Collecting Historical IO File Statistics

In a previous post, Collecting Historical Wait Statistics, I discussed how you can easily collect historical wait stats by using the DMV sys.dm_os_wait_stats. Well today, I'd like to cover the same concept, but this time collect historical IO file stats from the DMV, sys.dm_io_virtual_files_stats. However, I wanted to improve on the code to make it even easier to implement.

The data collection process is still implemented the same way.  First, we'll need to create a history table to store the data. The data is stored in time slices with the cumulative values as well as the difference (TimeDiff_ms, NumOfReadsDiff, NumOfWritesDiff, etc) in those values since the last collection time.

CREATE TABLE dbo.IoVirtualFileStatsHistory(
  SqlServerStartTime DATETIME NOT NULL
 ,CollectionTime DATETIME NOT NULL
 ,TimeDiff_ms BIGINT NOT NULL
 ,DatabaseName NVARCHAR(128) NOT NULL
 ,DatabaseId SMALLINT NOT NULL
 ,FileId SMALLINT NOT NULL
 ,SampleMs INT NOT NULL
 ,SampleMsDiff INT NOT NULL
 ,NumOfReads BIGINT NOT NULL
 ,NumOfReadsDiff BIGINT NOT NULL
 ,NumOfBytesRead BIGINT NOT NULL
 ,NumOfBytesReadDiff BIGINT NOT NULL
 ,IoStallReadMs BIGINT NOT NULL
 ,IoStallReadMsDiff BIGINT NOT NULL
 ,NumOfWrites BIGINT NOT NULL
 ,NumOfWritesDiff BIGINT NOT NULL
 ,NumOfBytesWritten BIGINT NOT NULL
 ,NumOfBytesWrittenDiff BIGINT NOT NULL
 ,IoStallWriteMs BIGINT NOT NULL
 ,IoStallWriteMsDiff BIGINT NOT NULL
 ,IoStall BIGINT NOT NULL
 ,IoStallDiff BIGINT NOT NULL
 ,SizeOnDiskBytes BIGINT NOT NULL
 ,SizeOnDiskBytesDiff BIGINT NOT NULL
 ,FileHandle VARBINARY(8) NOT NULL
 ,CONSTRAINT PK_IoVirtualFileStatsHistory PRIMARY KEY CLUSTERED
  (CollectionTime,DatabaseName,DatabaseId,FileId)
 )WITH (DATA_COMPRESSION = PAGE);
GO

The next step will be to get the start time of SQL Server, so that we can compare it to the previous collection. If the dates are different, then we must take that into account when calculating the diff values. Because if SQL Server is restarted, then all values in the DMV are reset back to zero. At this point, we know the diff values are actually the same value as the current counters, because this is the first collection after a restart.

IF @CurrentSqlServerStartTime <> ISNULL(@PreviousSqlServerStartTime,0)
BEGIN
 -- If SQL started since the last collection, then insert starter values
 -- Must do DATEDIFF using seconds instead of milliseconds to avoid arithmetic overflow.
 INSERT INTO dbo.IoVirtualFileStatsHistory
 SELECT
   @CurrentSqlServerStartTime
  ,CURRENT_TIMESTAMP
  ,CONVERT(BIGINT,DATEDIFF(SS,@CurrentSqlServerStartTime,CURRENT_TIMESTAMP))*1000
  ,@DatabaseName
  ,@DatabaseId
  ,file_id
  ,sample_ms
  ,sample_ms
  ,num_of_reads
  ,num_of_reads
  ,num_of_bytes_read
  ,num_of_bytes_read
  ,io_stall_read_ms
  ,io_stall_read_ms
  ,num_of_writes
  ,num_of_writes
  ,num_of_bytes_written
  ,num_of_bytes_written
  ,io_stall_write_ms
  ,io_stall_write_ms
  ,io_stall
  ,io_stall
  ,size_on_disk_bytes
  ,size_on_disk_bytes
  ,file_handle
 FROM sys.dm_io_virtual_file_stats(@DatabaseId,NULL);
END
GO

You may notice the DATEDIFF is using "seconds" instead of "milliseconds".  This is because DATEDIFF only returns an INT value. The largest number it can return is equal to about 24 days before it hits an arithmetic overflow error. By converting it to seconds, we can avoid that error. All of the following data collections will do a DATEDIFF using milliseconds.

If the current start time is the same as the previous collection, then we'll grab the difference in values and insert those into the history table.

WITH CurrentIoVirtualFileStats AS
(
 SELECT
   CURRENT_TIMESTAMP AS 'CollectionTime'
  ,@DatabaseName AS 'DatabaseName'
  ,*
 FROM sys.dm_io_virtual_file_stats(@DatabaseId,NULL)
)
INSERT INTO dbo.IoVirtualFileStatsHistory
SELECT
  @CurrentSqlServerStartTime
 ,CURRENT_TIMESTAMP
 ,CONVERT(BIGINT,DATEDIFF(MS,@PreviousCollectionTime,curr.CollectionTime))
 ,@DatabaseName
 ,@DatabaseId
 ,file_id
 ,sample_ms
 ,curr.sample_ms - hist.SampleMs
 ,num_of_reads
 ,curr.num_of_reads - hist.NumOfReads
 ,num_of_bytes_read
 ,curr.num_of_bytes_read - hist.NumOfBytesRead
 ,io_stall_read_ms
 ,curr.io_stall_read_ms - hist.IoStallReadMs
 ,num_of_writes
 ,curr.num_of_writes - hist.NumOfWrites
 ,num_of_bytes_written
 ,curr.num_of_bytes_written - hist.NumOfBytesWritten
 ,io_stall_write_ms
 ,curr.io_stall_write_ms - hist.IoStallWriteMs
 ,io_stall
 ,curr.io_stall - hist.IoStall
 ,size_on_disk_bytes
 ,curr.size_on_disk_bytes - hist.SizeOnDiskBytes
 ,file_handle
FROM CurrentIoVirtualFileStats curr INNER JOIN dbo.IoVirtualFileStatsHistory hist
 ON (curr.DatabaseName = hist.DatabaseName
  AND curr.database_id = hist.DatabaseId
  AND curr.file_id = hist.FileId)
  AND hist.CollectionTime = @PreviousCollectionTime;
GO

At this point, we're through collecting the raw data. However, as I mentioned earlier, I added a lot of functionality into this script. The script is actually a stored procedure that can run all of this code for you; including creation of the history table, data collection, historical data purging and finally reporting.

The stored procedure has 5 input parameters.

@Database - This is used to specify a single database, a list of databases, or a wildcard.

• '*' is the default value which selects all databases
• 'MyDatabase1'  will process only that single database
• 'MyDatabase1,MyDatabase2,MyDatabase3' Comma delimited list of databases
• 'USER_DATABASES' used to process all user databases
• 'SYSTEM_DATABASES' used to process only the system databases.

@GenerateReport - Flag. When off, the stored procedure will collect data. When on, it will generate an aggregated report of the historical data.

