AWS has had at least one documented incident where a region had an S3 failure that was not recoverable. They lost about 2% of all data. That might not sound like much but if you have a lot of data, partial restoration of that data doesn't necessarily leave your system in a functional state. If it loses my compiled CSS files I might be able to redeploy my app to fix it. Then again if I'm a SaaS company and that file was generated in part from user input, it might be more difficult to reconstruct that data.
Which incident is this? I can’t find it online. The closest I can recall is when they lost some number of EBS volumes. We were affected by that, but ran snapshots (to s3) to recover the affected servers.
