And it was never a question, Query Store was crowing for repair. You gave it space and direction but you couldn’t keep it there…

Yes! Back to Query Store! I have had this problem for months where one of my Query store databases grows by a gig each week! It completely fills up, goes into a Read only state (which sets off the an alarm that I built to tell me when it switches to read only) and the only way I could get it to work again was to add space. I would add a gig and think, “Surely that will be enough to feed the hunger”. The next week, the alarm would go off again and I would feed it again!

I adjusted how often stats were collected, how frequently data was flushed, the max plans per query and anything else I could think to do, and still, it was hungry.

I had searched, read, googled, and kept coming up with nothing. I finally found something on corruption in the query store. CORRUPTION? Could it be possible? It was worth a try, my query store was in need of a serious diet and I still needed it to function.

The next time it went in to read only mode, I turned it off (it has to be off to fix the corruption) and ran this:

sp_query_store_consistency_check

Guess what happened next?!!!! My query store had a full gig free! I have left it alone for a few weeks and today I was able to shrink it by 5 gb! It has been glorious to have it working and not being worried as to why it was growing out of control.

The song for this post is Toad the Wet Sprocket’s Crowing

Now a personal note about Toad the Wet Sprocket. They are one of my favorite bands. Last night as I was listening to “Crowing”. I looked up the lyrics to figure out one of the words and realized I had be singing along to the wrong words. I thought it was “crowing for her” when it is actually “crowing for repair”. That completely changed the meaning of the song for me and made me love that song even more. It also made me realize I need to read Toad lyrics more often.

This also took my mind to the time that Ryan surprised me with tickets to go see them at a Reunion show in Vegas. After the show, fortune shown on me and I got to meet Glen Phillips the lead singer. He was super kind and gracious and let us take a picture and right after I fan-girled out a lot and started crying while trying to tell him how much I appreciated his and the band’s music. Huge apology to all the people I have completely scared with a fan-girl episode. I promise I try not to, just sometimes I can’t word how important that moment is to me.

Give me a minute to restore my logs…

This is part 4 of my log-shipping journey, if you missed part 1, you can find it here, part 2 is here, and part 3 is here.

We have one more set up piece to finish up before we do our restores. We need to tell my scripts which logfile is going to be the final one for now, so it can restore with Standby and I can read my databases.

UPDATE RestoreFile
SET IsFinalLogFile = 1
FROM RestoreFile RF1
INNER JOIN (SELECT RF.Origin, MAX(RF.CreatedDate) AS CreatedDate
FROM RestoreFile RF
INNER JOIN	(SELECT ForDatabase, MAX(CreatedDate) AS CreatedDate
					FROM [dbo].[RestoreFile] RF
					WHERE RF.FileType = 'B' 
					GROUP BY ForDatabase) BF ON RF.ForDatabase = BF.ForDatabase AND RF.CreatedDate > BF.CreatedDate
WHERE FileType = 'L' 
GROUP BY RF.Origin)RF2 ON RF1.Origin = RF2.Origin AND RF1.CreatedDate = RF2.CreatedDate

There are a few things going on here. First, I need to make sure that my logfile is more recent than my most recently restored full backup. Because I am restoring more than one database with this script, I have to join on both the ForDatabase and the CreatedDate (making sure it is greater than or, in the logfile case, equal to Date and Time I need). When I find the last logfile that is greater than the full backup’s created date, I am marking it with a IsFinalLogFile = 1 (True).

Now we are ready to build our full database restores.

DECLARE @SQLToExecute nvarchar(max), @RestoreFileId int
DECLARE RestoreBackups CURSOR FAST_FORWARD
FOR
SELECT 
'RESTORE DATABASE ' + RF.ForDatabase + ' FROM DISK = N''' + RF.FileLocation + ''' WITH FILE =  1, MOVE N''' + MF.name + ''' TO N''D:\SQLData\' + RF.ForDatabase + '.mdf'', MOVE N''' + MF.name + '_log'' TO N''L:\SQLLogs\'+ RF.ForDatabase + '_log.ldf'', NORECOVERY,  NOUNLOAD,  REPLACE,  STATS = 5'
, Id

