SQL Server 2008 R2 Service Pack 1 provides a new set of Events in Extended Events to collect performance counter data from the Windows OS that would be really useful to monitoring SQL Server.  The first place I can find that they were mentioned is on a blog post by Mark Weber, a PFE for SQL and SAP at Microsoft.  However, a few weeks ago a question was asked about these counters one of the forums and the question was around how to use them.  I looked at the Events and found out that they aren’t really useable in their current implementation, something that is disappointing since being able to collect the data provided by these Events would really benefit most DBA’s out there. 

If you look at the Events and columns, these events collect information about the Logical Disk, Processor, Process for the SQL instance, and System at 15 second intervals and makes the data available through Extended Events.

SELECT name, description
FROM  sys.dm_xe_objects
WHERE name like 'perfobject_%'

If we look at the columns, we’ll see that the columns actually represent the individual counters under the categories exposed by the Event names. For example, the perfobject_logicaldisk event returns the following columns:

SELECT
    object_name, 
    name AS column_name, 
    description
FROM  sys.dm_xe_object_columns
WHERE object_name = 'perfobject_logicaldisk'
  AND column_type = 'data'
ORDER BY object_name, name

This all seems good, until we actually use the Events in an Event Session and take a look at the data being returned.

CREATE EVENT SESSION [XE_PerfCounters] 
ON SERVER 
ADD EVENT sqlserver.perfobject_logicaldisk 
ADD TARGET package0.ring_buffer;
GO

Unfortunately, the counters are returning Raw values for the Event and the necessary Base counters that are required to give these values any useful meaning have been left out of the Events data.  Looking at this in my test environment, it appears the counter values pulled for the perfobject_ Events are pulled directly from Win32_PerfRawData_PerfDisk_LogicalDisk, but if you look at the CookingType requirements for the counters in Win32_PerfFormattedData_PerfDisk_LogicalDisk the raw values have to be calculated by their base values for them to have meaning:

I’ve submitted Connect Item 725167 for this and I really hope that this one gets fixed in a future Cumulative Update or Service Pack.

In the first post in this series, SQL Server 2012 Extended Events Update - 1- Introducing the SSMS User Interface, we looked at how to use the New Session Wizard in SQL Server 2012 to define an Event Session.  In this post we’ll compare the Wizard to the standard New Event Session dialog that can also be used for creating and Event Session in SQL Server 2012.  The New Event Session dialog is the same dialog that is used for editing an existing Event Session on the server, and can be accessed from the Extended Events node in Object Explorer, just like the New Session Wizard can be.

Rather than opening up with an Introduction Page, the New Session dialog opens directly allowing you to begin configuring your session.  I like to think of the New Session dialog as the power user method of creating a new session, and as we’ll see in this post, it can actually takes less steps to configure an Event Session using the New Session dialog over the New Session Wizard.

The General page of the New Session dialog allows you to specify a name for the Event Session, as well as to select a template to create the Event Session from.  You also have the option to specify whether the Event Session should start automatically when SQL Server starts, whether to start the Event Session immediately after creating it with the dialog, and whether you want to immediate begin viewing the Event Session data in the Live Viewer.  If you compare this page to the first three, and final pages of the wizard you will see that we have a much more concise configuration using the dialog so far.  If you click on the Events page on the left hand side, the Events page will show in the dialog allowing you to customize the events being collected by the Event Session.

On first look, the Events page looks very similar to the Select Events To Capture page from the Wizard.  The Event search functionality, and the ability to filter the events by Category and Channel, as well as how you add and remove Events from the Event Session is identical.  However, in the New Session dialog, two additional buttons exist, a Configure button, (circled in red) that allows you to begin configuring the Events that have been added to the Event Session, and a Select button, (circled in green) that allows you to return to the Event selection screen from the configuration screen.  When the buttons are clicked, the screen will collapse/expand left and right.

The biggest difference between the functionality provided by the New Session Wizard versus the New Session dialog is the level of granularity that you have with assigning Actions or Global Fields, and Predicates or filtering, to the Events that have been added to the Event Session.  In the New Session Wizard, any Action or Predicate that is added to the Event Session, is added across the board to all of the Events in the session.  The same functionality can be achieved using the New Session dialog by using the multi-select functionality of the UI to select all of the Events, and then adding the appropriate Actions and Predicates.  However, typically we don’t actually need the Actions and Filters applied to every Event in the Event Session, and since Actions and global Predicates incur an overhead for data collection, even though it is incredibly small, as a performance best practice.  By selecting a single Event, new Actions can be added to the the Event, or as shown below, filtering can be applied at the individual Event level, which allows the filtering definition to be against any column on the Event, not just the shared subset of columns, or global files, across all Events.

If multiple events share the same columns, for example, the sqlserver.sql_batch_completed and sqlserver.sql_statement_completed Events in our session, you can also multi-select those Events and define filtering specific to both of those Events.

