I guess you have to read this kind of items hidden in careful language, running the instances had no problem, it is different matter they had limited connectivity!from AWS point of view they don't seem to see user impact but services from their point of view.

Perhaps that distinction has value if your workloads did not depend on network connectivity externally for example say S3 access without vpc and only compute some DS/ ML jobs perhaps.

How are you measuring this? Remember that cloudwatch was apparently also losing metrics, so aggregating CW metrics might show that kind of drop.

Yeah, I know. This was based off instance store logging on these instances. For better or worse, they're very simple ports of pre-AWS on-prem servers, they don't speak AWS once they're up and running.