Traffic 101: Packets Mostly Flow
slack.engineering
slack.engineering
EDIT: and, for any potential smart-arse who points out that I did not expand "IME" - TBH, that rule does not hold value (IMO) for abbreviations that are widely-known, e.g. "AWS", etc.
it seems obvious at the level of like an individual load balancer to consider the load of the nodes backing an upstream when it decides which one to send reqs to. i didn't think of lifting that to the record that points to the load balancers.
i'm trying to think how you'd do it with route53. maybe with health checks that take load into account? i don't think you'd get the low and high watermarks like in ns1s thing. maybe a weighted record with something wired up to to tweak weights based on load? sounds just annoying enough to get right and working safely that if route53 implemented it well, people would pay for it...