count, avg_page_space_used_in_percent, record_count , min_record_size_in_bytes, max_record_size_in_bytes FROM sys.dm_db_index_physical_stats(DB_ID(), OBJECT_ID(N'uniqueClustered'), Null, Null, N'Detailed') WHERE index_level = 0 UNION All /* Non-unique clustered index, but no duplicate values */ SELECT 'non-unique, no dups', page_count, avg_page_space_used_in_percent, record_count , min_record_size_in_bytes, max_record_size_in_bytes FROM sys.dm_db_index_physical_stats(DB_ID(), OBJECT_ID(N'nonUniqueNoDups'), Null, Null, N'Detailed') WHERE index_level = 0 UNION All /* Non-unique clustered index, duplicate values */ SELECT 'duplicates', page_count, avg_page_space_used_in_percent, record_count , min_record_size_in_bytes, max_record_size_in_bytes FROM sys.dm_db_index_physical_stats(DB_ID(), OBJECT_ID(N'nonUniqueDuplicates'), Null, Null, N'Detailed') WHERE index_level = 0;
Here’s the results:
type page_count avg_page_space_used_in_percent record_count min_record_size_in_bytes max_record_size_in_bytes ------------------- -------------------- ------------------------------ -------------------- ------------------------ ------------------------ unique 914 99.8055102545095 7305 1009 1009 non-unique, no dups 914 99.8055102545095 7305 1009 1009 duplicates 1044 88.066036570299 7305 1009 1017
I want to point out a couple of things. First, there is no difference in the number of pages between the non-unique clustered index with no duplicates ([nonUniqueNoDups]) and the unique clustered index ([uniqueClustered]). The table with duplicate clustered key values, however, requires 14% more pages to store the same amount of data. Secondly, the [max_record_size_in_bytes] of the [nonUniqueDuplicates] table is 8 bytes more than that of the other two. We’ll discuss why in a minute.
Now let’s take a look at the actual data pages. For this, I’m going to use my page internals proc.
Execute dbo.dba_viewPageData_sp @databaseName = 'sandbox' , @tableName = 'sandbox.dbo.uniqueClustered' , @indexName = 'CIX_uniqueClustered';
I’m not going to post the entire results here, but I want to draw your attention to “m_slotCnt = 8″, which is near the top of the page. That means 8 records are stored on this page. Also, when you look near the end of the first record (Slot 0), you should see the following results:
Slot 0 Offset 0x60 Length 1009 Record Type = PRIMARY_RECORD Record Attributes = NULL_BITMAP Record Size = 1009 Memory Dump @0x00A9C060 00000000: 1000ee03 c3150b01 00000064 61746120 †..î.Ã......data [...] 000003F0: 00†††††††††††††††††††††††††††††††††††. Slot 0 Column 1 Offset 0x4 Length 3 Length (physical) 3 myDate = 1990-01-01 Slot 0 Column 2 Offset 0x7 Length 4 Length (physical) 4 myNumber = 1 Slot 0 Column 3 Offset 0xb Length 995 Length (physical) 995 myColumn = data
Now let’s look at the table that has a non-unique clustered index but no duplicates:
EXECUTE dbo.dba_viewPageData_sp @databaseName = 'sandbox' , @tableName = 'sandbox.dbo.nonUniqueNoDups' , @indexName = 'CIX_nonUniqueNoDups';
The m_slotCnt count is also 8 for this page. This time, let’s glance at the first and second records (Slot 0 and 1 respectively):
Slot 0 Offset 0x60 Length 1009 Record Type = PRIMARY_RECORD Record Attributes = NULL_BITMAP Record Size = 1009 Memory Dump @0x62FDC060 00000000: 1000ee03 c3150b01 00000064 61746120 †..î.Ã......data [...] 000003F0: 00†††††††††††††††††††††††††††††††††††. Slot 0 Column 0 Offset 0x0 Length 4 Length (physical) 0 UNIQUIFIER = 0 Slot 0 Column 1 Offset 0x4 Length 3 Length (physical) 3 myDate = 1990-01-01 Slot 0 Column 2 Offset 0x7 Length 4 Length (physical) 4 myNumber = 1 Slot 0 Column 3 Offset 0xb Length 995 Length (physical) 995 myColumn = data Slot 1 Offset 0x451 Length 1009 Record Type = PRIMARY_RECORD Record Attributes = NULL_BITMAP Record Size = 1009 Memory Dump @0x62FDC451 00000000: 1000ee03 c4150b02 00000064 61746120 †..î.Ä......data [...] 000003F0: 00†††††††††††††††††††††††††††††††††††. Slot 1 Column 0 Offset 0x0 Length 4 Length (physical) 0 UNIQUIFIER = 0 Slot 1 Column 1 Offset 0x4 Length 3 Length (physical) 3 myDate = 1990-01-02 Slot 1 Column 2 Offset 0x7 Length 4 Length (physical) 4 myNumber = 2 Slot 1 Column 3 Offset 0xb Length 995 Length (physical) 995 myColumn = data
We now see a new addition to the row, “UNIQUIFIER = 0.” This is SQL Server’s way of managing row uniqueness internally. You’ll notice that, because the clustered key values are unique, the UNIQUIFIER is set to 0 and the row size is still 1009; for all intents and purposes, the UNIQUIFIER is not consuming any space.
