Friday, August 7, 2009

TSQL Challenges

I just like to thank Jacob Sebastian who is a fellow SQL Server MVP and founder of the www.tsqlchallenges.com for offering me an opportunity to be part of the TSQL Challenges team. I’m really happy to be a part of a team consists of people like Jacob, Alejandro Messa, Peter Larsson (all 3 are SQL Server MVPs), Adam Haines (a moderator of MSDN SQL Server forums), Rui Carvalho and many other talented people.

Here is a brief description about TSQL Challenges site: TSQL Challenges constantly aim at helping people to enhance their SET based query writing skills. With TSQL Challenges, sometimes you learn stuff that you don’t know, sometimes you will see better ways of doing stuff that you already know and sometimes you will be able to use your expertise to help others to learn TSQL querying skills. Even SQL Server experts love TSQL Challenges because every challenge inspires them to come up with new better ways of solving the given problem.

The Mission: The entire “TSQL Challenge” team will focus on fulfilling our mission; “helping people to enhance their SET based query writing skills”. We will come up with more and more interesting TSQL Challenges that encourages you to look for alternate logics and inspires you to think outside the regular thought process.

I would like to invite my readers to participate in a TSQL Challenge - www.tsqlchallenges.com

Also like to thank Jacob again for a warm welcome and kind word he has put in introduction post - Introducing new “TSQL Challenge” Team Members

So I hope I will come with up some interesting SQL puzzles that will challenge your SQL skills and also you will a fun solving them.

Mangal

Tuesday, August 4, 2009

SQL Server 2008 Service Pack (SP) 1 on Microsoft Update as a Required Automatic Update

SQL Server 2008 Service Pack 1 will soon be available through Automatic Update starting from September. 

For the latest information you can read it from SQL Server Setup blog - SQL Server 2008 Service Pack (SP) 1 on Microsoft Update as a Required Automatic Update

For better understanding of Automatic Update see - Update Your PC Automatically

Wednesday, July 22, 2009

UNION Vs UNION ALL

Many times you may have heard this “Use UNION ALL over UNION whenever possible.”  The question arises - why?  To answer this question in one statement  - UNION ALL performs faster compare to UNION. 
Then again question arises - Why UNION ALL performs faster?  Also - Why whenever possible, why not always?

Let me answer the 2nd question 1st – Though both UNION and UNION ALL combines the results of two or more queries into a single result set, there is fundamental difference between these two.  UNION returns only DISTINCT result set, while UNION ALL returns basically all rows available, this includes duplicates.

Lets see the following example:

-- create 2 tables A and B.
CREATE TABLE A
(
ID INT,
Names VARCHAR(10)
)
GO
CREATE TABLE B
(
ID INT,
Names VARCHAR(10)
)
GO
-- insert data into table A
INSERT INTO A VALUES(1,'Mangal');
INSERT INTO A VALUES(5,'Sham');
INSERT INTO A VALUES(2,'Ram');

-- insert data into table B
INSERT INTO B VALUES(2,'Ram');
INSERT INTO B VALUES(3,'Shiv');
INSERT INTO B VALUES(4,'John');

-- test sample data
SELECT id, Names
FROM A
GO
SELECT id, Names
FROM B
GO

Here is how the data of the table A and B looks like :

samle

Note that id=2 and names=Ram is there in both the tables.  That will help us in understanding the difference between UNION and UNION ALL.  Now lets execute the following 2 queries, 1st is with UNION and 2nd is with UNION ALL.

-- with UNION
SELECT id, Names
FROM A
UNION
SELECT id, Names
FROM B
GO

-- with UNION ALL
SELECT id, Names
FROM A
UNION ALL
SELECT
id, Names
FROM B
GO

The result: 

Result

Observations :
1. 1st query with UNION returns 5 rows, and UNION ALL query returns 6 rows.
2.  Row for ID=2(for Ram) appears twice in UNION ALL result set.
3. Result set for UNION is sorted on ID column.  For UNION ALL all the rows of table A appeared 1st followed by rows of table B(no sort).

As you can see, UNION eliminates any duplicate rows from final result set while UNION ALL returns basically all rows available including duplicates.  That is the cause of UNION being slow.  For each row UNION operator checks whether the entire row exists in previous rows or not.  And for making this validation UNION by default 1st sort the result set on the 1st available column of the result set.  In our example UNION has sorted the result set on ID column even though I haven’t specified any ORDER BY clause.  If you see Name “Sham” (which is in table A) appeared last in the UNION result because it has the highest id 5 while it appeared on 2nd row of UNION ALL result.  A look at the query execution plan can help you visualizing it better :

plan 

As you can see cost of the UNION query is 73% compare to 27% for UNION ALL.  And measure reason being the “Distinct Sort” that UNION operator performs to sort and to eliminate the duplicate rows.  While UNION ALL doesn’t really bother about sort and duplicates.  And that is why UNION is slow compare to UNION ALL. 

So again going back to question – why not use UNION ALL always?  And one more question to be added - when to use which one?

