Vaguely related: We use AWS S3 to poll for changes in a file from dozens of deployed devices regularly (polling is not the smartest way...). We get failures that the file header cannot be retrieved many times a week. Well, one could guess it's a connectivity problem of the device. Most are on cellular networks, but the problem also occurs from devices connected to one of the top US universities' network. The devices send an alarm right away when it happens via AWS SNS and we receive them, so we are forced to believe that AWS S3 reliability is not that high. If it really were a general connectivity issue we wouldn't get the alarm either; there is (light) retry for the S3 access, but none for the SNS alarm. These are authenticated accesses so more points of failure involved on S3 side. We use several regions relatively local to the devices and it happens in most if not all of them.