So when stackoverflow or github goes down, and they trace the bug to some legacy component, they should refactor the entire thing, taking a week, instead of increasing the memory on their VM, bypassing the release process, then doing full postmortem once the fire is out?

This isn't dev. Running a factory is ops. It can't take 6 days to renew your domain name when you realise it expires in 2. It can't take 6 days to patch your server when some script kiddie takes it down the first time. Not everyone has a job that gives them the luxury of making sure everything is perfect before they commit to anything.