- You should  use UNION when you don’t want the duplicates in your final result set, and you are not sure (or may be you are sure) that duplicate records exists in the different queries involved in the UNION.

- You should be using UNION ALL when :
1.  You are not bothered about the duplicate rows in the result.
2.  You are sure there are no duplicates in different queries involved in UNION.  e.g. if you are combining results from 2 or more different years(sales orders) with each query reruns result for individual year with some unique id for each row.  Or combining result for 2 or more different departments.

All this long I’m talking about UNION and UNION ALL as if they are 2 different things all together.  Are they?  Not exactly.  Reason I’m saying this because, when one of my friend asked me about UNION ALL and I advised him to look into the Books online, and he came back complaining me that “books online doesn’t say anything about UNION ALL”.  Reason – he was thinking that Books online must be having some separate section dedicated to UNION ALL, as if it is different from UNION.

Actually, the ALL is just an optional argument in the UNION syntax.  For more on UNION you can refer the books online - http://msdn.microsoft.com/en-us/library/ms180026.aspx.  

Mangal

Friday, July 3, 2009

DELETE Vs TRUNCATE

I know this has been done many times, but still here is something from my side. What is the difference between DELETE and TRUNCATE?
Well one reason I wanted to write this post was, on so many of the blogs, forum threads I keep seeing statements like “TRUNCATE cannot be rolled back”. This is also one of the most frequently asked questions in Interviews.  And since my target audience is people who are just started learning SQL (and NOT SQL Experts), I thought I should write this one.
Instead of just focusing on ROLLBCAK I will try to cover all the differences between DELETE and TRUNCATE.
  • Remove Data  : First thing first, both can remove the data from a table. 
    But a DELETE can be used, to remove the rows not only from a Table but also from a VIEW or the result of an OPENROWSET or OPENQUERY subject to provider capabilities.

  • FROM Clause : With DELETE you can also delete rows from one table/view/rowset_function_limited based on rows from another table by using another FROM clause.  In that FROM clause you can also write normal JOIN conditions.  Actually you can create a DELETE statement from a SELECT statement that doesn’t contain any aggregate functions by replacing SELECT with DELETE and removing column names. 
    With TRUNCATE you can’t do that.


  • WHERE : A TRUNCATE cannot have WHERE Conditions, but a DELETE can.  That means with TRUNCATE you can’t delete a specific row or specific group of rows.
    TRUNCATE TABLE is similar to the DELETE statement with no WHERE clause.

  • Performance : TRUNCATE TABLE is faster and uses fewer system and transaction log resources.
    And one of the reason is locks used by either statements. The DELETE statement is executed using a row lock, each row in the table is locked for deletion. TRUNCATE TABLE always locks the table and page but not each row.

  • Transaction log : DELETE statement removes rows one at a time and makes individual entries in the transaction log for each row. 
    TRUNCATE TABLE removes the data by deallocating the data pages used to store the table data and records only the page deallocations in the transaction log.

  • Pages : After a DELETE statement is executed, the table can still contain empty pages.
    TRUNCATE removes the data by deallocating the data pages used to store the table data.

  • Trigger : TRUNCATE does not activate the delete triggers on the table.  So you must be very careful while using TRUNCATE.  One should never use a TRUNCATE if delete Trigger is defined on the table to do some automatic cleanup or logging action when rows are deleted.
  • Identity Column : With TRUNCATE if the table contains an identity column, the counter for that column is reset to the seed value defined for the column.  If no seed was defined, the default value 1 is used.
    DELETE doesn’t reset the identity counter.  So if you want to retain the identity counter, use DELETE instead.

  • Replication : DELETE can be used against table used in transactional replication or merge replication. 
    While TRUNCATE cannot be used against the tables involved in transactional replication or merge replication.

  • Rollback : DELETE statement can be rolled back. 
    TRUNCATE can also be rolled back provided it is enclosed in a TRANSACTION block and session is not closed. 
    Once session is closed you won't be able to Rollback TRUNCATE.
  • Restrictions : The DELETE statement may fail if it violates a trigger or tries to remove a row referenced by data in another table with a FOREIGN KEY constraint. If the DELETE removes multiple rows, and any one of the removed rows violates a trigger or constraint, the statement is canceled, an error is returned, and no rows are removed. 
    And if DELETE is used against View, that View must be an Updatable view.

    TRUNCATE cannot be used against the table used in Indexed view. 

    TRUNCATE cannot be used against the table referenced by a FOREIGN KEY constraint, unless a table that has a foreign key that references itself.

Source :
DELETE -
http://msdn.microsoft.com/en-us/library/ms189835.aspx 
TRUCNATE -
http://msdn.microsoft.com/en-us/library/ms177570.aspx

I hope you liked my this post on the topic: the difference between DELETE and TRUNCATE in SQL Server.

Friday, June 26, 2009

PIVOT Multiple Columns

In one of my previous post I showed you how to UNPIVOT multiple columns.  On similar lines I also wanted to write on “How to PIVOT multiple columns?”, so this post was due for some time.  Actually I was looking for some efficient way of doing it.  Limitation of PIVOT operator is, it supports pivoting only on a single column.  But you can always have multiple PIVOT operators in the FROM clause. I was trying to create a PIVOT query with multiple columns with multiple PIVOT operators. But at the end of it I found that our old fashioned CASE expression is performing much better than a multiple PIVOT operator query. 

