Updating database schema or interfaces between services should take time if you want to limit the downtime. The extra work involved might mean you should accept a little downtime instead. It probably should be on a case by case basis rather than a blanket policy.