@HistoryRetention - The is the number of days to keep in the history table. Default is 365.

@ExcludedDBs - Is a comma delimited list of databases to exclude from processing. It should be used when @Database is set to '*'.

@Debug - Flag. When on, it will output TSQL commands that being executed.

Examples:

1. Collect data for all databases.

EXEC dbo.sp_CollectIoVirtualFileStats
 @Database = '*';
GO

2. Collect data for all databases except for AdventureWorks, and output all debug commands.

EXEC dbo.sp_CollectIoVirtualFileStats
  @Database = '*'
 ,@ExcludedDBs = 'AdventureWorks'
 ,@Debug = 1;
GO

3. Output an aggregated report of data collected so far for tempdb.

EXEC dbo.sp_CollectIoVirtualFileStats
  @Database = 'tempdb'
 ,@GenerateReport = 1;
GO

The report would look like this.

Finally, you can copy this report data into Excel and generate some easy to read charts.

From the chart, we can see there was a spike in write latency between 3PM and 4PM for tempdb. If we collect this data over time and identify a similar spike each day then we'd want to investigate further to find out what is causing it. But that can only be done if you're collecting these metrics and storing them for historical analysis. Hopefully, this stored procedure will help you be more proactive in collecting performance metrics for each of your servers.

The entire script is available on the resources page.

Collecting Historical Wait Statistics

As a DBA, I'm sure you've heard many times to always check the sys.dm_os_wait_stats DMV to help diagnose performance issues on your server. The DMV returns information about specific resources SQL Server had to wait for while processing queries. The counters in the DMV are cumulative since the last time SQL Server was started and the counters can only be reset by a service restart or by using a DBCC command. Since DMVs don't persist their data beyond a service restart, we need to come up with a way to collect this data and be able to run trending reports over time.

Collecting the data seems easy enough by simply selecting all rows into a permanent table. However, that raw data won't help us determine the time in which a particular wait type occurred. Think about it for a minute. If the raw data for the counters is cumulative, then how can you tell if a bunch of waits occurred within a span of a few minutes or if they occurred slowly over the past 6 months that SQL Server has been running. This is where we need to collect the data in increments.

First, we need to create a history table to store the data. The table will store the wait stat values as well as the difference (TimeDiff_ms, WaitingTasksCountDiff, WaitTimeDiff_ms, SignalWaitTimeDiff_ms) in those values between collection times.

CREATE TABLE dbo.WaitStatsHistory
(
     SqlServerStartTime DATETIME NOT NULL
    ,CollectionTime DATETIME NOT NULL
    ,TimeDiff_ms INT NOT NULL
    ,WaitType NVARCHAR(60) NOT NULL
    ,WaitingTasksCountCumulative BIGINT NOT NULL
    ,WaitingTasksCountDiff INT NOT NULL
    ,WaitTimeCumulative_ms BIGINT NOT NULL
    ,WaitTimeDiff_ms INT NOT NULL
    ,MaxWaitTime_ms BIGINT NOT NULL
    ,SignalWaitTimeCumulative_ms BIGINT NOT NULL
    ,SignalWaitTimeDiff_ms INT NOT NULL
    ,CONSTRAINT PK_WaitStatsHistory PRIMARY KEY CLUSTERED (CollectionTime, WaitType)
)WITH (DATA_COMPRESSION = PAGE);
GO

Next, we need to get a couple of timestamps when we collect each sample. The first will be the SQL Server start time. We need the SQL Server start time, so we can identify when the service was restarted.

SELECT @CurrentSqlServerStartTime = sqlserver_start_time FROM sys.dm_os_sys_info;
GO

The second set is the previous start time and previous collection time, if they exist in the history table.

SELECT
     @PreviousSqlServerStartTime = MAX(SqlServerStartTime)
    ,@PreviousCollectionTime = MAX(CollectionTime)
FROM msdb.MSDBA.WaitStatsHistory;
GO

The last timestamp is the collection time. We’ll also use this timestamp to calculate the difference in wait stat values between each collection.

SELECT GETDATE() AS 'CollectionTime',* FROM sys.dm_os_wait_stats;
GO

We need to compare the current SQL Server start time to the previous start time from the history table. If they don’t equal, then we assume the server was restarted and insert “starter” values. I call them starter values, because we just collect the current wait stat values and insert 0 for each of the diff columns.

IF @CurrentSqlServerStartTime <> ISNULL(@PreviousSqlServerStartTime,0)
BEGIN
    -- Insert starter values if SQL Server has been recently restarted
    INSERT INTO dbo.WaitStatsHistory
    SELECT
         @CurrentSqlServerStartTime
        ,GETDATE()
        ,DATEDIFF(MS,@CurrentSqlServerStartTime,GETDATE())
        ,wait_type
        ,waiting_tasks_count
        ,0
        ,wait_time_ms
        ,0
        ,max_wait_time_ms
        ,signal_wait_time_ms
        ,0
    FROM sys.dm_os_wait_stats;
END
GO

If the timestamps are the same, we will collect the current wait stats and calculate the difference (in milliseconds) in collection time as well as the difference in values.

INSERT msdb.MSDBA.WaitStatsHistory
SELECT
     @CurrentSqlServerStartTime
    ,cws.CollectionTime
    ,DATEDIFF(MS,@PreviousCollectionTime,cws.CollectionTime)
    ,cws.wait_type
    ,cws.waiting_tasks_count
    ,cws.waiting_tasks_count - hist.WaitingTasksCountCumulative
    ,cws.wait_time_ms
    ,cws.wait_time_ms - hist.WaitTimeCumulative_ms
    ,cws.max_wait_time_ms
    ,cws.signal_wait_time_ms
    ,cws.signal_wait_time_ms - hist.SignalWaitTimeCumulative_ms
FROM CurrentWaitStats cws INNER JOIN MSDBA.WaitStatsHistory hist
    ON cws.wait_type = hist.WaitType
    AND hist.CollectionTime = @PreviousCollectionTime;
GO

You could filter the collection to only specific wait stat counters that you want to track by just a where clause, but I prefer to collect them all and then filter at the reporting end.

At this point, we’re ready to schedule the job. The script could be run at any interval, but I usually leave it to collect data once a day. If I notice a spike in a specific wait stat counter, then I could easily increase the job frequency to once every few hours or even once an hour. Having those smaller, more granular data samples will allow us to isolate which time frame we need to concentrate on.

For example, if we notice the CXPACKET wait suddenly spikes when collecting the data each day, then we could schedule the collection every hour to see if it’s happening during a specific window.

SELECT * FROM msdb.MSDBA.WaitStatsHistory
WHERE WaitType = 'CXPACKET';
GO

Finally, we can use Excel to format this raw data into an easy to read chart.

From this chart, we can see at 5PM there was a spike in CXPACKET waits, but a low number of tasks that waited. In this case, I would be assume there is a single process running in parallel that caused these waits and from there I could dig further into finding the individual query.