  FROM [dbo].[RestoreFile] RF
	INNER JOIN sys.databases d ON RF.ForDatabase = d.name
	INNER JOIN sys.master_files mf on d.database_id = mf.database_id and mf.type = 0
  WHERE RF.FileType = 'B' AND RF.IsApplied = 0
OPEN RestoreBackups
FETCH NEXT FROM RestoreBackups INTO @SQLToExecute, @RestoreFileId
WHILE (@@FETCH_STATUS = 0)
BEGIN
    EXEC sp_executeSQL @SQLToExecute
	
	UPDATE [dbo].[RestoreFile]  SET IsApplied = 1 WHERE Id = @RestoreFileId

    FETCH NEXT FROM RestoreBackups INTO @SQLToExecute, @RestoreFileId
END
CLOSE RestoreBackups
DEALLOCATE RestoreBackups
GO

I am using a cursor to run each restore one at a time. I make sure that I am only doing FileType = ‘B’ so I know they are full backups and ones that haven’t been applied yet. I am building my restore string to include a move since the location on the new server may be different from what was used in the past. I update the applied status as I do the restore so I won’t apply this one twice, for this I am using the RestoreFileId which allows me to get the specific record that I want. This is the main reason I wanted a new table. If I would have stayed with one table, there is a chance I would have had duplicates on the filenames. That would have meant duplicate chances to restore records. It just made everything so much cleaner to make one a stage table and one a production table.

For the logfile restore, I followed a similar pattern:

DECLARE @SQLToExecute nvarchar(max), @RestoreFileId int
DECLARE RestoreBackups CURSOR FAST_FORWARD
FOR
SELECT 
'RESTORE LOG ' + RF.ForDatabase + ' FROM DISK = N''' + RF.FileLocation + ''' WITH NORECOVERY;'
, Id

  FROM [dbo].[RestoreFile] RF
	INNER JOIN sys.databases d ON RF.ForDatabase = d.name
	INNER JOIN sys.master_files mf on d.database_id = mf.database_id and mf.type = 0
	INNER JOIN	(SELECT ForDatabase, MAX(CreatedDate) AS CreatedDate
					FROM [dbo].[RestoreFile] RF
					WHERE RF.FileType = 'B' 
					GROUP BY ForDatabase) BF ON RF.ForDatabase = BF.ForDatabase AND RF.CreatedDate > BF.CreatedDate
  WHERE RF.FileType = 'L' AND RF.IsApplied = 0 AND RF.IsFinalLogFile = 0
  ORDER BY RF.ForDatabase, RF.CreatedDate
OPEN RestoreBackups
FETCH NEXT FROM RestoreBackups INTO @SQLToExecute, @RestoreFileId
WHILE (@@FETCH_STATUS = 0)
BEGIN

BEGIN TRY
    EXEC sp_executeSQL @SQLToExecute

		UPDATE [dbo].[RestoreFile]  SET IsApplied = 1 WHERE Id = @RestoreFileId
END TRY
BEGIN CATCH
    --SELECT   
        --ERROR_NUMBER() AS ErrorNumber,ERROR_MESSAGE() AS ErrorMessage
		UPDATE [dbo].[RestoreFile]  SET IsApplied = 0 WHERE Id = @RestoreFileId
END CATCH 

    FETCH NEXT FROM RestoreBackups INTO @SQLToExecute, @RestoreFileId
END
CLOSE RestoreBackups
DEALLOCATE RestoreBackups
GO

The pattern here is almost the same. I build my simple restore script but only for filetype = ‘L’ (Logfile) and only if it hasn’t been applied and if it isn’t the final logfile. I also check to make sure the full backup has restored and is earlier than the logfile. Then I execute the string of sql that I built and update the record to applied. I ended up having issues with the IsApplied being marked to true, even when it didn’t apply correctly so I added in error handling to change it back if that is the case. I may go back and add this to the Full Backup Restore script.

