More information about the session can be found on the presentation’s web page, including:

• The session’s abstract and goals
• The current release of the session materials, along with all past versions
• A list of past deliveries of the presentation

See you this afternoon!

Note: Before attempting to install PowerShell on a computer, confirm that it isn’t already installed!

PowerShell version 1.0 was available for installation by itself. PowerShell version 2.0 is installed as part of the Windows Management Framework. Don’t worry… WinRM is disabled by default, so this update can be safely installed without affecting the security of Windows.

Steps Common to Each Windows Operating System

Installation on Windows 7 or Windows Server 2008 R2

Installation on Windows Server 2008 or Windows Vista

Installation on Windows Server 2003

Installation on Windows XP

Installation on Windows 2000

Beginning with Windows 7 and Windows Server 2008 R2, PowerShell is installed and enabled by default. As people realize its significance, more and more administrators and power users are installing PowerShell on earlier operating systems. There’s a good chance that PowerShell is already installed on your computer, even if you didn’t put it there yourself!

Here are the steps for verifying that PowerShell is ready to use:

1. Open the Windows® Run dialog box
   1. Press the + keys
2. Type PowerShell.exe into the Open text box and press the key

• If PowerShell is installed, a console window will be displayed:

  

• If PowerShell is not installed, you will see the following error message:

  Windows cannot find 'PowerShell.exe'. Make sure you typed the name correctly, and then try again.

  That means it’s time install PowerShell!

I really enjoy seeing people talking passionately & with significant knowledge about #SQL Server. @SQLServerSleuth #codestock.

During the session, I state that the following two ideas are the most important things to take away from the presentation:

• Do not run SQL Server Profiler against a production SQL Server instance! Use a server-side trace instead
• Always place size and duration caps on your traces to limit their overhead and to protect against a run-away trace (see step three below)

I spend a long time during the presentation warning against using Profiler (in production) and explaining its performance impact (as well as the overhead of tracing). Still, Profiler can be a useful tool for defining trace sessions and analyzing trace data. Here is the proper workflow for reaping its benefits without shooting yourself in the foot:

1. On your workstation, use Profiler to define the trace (event classes, columns, and filters)
2. Use Profiler to generate a Transact-SQL script file with the system stored procedure calls for a server-side trace:

   

3. Edit the script, including placing limits on the size and duration of the trace (using the @maxfilesize and @stoptime parameters of the [sys].[sp_trace_create] system stored procedure, as demonstrated in the presentation’s “Demo 07_Create a server-side trace to provide information on the next deadlock(s),2005+.sql” file)
4. Execute the script on the server that is under investigation
5. If the trace automatically stops more quickly than expected, refine the trace definition to be more conservative
6. Once you have several trace files with useful data, compress them
7. Transfer the archive of compressed trace files from the target server to your workstation
8. Analyze the trace data using the [sys].[fn_trace_gettable] system function and/or using Profiler

• An Introduction to Profiler and SQL Trace at 2011-06-03 13:50-04:00. The session’s materials have been posted here.
• A PowerShell Cookbook for DBAs at 2011-06-04 09:50-04:00. The session’s materials have been posted here.

If you happen to be attending, please stop by and visit!

The customer has a two-node failover cluster with two instances of SQL Server 2005. Each clustered instance has dozens of databases, nearly all of them with identical schema (they have a Data-Dependent Routing/shard architecture), and each is in the full recovery model. They don’t have enough disk space allocated for the instances and, much too frequently, they encounter 9002 errors from SQL Server (when one or more of the transaction log files fill up and don’t have room to grow):

Msg 9002, Level 17, State 1, Line 1
The transaction log for database '____' is full. To find out why space in the log cannot be reused, see the log_reuse_wait_desc column in sys.databases

The customer reacts by looping through all of their databases and running BACKUP LOG … WITH NO_LOG against each. Let’s walk through the steps that they should be taking, instead of this knee-jerk overreaction.

The first problem is that, despite the hint in the 9002 error message, the customer doesn’t use the [log_reuse_wait_desc] column of the [sys].[databases] catalog view to determine which databases need attention. Instead, they resort to shotgun troubleshooting and truncate the transaction logs of all databases.