Data compression is enabled on this table to help keep it small. It can easily turn it off for the table by removing WITH (DATA_COMPRESSION = PAGE) from the CREATE TABLE statement. However, with page compression enabled, 24 collections (one per hour) only takes up 775KB of space. Without compression, the same sample of data consumes about 2.2MB. If you plan to keep a lot of history, then it's best to leave page compression enabled.

Hopefully, this script will help you keep track of your historical wait statistics, so you can have better knowledge of what has happened to your environment over time. The entire script is posted below. If you want to read further into what each wait statistics means, then check out Paul Randal’s article about wait stats. Additionally, if you want more info on using DMVs, check out Glenn Berry’s diagnostic queries.

/***********************************************
    Create the historical table
***********************************************/

USE msdb;
GO

-- Create the history table if it does not exist
IF OBJECT_ID('dbo.WaitStatsHistory') IS NULL
BEGIN
    CREATE TABLE dbo.WaitStatsHistory
    (
         SqlServerStartTime DATETIME NOT NULL
        ,CollectionTime DATETIME NOT NULL
        ,TimeDiff_ms INT NOT NULL
        ,WaitType NVARCHAR(60) NOT NULL
        ,WaitingTasksCountCumulative BIGINT NOT NULL
        ,WaitingTasksCountDiff INT NOT NULL
        ,WaitTimeCumulative_ms BIGINT NOT NULL
        ,WaitTimeDiff_ms INT NOT NULL
        ,MaxWaitTime_ms BIGINT NOT NULL
        ,SignalWaitTimeCumulative_ms BIGINT NOT NULL
        ,SignalWaitTimeDiff_ms INT NOT NULL
        ,CONSTRAINT PK_WaitStatsHistory PRIMARY KEY CLUSTERED (CollectionTime, WaitType)
    )WITH (DATA_COMPRESSION = PAGE);
END
GO

/***********************************************
    Schedule this section as an on-going job
***********************************************/

DECLARE
     @CurrentSqlServerStartTime DATETIME
    ,@PreviousSqlServerStartTime DATETIME
    ,@PreviousCollectionTime DATETIME;

SELECT @CurrentSqlServerStartTime = sqlserver_start_time FROM sys.dm_os_sys_info;

-- Get the last collection time
SELECT
     @PreviousSqlServerStartTime = MAX(SqlServerStartTime)
    ,@PreviousCollectionTime = MAX(CollectionTime)
FROM msdb.dbo.WaitStatsHistory;

IF @CurrentSqlServerStartTime <> ISNULL(@PreviousSqlServerStartTime,0)
BEGIN
    -- Insert starter values if SQL Server has been recently restarted
    INSERT INTO dbo.WaitStatsHistory
    SELECT
         @CurrentSqlServerStartTime
        ,GETDATE()
        ,DATEDIFF(MS,@CurrentSqlServerStartTime,GETDATE())
        ,wait_type
        ,waiting_tasks_count
        ,0
        ,wait_time_ms
        ,0
        ,max_wait_time_ms
        ,signal_wait_time_ms
        ,0
    FROM sys.dm_os_wait_stats;
END
ELSE
BEGIN
    -- Get the current wait stats
    WITH CurrentWaitStats AS
    (
        SELECT GETDATE() AS 'CollectionTime',* FROM sys.dm_os_wait_stats
    )
    -- Insert the diff values into the history table
    INSERT msdb.dbo.WaitStatsHistory
    SELECT
         @CurrentSqlServerStartTime
        ,cws.CollectionTime
        ,DATEDIFF(MS,@PreviousCollectionTime,cws.CollectionTime)
        ,cws.wait_type
        ,cws.waiting_tasks_count
        ,cws.waiting_tasks_count - hist.WaitingTasksCountCumulative
        ,cws.wait_time_ms
        ,cws.wait_time_ms - hist.WaitTimeCumulative_ms
        ,cws.max_wait_time_ms
        ,cws.signal_wait_time_ms
        ,cws.signal_wait_time_ms - hist.SignalWaitTimeCumulative_ms
    FROM CurrentWaitStats cws INNER JOIN dbo.WaitStatsHistory hist
        ON cws.wait_type = hist.WaitType
        AND hist.CollectionTime = @PreviousCollectionTime;
END
GO

In-Memory OLTP and the Identity Column

Over the past month I've been playing around with the new In-Memory OLTP (code name: "Hekaton") features within SQL Server 2014 CTP2. My organization is all about low latency applications, and this is one feature of SQL Server that I need to get familiar with ASAP.

To do this, I started my own little project that takes an existing database and converts parts of it into in-memory tables.  Once that step is complete, I could work on rewriting the TSQL code.

It might seem fairly simple, but with every new feature of SQL Server there are usually limitations. And one of the first ones I noticed was the use of an IDENTITY column. They are prohibited in Hekaton tables which means I had to find an alternative. This is where the new SEQUENCE object comes into play.

The CREATE SEQUENCE command allows you to create a user-defined numerical value that can be ascending or descending. This gives it much more flexibility than an IDENTITY column, and it's fully supported for use within an in-memory table.

Looking at the example below, we have a table with an IDENTITY value used for the OrderID column.

CREATE TABLE dbo.Orders (
     OrderID INT IDENTITY(1,1) NOT NULL
    ,OrderDate DATETIME NOT NULL
    ,CustomerID INT NOT NULL
    ,NetAmount MONEY NOT NULL
    ,Tax MONEY NOT NULL
    ,TotalAmount MONEY NOT NULL
);
GO

And we have a typical insert statement to insert a new order. Notice the IDENTITY column is not specified because it's value is automatically generated during at runtime.

INSERT INTO dbo.Orders (OrderDate,CustomerID,NetAmount,Tax,TotalAmount)
VALUES (GETDATE(),16,9.99,0.80,10.79);
GO

So how would this need to be rewritten to be turned into an in-memory table?  First we just need to create the table without the IDENTITY value.

CREATE TABLE dbo.Orders (
     OrderID INT NOT NULL
    ,OrderDate DATETIME NOT NULL
    ,CustomerID INT NOT NULL
    ,NetAmount MONEY NOT NULL
    ,Tax MONEY NOT NULL
    ,TotalAmount MONEY NOT NULL
);
GO

Then we'll need to create a SEQUENCE that produces the same order of values as the IDENTITY. In our example, it starts and 1 and increments by one.

CREATE SEQUENCE dbo.CountBy1 AS INT
    START WITH 1
    INCREMENT BY 1;
GO

The insert statement will look a little different, because we'll need to call the NEXT VALUE FOR function for the SEQUENCE we just created.

INSERT INTO dbo.Orders (OrderID,OrderDate,CustomerID,NetAmount,Tax,TotalAmount)
VALUES (NEXT VALUE FOR dbo.CountBy1,GETDATE(),16,9.99,0.80,10.79);
GO

You could also generate the next sequence number ahead of time and then insert the value in a later statement.