Update: The DBCC God himself, Paul Randal, explained that non-dupes actually have a NULL UNIQUIFIER, which DBCC PAGE displays as a 0. Thanks for explaining, Paul! I wondered about that but chalked it up to SQL voodoo.
Now let’s look at our final case, a non-unique clustered index with duplicate key values:
EXECUTE dbo.dba_viewPageData_sp @databaseName = 'sandbox' , @tableName = 'sandbox.dbo.nonUniqueDuplicates' , @indexName = 'CIX_nonUniqueDuplicates';
Here’s where things get interesting. The m_slotCnt value is now 7, which means we’re now storing 1 record less per page. Let’s look at the details:
Slot 0 Offset 0x60 Length 1009 Record Type = PRIMARY_RECORD Record Attributes = NULL_BITMAP Record Size = 1009 Memory Dump @0x00A9C060 00000000: 1000ee03 df300b01 00000064 61746120 †..î.ß0.....data [...] 000003F0: 00†††††††††††††††††††††††††††††††††††. Slot 0 Column 0 Offset 0x0 Length 4 Length (physical) 0 UNIQUIFIER = 0 Slot 0 Column 1 Offset 0x4 Length 3 Length (physical) 3 myDate = 2009-01-01 Slot 0 Column 2 Offset 0x7 Length 4 Length (physical) 4 myNumber = 1 Slot 0 Column 3 Offset 0xb Length 995 Length (physical) 995 myColumn = data Slot 1 Offset 0x451 Length 1017 Record Type = PRIMARY_RECORD Record Attributes = NULL_BITMAP VARIABLE_COLUMNS Record Size = 1017 Memory Dump @0x00A9C451 00000000: 3000ee03 df300b02 00000064 61746120 †0.î.ß0.....data [...] 000003F0: 000100f9 03010000 00†††††††††††††††††...ù..... Slot 1 Column 0 Offset 0x3f5 Length 4 Length (physical) 4 UNIQUIFIER = 1 Slot 1 Column 1 Offset 0x4 Length 3 Length (physical) 3 myDate = 2009-01-01 Slot 1 Column 2 Offset 0x7 Length 4 Length (physical) 4 myNumber = 2 Slot 1 Column 3 Offset 0xb Length 995 Length (physical) 995 myColumn = data
The first record, Slot 0, looks exactly the same as in the previous table; the UNIQUIFIER is 0 and the row size is 1009. The second record (Slot 1), however, now has a UNIQUIFIER value of 1 and the row size is 1017. If you notice, the “Record Attributes” of Slot 1 are also different, with the addition of “VARIABLE_COLUMNS.” This is because the UNIQUIFIER is stored as a variable column. The extra 8 bytes of overhead break down to 4 bytes to store the UNIQUIFIER, 2 bytes to store the variable column offset, and 2 bytes to store the variable count. The tables we created used all fixed-length columns; you may notice some minor overhead differences if your table already contains variable columns.
To summarize, there is indeed a difference in the page structure between a unique clustered index and a non-unique clustered index; however, there’s only a possible performance and space impact when storing duplicate clustered key values. So there you go, more detail than you ever wanted to know about clustered indexes and uniqueness!
Filtered Indexes: What You Need To Know
Filed under: Performance & Tuning, SQL 2008, Syndication
Filtered indexes are probably my favorite feature in 2008. That’s saying a lot, since there are so many great new features to choose from. In this post, I want to explore a little about how filtered indexes work, how they can be applied, and some of the “gotchas” to be aware of.
First, for those of you who may not yet know about filtered indexes, allow me enlighten you. In short, filtered indexes allow you to create an index on a subset of data using a filtering predicate. Filters can only be applied to non-clustered indexes. The general syntax of a filtered index is:
CREATE NONCLUSTERED INDEX [index_name] ON [table_name] ([column_list]) Include ([column_list]) WHERE [filtered_criteria];
For our purposes, we’re going to be working with the Sales.SalesOrderDetail table in the AdventureWorks database. Let’s look at a specific example. Suppose we have a query that regularly searches on the [SpecialOfferID] column.
