Comparing MS SQL Database Structures and Data

Filed under Code Garage, SQL

If you’ve messed with SQL much, invariably, you get into a situation where you have to ask “What has changed in the database from the last version to this version?”

There’s some good tools out there for this, to be sure.

AdeptSQL is my favorite, but RedGate SQL Compare is very good. And DBGhost is highly regarded for synchronizing db schemas (and data to a degree, I believe).

But sometimes you want (or need) to “roll your own”.

So, how would you do that?

First, the criteria:

  • I need to be able to compare both schema and some “seed data” in certain tables (for instance, pre-populated lookup tables, etc)
  • I’d like to see any significant changes, but everyone’s definition of significant can be different.
  • I’d like the changes to be “easy” to read. They need to pop out if possible

Next, the display.

Well, there’s plenty of file comparison utilities out there, and my favorite is Araxis, so why not just use it to do the “delta-generation” heavy lifting.

I’ll need two text files to compare, the “before” version and the after.

I need them grouped by schema object type (stored-procs in one section, table definitions in another, data values in another, etc).

And I need them ordered consistently so that any changes are easy to spot.

I ended up a with a stored proc that, at least right now, does most of what I need.

The basic idea is that you call DBOutput, and it then calls itself recursively to resolve and output all the various elements of the database that you might be interested in comparing. Why a recursive procedure instead of several separate procs? No earth-shattering reasons. It’s just easier to deal with one proc for this sort of thing that several separate ones.

The routine itself pretty simply, and it doesn’t yet have to resort to directly reading the system tables for any information. It makes heavy use of the INFORMATION_SCHEMA views to retrieve most information. It should run under MSSQL 2000, MSDE, 2005 Express and 2005, but I haven’t tested it under the 2000 variants. 

Probably most notable is the use of sp_helptext to retrieve the actual full definition text of stored procs and views.

You’ll have to excuse the formatting, pasting into HTML seems to have messed up my nice tabs.

ALTER PROCEDURE [dbo].[DBOutput] (
    @TableName sysname = null,
	@SPName sysname = null,
	@ViewName sysname = null
    )

AS

DECLARE @Rows as int

SET NOCOUNT ON

if	len(IsNull(@TableName, '')) = 0 and 
	Len(IsNull(@SPName,'')) = 0 and 
	Len(IsNull(@ViewName,'')) = 0 BEGIN

    -- do all tables
    declare @Table sysname
	declare @Routine sysname
	declare @View sysname
    declare @i int
    
    DECLARE cCursor CURSOR FOR
        select lower(Table_Name) from Information_Schema.Tables
        where table_type = 'base table' order by Table_Name
    
    OPEN cCursor
    FETCH NEXT FROM cCursor INTO @Table
    WHILE @@FETCH_STATUS = 0 BEGIN
        -- check for tables we don't care about
        if @Table in ('sysdiagrams'
                     )
            -- do nothing
            set @i = 0
        else if left(@Table,2) in ('ms', 'xx')
            set @i = 0
        else if left(@Table,3) in ('sys')
            set @i = 0
        else if left(@Table,6) in ('dtprop')
            set @i = 0
        else BEGIN
            exec DBOutput @Table, ''
            END
    
        FETCH NEXT FROM cCursor INTO @Table
        END

    CLOSE cCursor
    DEALLOCATE cCursor


	-- now Scan all the Stored Procs and functions
    DECLARE rCursor CURSOR FOR
        select lower(ROUTINE_NAME) from Information_Schema.Routines order by ROUTINE_NAME
    
    OPEN rCursor
    FETCH NEXT FROM rCursor INTO @Routine
    WHILE @@FETCH_STATUS = 0 BEGIN
        -- check for Stored Procs and functions we don't care about
        if @Routine in (''
                     )
            -- do nothing
            set @i = 0
        else if left(@Routine, 3) in ('sp_', 'fn_')
            set @i = 0
        else BEGIN
            exec DBOutput '', @Routine
            END
    
        FETCH NEXT FROM rCursor INTO @Routine
        END

    CLOSE rCursor
    DEALLOCATE rCursor


	-- now Scan all the Stored Procs and functions
    DECLARE vCursor CURSOR FOR
        select lower(TABLE_NAME) from Information_Schema.Views order By TABLE_NAME
    
    OPEN vCursor
    FETCH NEXT FROM vCursor INTO @View
    WHILE @@FETCH_STATUS = 0 BEGIN
        -- check for Stored Procs and functions we don't care about
        if @View in (''
                     )
            -- do nothing
            set @i = 0
        else if left(@View, 3) in ('sp_', 'fn_')
            set @i = 0
        else BEGIN
            exec DBOutput '', '', @View
            END
    
