Archive | January, 2011

Winter Nerd-Cation: I’m Talking about Isolation Levels in Cleveland on February 5 (#SQLSat60)

Cerealizable: Breakfast Isolation Level

A few weeks ago I was talking with a friend about upcoming plans. Without thinking, I said:

“Oh, and I’m going to go to Cleveland to talk about Isolation Levels. And partitioning. In February.”

There was a  pause. My friend laughed. I believe she called me a nerd.

I took it as a compliment.

What Makes A Great Nerd-Cation

To me, the best, most relaxing vacation is a nerd-cation. I like to go someplace new, meet some new people, and ideally see some friends I don’t get to see often. I enjoy speaking and learning from other speakers. I like driving around and seeing an unfamiliar city. It gives my brain things to work on, but in a different way than I use it most of the time. (If I traveled for work, this might be a bit less vacation-y. Since I don’t, it’s all ice cream.)

If you’ve never tried this: you should! Make sure to add an extra day or two to the trip so you’re not in a rush. Take a copilot or go solo- it’s fun either way.

Note on terminology: I mean ‘Nerd’ in the nicest way possible. I’m proud of my inner nerd!

Why I Picked Cleveland

I submitted sessions to SQLSaturday #60 in Cleveland because I know two of the organizers, Allen White (b | t ) and Erin Stellato (b | t) are going to put on a really great event. I was really happy to be accepted to give sessions.

And although I have been to the midwest in the winter (taking my warmest hat and mittens!) I’ve never been to Cleveland.  Things I’m looking forward to trying to fit in: pierogi (oh how I love pierogi!!!), the Algebra Tea House, and the West Side Market.

Come to SQL Saturday Cleveland!

It’s going to be a blast. The schedule is full of good stuff.

My sessions are:

Table Partitioning: Evaluation, Planning, Options
Does SQL Server’s table partitioning feature offer performance improvements, manageability benefits, or both? Table partitioning is a significant investment: it’s important to understand many aspects of the feature to evaluate where it will be useful. This session will cover basic concepts of table partitioning. We’ll discuss where the feature can be applied, its benefits, and changes that partitioning may require to existing database schema. Special considerations for management of partitioned tables will be discussed, as well as potential changes to query optimization and plans

(NOLOCK) for YESFUN: Games with Isolation Levels

Understanding transaction isolation levels is critical for developing concurrent software as well as administering databases. Unfortunately most of us learn about transaction isolation early in our careers and don’t maintain an in-depth knowledge of different isolation levels. This talk will help! We’ll cover all isolation levels and discuss problems and benefits.

Check the event out on Twitter here: #sqlsat60

And get full information on SQLSaturday60 here.

Comments { 3 }

Internals Matter: Why You Should Check Your Page Verify Settings, and I Should Go to Masters Immersion Training (SQL Skills Contest Entry)

This post is about two things:

1) Your Page Verification Settings are Critical

You should confirm you are running with the correct page verification settings on your SQL Server databases. It’s very important, and I’ll show you why.

2) I Should Attend the Awesome SQLSkills Master Immersion Event – Internals and Performance

My scripts to demo the importance of page verification settings are part of  my entry to win free registration to the SQLSkills Master Immersion Event on Internals and Performance in February. Read more about the training here.

Keep reading to see why I hope this helps me win.

Let’s Start with Page Verification

The default setting for the PAGE_VERIFY database option for SQL 2000 was TORN_PAGE_DETECTION. As Books Online explains here:

When TORN_PAGE_DETECTION is specified, a specific 2-bit pattern for each 512-byte sector in the 8-kilobyte (KB) database page is saved and stored in the database page header when the page is written to disk. When the page is read from disk, the torn bits stored in the page header are compared to the actual page sector information.

SQL 2005 introduced a new page verify database option, CHECKSUM. This was the default for new databases, but the page verification option was not automatically changed when databases were migrated or upgraded from SQL 2000 to SQL 2005. Books Online explains that when you run with CHECKSUM…

the Database Engine calculates a checksum over the contents of the whole page and stores the value in the page header when a page is written to disk. When the page is read from disk, the checksum is recomputed and compared to the checksum value stored in the page header.