SELECT SalesOrderID , COUNT(*) AS 'CountOfLineItem' , SUM(LineTotal) AS 'SumOfLineTotal' FROM Sales.SalesOrderDetail WHERE SpecialOfferID <> 1 GROUP BY SalesOrderID;
We notice that there’s no covering index for this query by looking at the actual execution plan:
Query Plan - Clustered Scan
If this is a commonly executed query, then we’d probably want to toss an index on it. Before we get started, let’s take a look at what the distribution of values are on that column:
SELECT SpecialOfferID , COUNT(*) AS 'rows' FROM Sales.SalesOrderDetail GROUP BY SpecialOfferID ORDER BY COUNT(*) DESC;
Our distribution of values is:
SpecialOfferID rows -------------- ----------- 1 115884 2 3428 3 606 13 524 14 244 16 169 7 137 8 98 11 84 4 80 9 61 5 2
As you can see, [SpecialOfferID] = 1 accounts for 96% of our values. In 2005, we’d create an index that may look something like this:
CREATE NONCLUSTERED INDEX IX_Sales_SalesOrderDetail_SpecialOfferID ON Sales.SalesOrderDetail(SpecialOfferID) Include (SalesOrderID, LineTotal);
Now if we re-run our original query, this is what we see:
Indexed Query Plan
So we’re now performing a non-clustered index seek instead of a clustered index scan. Already this results in some pretty significant performance improvements. To see this, we’re going to use the INDEX query hint to force an index scan. We’re also going to use the DBCC command DROPCLEANBUFFERS, which will allow us to clear the buffer cache and better examine what’s happening with our IO.
SET STATISTICS IO ON; DBCC DropCleanBuffers; SELECT SalesOrderID , COUNT(*) AS 'CountOfLineItem' , SUM(LineTotal) AS 'SumOfLineTotal' FROM Sales.SalesOrderDetail WITH (INDEX(PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID)) WHERE SpecialOfferID <> 1 GROUP BY SalesOrderID; DBCC DropCleanBuffers; SELECT SalesOrderID , COUNT(*) AS 'CountOfLineItem' , SUM(LineTotal) AS 'SumOfLineTotal' FROM Sales.SalesOrderDetail WHERE SpecialOfferID <> 1 GROUP BY SalesOrderID; SET STATISTICS IO OFF;
Clustered Index Scan: Table 'SalesOrderDetail'. Scan count 1, logical reads 1240, physical reads 17, read-ahead reads 1242... NonClustered Index Seek: Table 'SalesOrderDetail'. Scan count 2, logical reads 30, physical reads 4, read-ahead reads 480...
As you can see, the non-clustered (NC) index seek performs quite a bit better. Now let’s create a filtered index and explore what happens:
CREATE NONCLUSTERED INDEX FIX_Sales_SalesOrderDetail_SpecialOfferID_Filtered ON Sales.SalesOrderDetail(SalesOrderID) Include (LineTotal) WHERE SpecialOfferID <> 1;
First, let’s look at the pages consumed by each index:
SELECT i.name, ddips.index_depth, ddips.index_level , ddips.page_count, ddips.record_count FROM sys.indexes AS i Join sys.dm_db_index_physical_stats(DB_ID(), OBJECT_ID(N'Sales.SalesOrderDetail'), Null, Null, N'Detailed') AS ddips ON i.OBJECT_ID = ddips.OBJECT_ID And i.index_id = ddips.index_id WHERE i.name In ('IX_Sales_SalesOrderDetail_SpecialOfferID' , 'FIX_Sales_SalesOrderDetail_SpecialOfferID_Filtered' , 'PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID') AND ddips.index_level = 0;
name index_depth index_level page_count record_count ---------------------------------------------------------- ----------- ----------- ----------- -------------------- PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID 3 0 1234 121317 IX_Sales_SalesOrderDetail_SpecialOfferID 3 0 480 121317 FIX_Sales_SalesOrderDetail_SpecialOfferID_Filtered 2 0 19 5433
If you scroll over, you’ll see that the clustered index consumes the most pages, naturally. The non-filtered NC index consumes less pages than the clustered index because it’s narrower; however, it still consumes more pages than the filtered index because it’s storing every data row. The filtered index, with only 5433 rows stored, is by far our smallest index, consuming 96% less space than our non-filtered NC index.