DECLARE @NextValue INT = NEXT VALUE FOR dbo.CountBy1;

-- Do some other stuff here then insert --

INSERT INTO dbo.Orders (OrderID,OrderDate,CustomerID,NetAmount,Tax,TotalAmount)
VALUES (@NextValue,GETDATE(),16,9.99,0.80,10.79);
GO

So far, I think Microsoft has done a great job with the new Hekaton feature. They are definitely marketing it as a feature to implement with little to no changes in code, but I think that really depends on the existing code.  This is very basic rewrite, but one that only took a few minutes to implement.

Check out Books Online for more detailed information about both Hekaton and Sequence Numbers.

TSQL Tuesday #47 - Your Best SQL Server SWAG

The host for T-SQL Tuesday #47 is Kendal Van Dyke (blog|twitter), and his topic of choice is about the best SQL Server SWAG we ever received at a conference; specifically, the “good stuff”.

I’ve been doing a lot of work with SQL Server over the years, but I’ve only had the opportunity to attend the big conferences a few times. As a matter of fact, next week will be my first time attending the SQL PASS Summit. We're supposed to talk about the “good stuff” and not any of the “cheap tchotchkes” that are given away by the vendors, but I feel that I really have to include both.

First, I’d like to talk about a piece of swag that I received while at the SQL Server Connection conference in Las Vegas in November 2007. This wasn't my first trip to Las Vegas, but it was my first conference in there. And to make it better, one of my best friends from college was attending the same conference. So you could only imagine the fun I had while in Las Vegas with “college buddy”. In the vendor area, one of the representatives from Microsoft was handing out koozies with SQL Server 2008 printed on the side. These were not your normal koozies. They were slap koozies!

I actually own two other slap koozies, but this one was definitely going to be my new favorite. Like I said, it's cheap, but I love it, and it's great conversation starter.

Now let’s talk about the good stuff. 

The date was November 7, 2005.  The location was the Moscone Center in San Francisco, CA. The event was the launch party for SQL Server 2005, Visual Studio 2005, .NET Framework 2.0, and BizTalk Server 2006. Microsoft had just spent years rewriting SQL Server, and now they were throwing this elaborate party to celebrate the release. Unlike a week-long conference, this one-day event was completely free. I was living in San Francisco at the time, so it made it really easy to get to this event. All I had to do was hop on my scooter and head downtown. Microsoft didn’t disappoint for their launch party. The event boasted some big headliners. Microsoft CEO Steve Ballmer gave the keynote speech.

The live entertainment was also exciting.  The headliner band was Cheap Trick.  Although not at the height of their popularity, they are a talented rock band in any day.

There was also another all girl cover band that played nothing but AC/DC music.  They were called AC/DShe. Quite a catchy name.

The other highlight of the night was the presence of Paul Teutul, Sr. from the Orange County Choppers show on the Discovery Channel. Not only was Paul Sr. there hanging out in the crowd taking pictures with the attendees, but his team build a chopper for Microsoft with the SQL Server logo on it.

So finally to the swag. Each attendee was given a free copy of SQL Server 2005 Standard Edition and Visual Studio 2005 Professional Edition. Most software given away at conferences are evaluation or time-bombed copy, but these were fully licensed copies. In 2005, this was probably $1000 worth of software that was now mine.

It may sound anticlimactic, but for a guy on a shoe-string budget, living in one of the most expensive cities in the country, this was definitely the best swag I’ve ever received.

One Year Later

Wow!  It’s been one year since I launched my blog, and my how things have changed.

Accomplishments Over the Past Year
I’ve had a chance to interact with a lot of people relating to many of the posts on my blog, and even run into a few people that said “Hey I know you through your blog”. I’ve gotten much more involved in the #sqlfamily through Twitter, Stackexchange, as well as through my local SQL Server user group. Although I’ve attended meetings at my local group off and on over the past several years, I am now making a specific point to attend every meeting for both the Charlotte SQL Server User Group and the Charlotte BI Group. I’ve attended SQL Saturday’s. I’ve moved into a new job at my company, where I am now responsible for platform engineering of SQL Server for Wells Fargo Securities. I’ve gone from outdated MCP certifications to the more current MCITP: Database Administrator 2008. And most importantly, the Atlanta Braves won their first division title since 2005.

The Roadmap for the Upcoming Year
I plan to keep writing about SQL Server through my blog, as well as continue learning about SQL Server through reading other blogs. That’s one thing I learned quickly about blogging. The more I wrote about SQL Server, the more I have read. My wife keeps telling me “For someone who hates to read, you sure do read a lot."

I had hoped to eventually get to an MCM certification, but Microsoft derailed that recently by terminating the program. So for now, I’ll continue on with the development exams for SQL Server 2008 and then move to upgrade them to the SQL Server 2012 MCSE: Data Platform. For my new job, I’m not required to have certifications, but I do need to have a more holistic view of SQL Server, rather than have a more narrow view on just the database engine. Studying for the certifications has helped in those areas that I’m less familiar with, such as Analysis Services.

In just a few more weeks I’ll be attending my first SQL PASS Summit. I have been so excited about this ever since I found out it will be hosted in my town, Charlotte, NC. The Charlotte Convention Center is right next door to where I work, and I’m obviously familiar with the surrounding area. I’ve been to the DEV Connections conference in Las Vegas before, but this will be my first PASS Summit.

I also hope to start speaking at local events. I already do this within my company, so now I want to venture out and do it in a more public arena. I might start with my local user group and move up to SQL Saturdays and beyond.

I also want to make sure I set aside plenty of time for my own family. My wife has been incredibly supportive in my blogging, attending user group meetings, and studying for certifications. I want her to know how much I’m indebted to her.

Thanks to all who have read my blog, and I hope I can continue to provide quality information.


Go Braves!
</ <| <\ <| </ <| <\ <| </ <| <\ <|
(That’s the tomahawk chop)

The Case of the NULL Query_Plan

As a DBA, we're often asked to troubleshoot performance issues for stored procedures.  One of the most common tools at our disposal is the query execution plan cached in memory by SQL Server. Once we have the query plan, we can dissect what SQL Server is doing and hopefully find some places for improvement.

Grabbing the actual XML query plan for a stored procedure from the cache is fairly easy using the following query.

USE AdventureWorks2012;
GO
SELECT qp.query_plan FROM sys.dm_exec_procedure_stats ps
    JOIN sys.objects o ON ps.object_id = o.object_id
    JOIN sys.schemas s ON o.schema_id = s.schema_id
    CROSS APPLY sys.dm_exec_query_plan(ps.plan_handle) qp
WHERE ps.database_id = DB_ID()
    AND s.name = 'dbo'
    AND o.name = 'usp_MonsterStoredProcedure';
GO

From this point, we can open the XML query plan in Management Studio or Plan Explorer to start our investigation. But what happens if SQL Server returns NULL for the query plan?

Let's back up a little bit.  We were pretty sure the query plan is still in cache, right?  Let's verify it.