        FETCH NEXT FROM vCursor INTO @View
        END

    CLOSE vCursor
    DEALLOCATE vCursor

    RETURN
    END


-- handle a single table recursively
IF Len(@TableName) > 0 BEGIN
	DECLARE @PrimaryKeyColumn nvarchar(64)
	DECLARE @sql nvarchar(1000)
	DECLARE @Parms nvarchar(1000)
	DECLARE @Temp sysname

	-- Primary key info (we're not using yet)
	SELECT @PrimaryKeyColumn = [COLUMN_NAME] FROM INFORMATION_SCHEMA.KEY_COLUMN_USAGE
	   WHERE [TABLE_NAME] = @TableName

	-- dump a list of the columns, with pertinent info
	PRINT replicate('*', 60)
	PRINT replicate('*', 60)
	PRINT 'TABLE: ' + @TableName 
	EXEC ('SELECT ORDINAL_POSITION as ColNum, left(COLUMN_NAME, 20) as ColName, left(DATA_TYPE, 20) as DataType, Left(COLUMN_DEFAULT, 20) as DefaultValue, CHARACTER_MAXIMUM_LENGTH as Len, IS_NULLABLE from INFORMATION_SCHEMA.COLUMNS WHERE Table_name = ''' + @TableName + ''' order by ColNum')

	-- dump a list of the indexes, with pertinent info
	-- also includes RI information associated with the keys (if any)
	EXEC ('SELECT	Keys.ORDINAL_POSITION as KeyNum, 
					left(Keys.CONSTRAINT_NAME,30) as IDXName, 
					left(Keys.COLUMN_NAME, 20) as ColumnName, 
					Left(RI.UNIQUE_CONSTRAINT_NAME, 20) as PrimaryKey, 
					RI.Match_Option, 
					RI.Update_Rule, 
					RI.Delete_Rule,
					CHK.Check_Clause
			FROM information_schema.Key_Column_Usage as Keys 
					left outer join information_schema.referential_constraints as ri on keys.Constraint_Name = ri.Constraint_Name 
					left outer join information_schema.check_constraints as chk on keys.Constraint_name = chk.Constraint_Name
			WHERE Table_name = ''' + @TableName + ''' 
			order by KeyNum, IDXName')

	-- check if there's any data in the table. If so, dump it all
	set @SQL = 'SELECT @Cnt = Count(*) FROM ' + @TableName
	set @parms = '@Cnt int OUTPUT'
	EXEC sp_executesql @SQL, @Parms, @Cnt = @Rows OUTPUT
	IF @Rows > 0 BEGIN
		PRINT replicate('~', 60)
		PRINT 'TABLE DATA FOR:' + @TableName
		EXEC ('SELECT * FROM ' + @TableName)
		END
	PRINT ''
	PRINT ''
	END


-- Handle all stored procs here
IF LEN(IsNull(@SPName, '')) > 0 BEGIN
	PRINT replicate('*', 60)
	PRINT replicate('*', 60)
	PRINT 'STORED PROCEDURE/FUNCTION: ' + @SPName
 
	-- pull the entire text of the proc into a variable to let us print it nicely
	EXEC sp_helptext @SPName
	PRINT ''
	PRINT ''
	END


-- Handle all Views
IF LEN(IsNull(@ViewName, '')) > 0 BEGIN
	PRINT replicate('*', 60)
	PRINT replicate('*', 60)
	PRINT 'VIEW: ' + @ViewName
 
	-- pull the entire text of the proc into a variable to let us print it nicely
	EXEC sp_helptext @ViewName
	PRINT ''
	PRINT ''
	END


SET NOCOUNT OFF

Basically, you just open up Enterprise Manager, and exec DBOutput. It’s best to put the results window in TEXT mode as opposed to grid mode. Then you can simply save the output to a text file and use your favorite file compare to get a delta. All the PRINTs are there just to make things a little more legible in a typical “text editor” view.

Let me know what you think, or if there’s additional bits to report that might be useful.

Also, I plan to update this script over time, and I’ll post the updates here.

2 Comments

  1. Ralf says:

    Very cool! It’s going in my scrapbook of SQL Server code. 🙂

    Sometimes all you want to do is check to see if ANYTHING has changed in a SQL Server schema. Call it a quick pre-check to see if further digging & comparison is warranted. To detect global changes, I use:

    SELECT SUM(base_schema_ver) FROM sysobjects

    …and store the resulting huge-ass number away for future reference. Upon initialization of my app, I rerun the query and compare the results with my stored value. If they don’t match, I know something, someplace has changed. That might trigger a further check of required tables, or simply an alert to let the user knows their schema has been twiddled.

  2. Darin says:

    That’s not a bad technique. Kind of a crude checksum, but it’d work.

Post a Comment

Your email is never published nor shared. Required fields are marked *

*
*