My department currently relies heavily on CmdExec steps in SQL Server Jobs, called using a SQL Server Login.
osql /e /n /S SERVER01 /U USER /P PA$$$ /d DATABASE_01 /i "\\LOCATION\OF\SQL\SCRIPT\scriptfile.sql" /o "\\LOCATION\OF\SQL\LOG\scriptfile.log"
For a lot of reasons (not least security and the expectation that this method will become obsolete) I'd like us to move away from this model and replace with a mixture of stored procedures and SSIS packages.
I've proposed this several times and I'm always told that we can only move in this direction if we can recreate the detail of logging possible by the method above, which our department has come to rely on quite heavily. The method above outputs the contents of the script fed in along with any server messages including all rowcounts and error messages, in context.
This does admittedly mean that errors that wouldn't count as errors for SQL Server are far more easily picked up - Unexpectedly low row counts etc. You could put in as much error handling as you liked and never get something as easy to follow as the logging from this method, so I can see this side of the argument.
To sell my boss on a move away from cmdexec steps I'd like to find a method of recreating something like the output of the logging that cmdexec currently gives us - perhaps to SQL tables rather than files (although either would be fine) - in SSIS packages and stored procedures.
I'm familiar with using raiserror for error handling - I use this extensively in stored procedures - but the task here is not just to catch SQL errors, but to catch all that cmdexec logging catches and ideally in context of the script being executed... Short of writing a stored procedure and executing it after every step in every script I can't see a way of doing this.
Also I'm keen to hear about best practice approaches to logging for people who don't use cmdexec steps. I've learnt SQL in this department and this is the only way I've ever known of doing things.
Increasingly for my own development I try to write stored procs with error handling. And if I come across issues that don't throw errors, I do my troubleshooting by rerunning the code interractively on a dev server so I can monitor the step by step information. This is less convenient than the full logging we get from cmdexec but more secure.
Do other people do more than this? Is there a better way?