sp_MSForEachDB doesn't seem to like GO - sql

Question

I have used this SP before. Now, I am trying to permission a user to 50 odd databases that start with the same letters, using the code below. It looks like it does not like "GO" in the code. Why is that ? and what is the work around?
Thanks for your time.. :)
RM
exec sp_MSForEachDB
'
IF ''?'' LIKE ''MYDBNames%''
BEGIN
Use [?]
Go
CREATE USER [MYDOMAIN\Analysts] FOR LOGIN [MYDOMAIN\Analysts]
GO
EXEC sp_addrolemember N''db_owner'', N''MYDOMAIN\Analysts''
GO
END
'

Answer 1

I just explained this in another question yesterday (here). This essence is this: GO isn't a SQL statement, it's an SSMS/SQLCMD command that is used to separate batches (groups of SQL statements that are compiled together). So you cannot use it in things like stored procedures or Dynamic SQL. Also, very few statement contexts can cross over a GO boundary (transactions and session-level temp tables are about it).
However, because both stored procedures and Dynamic SQL establish their own separate batches/execution contexts, you can use these to get around the normal need for GO, like so:
exec sp_MSForEachDB
'
IF ''?'' LIKE ''MYDBNames%''
BEGIN
Use [?]
EXEC(''
CREATE USER [MYDOMAIN\Analysts] FOR LOGIN [MYDOMAIN\Analysts]
'')
EXEC('' EXEC sp_addrolemember N''''db_owner'''', N''''MYDOMAIN\Analysts'''' '')
END
'

Answer 2

The word GO is a batch separator and is not a SQL keyword. In SSMS you can go to options and change it to anything - COME for example.
Try this:
exec sp_MSForEachDB
'
IF ''?'' LIKE ''MYDBNames%''
BEGIN;
Use [?];
CREATE USER [MYDOMAIN\Analysts] FOR LOGIN [MYDOMAIN\Analysts];
EXEC sp_addrolemember N''db_owner'', N''MYDOMAIN\Analysts'';
END;'

Answer 3

Just stop using GO, it's not necessary (and it's not even T-SQL; it's just a batch separator for Management Studio).
Stop using sp_MSForEachDB. Oh, the problems (if you want proof, see here, here and here).
Here is code that does this without using any stored procedure:
DECLARE #sql NVARCHAR(MAX) = N'';
SELECT #sql += N'USE ' + QUOTENAME(name) + ';
CREATE USER [MYDOMAIN\Analysts] FOR LOGIN [MYDOMAIN\Analysts];
EXEC sys.sp_addrolemember N''db_owner'', N''MYDOMAIN\Analysts'';
-- actually should use ALTER ROLE now
'
FROM sys.databases
WHERE state = 0 -- online
AND name LIKE N'MyDBName%';
PRINT #sql;
-- EXEC sys.sp_executesql #sql;

