zlacker

Of those "correctly" architected apps, most are not properly tested for the failovers and won't actually work as architected (because of your own bugs or because aws failover stuff has bugs and you can't even test it).

Eg, falls over due to steep traffic spikes caused by outages when autoscaling mechanisms get previously unseen levels of load increases and enter some yoyo oscillation pattern, whole AZ is overloaded because all the failovers from the other failing AZ triggering at once, hit circuit breakers, spin up too slowly to ever pass health checks etc. Or can't detect something becoming glacially slow but not outright failing.

See eg https://www.theverge.com/2021/12/22/22849780/amazon-aws-is-d... & https://www.theverge.com/2020/11/25/21719396/amazon-web-serv... etc (many more examples are out there)