Next, I run the script that will do my final logfile restore for each database:

DECLARE @SQLToExecute nvarchar(max), @RestoreFileId int
DECLARE RestoreBackups CURSOR FAST_FORWARD
FOR
SELECT 
'RESTORE LOG ' + RF.ForDatabase + ' FROM DISK = N''' + RF.FileLocation + ''' WITH STANDBY=''Z:\Standby\' + RF.ForDatabase + '_Standby.bak'';'
, Id

  FROM [dbo].[RestoreFile] RF
	INNER JOIN sys.databases d ON RF.ForDatabase = d.name
	INNER JOIN sys.master_files mf on d.database_id = mf.database_id and mf.type = 0
	INNER JOIN	(SELECT ForDatabase, MAX(CreatedDate) AS CreatedDate
					FROM [dbo].[RestoreFile] RF
					WHERE RF.FileType = 'B' 
					GROUP BY ForDatabase) BF ON RF.ForDatabase = BF.ForDatabase AND RF.CreatedDate > BF.CreatedDate
  WHERE RF.FileType = 'L' AND RF.IsApplied = 0 AND RF.IsFinalLogFile = 1
  ORDER BY RF.ForDatabase, RF.CreatedDate
OPEN RestoreBackups
FETCH NEXT FROM RestoreBackups INTO @SQLToExecute, @RestoreFileId
WHILE (@@FETCH_STATUS = 0)
BEGIN

BEGIN TRY
    EXEC sp_executeSQL @SQLToExecute

		UPDATE [dbo].[RestoreFile]  SET IsApplied = 1 WHERE Id = @RestoreFileId
END TRY
BEGIN CATCH
    --SELECT   
        --ERROR_NUMBER() AS ErrorNumber,ERROR_MESSAGE() AS ErrorMessage
		UPDATE [dbo].[RestoreFile]  SET IsApplied = 0 WHERE Id = @RestoreFileId
END CATCH 

    FETCH NEXT FROM RestoreBackups INTO @SQLToExecute, @RestoreFileId
END
CLOSE RestoreBackups
DEALLOCATE RestoreBackups
GO

The only difference between these last two are the “WITH STANDBY” instead of “WITH NORECOVERY” and the IsFinalLogFile being set to true.

Last, but not least, I am cleaning old records out of my RestoreFile table so it doesn’t get large. I debated how much data to keep and for me, it makes sense to delete anything older than two days. I am doing a full restore daily and if I change that to weekly, I will keep this data longer.

  DELETE
  FROM [Maintenance].[dbo].[RestoreFile]
  WHERE CreatedDate < GETDATE()-2

This process allows me to stay flexible. If I don’t get a new full backup, this process will keep restoring logfiles until a new full backup file appears. If the backup chain is broken, this process will also break, but it will work as long as that chain is healthy and it gets all the files it needs.

Whew! We did it! That is my full poor girl log shipping process.

The song for this post is George Ezra’s Hold My Girl.

Load quick, got a list of files that need to be in a table…

Yes, I am trying to do a version of log-shipping that doesn’t require linking any servers and allows them to be in different domains. I also know I am crazy.

Now that we have that out of the way, let me show you some of the stuff I am doing! Loading a table from multiple file locations using Powershell. It will go and read what is in the directories and load them into a table for me (Oh my goodness, I knew Powershell was cool, but it took me so long to figure this out, that I was super proud with how fast it loaded).

First, the housekeeping. Create two tables in a DBA Database:

CREATE TABLE [dbo].[LogshippingFile](
	[FileName] [varchar](500) NULL,
	[FileLocation] [varchar](1000) NULL,
	[FileType] [char](1) NULL,
	[Origin] [varchar](50) NULL,
	[ForDatabase] [varchar](50) NULL,
	[CreatedDate] [datetime2](7) NULL,
	[Applied] [bit] NOT NULL
) ON [PRIMARY]
GO

ALTER TABLE [dbo].[LogshippingFile] ADD  CONSTRAINT [DF_LogshippingFiles_Applied]  DEFAULT ((0)) FOR [Applied]
CREATE TABLE [dbo].[Config](
	[Origin] [varchar](50) NULL,
	[FileLocation] [varchar](500) NULL,
	[Type] [char](1) NULL
) ON [PRIMARY]

In the Config table, I am going to put the common name for the database I am restoring so the rows will look like this:

OriginFileLocationType
Batman J:\FullBackups\Batman B
Batman J:\LogBackups\Batman L
Superman J:\FullBackups\Superman B
Superman J:\LogBackups\Superman L

Yay! Now we can get to the fun stuff. This is how I set up the Powershell:



# Credentials for connection to SQL Server. 
$username = 'Batgirl'
$password = '!Pennyworth54'
#Connecting to the SQL Server and Query to get FileLocation and Origin from the Config table. 
$DS = Invoke-Sqlcmd -ServerInstance "BatCave01" -Database "DBAStuff" -Username $username -Password $password -Query "SELECT FileLocation, Origin FROM Config" -As DataSet

#Magic that puts my two columns into variables
$data = $DS.Tables[0].Rows
$data
ForEach ($Row in $data) {
$FileLocation = $Row.FileLocation 
$Origin = $Row.Origin



#Build my insert statement so Powershell knows what to expect. 
#gets all the filenames and loads them into the table. 
$sqlstatement=@'
INSERT INTO LogshippingFile
(
FileName,	
FileLocation, 
FileType,
Origin,
ForDatabase 
) 
VALUES (
	'{0}',
	'{1}',
	'{2}',
	'{3}',
	'{4}'

   
)
'@
Get-ChildItem -Recurse $FileLocation  | 
select Name,FullName, 
    @{
    Label='FileType'
    Expression={($FileLocation.Substring(10,1))}}, @{
    Label='Origin'
    Expression={($Origin)}}, @{
    Label='ForDatabase'
    Expression={'Gryphon_'+($Origin)}} |
   # @{Label="Length";Expression={$_.Length / 1024KB -as [int] }}  |
ForEach-Object {
          $SQL = $sqlstatement -f  $_.name,$_.FullName, $_.FileType, $_.Origin, $_.ForDatabase #,  $_.Length		
    Invoke-sqlcmd -Username $username -Password $password -Query $SQL -ServerInstance “BatCave01” -database “DBAStuff” -Querytimeout 0
    }
    }

Now we have all this data in the table, what do we do with it? Guess you will have to tune in again to this same Bat channel.

The song for this post is Matt Nathanson’s Kiss Quick

Little Bit of Love, when your logs are rolling too much.

In January, the awesome Tim Radney (b|t) talked to the Utah user group about best practices. One that he mentioned was rolling over your error logs everyday and keeping 35 logs (a month plus 3 reboots). I loved this idea and implemented it using what I had done here and adding it to an agent job.

Then I realized we didn’t have any alerts on if our logs were rolling too much. Way back in my career, it used to be something that I would watch and it could mean someone was trying to hack your system and cover their tracks by rolling your logs over a bunch. I fought so much with figuring how to tell if my logs are rolling over, I had to save it for the future.

DROP TABLE IF EXISTS #EnumErrorLog;

CREATE TABLE #EnumErrorLog
(
    [Archive#] varchar(3) NOT NULL
        PRIMARY KEY CLUSTERED
    , [Date] datetime NOT NULL
    , [LogFileSizeByte] int NOT NULL
);

INSERT INTO #EnumErrorLog ([Archive#], [Date], [LogFileSizeByte])
EXEC sys.sp_enumerrorlogs;


SELECT CASE WHEN COUNT([Archive#]) > = 5 THEN 1 ELSE 0 END
FROM #EnumErrorLog
WHERE Date > DATEADD(hour, -3, GETDATE())

I create a temp table so I can execute a system stored proc to pull the information into a table and select it back out. I run this alert check once an hour, which means that for 3 hours if the alert condition has been met, it will alert me that something has rolled over too much (1 means to alert, 0 means to not do anything). I am using a third party tool right now, but I bet this could be set up with native SQL alerts or agent jobs.

The song for this post is Little Bit of Love by JP Cooper, it makes me smile even on the toughest days. *hugs*

Sometimes I get so high, so low, where did all my good plans go?

Greetings! Today I was playing with query store and noticed that I had some failing forced plans. How do you find failing forced plans? I asked this question over and over and finally found an answer.

I started on my main database and ran this query to look at query store:

SELECT *
FROM sys.query_store_plan
WHERE is_forced_plan = 1 and force_failure_count > 0

I had over 20 of them that were failing! Next, I had to figure out how to unforce the failing plans. Some of them were so old, they wouldn’t come up when I tried to look for them using the plan id in the GUI. I did more digging and found this:

EXEC sp_query_store_unforce_plan @query_id = Enter your queryid, @plan_id = Enter your planid

It was incredibly satisfying to watch each of the rows in the first query disappear as I ran them through the second query. This is now on my list of things to check so I can have a clean and healthy query store.

