A reliable way to verify T-SQL stored procedures

问题:

We're upgrading from SQL Server 2005 to 2008. Almost every database in the 2005 instance is set to 2000 compatibility mode, but we're jumping to 2008. Our testing is complete, but what we've learned is that we need to get faster at it.

I've discovered some stored procedures that either SELECT data from missing tables or try to ORDER BY columns that don't exist.

Wrapping the SQL to create the procedures in SET PARSEONLY ON and trapping errors in a try/catch only catches the invalid columns in the ORDER BYs. It does not find the error with the procedure selecting data from the missing table. SSMS 2008's intellisense, however, DOES find the issue, but I can still go ahead and successfully run the ALTER script for the procedure without it complaining.

So, why can I even get away with creating a procedure that fails when it runs? Are there any tools out there that can do better than what I've tried?

The first tool I found wasn't very useful: DbValidator from CodeProject, but it finds fewer problems than this script I found on SqlServerCentral, which found the invalid column references.

-------------------------------------------------------------------------
-- Check Syntax of Database Objects
-- Copyrighted work.  Free to use as a tool to check your own code or in 
--  any software not sold. All other uses require written permission.
-------------------------------------------------------------------------
-- Turn on ParseOnly so that we don't actually execute anything.
SET PARSEONLY ON 
GO

-- Create a table to iterate through
declare @ObjectList table (ID_NUM int NOT NULL IDENTITY (1, 1), OBJ_NAME varchar(255), OBJ_TYPE char(2))

-- Get a list of most of the scriptable objects in the DB.
insert into @ObjectList (OBJ_NAME, OBJ_TYPE)
SELECT   name, type
FROM     sysobjects WHERE type in ('P', 'FN', 'IF', 'TF', 'TR', 'V')
order by type, name

-- Var to hold the SQL that we will be syntax checking
declare @SQLToCheckSyntaxFor varchar(max)
-- Var to hold the name of the object we are currently checking
declare @ObjectName varchar(255)
-- Var to hold the type of the object we are currently checking
declare @ObjectType char(2)
-- Var to indicate our current location in iterating through the list of objects
declare @IDNum int
-- Var to indicate the max number of objects we need to iterate through
declare @MaxIDNum int
-- Set the inital value and max value
select  @IDNum = Min(ID_NUM), @MaxIDNum = Max(ID_NUM)
from    @ObjectList

-- Begin iteration
while @IDNum <= @MaxIDNum
begin
  -- Load per iteration values here
  select  @ObjectName = OBJ_NAME, @ObjectType = OBJ_TYPE
  from    @ObjectList
  where   ID_NUM = @IDNum 

  -- Get the text of the db Object (ie create script for the sproc)
  SELECT @SQLToCheckSyntaxFor = OBJECT_DEFINITION(OBJECT_ID(@ObjectName, @ObjectType))

  begin try
    -- Run the create script (remember that PARSEONLY has been turned on)
    EXECUTE(@SQLToCheckSyntaxFor)
  end try
  begin catch
    -- See if the object name is the same in the script and the catalog (kind of a special error)
    if (ERROR_PROCEDURE() <> @ObjectName)
    begin
      print 'Error in ' + @ObjectName
      print '  The Name in the script is ' + ERROR_PROCEDURE()+ '. (They don''t match)'
    end
    -- If the error is just that this already exists then  we don't want to report that.
    else if (ERROR_MESSAGE() <> 'There is already an object named ''' + ERROR_PROCEDURE() + ''' in the database.')
    begin
      -- Report the error that we got.
      print 'Error in ' + ERROR_PROCEDURE()
      print '  ERROR TEXT: ' + ERROR_MESSAGE() 
    end
  end catch

  -- Setup to iterate to the next item in the table
  select  @IDNum = case
            when Min(ID_NUM) is NULL then @IDNum + 1
            else Min(ID_NUM)
          end  
  from    @ObjectList
  where   ID_NUM > @IDNum

end
-- Turn the ParseOnly back off.
SET PARSEONLY OFF 
GO

回答1:

You can choose different ways. First of all SQL SERVER 2008 supports dependencies which exist in DB inclusive dependencies of STORED PROCEDURE (see http://msdn.microsoft.com/en-us/library/bb677214%28v=SQL.100%29.aspx, http://msdn.microsoft.com/en-us/library/ms345449.aspx and http://msdn.microsoft.com/en-us/library/cc879246.aspx). You can use sys.sql_expression_dependencies and sys.dm_sql_referenced_entities to see and verify there.