Next, the customer should perform appropriate capacity planning and provision enough disk space for their databases (now and in the future).

Since the customer won’t acquire sufficient disk space, I have asked whether their databases really belong in the full recovery model. They expect point-in-time-recoverability, but their frequent and ill-advised reaction to 9002 errors trashes the databases’ backup log chains. It will be an upsetting day when they need to restore to a certain time and cannot do so. It would be better to review the Recovery Point Objective (RPO) for each database (or define it for the first time) and then place the majority of the databases in the simple recovery model, saving the resources and administrative overhead associated with transaction log backups. Differential backups could be used to reduce data loss to levels consistent with the RPOs (YMMV).

As bad as it is to rely on sabotaged point-in-time-recoverability, the bigger issue is that the customer will follow-up the transaction log truncations with a round of shrinking the transaction log files! I don’t have time right now to delve into the perils of shrinking transaction log files, so here are some great posts from others in the community:

• BACKUP LOG WITH NO_LOG – use, abuse, and undocumented trace flags to stop it by Paul Randal (blog, Twitter)
• Don’t Touch that Shrink Button! by Mike Walsh (blog, Twitter)
• Leaking roof and file shrinking by Tibor Karaszi (blog, Twitter)

If, despite careful planning and database administration, the transaction log for one of your databases fills up (and you receive a 9002 error), take a deep breath and simply back up the transaction log for that database. It takes slightly longer than throwing away all of the committed transaction log records (via BACKUP LOG … WITH NO_LOG/TRUNCATE_ONLY) but it also avoids the problems discussed in this post and the last one. If absolutely necessary, discard the committed transactions using the following process (which still works even after the wretched BACKUP LOG … WITH NO_LOG/TRUNCATE_ONLY functionality was removed from SQL Server):

1. Put the database in simple recovery model:

   ALTER DATABASE [YourDatabase] SET RECOVERY SIMPLE;

2. Wait for the transaction log to automatically truncate… or force it:

   USE [YourDatabase];
   GO
   CHECKPOINT;
   GO

3. If the database truly belongs in full recovery model:
   1. Change the recovery model back:

      ALTER DATABASE [YourDatabase] SET RECOVERY FULL;

   2. Immediately perform a full backup of the database to reestablish the backup chain

Note: While performing research for this post, I came across a similar one from Brent Ozar (blog, Twitter). Despite the overlap, I decided to publish anyway, hoping that we rank higher in search engine results than all of the bad Internet advice that advocates forcibly clearing the transaction log!

While discussing THREADPOOL waits, I missed the name of a helpful DMV for identifying a lack of worker threads. I did hear the name of the relevant column though… “work_queue_count.” Thanks to SQL Server’s catalog views, it was easily to track down the DMV in question, [sys].[dm_os_schedulers]:

SELECT [_Objects].[object_id] AS [ObjectID],
    [_Objects].[name] AS [ObjectName],
    [_Objects].[type_desc] AS [ObjectType],
    [_Columns].[column_id] AS [ColumnID],
    [_Columns].[name] AS [ColumnName]
FROM [master].[sys].[system_objects] AS [_Objects]
INNER JOIN [master].[sys].[system_columns] AS [_Columns]
ON ([_Objects].[object_id] = [_Columns].[object_id])
WHERE (
    ([_Objects].[name] LIKE N'dm_%')
    AND
    ([_Objects].[type] IN ('IF', 'TF', 'V'))
    AND
    ([_Objects].[schema_id] = Schema_ID('sys'))
    AND
    ([_Objects].[is_ms_shipped] = 1)
    AND
    ([_Columns].[name] LIKE N'%work%queue%count%')
)
ORDER BY [_Objects].[name] ASC,
    [_Columns].[name] ASC;

This type of query is also useful when searching for gems in SQL Server’s meta data:

SELECT [_Objects].[object_id] AS [ObjectID],
    [_Objects].[name] AS [ObjectName],
    [_Objects].[type_desc] AS [ObjectType],
    [_Columns].[column_id] AS [ColumnID],
    [_Columns].[name] AS [ColumnName]
FROM [master].[sys].[system_objects] AS [_Objects]
INNER JOIN [master].[sys].[system_columns] AS [_Columns]
ON ([_Objects].[object_id] = [_Columns].[object_id])
WHERE (
    ([_Objects].[type] IN ('IF', 'TF', 'V'))
    AND
    ([_Objects].[schema_id] = Schema_ID('sys'))
    AND
    ([_Objects].[is_ms_shipped] = 1)
)
ORDER BY [_Objects].[name] ASC,
    [_Columns].[column_id] ASC;