It is quite common to find databases on SQL 2005 instances and higher still using the TORN_PAGE_DETECTION setting for page verification. When a change to the setting is suggested, people naturally ask if the extra protection is worth the small amount of extra overhead of computing the checksum.

It certainly is! This post demonstrates why.

To Play Along At Home…

You may want to read my prior post on causing database corruption with a hex editor here. This post explains how to edit data in unsupported ways, and these techniques should never be used near a production server, or a database of any value at all. Make sure to wash your hands before and after using these practices, and having a change of clothes handy is also advisable.

You’ll also need to download the XVI32 hex editor by Christian Maas. No installer is necessary: download the zip file, then unzip all files to a directory and run XVI32.exe

Our Victim Database

Let’s say we have a database with entry data for a contest. The contest is, quite coincidentally, the SQLSkills Master Immersion Event on Internals and Performance in February!

Let’s create our database and schema for our contest entries.

USE master ;
IF DB_ID('CorruptMe') IS NOT NULL
    BEGIN
        ALTER DATABASE CorruptMe SET SINGLE_USER WITH ROLLBACK IMMEDIATE
        DROP DATABASE CorruptMe
    END
CREATE DATABASE CorruptMe ;
GO
ALTER DATABASE CorruptMe SET PAGE_VERIFY TORN_PAGE_DETECTION ;
GO
USE CorruptMe ;
--This schema might have an issue or two ;) Just play along.
CREATE TABLE dbo.Contest (
    ContestId INT IDENTITY,
    ContestName NVARCHAR(256),
    CONSTRAINT cxContestId_Contest UNIQUE CLUSTERED (ContestId)
 )
CREATE TABLE dbo.ContestEntry (
    ContestEntryId INT IDENTITY,
    ContestId INT REFERENCES dbo.Contest ( ContestId ) ,
    FirstName NVARCHAR(256) ,
    LastName NVARCHAR(256) ,
    ContestStatus NVARCHAR(256),
    CONSTRAINT cxContestEntryId_ContestEntryId UNIQUE CLUSTERED (ContestEntryId)
)

INSERT  dbo.Contest ( ContestName)
VALUES ( '5-Day Internals and Performance class in Dallas, February 21-25')

INSERT  dbo.ContestEntry ( ContestId, FirstName, LastName, ContestStatus )
VALUES ( 1, 'Hello', 'Kitty', 'Win!' ),
	( 1, 'PePe', 'LePeu', 'Loss' ),
	( 1, 'Kendra', 'Little', 'Sigh' )

It looks like the contest has already been judged. How did everyone do?

SELECT c.ContestName, ce.FirstName, ce.LastName, ce.ContestStatus
FROM dbo.Contest c
JOIN dbo.ContestEntry ce ON c.ContestId=ce.ContestId

Well, it figures. Hello Kitty can be pretty persuasive. Congrats, Kitty.

But What If…

What if something happened to the data? The database is running with TORN_PAGE_DETECTION– let’s see what happens if we change some of the values  in the physical data (mdf) file with a hex editor.

First, take a look at the pages for dbo.ContestEntry:

DBCC IND ('CorruptMe', 'ContestEntry', 1)

On my instance, we have one data page in our clustered index (PageType1). It’s PagePID 147. Let’s see if we can make some changes on that page and if SQL Server will notice.

To do this, take the database offline, and figure out your physical file name and your offset:

USE master ;
ALTER DATABASE CorruptMe SET OFFLINE ;

SELECT physical_name FROM sys.master_files WHERE name = 'CorruptMe' ;
SELECT 147 * 8192 AS [My Offset]

Then run the XV134 Hex editor with administrative privileges, and open the file.
Use Address -> GoTo and enter your offset as calculated above. (In my case: 1204224).

Here’s the data as I found it on the page (click to view in a new window):

And here is the data after I made quite a few changes in the hex editor, just by typing in the right pane (click to view in a new window):

That couldn’t possibly work, could  it?

Let’s put the database back online and see.

ALTER DATABASE CorruptMe SET ONLINE
GO
USE CorruptMe
GO
SELECT c.ContestName, ce.FirstName, ce.LastName, ce.ContestStatus
FROM dbo.Contest c
JOIN dbo.ContestEntry ce ON c.ContestId=ce.ContestId

This returns the following. The values on the page aren’t just in a different order, they’re different than the original values:

What About When I Run the Next CheckDb?