Complex Predicates can be defined through the use of the right-click context menu in the Filter table.  Keep in mind that Predicates in Extended Events allow short circuiting logic to occur, so the order of Predicates matters during evaluation time.  The context menu will allow you to insert a new Predicate above or below the currently selected Predicate in the UI, add or delete additional clauses to the existing Predicate list, group subsets of clauses together so that they evaluate as a complete set, ungroup previously grouped sets of clauses, and to toggle the Not operator which evaluates for the negation of the clause being configured.

A really good example of a complex predicate configuration can be seen by looking at the system_health Event Session and the Predicate on the sqlserver.error_reported event. 

The Event Fields tab, will allow you to turn on/off the collection of any customizable columns for the Event that is currently selected.  For example, the sql_batch_completed Event has a customizable column for the batch_text, which is turned on by default.  If you don’t need the batch text, for example, you may be collecting the tsql_frame action which is much smaller because you know that you will be able to get the batch information from the cache at the point you are analyzing the results, you can turn it off by unchecking the checkbox next to it.

The Data Storage page of the dialog, allows you to configure the targets for the Event Session.  The biggest difference here is that you get full use of all of the targets available in Extended Events, not just the event_file and ring_buffer targets provided by the Wizard, though these will typically be the targets that you will use the most.

The Advanced page, allows you to customize the Event Session Options to define how the session will be setup in the Extended Events Engine when it starts.

Once all of the configuration for an Event Session has been completed, you can script the Event Session DDL using the standard SSMS Script button at the top of the UI, or you can create the Event Session immediately by clicking OK.  If you need to change the Event Session definition after creating it, the Session Properties dialog can be opened from the right-click context menu in Object Explorer for the session.  The Session Properties dialog is exactly the same as the New Session dialog.

In the next post we’ll look at the target data viewer in SLQ Server 2012 and how to use it for analyzing the captured Events from an Event Session.

Yesterday I was asked by email about a problem that someone encountered associated with a SQL Server Failover Cluster configuration that I have run into a number of times myself, and I have had questions about repeatedly in the past.  The problem is that during the SQL Server Setup Validation of the environment, a warning is raised stating that the Network Binding Order is incorrect for the environment.  If you click on the specific Setup Validation Warning you will get a box like the following:

What this warning is telling you, is that you have multiple network interfaces configured on the server, and the default binding order for the interfaces places the current Domain access interface in a position that is not the first interface for the server.  This can be changed by adjusting the network binding order for the server in the Network Properties.  To make this change, open up the Network and Sharing Center by clicking on the network connection on the system tray and then clicking the Open Network and Sharing Center link.

When the Network and Sharing Center opens, click on the Change adapter settings link on the left hand side.

Then when the Network Connections window opens, you have to press the Alt key to open up the window context menu so that you can then click on Advanced and Advanced Settings.

When the Advanced Settings window opens, you will be on the Adapters and Bindings tab, which allows you to change the binding order for the network interfaces on the server.

If you click on the appropriate interface for the domain connection, you can move it up to the first position in the binding order which will eliminate the warning in SQL Server Setup.

If changing the binding order does not resolve the warning, there may be a disabled or ghost network adapter in the system as discussed in the following KB article. (http://support.microsoft.com//kb/955963)

For the most part I have been relatively quiet about the coming changes in SQL Server 2012 with regards to Extended Events.  Primarily this has been to allow the new features of the product to become fully baked to ensure that the information would continue to be applicable as the product lifecycle progressed, and there have been a number of major changes that have made this decision a really good one.  With SQL Server 2012 in it’s RC0 phase, and based on the responses I have seen to a number of feedback items for bugs on Microsoft SQL Server Connect, like the recent one for the template issue I blogged about on my blog post, Workaround for Bug in Activity Tracking Event Session Template in 2012 RC0, I’ve decided to go ahead and start a new series of posts that outline the new features of Extended Events in SQL Server 2012.  I can’t think of a better way to start off a series on the new features in SQL Server 2012 for Extended Events than the new SQL Server Management Studio User Interface for Extended Events in a couple of blog posts.

For this initial post, we’ll take a look at some of the features that replace existing SQL Profiler functionality that most DBA’s tend to use in their day to day operations.  To start off this topic, the first thing you need to know is that there is a new node for Extended Events under the Management folder in Object Explorer for SQL Server 2012.

If you happen to be connecting to a SQL Server 2008 server using SSMS from SQL Server 2012, this node will not exist.  This is due to the fact that UI for SSMS in SQL Server 2012 are not backwards compatible with SQL Server 2008, even though Extended Events exist in SQL Server 2008.  At some point when SQL Server 2012 actually releases to manufacturing (RTM), I will release an update to my SSMS Addin for Extended Events that back ports compatibility for SQL Server 2008 to Management Studio 2012, allowing full integration between SSMS 2012 and SQL Server 2008.

Within the scope of SQL Server 2012, the UI provides a lot of new functionality that should simplify the implementation and usage of Extended Events for most DBA’s.  One of the best enhancements is the ability to create an event session using the New Session Wizard to define an event session based with the least number of steps possible, possibly using an template for the event session, or manually defining a custom configuration that is applied to all of the events configured for the session.