This year, the session selection committee is trying something new and has requested input on the submitted abstracts. I strongly encourage you to take a few minutes and vote for abstracts that appeal to you in the PASS Summit 2011 Session Preference Survey. You can vote even if you might not be able to attend the conference… all that you need is an account on the SQLPASS.org web site. If you don’t already have one, registration is easy and free.

If you have attended any of my presentations and enjoyed them/learned from them, I would greatly appreciate the support of voting for some of my sessions. I submitted the maximum of four abstracts, but I would prefer to deliver either of my “cookbook” presentations:

• A PowerShell Cookbook for DBAs
  • I have been delivering this presentation since SQLSaturday #22 on 2010-06-05
  • As a production/escalation DBA, I use PowerShell more than any other single program, especially when interacting with SQL Server
  • My goal is to convince fellow DBAs to try PowerShell by showing a variety of ways that PowerShell makes our jobs easier and by providing attendees with a collection of ready-to-use scripts for getting started
• An XQuery Cookbook for DBAs
  • This is a new presentation that is under development. I will probably deliver it for the first time at the July meeting of the Fort Worth SQL Server Users Group
  • A lot of valuable diagnostic data is exposed as XML these days
  • I have written many T-SQL/XQuery scripts to analyze and manipulate the data, but I would not describe the experience as being pleasant
  • I would like to give other DBAs these scripts, along with a little bit of background understanding, so that they can access this very powerful meta data without the many hours of frustration in developing the scripts
  • Some of the “recipes” included in the presentation will work with the following data sets:
    • DDL trigger event data
    • Deadlock graphs
    • Extended Events payloads
    • Output from the Performance Analysis of Logs (PAL) Tool
    • The [sys].[dm_os_ring_buffers] DMV’s contents
    • Windows event logs
    • XML statistics profiles (query plans)

Note: The deadline for voting is tomorrow, so please do so as soon as possible!

Msg 4214, Level 16, State 1, Line 1
BACKUP LOG cannot be performed because there is no current database backup.
Msg 3013, Level 16, State 1, Line 1
BACKUP LOG is terminating abnormally.

In a future blog post, I will go into why users resort to this bad practice and what they should be doing instead.

This solution is based on a great blog post from Kalen Delaney (blog, Twitter), that showed how the [sys].[database_recovery_status] catalog view can reveal databases suffering from this problem, using a query like so:

SELECT DB_Name([database_id]) AS [DatabaseName]
FROM [sys].[database_recovery_status]
WHERE (
    ([last_log_backup_lsn] IS NULL)
    AND
    (Cast(DatabasePropertyEx(DB_Name([database_id]), 'Recovery') AS NVarChar(16)) = N'FULL')
    AND
    (Cast(DatabasePropertyEx(DB_Name([database_id]), 'Status') AS NVarChar(16)) = N'ONLINE')
)
ORDER BY [DatabaseName] ASC;

The following PowerShell function uses the query to retrieve a list of problematic databases and then automatically backs up each one:

Function Repair-LogBackupChain (
    [String]$ServerName = $(Throw (New-Object -TypeName 'System.ArgumentNullException' -ArgumentList '$ServerName')),
    [String]$BackupDirectoryPath = $(Throw (New-Object -TypeName 'System.ArgumentNullException' -ArgumentList '$BackupDirectoryPath')),
    [String]$OutputDirectoryPath = $(Throw (New-Object -TypeName 'System.ArgumentNullException' -ArgumentList '$OutputDirectoryPath'))
) {
    [Int]$BackedUpDatabaseCount = 0;
    [String]$BackupFilePath = '';
    [String]$DatabaseName = '';
    [String]$ErrorMessage = '';
    [String]$OutputFilePath = '';
    [String]$Query = "SET NOCOUNT ON; SELECT DB_Name([database_id]) AS [DatabaseName] FROM [sys].[database_recovery_status] WHERE (([last_log_backup_lsn] IS NULL) AND (Cast(DatabasePropertyEx(DB_Name([database_id]), 'Recovery') AS NVarChar(16)) = N'FULL') AND (Cast(DatabasePropertyEx(DB_Name([database_id]), 'Status') AS NVarChar(16)) = N'ONLINE')) ORDER BY [DatabaseName] ASC;";
    [String]$TSQLCode = '';