Let’s see:

DBCC CHECKDB('CorruptMe')

This comes back clean. An excerpt from the output is:

What Would Happen if We Were Using Page Verify CHECKSUM?

You can see this by running through the demo again. This time, use CHECKSUM as your database PAGE_VERIFY option, either by default or by running this after it is created:

ALTER DATABASE CorruptMe SET PAGE_VERIFY CHECKSUM;

After editing the page as described in the hex editor, you will be able to bring the database back online successfully. However, as soon as you read the records from the dbo.ContestEntry table, you’ll get an 824 error for the page we edited, like this:

Msg 824, Level 24, State 2, Line 1
SQL Server detected a logical consistency-based I/O error: incorrect checksum (expected: 0x54bc84f5; actual: 0x41298172). It occurred during a read of page (1:147) in database ID …

CheckDb will also alert you to the error the first time it is run.

The Moral of the Story Is: CHECKSUM is the Way to Go For Page Verify

This isn’t about security– this is about you knowing if changes to SQL Server’s files are happening outside of its control, whether by corruption or any other means.

It’s very easy to check if you’re using CHECKSUM. Take a look at your settings today, and make a plan to move to checksum for any databases which need it!

SELECT NAME
FROM sys.databases
WHERE page_verify_option_desc <> 'CHECKSUM'

Ideally you will include an index rebuild in your upgrade plan, or shortly thereafter– because the checksum needs to be written to each page, this doesn’t start working for a page  until it’s had a write.

For all sorts of more information, see Paul Randal’s post on Myths around Checksums.

Meanwhile, I’m Entering A Contest

Back to our contest for the SQLSkills Master Immersion Event on Internals and Performance in February (info).

My challenge was to express why I’d like to the go to the class, and why I’ll make the best use of the knowledge I gain.

I’d love to go because:

  • I’m passionate about learning about internals and performance. Hopefully I’ve demonstrated that here!
  • I still have a lot to learn, and there’s no replacement for a great classroom experience.
  • I know SQLSkills training is highly interactive and is a great place to ask questions and get real-world, in-depth answers.
  • This training will help me grow as a professional.

I’ll make the best use of the knowledge because:

  • I love using what I learn to explore more and write blog posts like this– it helps me learn more and share with the community.
  • I’m now the *only DBA* at a small, clever internet business. Every bit of expertise I can gain can help me work as part of a great team and make a real difference in our business.
  • I’m good at asking questions and helping others learn– I work hard to be a great fellow student and learn together with others.
  • I’ve clearly learned enough to be dangerous ;)

What I’d Like to Do at the Training

I would love to blog my way through my first immersion event. I certainly couldn’t (and wouldn’t) capture all the content we cover, but I’d like to capture what the experience is like. I’d also like to characterize how it  helps me think about, and work  with, SQL Server differently.

I hope whomever is selected as the winner chooses to share their experience.

Comments { 5 }

Corrupting Databases for Dummies- Hex Editor Edition

Corruption is so ugly it gets a lolworm instead of a lolcat.

Let’s make one thing clear from the start:

This Post Tells You How To Corrupt a SQL Server Database with a Hex Editor in Gruesome Detail

And that’s all this post tells you. Not how to fix anything, just how to break it.

If you aren’t familiar with corruption, corruption is bad. It is no fun at all on any data, or any server, that you care about.

Where You (Possibly) Want To Do This

You only want to do this on a test database, in a land far far away from your customers, for the purpose of practicing dealing with corruption.

When things go badly, you want to  be  prepared. This post gives you the tools in a simple, step by step fashion, to create different types of corruption so that you can practice resolving them.

Big Disclaimer: Do not run this in production. Or anywhere near production, or anything important. Ever. Only use this at home, in a dark room, alone, when not connected to your workplace, or anything you’ve ever cared about. If you corrupt the wrong pages in a user database, you may not be able to bring it back online. If you corrupt a system database, you may be reinstalling SQL Server.