USE AdventureWorks2012;
GO
SELECT * FROM sys.dm_exec_procedure_stats ps
    JOIN sys.objects o on ps.object_id = o.object_id
WHERE o.name = 'usp_MonsterStoredProcedure';
GO

Sure enough.  The query plan is still cached in memory, and we even can even see the plan_handle.

So why did our first query not return the XML plan?  Let's copy the plan_handle and manually run it through the sys.dm_exec_query_plan function.

SELECT * FROM sys.dm_exec_query_plan(0x05000500DD93100430BFF0750100000001000000000000000000000000000000000000000000000000000000);
GO

Why are we getting NULL returned for the XML query plan when we know is in the cache?  In this case, because the query plan is so large and complex, we're hitting an XML limitation within SQL Server.  "XML datatype instance has too many levels of nested nodes. Maximum allowed depth is 128 levels".  

Let's try to pull the text version of query plan.

SELECT * FROM sys.dm_exec_text_query_plan(0x05000500DD93100430BFF0750100000001000000000000000000000000000000000000000000000000000000,DEFAULT,DEFAULT);
GO

It looks as though we have solved the issue; however, we didn't.  Management Studio has a 65535 character limit in grid results and 8192 character limit in text results.  Our query plan has been truncated far from the end.  Now it seems we are back to square one.  

We still know the query plan is in cache, but we just need a tool other than Management Studio to retrieve it.  This is where Powershell enters the picture.

With Powershell, we can create a simple script to execute the sys.dm_exec_text_query_plan function and then output the data to a file.  All we need is to pass two variables.  The first is the SQL Server name where the plan is cached, and the second is the plan_handle. 

param (    
    [Parameter(Mandatory=$true)]
    [ValidateNotNullOrEmpty()]
    [string]
    $SqlInstance

   ,[Parameter(Mandatory=$true)]
    [ValidateNotNullOrEmpty()]
    [string]
    $PlanHandle
)

The script will simply execute a TSQL script and capture the output into a string variable.

$SqlCommand = "SELECT query_plan FROM sys.dm_exec_text_query_plan(" 
    + $PlanHandle + ",DEFAULT,DEFAULT);"
$QueryPlanText = $cmd.ExecuteScalar()

The final step will use System.IO.StreamWriter() to output the data to a file.

$stream = New-Object System.IO.StreamWriter($FileName)
$stream.WriteLine($QueryPlanText)

The Powershell script will save the entire XML query plan to a file named output.sqlplan.  As you can see below, the actual plan was over 5MB.

Finally we're able to view the entire query plan in our favorite tool and see the complexity of the stored procedure.

This is just another example of why  DBAs need to set aside some time to learn Powershell.  The entire script is posted below.  Feel free to modify it as needed to fit your environment.

######################################################################################
#
#   File Name:    Get-QueryPlan.ps1
#
#   Applies to:   SQL Server 2008
#                 SQL Server 2008 R2
#                 SQL Server 2012
#
#   Purpose:      Used to retrieve an XML query plan from cache.
#
#   Prerequisite: Powershell must be installed.
#                 SQL Server components must be installed.
#
#   Parameters:   [string]$SqlInstance - SQL Server name (Ex: SERVER\INSTANCE)
#                 [string]$PlanHandle - Binary query handle
#
#   Author:       Patrick Keisler
#
#   Version:      1.0.0
#
#   Date:         08/30/2013
#
#   Help:         http://www.patrickkeisler.com/
#
######################################################################################

#Define input parameters
param ( 
  [Parameter(Mandatory=$true)]
  [ValidateNotNullOrEmpty()]
  [string]
  $SqlInstance
  
  ,[Parameter(Mandatory=$true)]
  [ValidateNotNullOrEmpty()]
  [string]
  $PlanHandle
  )

Write-Host "Script starting."

#Grab the path where the Powershell script was executed from.
$path = Split-Path $MyInvocation.MyCommand.Path

#Build the SQL Server connection objects
$conn = New-Object System.Data.SqlClient.SqlConnection
$builder = New-Object System.Data.SqlClient.SqlConnectionStringBuilder
$cmd = New-Object System.Data.SqlClient.SqlCommand

#Build the TSQL statement & connection string
$SqlCommand = "SELECT query_plan FROM sys.dm_exec_text_query_plan(" + $PlanHandle + ",DEFAULT,DEFAULT);"
$builder.psBase.DataSource = $SqlInstance
$builder.psBase.InitialCatalog = "master"
$builder.psBase.IntegratedSecurity = $true
$builder.psBase.ApplicationName = "Get-QueryPlan"
$builder.psBase.Pooling = $true
$builder.psBase.ConnectTimeout = 15
$conn.ConnectionString = $builder.ConnectionString
$cmd.Connection = $conn
$cmd.CommandText = $SqlCommand

try
{
 if ($conn.State -eq "Closed")
 {
  #Open a connection to SQL Server
  $conn.Open()
 }
 
 #Execute the TSQL statement
 [string]$QueryPlanText = $cmd.ExecuteScalar()

 #Write the output to a file
 $FileName = $path + "\output.sqlplan"
 $stream = New-Object System.IO.StreamWriter($FileName)
 $stream.WriteLine($QueryPlanText)

 if ($stream.BaseStream -ne $null)
 {
  #Close the stream object
  $stream.close()
 }

 if ($conn.State -eq "Open")
 {
  #Close the SQL Server connection
  $conn.Close()
 }
 
 Write-Host "Script completed successfully."
}
catch
{
 #Capture errors if needed
 if ($_.Exception.InnerException)
 {
  $Host.UI.WriteErrorLine("ERROR: " + $_.Exception.InnerException.Message)
  if ($_.Exception.InnerException.InnerException)
  {
   $Host.UI.WriteErrorLine("ERROR: " + $_.Exception.InnerException.InnerException.Message)
  }
 }
 else
 {
  $Host.UI.WriteErrorLine("ERROR: " + $_.Exception.Message)
 }
 
 Write-Host .
 Write-Host "ERROR: Script failed."
}

How to Tell If Your Users are Connecting to the Availability Group Listener

You've spent a lot of time planning and building out a new SQL Server 2012 environment complete with Availability Group Listeners, but how can you be sure the end users are connecting to the listener and not directly to the SQL Server instance?

So why would we care about this?  To begin with, if the users are not connecting to the listener, then upon a failover to another replica, those users would have to connect to a different SQL Server instance name.  Having a single point of connection is crucial for the high availability process to work correctly.

In a previous blog post, we setup an Availability Group Listener, AdventureWorks.mcp.com, with two IP addresses:   192.168.1.55 & 192.168.2.55.  We'll use this one for our example.

The DMV, sys.dm_exec_connections, contains information about each connection to a SQL Server instance, and can be used to answer our question.

Open a TSQL connection to either the Availability Group listener, and execute the following command.

SELECT
     session_id
    ,local_net_address
    ,local_tcp_port
FROM sys.dm_exec_connections;
GO

The local_net_address and local_tcp_port columns will display the IP address and port number of the client's connection target.  This will be the connection string the users entered to connect to the SQL Server instance.