By opening the New Session Wizard, immediately a Introductory page is presented that can be bypassed by selecting the option to Do not show this page again:

Since this page will typically slow down the creation of an event session for use, I would typically check this option before clicking on Next. The Set Session Properties page will allow you to specify a name for the session as well as whether or not the event session will startup automatically when SQL Server starts.  This can be very useful for troubleshooting an infrequent problem that does not predictably occur and you need to ensure that the session data is collected whenever the problem next occurs.  Some examples of where this might be applied will be shown in future posts in this series.

The New Session Wizard provides the ability to create the new event session based on a previously created template, or one of the templates provided by default with SQL Server 2012.  Keep in mind that in the RC0 build, and unfortunately the RTM release of SQL Server 2012, the Activity Tracking template has a bug in the XML definition that will cause this UI to error out.  This was documented by Mike Wachal, the PM for Extended Events at Microsoft on his blog post Activity Tracking event session template is broken, but a fix for the problem in the XML is available in my blog post Workaround for Bug in Activity Tracking Event Session Template in 2012 RC0.  After replacing the broken template with the one attached to my previous blog post we can select it in the UI.

The alternative to using an existing template is to create a blank event session by selecting the option to Do not use a template.  If a template is selected, once you click Next the events from the template will be displayed in the Select Events To Capture page, which also displays the events available in Extended Events and information about the data returned by the events.  The event library can be searched a number of ways to simplify finding the correct events for the session.  The most common way to search, once you start learning the events that are available, would be to start typing the event name in the textbox (green circle below) and the results will dynamically begin filtering out the events that don’t match the search text.  If you click on a specific event in the table, the event description and information about the columns returned by the event will be displayed (purple box below).

However, if you don’t know the specific events that you want, but you know the general category that the events apply to you can make use of other search options in the UI as well.  Events in Extended Events are broken down by two attributes, a Category (Keyword in the DMVs) and a Channel, that make them compatible with Event Tracing for Windows (ETW).  The Category is similar to the trace category that the existing Trace events have and can be used for logically grouping events similar to the way SQL Profiler groups events in previous versions of SQL Server.  The Channel aligns with the channels you would see with ETW.  By default one of the Channels is excluded in the UI, the Debug Channel.  Debug events are focused towards internal debugging tasks and are not of general purpose use by most DBAs.  These events tend to be counting in nature, or can fire incredibly frequently. If you want to see the Debug events in the UI, you can click the drop down and check the checkbox for the Debug Channel and they will be available.

Additionally you can filter out specific categories by clicking the Category dropdown and unchecking specific Category names from the selection.  To add an Event to the session, you can double click on the event, or you can select multiple events using Ctrl or Shift + click on the event names and then clicking the arrow that points to the right.  Alternately, you can also remove events using a double click or by highlighting the event and clicking the arrow that points to the left.

After adding events to the session, the next page allows you to specify the Global Fields, known as actions in Extended Events, that you want added to each of the events in the session.  If you look at the columns being returned by the events in Extended Events, there are significantly fewer data elements being returned at the individual event level.  Many of the trace columns map to the Global Fields (actions) in Extended Events and can be added as needed to the events.  The goal was to minimize the size of the firing events allowing additional information to be added as needed to maximize the performance of Extended Events firing.

After selecting the global fields to add to the Event Session and clicking Next, the Set Session Event Filters page is displayed where you can define filtering (known as predicates) on the events in the session.  Any filters that were configured as a part of the template will be displayed in the upper table, while new filters that will be applied to all of the events in the session can be added to the bottom table.  New filters can only be created on the columns that are available for all of the events in the event session, which typically there won’t be any if using multiple events, or on the global fields available to Extended Events.

This is a very restrictive functionality of the New Session Wizard that was put in place to provide a parity for session creation to what most users would expect from SQL Server Profiler.  I’ll show more about how this is not the best thing when we look at the other parts of the new UI in SSMS in other posts.  Once the filters have been created, the next step is to define the event storage.  The New Session Wizard restricts you to the event_file and ring_buffer targets, which are going to be the most commonly used targets by most DBAs since they retain the full event data and do not apply additional processing to the events.  For an event session that is going to be collecting data for a long period of time, or data that generates at a fast rate, the event_file target should be used, and similar to SQL Trace you can setup the maximum file size and whether or not file rollover should occur.  If the event session is going to be collecting data for a short period of time or where the event predicates will restrict the session to only collecting a small amount of data, the ring_buffer target will generally be a good choice.

Once the data storage has been configured the session can be created by clicking Finish, or you can click Next to get to the Summary page which will allow you to review all the configured options for the Event Session and Script the session definition for further changes to the DDL if necessary.

Once the Event Session is created, the last page provides you the opportunity to start the event session and to open the Live Data Viewer for the event session which is similar to the SQL Server Profiler view from SQL Trace.

In the next blog post I’ll show the New Session dialog which is not a wizard based implementation and why it provides a much more robust method of creating an event session in SQL Server 2012.

