Now repeat all the tests developed earlier. Follow the usual debugging process if they fail. The tests can be repeated time and time again as the problem is debugged. It is natural to run a failing test over again each time a fix is attempted. However, since many server processes are interrelated, be sure to run the full suite before declaring the upgrade a success. The fix for the test that failed may have broken a previously successful test!
Customers should be involved here. As with the helpdesk model in Chapter 14, the job isn’t done until customers have verified that everything is complete. This may mean having the customer called at a prearranged time, or the customer may agree to report back the next day, after the maintenance window has elapsed. In that case, getting the automated tests right is even more critical.