But the most simple way to do verification of all STORED PROCEDURE is following:

1. export all STORED PROCEDURE
2. drop old existing STORED PROCEDURE
3. import just exported STORED PROCEDURE.

If you upgrade DB the existing Stored Procedure will be not verified, but if you create a new one, the procedure will be verified. So after exporting and exporting of all Stored Procedure you receive all existing error reported.

You can also see and export the code of a Stored Procedure with a code like following

SELECT definition
FROM sys.sql_modules
WHERE object_id = (OBJECT_ID(N'spMyStoredProcedure'))

UPDATED: To see objects (like tables and views) referenced by Stored Procedure spMyStoredProcedure you can use following:

SELECT OBJECT_NAME(referencing_id) AS referencing_entity_name 
    ,referenced_server_name AS server_name
    ,referenced_database_name AS database_name
    ,referenced_schema_name AS schema_name
    , referenced_entity_name
FROM sys.sql_expression_dependencies 
WHERE referencing_id = OBJECT_ID(N'spMyStoredProcedure');

UPDATED 2: In the comment to my answer Martin Smith suggested to use sys.sp_refreshsqlmodule instead of recreating a Stored Procedure. So with the code

SELECT 'EXEC sys.sp_refreshsqlmodule ''' + OBJECT_SCHEMA_NAME(object_id) +
              '.' + name + '''' FROM sys.objects WHERE type in (N'P', N'PC')

one receive a script, which can be used for verifying of Stored Procedure dependencies. The output will look like following (example with AdventureWorks2008):

EXEC sys.sp_refreshsqlmodule 'dbo.uspGetManagerEmployees'
EXEC sys.sp_refreshsqlmodule 'dbo.uspGetWhereUsedProductID'
EXEC sys.sp_refreshsqlmodule 'dbo.uspPrintError'
EXEC sys.sp_refreshsqlmodule 'HumanResources.uspUpdateEmployeeHireInfo'
EXEC sys.sp_refreshsqlmodule 'dbo.uspLogError'
EXEC sys.sp_refreshsqlmodule 'HumanResources.uspUpdateEmployeeLogin'
EXEC sys.sp_refreshsqlmodule 'HumanResources.uspUpdateEmployeePersonalInfo'
EXEC sys.sp_refreshsqlmodule 'dbo.uspSearchCandidateResumes'
EXEC sys.sp_refreshsqlmodule 'dbo.uspGetBillOfMaterials'
EXEC sys.sp_refreshsqlmodule 'dbo.uspGetEmployeeManagers'

回答2:

Here is what worked for me:

-- Based on comment from http://blogs.msdn.com/b/askjay/archive/2012/07/22/finding-missing-dependencies.aspx
-- Check also http://technet.microsoft.com/en-us/library/bb677315(v=sql.110).aspx

select o.type, o.name, ed.referenced_entity_name, ed.is_caller_dependent
from sys.sql_expression_dependencies ed
join sys.objects o on ed.referencing_id = o.object_id
where ed.referenced_id is null

You should get all missing dependencies for your SPs, solving problems with late binding.

Exception: is_caller_dependent = 1 does not necessarily mean a broken dependency. It just means that the dependency is resolved on runtime because the schema of the referenced object is not specified. You can avoid it specifying the schema of the referenced object (another SP for example).

Credits to Jay's blog and the anonymous commenter...

回答3:

I am fond of using Display Estimated Execution Plan. It highlights many errors reasonably without ever having to really run the proc.

回答4:

I had the same problem in a previous project and wrote an TSQL checker on SQL2005 and later a Windows program implementing the same functionality.

回答5:

When I came across this question I was interested in finding a safe, non-invasive, and fast technique for validating syntax and object (table, column) references.

While I agree that actually executing each stored procedure will likely turn up more issues than just compiling them, one must exercise caution with the former approach. That is, you need to know that it is, in fact, safe to execute each and every stored procedure (i.e. does it erase some tables, for example?). This safety issue can be addressed by wrapping the execution in a transaction and rolling it back so no changes are permanent, as suggested in devio's answer. Still, this approach could potentially take quite a long time depending on how much data you are manipulating.