My latest article on Simple-Talk was published this morning.  In this article I dig into the actual meaning of one of the performance counters I often see mentioned on the forums, but in a completely incorrect context.

Great SQL Server Debates: Buffer Cache Hit Ratio

In a word; YES! In a lot more words, not always in the way that we want it to, but there are plenty of cases where it actually works and changes are made to the product as a result.

Now with that said, it doesn’t work all the time, and it helps to realize that what is important to us as an individual user might not be important to the product as a whole.  Before you post comments, I am sure that there are plenty of cases out there where people can say that Microsoft Connect for SQL Server is broken.  I have personally been irritated to the point of posting negative comments on Twitter about the whole Connect process.  I feel that it is about time that I show the other side of the story as well and talk about some Connect successes that have occurred in the past year, and of course what better topic to do this with than Extended Events.  Over the next few weeks, I’ll post a couple of different examples of Connect actually working and bringing about changes to the product that are beneficial to the community, starting with this post.

Extended Events does not track insert statements

This Connect item is actually incorrectly titled and is based on some confusion about what the sqlserver.plan_handle action actually returns when executed in the engine.  I blogged about this with much more detail last year in my blog post; What plan_handle is Extended Events sqlserver.plan_handle action returning?

If we revisit the Connect item there is a note that the sql_statement_completed event in SQL Server 2012 now includes a parameterized_plan_handle customizable column that can be used to retrieve the parameterized plan handle for statements that are auto-parameterized by SQL Server during their execution.  Taking the same original demo code from my previous blog post, we can now see how this Connect item has improved the ability to find information about plan caching in SQL Server 2012:

-- Create the Event Session
IF EXISTS(SELECT * 
          FROM sys.server_event_sessions 
          WHERE name='SQLStmtEvents')
    DROP EVENT SESSION SQLStmtEvents 
    ON SERVER;
GO
CREATE EVENT SESSION SQLStmtEvents
ON SERVER
ADD EVENT sqlserver.sql_statement_completed(
    SET collect_parameterized_plan_handle = 1
    ACTION (sqlserver.client_app_name,
            sqlserver.plan_handle,
            sqlserver.sql_text,
            sqlserver.tsql_stack,
            package0.callstack,
            sqlserver.request_id)
--Change this to match the AdventureWorks, 
--AdventureWorks2008 or AdventureWorks2008 SELECT DB_ID('AdventureWorks2008R2')
WHERE sqlserver.database_id=9
)
ADD TARGET package0.ring_buffer
WITH (MAX_DISPATCH_LATENCY=5SECONDS, TRACK_CAUSALITY=ON)
GO
 
-- Start the Event Session
ALTER EVENT SESSION SQLStmtEvents 
ON SERVER 
STATE = START;
GO
 
-- Change database contexts and insert one row
USE AdventureWorks2008R2;
GO
INSERT INTO [dbo].[ErrorLog]([ErrorTime],[UserName],[ErrorNumber],[ErrorSeverity],[ErrorState],[ErrorProcedure],[ErrorLine],[ErrorMessage])
VALUES(getdate(),SYSTEM_USER,-1,-1,-1,'ErrorProcedure',-1,'An error occurred')
GO 10
 
-- Drop the Event
ALTER EVENT SESSION SQLStmtEvents
ON SERVER
DROP EVENT sqlserver.sql_statement_completed;
GO

-- Retrieve the Event Data from the Event Session Target
SELECT
    event_data.value('(event/@name)[1]', 'varchar(50)') AS event_name,
    event_data.value('xs:hexBinary((event/data[@name="parameterized_plan_handle"]/value)[1])', 'varbinary(64)') as parameterized_plan_handle,
    event_data.value('xs:hexBinary((event/action[@name="plan_handle"]/value)[1])', 'varbinary(64)') as plan_handle,
    event_data.value('(event/action[@name="sql_text"]/value)[1]', 'varchar(max)') AS sql_text
FROM(    SELECT evnt.query('.') AS event_data
        FROM
        (   SELECT CAST(target_data AS xml) AS TargetData
            FROM sys.dm_xe_sessions AS s
            JOIN sys.dm_xe_session_targets AS t
                ON s.address = t.event_session_address
            WHERE s.name = 'SQLStmtEvents'
              AND t.target_name = 'ring_buffer'
        ) AS tab
        CROSS APPLY TargetData.nodes ('RingBufferTarget/event') AS split(evnt) 
     ) AS evts(event_data)

If we look at the output of this, we will get the parameterized plan handle for each subsequent call of the statement after the initial call caches the parameterized plan into the cache.