Answer 4

Like #RBarryYoung's answer, an embedded exec() can be used to avoid needing a "go" when the command carries the requirement that it must be the "first within a batch." I don't think the question's example needs a "go" but the headline itself may lead people here.
Here is a basic SP to wrap this that support's DDL statements like "create proc" and provides some error messages. I'm probably the ~5th person to do this and this one isn't the end all.
Blessing/Curse: There is a call to Common.ufn_UsvToNVarcharKeyTable() which you may need to patch in your own CSV splitter or remove:
alter proc [Common].[usp_ForEachDatabase]
#pSql nvarchar(max),
#pDatabaseNameLikeOpt nvarchar(300) = null, -- Optional pattern to match with like
#pDatabaseNameCsvOpt nvarchar(max) = null, -- Optional list of DBs
#pJustPrintDbNames bit = 0, -- Don't exec, just print database names
#pDebug bit = 1 -- may add additional print statements
as
/*-----------------------------------------------------------------------------------------------------------------
Purpose: Execute SQL on each database. Replacement for the standard sp_MSForEachDB which has a size character
limit and requires an exec() within for DDL commands which must be the first in a batch.
Sources: Ideas from http://www.experts-exchange.com/Database/MS-SQL-Server/SQL-Server-2005/Q_26337388.html
http://stackoverflow.com/questions/20125430
http://stackoverflow.com/questions/1819095/
Modified By Description
---------- ---------- -----------------------------------------------------------------------------------
2014.10.14 crokusek Initial version. Pieces from internet.
--------------------------------------------------------------------------------------------------------------*/
begin try
declare databaseCursor cursor local forward_only static for
select IsNull(sd.Name, ud.Name) as Name,
--
-- If a list was specified, flag when a name was not found
--
convert(bit, iif(sd.Name is null, 1, 0)) as IsNotFound
from
(
select Name from sys.databases
where Name not in ('master', 'tempdb', 'model', 'msdb')
and is_distributor = 0 -- http://stackoverflow.com/questions/1819095/
) sd
full outer join ( select Value as Name from Common.ufn_UsvToNVarcharKeyTable(#pDatabaseNameCsvOpt, ',' ) ) ud
on ud.Name = sd.Name
where (#pDatabaseNameLikeOpt is null or IsNull(sd.Name, ud.Name) like #pDatabaseNameLikeOpt)
and (#pDatabaseNameCsvOpt is null or ud.Name is not null)
order by IsNull(sd.Name, ud.Name);
declare
#matchingDatabaseNames nvarchar(max),
#databaseName nvarchar(300),
#isNotFound bit,
#errorCount int = 0,
#successCount int = 0,
--
-- Use an embedded exec() to place the user command(s) in a separate context to avoid errors like:
-- CREATE/ALTER PROCEDURE must be the first statement in a query batch.
--
#sqlExec nvarchar(max) = N'exec (''' + replace(#pSql, '''', '''''') + ''')';
open databaseCursor
fetch next from databaseCursor into #databaseName, #isNotFound;
while ##fetch_status = 0
begin
if (#isNotFound = 1)
begin
print 'Error: Database ' + #databaseName + ' was not found.';
set #errorCount += 1;
end
else
begin
set #matchingDatabaseNames = coalesce(#matchingDatabaseNames + ',', '') + #databaseName; -- Create Csv
print 'Database: ' + #databaseName;
if (#pJustPrintDbNames = 0)
begin
declare
#execSql nvarchar(max) = 'use ' + #databaseName + ';' + char(10) + #sqlExec;
begin try
exec (#execSql)
set #successCount += 1;
end try
begin catch
select #databaseName as [Database],
error_number() as ErrorNumber,
error_severity() as ErrorSeverity,
error_state() as ErrorState,
error_procedure() as ErrorProcedure,
error_line() as ErrorLine,
error_message() as ErrorMessage;
set #errorCount += 1;
end catch
end
end
fetch next from databaseCursor into #databaseName, #isNotFound;
end
if (#pJustPrintDbNames = 1)
print #matchingDatabaseNames; -- this can then be used as input
else
print 'Completed with ' + convert(varchar(30), #errorCount) + ' Errors ' +
'and ' + convert(varchar(30), #successCount) + ' Successes.';
end try
begin catch
if (xact_state() = -1)
rollback;
-- Log / Rethrow
end catch
go
Usage:
declare
#sql nvarchar(max) = N'
create proc usp_ReplaceEachSingleQuoteWithTwoSingleQuotesInTheDefinition
...';
exec [Common].[usp_ForEachDatabase]
#pSql = #sql,
#pDatabaseNameLikeOpt = null,
#pDatabaseNameCsvOpt = null,
#pJustPrintDbNames = 0,
#pDebug = 1;

Answer 5

Why not use a cursor?
They are great for management tasks like this. As long as the record set is small, they perform well.
The code below creates a SQL statement for each database that uses the ; to combine multiple tasks.
Unlike the code above, it remove any existing users before creating a new one. Avoids a possible issue.
Then it executes the code. If you want, you could even log these adhoc requests to a internal table and add error checking.
Good luck.
--
-- EXEC same statements against several like databases
--
-- Declare local variables
DECLARE #STMT NVARCHAR(4000);
-- Allocate cursor, return table names
DECLARE MYTRG CURSOR FAST_FORWARD FOR
SELECT
' use [' + ltrim(rtrim(d.name)) + ']; ' +
' IF EXISTS (SELECT * FROM sys.database_principals WHERE name = N''[MYDOMAIN\Analysts]'') DROP USER [MYDOMAIN\Analysts]; ' +
' CREATE USER [MYDOMAIN\Analysts] FOR LOGIN [MYDOMAIN\Analysts] WITH DEFAULT_SCHEMA=[DBO]; ' +
' sp_addrolemember N''db_owner'', N''MYDOMAIN\Analysts''; ' as STMT
FROM master.dbo.sysdatabases d
WHERE d.name like 'MYDBNames%'
ORDER BY d.name;
-- Open cursor
OPEN MYTRG;
-- Get the first row
FETCH NEXT FROM MYTRG INTO #STMT;
-- While there is data
WHILE (##FETCH_STATUS = 0)
BEGIN
-- Show detail database info
EXEC sp_executesql #STMT;
-- Get the first row
FETCH NEXT FROM MYTRG INTO #STMT;
END;
-- Close the cursor
CLOSE MYTRG;
-- Release the cursor
DEALLOCATE MYTRG;