@futurebird
Maybe all the exchanges run on aws so didn't process orders.

@xinit

@futurebird just look at the math. They were down for what 3 hours? And can you remember when that last happened? Right. Their uptime is beyond impressive. AWS is a beast.

@futurebird Moving a service out of AWS is something I’ve helped do. It burns a whole lot of time and has impacts on costs and revenue that are hard to measure let alone predict.

@noplasticshower @futurebird things weren't totally back to normal in us-east-1 for nearly 12 hours.

Even 3 hours of downtime is really bad. If you're up over the rest of the month, that's 99.6% uptime. Typical availability targets for a SaaS company are more like 99.9% (43 minutes of downtime a month), 99.99% being achievable if you are careful.

AWS gives SLAs of 99.99% for a number of single-region services, and 99.5% for individual EC2 instances. This outage took their overall availability for an entire region from the level of a careful highly-available service to that of a single host with no redundancy. Anyone who relied on those SLAs got screwed, and Amazon is going to be getting a lot of requests for significant refunds per the terms of those SLAs.

AWS in general is a beast but anyone who's worked with it at scale for a long time knows us-east-1 is best avoided, due to a long history of incidents like this (though usually not quite as bad).