If we plug one of the original plan_handle values from the sqlserver.plan_handle action into a query of sys.dm_exec_cached_plans() it will return nothing, but using the new parameterized_plan_handle value from the customizable column will give us the appropriate cached plan for the statement from cache:

-- Use the plan_handle from one of the Events action to get the query plan
DECLARE @plan_handle varbinary(64) = 0x06000900DFC9DD12608B18EE0100000001000000000000000000000000000000000000000000000000000000
SELECT * 
FROM sys.dm_exec_query_plan(@plan_handle)
GO

-- Use the parameterized_plan_handle from the same Events to get the query plan
DECLARE @plan_handle varbinary(64) = 0x06000900DD8D6D08601E70EE01000000010000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
SELECT * 
FROM sys.dm_exec_query_plan(@plan_handle)
GO

Now, you might point out the different lengths of the plan handles in the above two queries.  If you look back at the source, the same code is being used to perform the xhexBinary conversion in the XML so the values are exactly the same as what was originally provided by the event and the action.  The non-parameterized plan is not cached because it is not likely to be reused, which is why we have the auto-parameterized plan in cache.

In SQL Server 2012 RC0 there are a number of event session templates provided that make creating a commonly used session easier using the Event Session Wizard in SQL Server Management Studio.  One of these has a bug in it’s definition XML file that was filed in the following connect item:

https://connect.microsoft.com/SQLServer/feedback/details/705840/the-object-sqlserver-event-sequence-does-not-exist#tabs

If you attempt to pick the Activity Tracking template you will get the following error:

The error is occurring because the event_sequence action is provided by package0 and not sqlserver. To work around this, you can edit the template file and replace the sqlserver package references with package0. The template is saved in the following location:

C:\Program Files (x86)\Microsoft SQL Server\110\Tools\Templates\sql\xevent\xe_activity.xml

If you do a find for:

<action package="sqlserver" name="event_sequence" />

and replace it with:

<action package="package0" name="event_sequence" />

th template will work correctly once saved. This will at least let you play around with this while Microsoft works out the bug in the template XML.  A copy of the corrected file is attached to this blog post as well.

xe_activity.xml (18.28 kb)

After a recent discussion about Lock Pages in Memory on Twitter, I wrote an article that talks about the history of Lock Pages in Memory and the differing opinions on the appropriate usage of Lock Pages in Memory for 64 bit instances of SQL Server. 

Great SQL Server Debates: Lock Pages in Memory

Thanks to Brent Ozar (Blog|Twitter) and Glenn Berry (Blog|Twitter) for their part in discussing this on twitter, by email, and for reviewing the article before I submitted it.

Just over a year ago I blogged about the enhancements that were made to the sqlserver.page_split Event in SQL Server 2012 to make it easier to identify what the splitting object was and the type of split that was being performed.  Sadly what I discovered writing that post was that even with the extra information about the split type, the event didn’t give you enough information to really focus on the problematic splits that lead to fragmentation and page density issues in the database.  I didn’t do a whole lot with this again until recently when a question was posted by Ami Levin (Blog | Twitter) on the MVP email list that commented that the page_split event was broken in SQL Server 2012 based on a presentation he’d seen by Guy Glantser (Blog | Twitter).

Let me start off by saying, the event isn’t broken, it tracks page splits, but it doesn’t differentiate between an end page split that occurs for an ever increasing index, versus a mid-page split for a random index that leads to fragmentation and page density issues in the database.  Both of these are technically splits inside the storage engine, even if we as DBA’s don’t really care about the end-page split for a increasing key value like an IDENTITY column in the database.  I had Ami pass my information along to the presenter and we traded a few emails on the subject of tracking splits with the specific focus on trying to pull out the mid-page, fragmenting splits.  While going through things for the third time, it dawned on me that this is incredibly simple, based one of the demo’s that was sent to me.  Just over a year ago, I also blogged about tracking transaction log activity in SQL Server 2012 using the sqlserver.transaction_log event, which can be used to track mid-page splits in a database.

Last year when I wrote about the sqlserver.transaction_log event, there were 10 columns output by the event in CTP1, but as of RC0, the events output has changed and only 9 columns are output by the event.

SELECT 
    oc.name, 
    oc.type_name, 
    oc.description
FROM sys.dm_xe_packages AS p
INNER JOIN sys.dm_xe_objects AS o
    ON p.guid = o.package_guid
INNER JOIN sys.dm_xe_object_columns AS oc
    ON oc.object_name = o.name
        AND oc.object_package_guid = o.package_guid
WHERE o.name = 'transaction_log'
  AND oc.column_type = 'data';

For the purposes of identifying the mid-page splits, we want to look at the operation column that is output by the event, which contains the specific operation being logged.  In the case of a mid-page split occurring, the operation will be a LOP_DELETE_SPLIT, which marks the delete of rows from a page as a result of the split.  To build our event session, we are going to need the map_key for the LOP_DELETE_SPLIT log_op map.  This can be obtained from the sys.dm_xe_map_values DMV:

SELECT *
FROM sys.dm_xe_map_values
WHERE name = 'log_op'
  AND map_value = 'LOP_DELETE_SPLIT';

With the map_key value, we have a couple of ways to collect the information with our targets.  We could collect everything into an event_file, but that doesn’t really make sense for this event.  Instead the best target for this type of information is the histogram target which will bucket our results based on how we configure the target and tell us how frequently the event fires based on our bucketing criteria.  If we don’t know anything about the server in question, we can start off with a very general event session that has a predicate on the operation only, and then aggregate the information in the histogram target based on the database_id to find the databases that have the most mid-page splits occurring in them in the instance.