If the IP address and port number match the Availability Group IP, then you're in good shape.  If they do not match, then some users are likely connecting directly to the SQL Server instance, and that will need to be changed.

By joining the sys.dm_exec_sessions DMV, you'll also be able to get the hostname and program name of each connection.

SELECT
     ec.session_id
    ,es.host_name
    ,es.program_name
    ,local_net_address
    ,local_tcp_port
FROM sys.dm_exec_connections ec
    JOIN sys.dm_exec_sessions es ON ec.session_id = es.session_id;
GO

As you can see in this picture, we have one connection on session_id 62 that is connecting directly to the SQL Server instance and not the to the Availability Group Listener.  At this point, I would track down that user, and have them use the correct connection string.

Using this DMV will allow you to verify the users are connecting to SQL Server using the correct connection strings, and help prevent unneeded outages during a failover between replicas.

PASS Summit 2013 - You Ain't From Around Here Are Ya?

I know what y'all are thinkin', what's Charlotte got to do with SQL Server?  Just hear me out.  There's a lot more to Charlotte than NASCAR, fried chicken, and rednecks. I assume most of the 5000 attendees have never been to Charlotte, and probably don't know much about the area.  To help everyone out, I have made a list of useful tips.

My history and why you should listen to me. I begged my management for nearly a decade to send me to the PASS Summit, and this year they finally granted my request.  And to top it off even more, I just happen to live in Charlotte and work in the building right across the street from the Charlotte Convention Center.  I'm native to North Carolina and I have lived in Charlotte for about 17 years.  I even graduated from The University of NorthCarolina at Charlotte.

Queen City History. Charlotte is named in honor of Charlotte of Mecklenburg-Strelitz who was married to King George III of Great Britain.  This is why the city is nicknamed the "Queen City".  It is currently the 17th largest city in the US and it's the 2nd largest financial city, trailing only New York City. The city center is called uptown instead of downtown. The term downtown gives off a negative vibe; hence the term Uptown Charlotte.

Hotels. Just pick one, they're all about the same. However, if you are staying in a hotel on the south side of town near Pineville or Ballentyne, be prepared for I-77 and I-485 to be a parking lot during rush hour. Trust me on this one.

Transportation. The good news for anyone staying on the south side of town is the Lynx light rail. There is only one rail line but it runs from the center of town all the way south to Pineville. My suggestion is to take the light rail if it's near your hotel. Just get off at the 3rd St/Convention Center station, and the convention center is right across the street.

The CATS bus sytem is also not a bad option. The main transit center in uptown is only 3 blocks from the convention center. Any of the bus lines that end in an X are express routes (i.e. 54X) that pick you up from the commuter lots and head directly uptown. In uptown, there is a free bus line called the Goldrush. It different buses and only runs east/west along Trade Street.  It's helpful if you are staying in one of the hotels along that street.  And the best part is it's free.  Check out RideTransit.org for a complete system map.

If you like riding bicycles, the you'll want to check out CharlotteBcycle. There are about a dozen bicycle rentals around uptown. You just pay a small fee at the automated kiosk to share a bike, even if it's for a one way trip.

For those of you driving uptown, you'll need a place to park. There are over 40,000 parking spaces uptown, but you will have to compete with the daily workforce, like me. Most parking decks will run you about $15-20 per day. Once you get uptown, look for the giant "P" signs outside each of the parking decks. The signs will tell you the number of spaces available.

The parking lots are usually cheaper than the decks, $3-10 per day, and most of those you can pay by credit card at the kiosk. Some lots even allow you to pay using the Park Mobile app (Apple | Android | Windows). Just look for the Park Mobile sign near the kiosk for the lot number.

 

You might wonder what these over-street walkways are used for.  This is part of the Overstreet Mall.  It's a maze of walkways that interconnect some of the buildings and it's full of restaurants and shops.  Even if you're not interested in the shops, it's a nice way to get from building to building when it's raining.

While walking around uptown, you'll see these "You Are Here" street signs. The maps divide uptown four color-coded regions, North, South, East, and West. Each map provide you with information about attractions, hotels, and parking. 

Dining. You shouldn't have any issue finding a place to eat uptown; however, there are a few places of interest you should try out.  

For breakfast:

• Amelie's French Bakery
• Owen's Bagle and Deli (just south of uptown)
• Just Fresh
• Hazelnuts Creperie

For lunch:

• Price's Chicken Coop (just south of uptown, but THE BEST fried chicken in town!) CASH ONLY and TAKE-OUT ONLY!!
• French Quarter Restaurant
• Just Fresh
• Mert's Heart and Soul
• Johnny Burritos
• Fuel Pizza
• American Roadside Burgers
• Pike's Old Fashioned Soda Shop (just south of uptown)

For dinner:

• E2 Emeril's Eatery
• Sullivan's Steakhouse (just south of uptown)
• Dandelion Market
• Caffe Siena
• Bentley's on 27 (great view of Charlotte's skyline)
• Blue Restaurant and Bar
• Chima Brazilian Steakhouse

Also, if you're thinking of going to Ruth's Chris Steakhouse, then chose Sullivan's Steakhouse or Morton's Steakhouse instead.  I've never had a good experience at the uptown location, but that's just my opinion.

On a side note, when eating out, just keep in mind that you're in the south.  If you order iced tea, it WILL be sweet tea.  If you want unsweet tea, then ask for it.

Entertainment. There's plenty to do uptown as well as around town after the conference is over.  Next door to the convention center is the NASCAR Hall of Fame.  There are several other museums: Mint Museum, Bechtler Museum of Modern Art, etc.  The EpiCentre is a multi-use entertainment complex only 2 blocks from the convention center. There are restaurants, bars, and other entertainment there.  For beer lovers, there are plenty of bars uptown.  There are way too many to list, but a few are:

• The Carolina Ale House
• Prohibition Bar
• Connolly's on Fifth
• Ri Ra Irish Pub
• Picasso's Sports Bar
• Fitzgerald's Irish Pub
• Alexander Michael's Tavern

For wine lovers, check out Threes and The Wooden Vine.  Both have a wide range of selections.

The NC Music Factory is about 2 mile walk north from the convention center or only a 4 or 5 minute drive, but they do have free parking.  It's an entertain complex with live music, restaurants, bars, and even stand up comedy at The ComedyZone.  If you head over that way, be sure to visit the VBGB Beer Hall and Garden; definitely the best bar at the music factory.

Don't forget about the Carolina Panthers.  They'll have a home game on Sunday, October 20th at 1PM.  It might be your only chance to see the future superbowl champions in action!  

I know some of you might health nuts and would like find a place to workout besides your hotel gym.  The YMCA has a location uptown in my building.  $10 will get you a day pass, and $20 will get a 7-day pass.