    #   Create the necessary directories, if they don't already exist.
    $BackupDirectoryPath = Add-Directory -Path $BackupDirectoryPath;
    $OutputDirectoryPath = Add-Directory -Path $OutputDirectoryPath;

    #   Lookup the names of the databases, in the full recovery model, that 
    #   lack an intact transaction log backup chain.
    [String[]]$DatabaseNames = @(& 'SQLCmd.exe' -E -S $ServerName -d 'tempdb' -Q $Query -b -h-1 -W);
    $DatabaseNames | Write-Debug;

    #   Verify that the list of databases was successfully retrieved.
    If ((Get-Item -Path Variable:'LastExitCode').Value -eq 0) {
        #   Iterate through the databases that need backups.
        $DatabaseNames | ForEach-Object {
            $DatabaseName = $_;
            Write-Host "Backing up the [$($DatabaseName)] database..." -ForegroundColor 'Cyan';

            $OutputFilePath = "$($OutputDirectoryPath)$(Get-EscapedFileNameComponent -FileNameComponent $ServerName),Backup database,$(Get-EscapedFileNameComponent -FileNameComponent $DatabaseName),$([DateTime]::Now.ToString('yyyyMMddTHHmmssfff')).out";
            Write-Debug "`$OutputFilePath = `"$($OutputFilePath)`"";

            #   Construct the name of the backup file.
            $BackupFilePath = [String]::Format(
                '{0}{1}_{2}_{3}.bak',
                $BackupDirectoryPath,
                (Get-EscapedFileNameComponent -FileNameComponent $ServerName),
                (Get-EscapedFileNameComponent -FileNameComponent $DatabaseName),
                ([DateTime]::Now.ToString('yyyyMMddTHHmmssfff'))
            );

            #   Generate the T-SQL for backing up the database.
            $TSQLCode = [String]::Format(
                "BACKUP DATABASE [{0}] TO DISK = N'{1}';",
                $DatabaseName,
                $BackupFilePath
            );
            Write-Debug "`$TSQLCode = `"$($TSQLCode)`"";

            #   Backup the database.
            & 'SQLCmd.exe' -E -S $ServerName -d 'tempdb' -Q $TSQLCode -o $OutputFilePath -b;

            If (Test-Path -LiteralPath $BackupFilePath -PathType 'Leaf') {
                $BackedUpDatabaseCount++;
            }
            Else {
                $ErrorMessage = "There was a problem backing up the [$($DatabaseName)] database.";
                Throw (New-Object -TypeName 'System.InvalidOperationException' -ArgumentList $ErrorMessage);
            }
        }
    }
    Else {
        $ErrorMessage = [String]::Format(
            "There was a problem querying the [sys].[database_recovery_status] catalog view:`r`n{0}",
            (
                [String]::Join(
                    "`r`n",
                    $DatabaseNames
                )
            )
        );
        Throw (New-Object -TypeName 'System.InvalidOperationException' -ArgumentList $ErrorMessage);
    }

    Write-Host "A total of $($BackedUpDatabaseCount) database(s) was/were backed up." -ForegroundColor 'Green';

    Return;
}

Here’s an example of using the function:

Repair-LogBackupChain -ServerName 'DatabaseServer01\InstanceA' -BackupDirectoryPath 'C:\Temp\' -OutputDirectoryPath 'C:\Temp\';

Note: The code snippet is dependent on two functions that are included in the download for my PowerShell cookbook presentation, Add-Directory and Get-EscapedFileNameComponent.

Maybe by the time that I write the follow-up blog post, I will have received permission to break the customer of this nasty habit (using a tip from Paul Randal (blog, Twitter)).

Hopefully this PowerShell snippet will save you as much time as it has for me!