-- If the Event Session exists DROP it
IF EXISTS (SELECT 1 
            FROM sys.server_event_sessions 
            WHERE name = 'SQLskills_TrackPageSplits')
    DROP EVENT SESSION [SQLskills_TrackPageSplits] ON SERVER

-- Create the Event Session to track LOP_DELETE_SPLIT transaction_log operations in the server
CREATE EVENT SESSION [SQLskills_TrackPageSplits]
ON    SERVER
ADD EVENT sqlserver.transaction_log(
    WHERE operation = 11  -- LOP_DELETE_SPLIT 
)
ADD TARGET package0.histogram(
    SET filtering_event_name = 'sqlserver.transaction_log',
        source_type = 0, -- Event Column
        source = 'database_id');
GO
        
-- Start the Event Session
ALTER EVENT SESSION [SQLskills_TrackPageSplits]
ON SERVER
STATE=START;
GO

This event session will allow you to track the worst splitting database on the server, and the event data can be parsed out of the histogram target.  To demonstrate this, we can create a database that has tables and indexes prone to mid-page splits and run a default workload to test the event session:

USE [master];
GO
-- Drop the PageSplits database if it exists
IF DB_ID('PageSplits') IS NOT NULL
BEGIN
    ALTER DATABASE PageSplits SET SINGLE_USER WITH ROLLBACK IMMEDIATE;
    DROP DATABASE PageSplits;
END
GO
-- Create the database
CREATE DATABASE PageSplits
GO
USE [PageSplits]
GO
-- Create a bad splitting clustered index table
CREATE TABLE BadSplitsPK
( ROWID UNIQUEIDENTIFIER NOT NULL DEFAULT NEWID() PRIMARY KEY,
  ColVal INT NOT NULL DEFAULT (RAND()*1000),
  ChangeDate DATETIME2 NOT NULL DEFAULT CURRENT_TIMESTAMP);
GO
--  This index should mid-split based on the DEFAULT column value
CREATE INDEX IX_BadSplitsPK_ColVal ON BadSplitsPK (ColVal);
GO
--  This index should end-split based on the DEFAULT column value
CREATE INDEX IX_BadSplitsPK_ChangeDate ON BadSplitsPK (ChangeDate);
GO
-- Create a table with an increasing clustered index
CREATE TABLE EndSplitsPK
( ROWID INT IDENTITY NOT NULL PRIMARY KEY,
  ColVal INT NOT NULL DEFAULT (RAND()*1000),
  ChangeDate DATETIME2 NOT NULL DEFAULT DATEADD(mi, RAND()*-1000, CURRENT_TIMESTAMP));
GO
--  This index should mid-split based on the DEFAULT column value
CREATE INDEX IX_EndSplitsPK_ChangeDate ON EndSplitsPK (ChangeDate);
GO
-- Insert the default values repeatedly into the tables
WHILE 1=1
BEGIN
    INSERT INTO dbo.BadSplitsPK DEFAULT VALUES;
    INSERT INTO dbo.EndSplitsPK DEFAULT VALUES;
    WAITFOR DELAY '00:00:00.005';
END
GO

If we startup this workload and allow it to run for a couple of minutes, we can then query the histogram target for our session to find the database that has the mid-page splits occurring.

-- Query the target data to identify the worst splitting database_id
SELECT 
    n.value('(value)[1]', 'bigint') AS database_id,
    DB_NAME(n.value('(value)[1]', 'bigint')) AS database_name,
    n.value('(@count)[1]', 'bigint') AS split_count
FROM
(SELECT CAST(target_data as XML) target_data
 FROM sys.dm_xe_sessions AS s 
 JOIN sys.dm_xe_session_targets t
     ON s.address = t.event_session_address
 WHERE s.name = 'SQLskills_TrackPageSplits'
  AND t.target_name = 'histogram' ) as tab
CROSS APPLY target_data.nodes('HistogramTarget/Slot') as q(n)

With the database_id of the worst splitting database, we can then change our event session configuration to only look at this database, and then change our histogram target configuration to bucket on the alloc_unit_id so that we can then track down the worst splitting indexes in the database experiencing the worst mid-page splits.

-- Drop the Event Session so we can recreate it 
-- to focus on the highest splitting database
DROP EVENT SESSION [SQLskills_TrackPageSplits] 
ON SERVER

-- Create the Event Session to track LOP_DELETE_SPLIT transaction_log operations in the server
CREATE EVENT SESSION [SQLskills_TrackPageSplits]
ON    SERVER
ADD EVENT sqlserver.transaction_log(
    WHERE operation = 11  -- LOP_DELETE_SPLIT 
      AND database_id = 8 -- CHANGE THIS BASED ON TOP SPLITTING DATABASE!
)
ADD TARGET package0.histogram(
    SET filtering_event_name = 'sqlserver.transaction_log',
        source_type = 0, -- Event Column
        source = 'alloc_unit_id');
GO

-- Start the Event Session Again
ALTER EVENT SESSION [SQLskills_TrackPageSplits]
ON SERVER
STATE=START;
GO