If you prefer jogging outdoors, any of the streets uptown will work nicely.  However, if you like a little more scenery for your job, then head over to the Little Sugar Creek greenway.  The Charlotte Parks and Recreation built 35 miles of greenways around town.

This one is a beautiful, winding route nearly 6 miles long, and located just outside the south side of the I-277 belt loop uptown.

Finally, for the super adventurous attendees, the US National Whitewater Center is about 15 miles west of uptown, or head north to take a ride at 150mph at the Richard Petty Driving Experience.  It's only about 20 miles north of uptown at the Charlotte Motor Speedway.

As a bonus item, the very popular Showtime original Homeland is filmed right here in Charlotte.  If you have the time, why not try out as an extra for the show.

Other links with information about Charlotte:

• Charlotte Center City Partners
• Charlotte Visitor's Guide
• Charlotte Observer
• Charlotte Magazine
• Creative Loafing

I think I covered a lot, but if anyone has questions about Charlotte, please don't hesitate to contact me.

Are You the Primary Replica?

UPDATED -- Jul 3, 2015 -- To verify database exists, per comments by Konstantinos Katsoridis. Thanks for finding the bug!

In my recent adventures with AlwaysOn Availability Groups, I noticed a gap in identifying whether or not a database on the current server is the primary or secondary replica.  The gap being Microsoft did not provide a DMO to return this information.  The good news is the documentation for the upcoming release of SQL Server 2014 looks to include a DMO, but that doesn't help those of us who are running SQL Server 2012.

I've developed a function, dbo.fn_hadr_is_primary_replica, to provide you with this functionality.  This is a simple scalar function that takes a database name as the input parameter and outputs one of the following values.

 0 = Resolving
 1 = Primary Replica
 2 = Secondary Replica
-1 = Database Does Not Exist

The return values correspond to the role status listed in sys.dm_hadr_availability_replica_states.

In this example, I have setup 2 SQL Servers (SQLCLU1\SPIRIT1 and SQLCLU2\SPIRIT2) to participate in some Availability Groups.  I have setup 2 Availability Groups; one for AdventureWorks2012 and a second for the Northwind database.  SQLCLU1\SPIRIT1 is the primary for AdventureWorks2012 and secondary for Northwind.  SQLCLU2\SPIRIT2 is the primary for Northwind and secondary for AdventureWorks2012.

First let's run the function for both databases on SQLCLU1\SPIRIT1.

On this server, the function returns 1 because it's the primary for AdventureWorks2012, and returns 2 because it's the secondary for Northwind.

Now let's run it again on SQLCLU2\SPIRIT2.

As expected we get the opposite result.

This function does not take into account the preferred backup replica; it only returns information based on whether it is the primary or secondary replica.  It was created to use within other scripts to help determine a database's role if it's part of an Availability Group.  I hope this script can help you as well.

USE master;
GO

IF OBJECT_ID(N'dbo.fn_hadr_is_primary_replica', N'FN') IS NOT NULL
    DROP FUNCTION dbo.fn_hadr_is_primary_replica;
GO

CREATE FUNCTION dbo.fn_hadr_is_primary_replica (@DatabaseName SYSNAME)
RETURNS TINYINT
WITH EXECUTE AS CALLER
AS
/********************************************************************

  File Name:    fn_hadr_is_primary_replica.sql

  Applies to:   SQL Server 2012

  Purpose:      To return either 0, 1, 2, or -1 based on whether this 
                @DatabaseName is a primary or secondary replica.

  Parameters:   @DatabaseName - The name of the database to check.

  Returns:      0 = Resolving
                1 = Primary
                2 = Secondary
               -1 = Database does not exist

  Author:       Patrick Keisler

  Version:      1.0.1 - 07/03/2015

  Help:         http://www.patrickkeisler.com/

  License:      Freeware

********************************************************************/

BEGIN
    DECLARE @HadrRole TINYINT;

    IF EXISTS (SELECT 1 FROM sys.databases WHERE name = @DatabaseName)
    BEGIN
        -- Return role status from sys.dm_hadr_availability_replica_states
        SELECT @HadrRole = ars.role
        FROM sys.dm_hadr_availability_replica_states ars
        INNER JOIN sys.databases dbs 
            ON ars.replica_id = dbs.replica_id
        WHERE dbs.name = @DatabaseName;
    
        -- @DatabaseName exists but does not belong to an AG so return 1
        IF @HadrRole IS NULL RETURN 1;

        RETURN @HadrRole;
    END
    ELSE
    BEGIN
        -- @DatabaseName does not exist so return -1
        RETURN -1;
    END    
END;
GO

Setup an Availability Group with Multiple Subnets in VMware Workstation

Before we get started, I want to make it clear this is NOT how you would normally configure all these items in a production environment.  This is meant for a lab or demo area to play with Availability Groups over multiple subnets.

I use VMware a lot for demos at work as well as tooling around with various Windows and SQL Server related stuff.  In working with Availability Groups, one of the things I would like to do for my demos is have multiple subnets in VMware Workstation, so I can simulate a site failover.

Just to test Availability Groups requires at least three VMs; one for the Active Directory domain controller, one for the primary replica, and one for the secondary replica.  For this demo, we'll still just need those three VMs.

I'm not going to cover all the steps to setup an Active Directory domain controller or install SQL Server.  I'll assume you have already completed those steps on each of the VMs.  All three of my VMs are running Windows Server 2008 R2 Enterprise Edition.  If you are running a different version, then some of these screenshots could be different.

Here is the setup for each VM.

PDC

1. Windows Active Directory Domain Controller (MCP domain)
2. DNS server (mcp.com)
3. Network Policy and Remote Access (used for routing)
4. Connected to both 192.168.1.x and 192.168.2.x subnets

SQLCLU1

1. SQL Server 2012 Enterprise Edition
2. SPIRIT1 is the named instance listening on port 1433
3. Connected to 192.168.1.x subnet

SQLCLU2

1. SQL Server 2012 Enterprise Edition
2. SPIRIT2 is the named instance listening on port 1433
3. Connected to 192.168.2.x subnet

AdventureWorks2012AG

1. Availability Group for the AdventureWorks2012 database
2. Listening on port 1433
3. Mapped to 192.168.1.55 and 192.168.2.55

Now that you see how the finished environment is setup, let's see how to get there.

The first thing we need to do is setup each of the custom networks.   From the VMware Workstation menu, open the Virtual Network Editor.  Click on "Add Network" and select VMnet2.  Select Host-only and uncheck both the "Connect to Host Virtual Adapter" and "Use local DHCP" options.  Set the subnet IP to 192.168.1.0 and the subnet mask to 255.255.255.0.

Click "Add Network" and select VMnet3.  Make all the same setting changes, but this time set the subnet IP to 192.168.2.0. 

On the VM that is your Active Directory Domain Controller (PDC):

Edit the settings of the VM.  Make sure you have only ONE network card, and assign it to VMnet2.

