One time, in my entire career, we had a web app that started getting strange errors that had us baffled. We tried a dozen things to try and figure out what was wrong, but we were panicking and needed an answer immediately. So, we attached a debugger to the production instance and set up a few watch/break points. It helped us track down the errors and fix the problem.
Naturally, it hung the server during our debugging session, and made people mad, but no more mad than they already were, because of the problem we had.
It would not have been necessary if the code had been written better, with error logging and diagnostic points. I don't expect to ever do it again.