With the new event session definition, we can now rerun our problematic workload for a 2 minute period and look at the worst splitting indexes based on the alloc_unit_id’s that are in the histogram target:

 

-- Query Target Data to get the top splitting objects in the database:
SELECT
    o.name AS table_name,
    i.name AS index_name,
    tab.split_count,
    i.fill_factor
FROM (    SELECT 
            n.value('(value)[1]', 'bigint') AS alloc_unit_id,
            n.value('(@count)[1]', 'bigint') AS split_count
        FROM
        (SELECT CAST(target_data as XML) target_data
         FROM sys.dm_xe_sessions AS s 
         JOIN sys.dm_xe_session_targets t
             ON s.address = t.event_session_address
         WHERE s.name = 'SQLskills_TrackPageSplits'
          AND t.target_name = 'histogram' ) as tab
        CROSS APPLY target_data.nodes('HistogramTarget/Slot') as q(n)
) AS tab
JOIN sys.allocation_units AS au
    ON tab.alloc_unit_id = au.allocation_unit_id
JOIN sys.partitions AS p
    ON au.container_id = p.partition_id
JOIN sys.indexes AS i
    ON p.object_id = i.object_id
        AND p.index_id = i.index_id
JOIN sys.objects AS o
    ON p.object_id = o.object_id
WHERE o.is_ms_shipped = 0;

With this information we can now go back and change our FillFactor specifications and retest/monitor the impact to determine whether we’ve had the appropriate reduction in mid-page splits to accommodate the time between our index rebuild operations:

-- Change FillFactor based on split occurences
ALTER INDEX PK__BadSplit__97BD02EB726FCA55 ON BadSplitsPK REBUILD WITH (FILLFACTOR=70)
ALTER INDEX IX_BadSplitsPK_ColVal ON BadSplitsPK REBUILD WITH (FILLFACTOR=70)
ALTER INDEX IX_EndSplitsPK_ChangeDate ON EndSplitsPK REBUILD WITH (FILLFACTOR=80)
GO

-- Stop the Event Session to clear the target
ALTER EVENT SESSION [SQLskills_TrackPageSplits]
ON SERVER
STATE=STOP;
GO

-- Start the Event Session Again
ALTER EVENT SESSION [SQLskills_TrackPageSplits]
ON SERVER
STATE=START;
GO

With the reset performed we can again start up our workload generation and begin monitoring the effect of the FillFactor specifications on the indexes with our code.  After another 2 minute period, the following splits were noted.

With this information we can go back and again attempt to tune our FillFactor values for the worst splitting indexes and rinse/repeat until we determine the best FillFactor for each of the indexes to minimize splits.  This is an incredibly powerful tool for the DBA moving into SQL Server 2012, and will definitely change how we perform index fragmentation analysis and troubleshoot problems with excessive log generation in SQL Server 2012 onwards.

Cheers!

This afternoon, Orson Weston (Twitter), asked how to find the difference between two binary sets of schema options for replication on the #SQLHelp hash tag on twitter.  The valid values for the @schema_options parameter in replication are documented in the BOL Topic for sp_addarticle (http://msdn.microsoft.com/en-us/library/ms173857.aspx).  However, just having a table of values doesn’t really help you figure out what is different between two binary @schema_options values without doing some bitwise operations on the values.  To do this, we can create a table variable to hold the valid values and descriptions for the schema options and then use the & (Bitwise AND) operator to find which options are set for each of the binary values.

DECLARE @options TABLE
(Value VARBINARY(8), [Description] VARCHAR(1000))