References, and Thanks to Paul Randal

Everything I’m doing here I learned from Paul Randal’s blog posts. It just took me a little bit to understand how to use the hex editor and make sure I was doing it properly, so I thought I’d put down the steps I used here in detail. If you’d like to go straight to the source:

First, Get Your Hex Editor

Download XVI32 by Christian Maas. No installer is necessary: download the zip file, then unzip all files to a directory and run XVI32.exe

Create a Database to Corrupt

For our adventure, our database is named CorruptMe. We’ll create a single table, insert some data, and create a clustered index and nonclustered index on it.

(Note: Data generation technique found on Stack Overflow, attributed to Itzik Ben-Gan.)

USE master;
IF db_id('CorruptMe') IS NOT NULL
BEGIN
	ALTER DATABASE CorruptMe SET SINGLE_USER WITH ROLLBACK IMMEDIATE
	DROP DATABASE CorruptMe
END	

CREATE DATABASE CorruptMe;
GO

--Make sure we're using CHECKSUM as our page verify option
--I'll talk about other settings in a later post.
ALTER DATABASE CorruptMe SET PAGE_VERIFY CHECKSUM;

USE CorruptMe;

--Insert some dead birdies
CREATE TABLE dbo.DeadBirdies (
    birdId INT NOT NULL ,
    birdName NVARCHAR(256) NOT NULL,
    rowCreatedDate DATETIME2(0) NOT NULL )

;WITH
  Pass0 AS (SELECT 1 AS C UNION ALL SELECT 1),
  Pass1 AS (SELECT 1 AS C FROM Pass0 AS A, Pass0 AS B),
  Pass2 AS (SELECT 1 AS C FROM Pass1 AS A, Pass1 AS B),
  Pass3 AS (SELECT 1 AS C FROM Pass2 AS A, Pass2 AS B),
  Pass4 AS (SELECT 1 AS C FROM Pass3 AS A, Pass3 AS B),
  Pass5 AS (SELECT 1 AS C FROM Pass4 AS A, Pass4 AS B),
  Tally AS (SELECT ROW_NUMBER() OVER(ORDER BY C) AS NUMBER FROM Pass5)
INSERT dbo.DeadBirdies (birdId, birdName, rowCreatedDate)
SELECT NUMBER AS birdId ,
    'Tweetie' AS birdName ,
    DATEADD(mi, NUMBER, '2000-01-01')
FROM Tally
WHERE NUMBER <= 500000

--Cluster on BirdId.
CREATE UNIQUE CLUSTERED INDEX cxBirdsBirdId ON dbo.DeadBirdies(BirdId)
--Create a nonclustered index on BirdName
CREATE NONCLUSTERED INDEX ncBirds ON dbo.DeadBirdies(BirdName)
GO

Now we can take a look at the pages our table and nonclustered index got created on. I wanted to specifically corrupt a page in the nonclustered index on the DeadBirdies table. Of course if you wanted the clustered index, you could use index Id 1.

DBCC IND ('CorruptMe', 'DeadBirdies', 2)

I want to pick a data page for this nonclustered index, so I pick a PagePID where PageType=2. (The reference I use for DBCC IND is here.)

I pick PagePID 2784.

Note: If you’re following along, you may get a different PagePID if you use a different default fill factor.

Optional: Check out the page with DBCC PAGE

If you’d like to take a look at the page you’re about to corrupt, you can do so with the following command.

--Turn on a trace flag to have the output of DBCC PAGE return in management studio
--Otherwise it goes to the error log
DBCC TRACEON (3604);
GO
DBCC PAGE('CorruptMe', 1,2784,3);

Set the database offline

You must take your victim database offline to render it fully helpless accessible to your hex editor.

USE master;
ALTER DATABASE CorruptMe SET OFFLINE;

Also, get the name of your physical data file which you’ll open in your hex editor. Copy this to your clipboard.

SELECT physical_name FROM sys.master_files WHERE name='CorruptMe';

Figure out the starting offset of the page you want to corrupt. You do this simply by multiplying the page ID (PagePid) by 8192 (the number of bytes on a page).

SELECT 2784*8192 AS [My Offset]

It’s the Moment We’ve Been Waiting For: Trash That Page

Fire up your hex editor: run XVI32.exe.

Depending on your operating system, you may want to run this with elevated privileges / right click and “run as administrator”.

