Adding a column to a table is common task for DBAs. You can add a column to a table which is a nullable column or which has default values. But are these two operations are similar internally and which method is optimal?
Let us start this with an example.
I created a database and a table using following script:
USE master Go --Drop Database if exists IF EXISTS (SELECT 1 FROM SYS.databases WHERE name = 'AddColumn') DROP DATABASE AddColumn --Create the database CREATE DATABASE AddColumn GO USE AddColumn GO --Drop the table if exists IF EXISTS ( SELECT 1 FROM sys.tables WHERE Name = 'ExistingTable') DROP TABLE ExistingTable GO --Create the table CREATE TABLE ExistingTable (ID BIGINT IDENTITY(1,1) PRIMARY KEY CLUSTERED, DateTime1 DATETIME DEFAULT GETDATE(), DateTime2 DATETIME DEFAULT GETDATE(), DateTime3 DATETIME DEFAULT GETDATE(), DateTime4 DATETIME DEFAULT GETDATE(), Gendar CHAR(1) DEFAULT 'M', STATUS1 CHAR(1) DEFAULT 'Y' ) GO -- Insert 100,000 records with defaults records INSERT INTO ExistingTable DEFAULT VALUES GO 100000
Before adding a Column
Before adding a column let us look at some of the details of the database.
DBCC IND (AddColumn,ExistingTable,1)
By running the above query, you will see 637 pages for the created table.
Adding a Column
You can add a column to the table with following statement.
ALTER TABLE ExistingTable Add NewColumn INT NULL
Above will add a column with a null value for the existing records.
Alternatively you could add a column with default values.
ALTER TABLE ExistingTable Add NewColumn INT NOT NULL DEFAULT 1
The above statement will add a column with a 1 value to the existing records.
In the below table I measured the performance difference between above two statements.
If you look at the RowCount parameter, you can clearly see the difference. Though column is added in the first case, none of the rows are affected while in the second case all the rows are updated. That is the reason, why it has taken more duration and CPU to add column with Default value.
We can verify this by several methods.
Number of Pages
The number of data pages can be obtained by using DBCC IND command. Though, this an undocumented dbcc command, many experts are ok to use this command in production. However, since there is no official word from Microsoft, use this “at your own risk”.
DBCC IND (AddColumn,ExistingTable,1)
Before Adding the Columns
Adding a Column with NULL
Adding a column with DEFAULT value
This clearly shows that pages are physically modified. Please note, a high value indicated in the Adding a column with DEFAULT value column is also a result of page splits.
The next thing you could do is, examine the data pages. For this as in the previous case, the DBCC PAGE is an undocumented DBCC command.
DBCC TRACEON(3604) DBCC PAGE (AddColumn, 1, 154, 3);
You need to switch on trace switch 3604 for DBCC Page command.
If you analyze page number 154 as shown in the example above, you will see the newly added column.
What does this mean? Is to say that even a Nullable column will modify the data pages? If that is the case, there shouldn’t be much different from adding a nullable column or adding a column with a default value, which is not what we saw in the first table.
Let us analyse lsn number of the page. Lsn is the Log Sequence Number of the last log record that changed the page.
You will see that there is no change to the lsn before and after adding the nullable column while there is a change to the lsn value, if you add a column with a default value.
This tells us two things.
1. This confirms that there are no physical changes to the data pages when you are adding a nullable column.
2. DBCC PAGE is not simply showing data dump of the data page.
When adding a column to a large table, the best practice would be adding a nullable column. However, if you want to update existing rows, consider using the Update statement after adding a nullable column.