INSERT INTO @options (Value, Description)
VALUES 
    (0x00, 'Disables scripting by the Snapshot Agent and uses creation_script.'),
    (0x01, 'Generates the object creation script (CREATE TABLE, CREATE PROCEDURE, and so on). This value is the default for stored procedure articles.'),
    (0x02, 'Generates the stored procedures that propagate changes for the article, if defined.'),
    (0x04, 'Identity columns are scripted using the IDENTITY property.'),
    (0x08, 'Replicate timestamp columns. If not set, timestamp columns are replicated as binary.'),
    (0x10, 'Generates a corresponding clustered index. Even if this option is not set, indexes related to primary keys and unique constraints are generated if they are already defined on a published table.'),
    (0x20, 'Converts user-defined data types (UDT) to base data types at the Subscriber. This option cannot be used when there is a CHECK or DEFAULT constraint on a UDT column, if a UDT column is part of the primary key, or if a computed column references a UDT column. Not supported for Oracle Publishers.'),
    (0x40, 'Generates corresponding nonclustered indexes. Even if this option is not set, indexes related to primary keys and unique constraints are generated if they are already defined on a published table.'),
    (0x80, 'Replicates primary key constraints. Any indexes related to the constraint are also replicated, even if options 0x10 and 0x40 are not enabled.'),
    (0x100, 'Replicates user triggers on a table article, if defined. Not supported for Oracle Publishers.'),
    (0x200, 'Replicates foreign key constraints. If the referenced table is not part of a publication, all foreign key constraints on a published table are not replicated. Not supported for Oracle Publishers.'),
    (0x400, 'Replicates check constraints. Not supported for Oracle Publishers.'),
    (0x800, 'Replicates defaults. Not supported for Oracle Publishers.'),
    (0x1000, 'Replicates column-level collation.'),
    (0x2000, 'Replicates extended properties associated with the published article source object. Not supported for Oracle Publishers.'),
    (0x4000, 'Replicates UNIQUE constraints. Any indexes related to the constraint are also replicated, even if options 0x10 and 0x40 are not enabled.'),
    (0x8000, 'This option is not valid for SQL Server 2005 Publishers.'),
    (0x10000, 'Replicates CHECK constraints as NOT FOR REPLICATION so that the constraints are not enforced during synchronization.'),
    (0x20000, 'Replicates FOREIGN KEY constraints as NOT FOR REPLICATION so that the constraints are not enforced during synchronization.'),
    (0x40000, 'Replicates filegroups associated with a partitioned table or index.'),
    (0x80000, 'Replicates the partition scheme for a partitioned table.'),
    (0x100000, 'Replicates the partition scheme for a partitioned index.'),
    (0x200000, 'Replicates table statistics.'),
    (0x400000, 'Default Bindings'),
    (0x800000, 'Rule Bindings'),
    (0x1000000, 'Full-text index'),
    (0x2000000, 'XML schema collections bound to xml columns are not replicated.'),
    (0x4000000, 'Replicates indexes on xml columns.'),
    (0x8000000, 'Create any schemas not already present on the subscriber.'),
    (0x10000000, 'Converts xml columns to ntext on the Subscriber.'),
    (0x20000000, 'Converts large object data types (nvarchar(max), varchar(max), and varbinary(max)) introduced in SQL Server 2005 to data types that are supported on SQL Server 2000. For information about how these types are mapped, see the "Mapping New Data Types for Earlier Versions" section in Using Multiple Versions of SQL Server in a Replication Topology.'),
    (0x40000000, 'Replicate permissions.'),
    (0x80000000, 'Attempt to drop dependencies to any objects that are not part of the publication.'),
    (0x100000000, 'Use this option to replicate the FILESTREAM attribute if it is specified on varbinary(max) columns. Do not specify this option if you are replicating tables to SQL Server 2005 Subscribers. Replicating tables that have FILESTREAM columns to SQL Server 2000 Subscribers is not supported, regardless of how this schema option is set. '),
    (0x200000000, 'Converts date and time data types (date, time, datetimeoffset, and datetime2) introduced in SQL Server 2008 to data types that are supported on earlier versions of SQL Server. For information about how these types are mapped, see the "Mapping New Data Types for Earlier Versions" section in Using Multiple Versions of SQL Server in a Replication Topology.'),
    (0x400000000, 'Replicates the compression option for data and indexes. For more information, see Creating Compressed Tables and Indexes.'),
    (0x800000000, 'Set this option to store FILESTREAM data on its own filegroup at the Subscriber. If this option is not set, FILESTREAM data is stored on the default filegroup. Replication does not create filegroups, therefore, if you set this option, you must create the filegroup before you apply the snapshot at the Subscriber. For more information about how to create objects before you apply the snapshot, see Executing Scripts Before and After the Snapshot Is Applied.'),
    (0x1000000000, 'Converts common language runtime (CLR) user-defined types (UDTs) that are larger than 8000 bytes to varbinary(max) so that columns of type UDT can be replicated to Subscribers that are running SQL Server 2005.'),
    (0x2000000000, 'Converts the hierarchyid data type to varbinary(max) so that columns of type hierarchyid can be replicated to Subscribers that are running SQL Server 2005. For more information about how to use hierarchyid columns in replicated tables, see hierarchyid (Transact-SQL).'),
    (0x4000000000, 'Replicates any filtered indexes on the table. For more information about filtered indexes, see Filtered Index Design Guidelines.'),
    (0x8000000000, 'Converts the geography and geometry data types to varbinary(max) so that columns of these types can be replicated to Subscribers that are running SQL Server 2005.'),
    (0x10000000000, 'Replicates indexes on columns of type geography and geometry.'),
    (0x20000000000, 'Replicates the SPARSE attribute for columns. For more information about this attribute, see Using Sparse Columns.')

DECLARE @schema_option VARBINARY(8) = 0x000000000807509F;

SELECT CONVERT(INT, @schema_option, 1) & CONVERT(INT,value,1), *
FROM @options
WHERE CONVERT(INT, @schema_option, 1) & CONVERT(INT,value,1) <> 0;

SET @schema_option = 0x000000000803509F;

SELECT CONVERT(INT, @schema_option, 1) & CONVERT(INT,value,1), *
FROM @options
WHERE CONVERT(INT, @schema_option, 1) & CONVERT(INT,value,1) <> 0;

Using Orson’s two values we can see that the difference between the two is, the first value includes the “Replicates filegroups associated with a partitioned table or index.” option where the second value does not.