This post’s song is High Low by The Unlikely Candidates

I’m Going Under and This Time DBCC Can’t Save Me…

Over the weekend I was troubleshooting an issue and found that part of it was being caused by corruption on one of my databases.  I have seen DBCC CheckDB fix corruption a few times and thought, “No problem, I’ve got this.”

I ran:


ALTER DATABASE CorruptDB SET SINGLE_USER WITH ROLLBACK IMMEDIATE;
BEGIN TRANSACTION;
DBCC CHECKDB ('CorruptDB', REPAIR_REBUILD);
ALTER DATABASE CorruptDB SET MULTI_USER;
COMMIT

I waited for it to finish because the rule is, “Let it finish”.

There were still errors. So I ran it again.

The same errors appeared saying that some errors were waiting on others to be fixed first.  But nothing had fixed and my heart sank.  I had a moment of panic as I furiously googled.  I saw post after post telling me to restore the backup…Did I mention the small problem that the corruption had prevented the backups from completing for this database but didn’t throw an error? My backups were over two weeks old and a roll back was not an option.

I found a post by Paul Randal from 2006 talking about when DBCC CheckDB won’t solve the issue.  He mentioned that sometimes it is just an index or statistic that is corrupted.  This gave me hope and I started tracing my error.  I found it was on a Change Data Capture table that is being used in our replication.  I began pondering an idea.  I removed replication and ran the DBCC CheckDB.  It was clean! No corruption! I breathed and took a backup. Then I added replication back on and ran the DBCC CheckDB again.  It was clean again! Everything was healthy and fine!

Thank you to everyone for your past posts that save my sanity and thank you for DBCC CheckDB!!!

 

The song for this post is Someone You Loved by Lewis Capaldi

If I need to rearrange my fragmentation, I will for you….

Oh my goodness, I have been buried and learning tons about Indexing.  Something super cool that I learned was how to know if you need to adjust your fillfactors on your indexes.  My whole DBA career I have usually set it 85 and forgotten it. I learned that I can check my CommandLog and see how often that index is being rebuilt.   I am currently using Ola and when my indexes are rebuilt for maintenance (rebuilt nightly in this case), it is all logged to a table named CommandLog in the master database.  If an index is rebuilt everyday, adjust the fillfactor  down (75) so that it will rebuild less often.  If it nearly never rebuilds, adjust the fillfactor up (90) so it gets rebuilt once in a while and to avoid wasting space in an index. Here is the query I am using to see how often an index has rebuilt in the last 10 days.


SELECT Command, COUNT(StartTime) AS Rebuilt
FROM master.[dbo].[CommandLog]
WHERE CommandType LIKE '%Index%' AND StartTime > GETDATE()-10
GROUP BY Command
ORDER BY Rebuilt DESC

When I see ones that have a rebuilt number of 10, 9 or 8, I know those are rebuilding nearly everyday.  I will adjust their fillfactors down so that I don’t rebuild as often.  My sweet spot right now is once a week, but that doesn’t work for every index (or environment).  This is where the art part comes into the DBA world and I have to think about what is best for the system.  If I have already adjusted something down to 75 and it is still rebuilding frequently, I will adjust that number down more (65, then 55), and back up if it is too low.  I also evaluate how much that index is used and how important it is to keep healthy. If it is hardly ever used, do I want to waste fillfactor space?

Indexing is absolutely an art, but now I have a new brush for my kit by being able to query the CommandLog.

This post’s song is Particles by Nothing But Thieves

All of my friends say, “How much longer will it run?”

I have frequently needed to see how much longer a backup was going to run, or how much longer a restore was going to run.  This little bit of code is something I use to help me know how to plan.  I replace “Backup” with “Restore” if I want to see how much longer to plan for a restore.  I also use it for when I am tracking a rollback.  I remove the where clause and get a large list, the one that has a percentage and not a clear explanation is usually the one I want.   Sometimes the time remaining is lie.  I have had it imply 4 hours, when it took 13.  The comfort is that I can tell it is still working.