Because we’re using less space to store this index, we should also see an equivalent performance boost. Let’s verify that this is the case:
SET STATISTICS IO ON; DBCC DropCleanBuffers; SELECT SalesOrderID , COUNT(*) AS 'CountOfLineItem' , SUM(LineTotal) AS 'SumOfLineTotal' FROM Sales.SalesOrderDetail WITH (INDEX(IX_Sales_SalesOrderDetail_SpecialOfferID)) WHERE SpecialOfferID <> 1 GROUP BY SalesOrderID; DBCC DropCleanBuffers; SELECT SalesOrderID , COUNT(*) AS 'CountOfLineItem' , SUM(LineTotal) AS 'SumOfLineTotal' FROM Sales.SalesOrderDetail WHERE SpecialOfferID <> 1 GROUP BY SalesOrderID; SET STATISTICS IO OFF;
NonClustered Index Seek: Table 'SalesOrderDetail'. Scan count 2, logical reads 30, physical reads 4, read-ahead reads 480 Filtered Index Scan: Table 'SalesOrderDetail'. Scan count 1, logical reads 24, physical reads 2, read-ahead reads 22
Filtered Query Plan
As expected, we get the best results with our filtered index scan.
You’ll notice that I did *not* create the index on the [SpecialOfferID] column like I did in [IX_Sales_SalesOrderDetail_SpecialOfferID]. This is because my query doesn’t care what my [SpecialOfferID] value is, just as long as it’s not equal to 1. My non-filtered NC index was created on [SpecialOfferID] because it needed to navigate the B-TREE to find the records where [SpecialOfferID] <> 1. With my filtered index, the query optimizer knows that all of my records already meet the criteria, so doesn’t need to navigate through the index to find the matching results.
We could choose to include the [SpecialOfferID] data in our filtered index, but we’d most likely want to make it an included column rather than part of the index key. In fact, it’s important to note that, if I don’t add [SpecialOfferID] as an included column and I want to return it in the results, i.e.
SELECT SalesOrderID , SpecialOfferID , COUNT(*) AS 'CountOfLineItem' , SUM(LineTotal) AS 'SumOfLineTotal' FROM Sales.SalesOrderDetail WHERE SpecialOfferID <> 1 GROUP BY SalesOrderID , SpecialOfferID;
my filtered index will not be used and I will instead scan on the clustered index once more (assuming [IX_Sales_SalesOrderDetail_SpecialOfferID] does not exist). This is because the filtering criteria is not included anywhere on the actual index page. This is actually good news, in my opinion, since it allows you to create even leaner indexes. And like I already mentioned, if you do need the data returned, you can always add the filtering criteria as included columns.
What if you’re trying to find out whether or not an index is filtered, and what it’s filtered on? The sys.indexes catalog view has been updated in 2008 to include this information:
SELECT name, has_filter, filter_definition FROM sys.indexes WHERE name In ('IX_Sales_SalesOrderDetail_SpecialOfferID' , 'FIX_Sales_SalesOrderDetail_SpecialOfferID_Filtered' , 'PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID');
name has_filter filter_definition ------------------------------------------------------ ---------- ------------------------- FIX_Sales_SalesOrderDetail_SpecialOfferID_Filtered 1 ([SpecialOfferID]<>(1)) IX_Sales_SalesOrderDetail_SpecialOfferID 0 NULL PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID 0 NULL
I personally recommend Kimberly Tripp’s system stored proc, sp_helpindex2. It returns a lot of good information about your indexes, such as included columns and filtering criteria.
That’s all I have for today. Hopefully, you now understand how powerful filtered indexes can be. When used properly, filtered indexes can use less space, consume less IO, and improve overall query performance.
Registered Servers in SSMS
Filed under: Miscellaneous, SQL 2008, SQL Tips, Syndication
In my last blog post, I discussed changing the color of the status bar in SSMS 2008. I received a couple of comments and even an e-mail discussing how this doesn’t seem to always work. After playing with it for a little bit, I’ve found that the status bar color needs to be set in both Query->Connection->Connect/Change Connection… (here-in referred to as simply the Query menu) and Registered Servers.
Let’s run through this. First, connect to an instance with any color using the Query menu.
Now, create a new registered server. Make sure to use the same server.
New Server Registration
Pick a color, but make sure that it’s different than the previous color. This is just for demonstration purposes only. Since the whole point is to have a consistent color, you would normally use the same color in both connection methods for the same server.
Pick a color
Open a new query window via Registered Servers.
New Query Window
Registered Server Query Window
Here’s what happens when I connect to the same server using both Registered Servers (left) and another window using the Query menu (right).
SSMS - Same Server, Different Colors
For anyone who’s using both the Query menu and Registered Servers to connect to servers, then you should walk through the process of connecting to each server via both means and changing the colors to ensure consistency. I did this for 22 servers and it took me less than 10 minutes.
I hope that helps clear up some of the confusion.
Source: http://sqlfool.com/2009/03/registered-servers-in-ssms/
Subscribe to my blog!
It's easy, free, and will keep you informed of my latest SQL tips and scripts.Like what you see? Subscribe! 
Around the Web
SQLRockstar Blog Ranking