Power on your VM domain controller.  Once it's up, edit the IPv4 settings of your network card.  Since we're not using DHCP, we'll need to hard code the IP address.  I have my domain controller IP set to 192.168.1.50.  You can set yours to any IP as long as it's on the same subnet.  Set the subnet mask to 255.255.255.0 and then leave the gateway blank.  Set the preferred DNS server to 192.168.1.50, because this is also your DNS server.  Save the changes and then shutdown the VM.

Edit the settings if the VM, add a 2nd network card and assign it to VMnet3. 

Power on the VM.  Once it's up, edit the IPv4 settings of the new network card.  This time set the IP to 192.168.2.50, the subnet mask to 255.255.255.0, and the Preferred DNS server to 192.168.1.50. Save the changes.

Your PDC will act as a router between the two subnets, but it will need software to make it happen.  Open Server Manager, select roles, and then "Add Role".  Select "Network Policy and Access Services".

For the service role, select "Routing".  It will automatically select the other required services.

Click next and then install. Once the installation is complete, go to Administrative Tools and open "Routing and Remote Access".  Right click on the domain controller and select "Configure and Enable Routing and Remote Access".   From the wizard, choose "Custom Configuration" then click Next.  Select "LAN Routing" then click next to complete the configuration. 

When a pop up asks to start the service, click "Start Service".  Once the configuraiton is complete, you now have software routing enabled on your domain controller.  The routing should be automatically configured between the two subnets. 

You would normally use a hardware router for this job, but the Routing and Remote Access service functions just fine for a lab running on VMware.  The next step is to configure the network and IP settings for each of our SQL Servers. 

On the first SQL Server VM (SQLCLU1):

Open the VM properties and make sure your network card is assigned to VMnet2. 

Save the settings and then power on the VM.  Once it's up, edit the IPv4 settings of the network card.  Set the IP address to 192.168.1.51.  Set the subnet mask to 255.255.255.0 and the default gateway to 192.168.1.50.  The default gateway needs to be the IP address of the server that is running the Routing and Remote Access service.  In this case, it's the IP of the domain controller.  Set the Preferred DNS server to 192.168.1.50.  Click OK to save the settings.

Additionally, you will need to open firewall ports TCP 5022 for the HADR service, TCP 1433 for the SQL Server service, and UDP 1434 for the SQL Server Browser service.

On the second SQL Server VM (SQLCLU2):

Open the VM properties and make sure your network card is assigned to VMnet3. 

Save the settings and then power on the VM.  Once it's up, edit the IPv4 settings of the network card.  Set the IP address to 192.168.2.52.  Set the subnet mask to 255.255.255.0 and the default gateway to 192.168.2.50.  The default gateway needs to be the IP of the 2nd network care we setup earlier on the domain controller.  Set the Preferred DNS server to 192.168.1.50.  Click OK to save the settings.

Additionally, you will need to open firewall ports TCP 5022 for the HADR service, TCP 1433 for the SQL Server service, and UDP 1434 for the SQL Server Browser service.

Your two subnets should be working now.  If you want to test it, just open a command prompt from SQLCLU1 and issue a "PING SQLCLU2".  You can do the same test from SQLCLU2.

Setting up the Windows Cluster

Open Failover Cluster Manager and click "Create a Cluster".  Step through the setup wizard by selecting the two cluster nodes: SQLCLU1 and SQLCLU2. 

Key in the name of the cluster, SQLCLUV1.  Select the 192.168.1.0/24 subnet and enter the IP address of 192.168.1.54.  Make sure to uncheck the other subnet.  Click next to finish the setup.

At this point we would normally configure the quorum; however, since this is just for a lab setup, we'll leave the quorum set to Node Majority.  When setting this up in a production environment, you'll want to configure the quorum based on the number voting nodes.  This link will guide you through what changes are needed.

Look at the settings of each of the cluster networks.  Cluster Network 1 is the 192.168.1.x subnet and is connected to SQLCLU1. 

Cluster Network2 is the 192.168.2.x subnet and is connect to SQLCLU2.

Setting up the Availability Group

Now comes the easy part.  First we'll need to enable the Availability Group feature on each SQL Server instance.  On SQLCLU1, open the SQL Server Configuration Manger.  Right click on the SQL Server service and select properties.  Select the "AlwaysOn High Availability" tab, and check the box to enable it.  Click OK to save the changes, and then stop and restart the SQL Service services. 

Make the change on the second SQL Server, SQLCLU2.

Now make sure the AdventureWorks2012 database is in FULL recovery mode.  Within SQL Server Management Studio we'll setup the Availability Group for the AdventureWorks2012 database.  Open Object Explorer to SQLCLU1\SPIRIT1. Right click on the "AlwaysOn High Availability" node and select "New Availability Group Wizard".  Enter a name for the group, AdventureWorks2012AG and click next.

Check the box next to the AdventureWorks2012 database and click next.

Click Add Replica and add SQLCLU2\SPIRT2 to the list of replicas.  Check all the boxes for Automatic Failover and Synchronous Commit.

Click the Listener tab.  Select the "Create an Availability Group Listner" radio button, then enter a listener name and port number and make sure Static IP is selected for Network Mode.

Click the Add button.  Select the 192.168.1.0/24 subnet and enter the IP of the listener, 192.168.1.55, then click OK.

Click the Add button again.  Select the 192.168.2.0/24 subnet and enter the second IP of the listener, 192.168.2.55, then click OK.  You should now see 2 separate IP address for the listener.  Click Next to continue.

Select FULL data synchronization and specify a backup file share, \\SQLCLU1\BACKUP, and click next.  The file share is only needed to complete the initial data synchronization.

Verify all the validation checks are successful and then click next.  Click finish to complete the Availability Group setup.

Once the setup is complete, go back into Failover Cluster Manager to check out the Availability Group resource that was added.  What you'll notice is two IP addresses associated to the Availability Group listener.  One is currently online and the other is offline.  The IP that's online is associated to the subnet that SQLCLU1 is on, because it's currently the primary replica.

Now let's failover the Availability Group to SQLCLU2\SPIRIT2 to see what happens to the listener.  Open a query window to SQLCLU2\SPIRIT2 and run the following code.

ALTER AVAILABILITY GROUP AdventureWorks2012AG FAILOVER;

Once the failover is complete, go back into Failover Cluster Manager to check out the properties of the Availability Group.  You'll notice the IP resources have switched.  The IP 192.168.1.55 is offline and 192.168.2.55 is online.

SQLCLU2\SPIRIT2 is now the primary replica for the Availability Group and it's on the 192.168.2.x subnet.  You can also go back to the domain controller and open up the DNS Manager.  There you will see the two DNS entries for the the Availability Group listener; one for each IP address.

What we've covered here is a quick and easy way to setup an Availability Group on multiple subnets within VMware Workstation.  Remember this is not how you would normally setup everything within a production environment.  In a production environment we'd use a hardware router instead of the routing service, the separate subnets would likely be in different data centers, and the quorum would be configured according the number voting nodes.  However, this provides you with a platform for doing multisubnet failovers with Availability Groups.