Even though I’m writing this post on how to write a multiple PIVOT operator query, my suggestion is use CASE expressions instead for getting better performance.  Though personally I like to avoid CASE also.   Normally I like to do it in Reporting Services, by creating a Matrix report.  Now a days almost all Reporting Tools provides you an option of creating Matrix report.  And good thing about Matrix report is unlike PIVOT operator you don’t need to hard code any column value.

If you try to write a PIVOT query with 2 PIVOT operators, and use same column in FOR clause you will get an error : Invalid column name NameOfColumn.

Or if you use same column, but by declaring it again and using a different alias name, you still get an error : The column name ValueOfColumn specified in the PIVOT operator conflicts with the existing column name in the PIVOT argument.

So what’s the solution?  Solution is, declare the same column again, change the values in the column by some constant(you can add some constant, or you can concat some identifier ) and assign a new alias name to column.

Lets see the following example, I have used the AdventureWorks database of SQL Server 2005.

USE AdventureWorks
GO
SET ANSI_WARNINGS OFF
SELECT
CustomerId,
        SUM([Q2001]) AS Qty2001,
        SUM([Q2002]) AS Qty2002,
        SUM([V2001]) AS Val2001,
        SUM([V2002]) AS Val2002
FROM (
        SELECT     H.CustomerId,
                SUM(D.OrderQty) AS TotalQty,
                SUM(D.LineTotal) AS TotalVal,
                'Q'+CONVERT(VARCHAR(4),H.OrderDate,120)  AS QYear,
                'V'+CONVERT(VARCHAR(4),H.OrderDate,120)  AS VYear
        FROM Sales.SalesOrderDetail AS D INNER JOIN
        Sales.SalesOrderHeader AS H ON D.SalesOrderId = H.SalesOrderId
        WHERE D.ProductId=771
        AND H.OrderDate >='20010101'
        AND H.OrderDate <'20030101'
        GROUP BY H.CustomerId,
                CONVERT(VARCHAR(4),H.OrderDate,120)
    )Main
PIVOT
    (
        SUM(TotalQty)
        FOR QYear IN ([Q2001],[Q2002])
    ) PQ
PIVOT
    (
        SUM(TotalVal)
        FOR VYear IN ([V2001],[V2002])
    ) PV
GROUP BY CustomerId
ORDER BY CustomerId
GO

The query returns total quantity and line amount for year 2001 and 2002 for the product id 771 for all customers.  If look at the query carefully in Main sub query, CONVERT(VARCHAR(4),H.OrderDate,120) this convert statement will take out the Year part from the OrderDate column.  I have declared the same column twice, at first I concatenated Q to the Year, and at second time I concatenated the V.  Just execute the Main sub query, so it will be easy to understand for you.

SELECT     H.CustomerId,
                SUM(D.OrderQty) AS TotalQty,
                SUM(D.LineTotal) AS TotalVal,
                'Q'+CONVERT(VARCHAR(4),H.OrderDate,120)  AS QYear,
                'V'+CONVERT(VARCHAR(4),H.OrderDate,120)  AS VYear 
FROM Sales.SalesOrderDetail AS D INNER JOIN 
Sales.SalesOrderHeader AS H ON D.SalesOrderId = H.SalesOrderId 
WHERE D.ProductId=771 
                AND H.OrderDate >='20010101' 
                AND H.OrderDate <'20030101' 
GROUP BY H.CustomerId,
                CONVERT(VARCHAR(4),H.OrderDate,120)

Now we have 2 columns, with different values, and we can use them in different PIVOT with same effect, and that’s what I have done in my 1st query.

Here is a CASE expression version of same query, which gives much better performance if you scale it for large amount data.

USE AdventureWorks
GO
SELECT     H.CustomerId,
               SUM(CASE YEAR(H.OrderDate)
                      WHEN 2001
                      THEN D.OrderQty
                      END) AS Qty2001,
               SUM(CASE YEAR(H.OrderDate)
                      WHEN 2002
                      THEN D.OrderQty
                      END) AS Qty2002,
              SUM(CASE YEAR(H.OrderDate)
                     WHEN 2001
                     THEN D.LineTotal
                     END) AS Val2001,
              SUM(CASE YEAR(H.OrderDate)
                     WHEN 2002
                     THEN D.LineTotal
                     END) AS Val2002
FROM Sales.SalesOrderDetail AS D INNER JOIN
Sales.SalesOrderHeader AS H ON D.SalesOrderId = H.SalesOrderId
WHERE D.ProductId=771
          AND H.OrderDate >='20010101'
          AND H.OrderDate <'20030101'
GROUP BY H.CustomerId
ORDER BY H.CustomerId
GO

You can test the performance of both queries.  If you want to scale it for larger data you can remove the WHERE conditions added by me.  Total execution time for CASE query is almost half to that of PIVOT query.