Open the database file by using File ? Open, and then the data file name you copied to the clipboard. (If you didn’t set the database offline, you’ll get an error that it’s in use. If you got an error that you don’t have permissions to view the file, make sure you do have permissions and that you ran XVI32.exe with elevated privileges.)

Go to the page you want to corrupt by using Address ? GoTo (or Ctrl + G), then paste in your Offset Value. You want to search for this as a decimal.

XVI43.exe will take to right to the beginning of that page.

You can see the ASCII representation of the data in the right pane. For our example, you should be able to see the word ‘Tweetie’ represented.

I like to put the cursor  in the right pane at the beginning of the word ‘Tweetie’. XVI32.exe will automatically move the cursor in the left pane, to the appropriate location.

You can corrupt the data  by editing in the right pane or left pane.

For my example, I am replacing the ASCII ‘T’ in the first occurrence of the word ‘Tweetie’ with an ‘S’. You can edit more, but a little tiny corruption goes a long way.

Save the file, and you’re done!

Admire Your Own Corruption

First, bring your database back online. If you correctly edited pages in the data, this should work just fine.

Note: If you corrupted critical system tables early in the database, this may not work! If so, go back to the steps above to identify a good page offset.

ALTER DATABASE CorruptMe SET ONLINE;

You can see the corruption in a couple of different ways. If you have checksums enabled on the database, you can see the corruption by reading the page with the data on it.

Since I corrupted a page in a nonclustered index in my example, I need to make sure I use that index. So I can see it with this query:

Use CorruptMe;
SELECT birdName FROM dbo.deadBirdies;

That returns this big scary error, which confirms I did indeed corrupt page 2784:

Msg 824, Level 24, State 2, Line 1
SQL Server detected a logical consistency-based I/O error: incorrect checksum (expected: 0xb633a8e1; actual: 0xaeb39361). It occurred during a read of page (1:2784) in database ID 18 at offset 0x000000015c0000 in file ‘D:\BlahBlahBlah\CorruptMe.mdf’. Additional messages in the SQL Server error log or system event log may provide more detail. This is a severe error condition that threatens database integrity and must be corrected immediately. Complete a full database consistency check (DBCC CHECKDB). This error can be caused by many factors; for more information, see SQL Server Books Online.

You can also see this by running a CHECKDB or CHECKTABLE command.

DBCC CHECKDB('CorruptMe')

An excerpt from its output:

Msg 8928, Level 16, State 1, Line 1
Object ID 2105058535, index ID 2, partition ID 72057594038910976, alloc unit ID 72057594039828480 (type In-row data): Page (1:2784) could not be processed.  See other errors for details.
Msg 8939, Level 16, State 98, Line 1
Table error: Object ID 2105058535, index ID 2, partition ID 72057594038910976, alloc unit ID 72057594039828480 (type In-row data), page (1:2784). Test (IS_OFF (BUF_IOERR, pBUF->bstat)) failed. Values are 12716041 and -4.

Now Sit Back and Laugh Maniacally. And Then Fix It.

So, the whole point of this was probably to test something.

So take a moment to enjoy the fact that FOR ONCE you don’t have to panic when you see these errors, because it’s all part of your master plan.

Then go out and fix the corruption, and run your tests.

Comments { 18 }

Read from the Right End of the Index: BACKWARD Scans

Optimizing queries is the most fun when you don’t need to add indexes. There’s nothing quite so nice as finding a way to make reading data faster, without slowing down writes or creating new data structures that need to be maintained.

Here’s one way you can use BACKWARD scans to do this.

The Scenario: Clustered index on an increasing integer, and you’d like recently created rows

This is a common enough situation: you have a table with a clustered index on an integer value which increases with each row. You have another column which records the date the row was created.

You’d like frequently query the most recently created rows over some period of time.

The table has very frequent inserts, so for performance reasons you want to use the minimal indexes required. (And in general, this is the best practice.)

Question: Do you need to add a nonclustered index on the column containing the date the row was created?

Answer: Maybe not!

Getting the right clustered index scan

Say we’re working with the following table, which we have filled with five million rows of Tweetie birds. (Note: This generation technique is a tally table population technique which I found on Stack Overflow, which is attributed to Itzik Ben-Gan.)

