[joeyi]

I share your paranoia in general (as ops), but can assure you that regions are very isolated from one and other. I know that releases are rolled out on a very long schedule (think quarter long release), and that is to prevent what you describe.

I would argue that the application (ie: the application being hosted on AWS) probably is going to fail before multiple regions do simultaneously and that should be addressed, before thinking about going multi-provider.

[justinsb]

Do you work for AWS as well? If so, I'd ask that team AWS spend less time astro-turfing on HN, and more time documenting your systems, so we can assess these risks for ourselves.

For example, I haven't heard of any precautions taken against a thundering herd of clients retrying requests in other regions if us-east goes down. What does AWS have there? How much spare capacity do you run in each region?