SELECT command, percent_complete,
'elapsed' = total_elapsed_time / 60000.0,
'remaining' = estimated_completion_time / 60000.0
FROM sys.dm_exec_requests
WHERE command like 'BACKUP%'

The song for this post is “Just Another Girl” by The Killers

Dynamic Data Masking keeps playing…keep your hands off my data!

As promised, I have been playing with Dynamic Data Masking and here are some things I have learned.  I downloaded World Wide Importers so I would have a place to play and there were masked columns already included.

This query will show us what has already been masked:

SELECT mc.name, t.name as table_name, mc.is_masked, mc.masking_function
FROM sys.masked_columns AS mc
JOIN sys.tables AS t
 ON mc.[object_id] = t.[object_id]
WHERE is_masked = 1;

Here we can see the column and the table that is being masked and what masking function is being used.

masking 1

This is a great time to talk about the different masking functions and what they do.  The four types in 2016 are Default, Email, Random and Custom String.

Default – For numeric and binary it will show a “0” For a date it will show 01/01/1900 and for strings it will show xxxx’s (more or less depending on the size of the field).

Email – It will expose the first letter of the email address and the suffix at the end of the email (.com, .net, .edu etc.) For example Batgirl@DC.com  would now be bxxx@xxxx.com.

Random – Number randomly generated between a set range. Kind of like the game, “Pick a number between 1 and 10” but for SQL.

Custom String – Lets you get creative with how much you show or cover and what you use to cover (not stuck with just xxxx’s).

Now for fun, let’s create a table that will be masked.

CREATE TABLE SuperHero
(HeroId INT IDENTITY PRIMARY KEY
,HeroName VARCHAR(100)
,RealName VARCHAR(100) MASKED WITH (FUNCTION = 'partial(1,"XXXXXXX",0)') NULL
,HeroEmail VARCHAR(100) MASKED WITH (FUNCTION = 'email()') NULL
,PhoneNumber VARCHAR(10) MASKED WITH (FUNCTION = 'default()') NULL);

Let’s add some data that we will want to mask:

INSERT SuperHero (HeroName, RealName, HeroEmail, PhoneNumber) VALUES
('Batman', 'Bruce Wayne', 'batsy@heros.com', '5558675309' ),
('Superman', 'Clark Kent', 'manofsteel@heros.com','5558675308' ),
('Spiderman', 'Peter Parker', 'spidey@heros.com','5558675307' );

SELECT * FROM SuperHero;

and finally we add some low level permissions of people who will look at the masked version of the data:

CREATE USER CommonPeople WITHOUT LOGIN; 
GRANT SELECT ON SuperHero TO CommonPeople; 

Now the test to see if CommonPeople have access to see all of our Superhero secrets:

EXECUTE AS USER = 'CommonPeople';
SELECT * FROM SuperHero; 
REVERT;

Try it out and see for yourself how it looks. Now you have experienced Dynamic Data Masking 101 in SQL Server 2016!

The song for this post is Good Charlotte – Keep Your Hands Off My Girl

All the Masking in the World Can Maybe Cover Your Dirty Laundry….

I have spent the last week learning about new features in SQL Server 2016 and one that I want to play with is Dynamic Data Masking (DDM).

What is data masking? It is a way to change or hide sensitive data. If I want to hide an email address that is Batgirl@DC.com,  I could either change it to be Batwoman@Heros.com using a masking software or I could use DDM to cover it like this BXXXXX@XXXXXX.com. I can also determine how many letters I want to cover with the masking in DDM.

If you want to permanently mask it for security purposes and force it to never link back to your production data, SQL Server Dynamic Data Masking (DDM) is not for you.  The built-in feature only applies a mask over the top, it doesn’t actually change the data that is stored in the database.   Think of SQL Servers’ version of data masking like a Halloween mask that sits on your face as opposed to plastic surgery that will forever change the way you look.

SQL Servers’ DDM will mask data to users that you set up to see the mask.  This is helpful for reporting or for curious people who want to look at data they shouldn’t be viewing.  It will not hide the data from privileged users.  It will not protect your data from someone taking a backup and restoring it somewhere else (If you want that, try Alway Encrypted instead). As a side note, DDM and Alway Encrypted won’t work together on the same column.

Now let’s get ready to play with Dynamic Data Masking in SQL Server.  (Coming next month)

Today’s song is Dirty Laundry by Carrie Underwood.