The code in the question, and the first portion of Oleg's answer, both suggest re-instantiating each stored procedure, as that action recompiles the procedure and does just such syntactic validation. But this approach is invasive--it's fine for a private test system, but could disrupt the work of other develoeprs on a heavily used test system.

I came across the article Check Validity of SQL Server Stored Procedures, Views and Functions, which presents a .NET solution, but it is the follow-up post at the bottom by "ddblue" that intrigued me more. This approach obtains the text of each stored procedure, converts the create keyword to alter so that it can be compiled, then compiles the proc. And that accurately reports any bad table and column references. The code runs, but I quickly ran into some issues because of the create/alter conversion step.

The conversion from "create" to "alter" looks for "CREATE" and "PROC" separated by a single space. In the real-world, there could spaces or tabs, and there could be one or more than one. I added a nested "replace" sequence (thanks, to this article by Jeff Moden!) to convert all such occurrences to a single space, allowing the conversion to proceed as originally designed. Then, since that needed to be used wherever the original "sm.definition" expression was used, I added a common table expression to avoid massive, unsightly code duplication. So here is my updated version of the code:

DECLARE @Schema NVARCHAR(100),
    @Name NVARCHAR(100),
    @Type NVARCHAR(100),
    @Definition NVARCHAR(MAX),
    @CheckSQL NVARCHAR(MAX)

DECLARE crRoutines CURSOR FOR
WITH System_CTE ( schema_name, object_name, type_desc, type, definition, orig_definition)
AS -- Define the CTE query.
( SELECT    OBJECT_SCHEMA_NAME(sm.object_id) ,
            OBJECT_NAME(sm.object_id) ,
            o.type_desc ,
            o.type,
            REPLACE(REPLACE(REPLACE(LTRIM(RTRIM(REPLACE(sm.definition, char(9), ' '))), '  ', ' ' + CHAR(7)), CHAR(7) + ' ', ''), CHAR(7), '') [definition],
            sm.definition [orig_definition]
  FROM      sys.sql_modules (NOLOCK) AS sm
            JOIN sys.objects (NOLOCK) AS o ON sm.object_id = o.object_id
  -- add a WHERE clause here as indicated if you want to test on a subset before running the whole list.
  --WHERE     OBJECT_NAME(sm.object_id) LIKE 'xyz%'
)
-- Define the outer query referencing the CTE name.
SELECT  schema_name ,
        object_name ,
        type_desc ,
        CASE WHEN type_desc = 'SQL_STORED_PROCEDURE'
             THEN STUFF(definition, CHARINDEX('CREATE PROC', definition), 11, 'ALTER PROC')
             WHEN type_desc LIKE '%FUNCTION%'
             THEN STUFF(definition, CHARINDEX('CREATE FUNC', definition), 11, 'ALTER FUNC')
             WHEN type = 'VIEW'
             THEN STUFF(definition, CHARINDEX('CREATE VIEW', definition), 11, 'ALTER VIEW')
             WHEN type = 'SQL_TRIGGER'
             THEN STUFF(definition, CHARINDEX('CREATE TRIG', definition), 11, 'ALTER TRIG')
        END
FROM    System_CTE
ORDER BY 1 , 2;

OPEN crRoutines

FETCH NEXT FROM crRoutines INTO @Schema, @Name, @Type, @Definition

WHILE @@FETCH_STATUS = 0 
    BEGIN
        IF LEN(@Definition) > 0
            BEGIN
                -- Uncomment to see every object checked.
                -- RAISERROR ('Checking %s...', 0, 1, @Name) WITH NOWAIT
                BEGIN TRY
                    SET PARSEONLY ON ;
                    EXEC ( @Definition ) ;
                    SET PARSEONLY OFF ;
                END TRY
                BEGIN CATCH
                    PRINT @Type + ': ' + @Schema + '.' + @Name
                    PRINT ERROR_MESSAGE() 
                END CATCH
            END
        ELSE
            BEGIN
                RAISERROR ('Skipping %s...', 0, 1, @Name) WITH NOWAIT
            END
        FETCH NEXT FROM crRoutines INTO @Schema, @Name, @Type, @Definition
    END

CLOSE crRoutines
DEALLOCATE crRoutines