CREATE TABLE dbo.Birds (
    birdId INT NOT NULL ,
    birdName NVARCHAR(256) NOT NULL,
    rowCreatedDate DATETIME2(0) NOT NULL )
GO	

--Insert 5 million Tweetie birds
--Make them as if they were all created a minute apart.
;WITH
  Pass0 as (select 1 as C union all select 1),
  Pass1 as (select 1 as C from Pass0 as A, Pass0 as B),
  Pass2 as (select 1 as C from Pass1 as A, Pass1 as B),
  Pass3 as (select 1 as C from Pass2 as A, Pass2 as B),
  Pass4 as (select 1 as C from Pass3 as A, Pass3 as B),
  Pass5 as (select 1 as C from Pass4 as A, Pass4 as B),
  Tally as (select row_number() over(order by C) as Number from Pass5)
INSERT dbo.Birds (birdId, birdName, rowCreatedDate)
SELECT Number AS birdId ,
    'Tweetie' AS birdName ,
    DATEADD(mi, number, '2000-01-01')
FROM Tally
WHERE Number <= 5000000

--Cluster on BirdId. We won't add any other indexes.
CREATE UNIQUE CLUSTERED INDEX cxBirdsBirdId ON dbo.Birds(BirdId)

Say we would just like to see the maximum value in the rowCreatedDate column.

The most basic way to get this row is with this query:

SELECT MAX(rowCreatedDate)
FROM dbo.Birds

However, that leads to a table scan. We get lots of reads: 22,975 logical reads and 201 physical reads.

If we know we have a strong association between the BirdId column and the RowCreatedDate column, and that the highest ID in the table is the most recent row, we can rewrite the query like this:

SELECT MAX(rowCreatedDate)
FROM dbo.Birds
WHERE birdId = (SELECT MAX(birdId) FROM dbo.Birds)

This query still does a clustered index scan. But yet it does only 3 logical reads and 2 physical reads.

Looking in the execution plan, our query was able to use the extra information we provided it to scan the index backwards. It stopped when it had everything it needed, which was after a short distance– after all, it only needed recent rows, and those are all at one end of the table.

This backwards scan can be very useful, and can make using the MAX aggregate very useful.

But you usually need more than just the max value…

To see a bit more about how you extend this logic, compare these three queries:

Query A

This makes you think you need that non-clustered index: it does 22,975 logical reads, 305 physical reads, and 22968 read-ahead reads.

--Only run against a test server, not good for production
DBCC DROPCLEANBUFFERS

SELECT birdId, birdName, rowCreatedDate
FROM dbo.Birds
WHERE rowCreatedDate >= '2009-07-01 05:00:00'

Query B

We can introduce the backwards scan by adding an ORDER BY BIrdId DESC to the query. Now we get 23019 logical reads, 47 physical reads, and 22960 read-ahead reads.

--Only run against a test server, not good for production
DBCC DROPCLEANBUFFERS

SELECT birdId, birdName, rowCreatedDate
FROM dbo.Birds
WHERE rowCreatedDate >= '2009-07-01 05:00:00'
ORDER BY birdid desc

Query C

The this last query gives the optimizer extra information about using BirdId to do a BACKWARD scan to grab the maximum BirdId, and then use that to do a BACKWARD seek of the clustered index in nested loops to get the data. It does only 50 logical reads, 4 physical reads, and 817 read-ahead reads.

--Only run against a test server, not good for production
DBCC DROPCLEANBUFFERS

SELECT birdId, birdName, rowCreatedDate
FROM dbo.Birds
WHERE birdId >=
	(SELECT MAX(birdId)
	FROM dbo.Birds
	WHERE rowCreatedDate <= '2009-07-01 05:00:00')
AND rowCreatedDate >= '2009-07-01 05:00:00'
ORDER BY birdId DESC

Be Careful Out There

The examples I’m using work because there is a correlation between the integer field and the date field. Not all tables may be like this. As with all queries, you need to be familiar with your data.

Consider Your Options– Even the Ones You Don’t Think Are Great

I’m quite sure BACKWARD index reads are covered in some talks and publications on tuning.  But I learned about this by considering multiple approaches, even those I didn’t think would work at first. It pays to try things out, and you can look a lot by looking carefully at execution plans (including the properties) and your Statistics IO output.

What this means to me: it’s good to keep an open mind.

Comments { 4 }