Hacker News Reader

This seems like a useful feature but it would be nice to have a better understanding of how it affects rate-limiting.

Historically Route53 has been pretty useless for service discovery because of the heavy rate-limiting. Something around 5 updates per second, on the whole account with a window of 60 seconds. I don't know if it's still the case but it used to be quite painful, especially when you need to scale up the number of instances quickly.

View on HN

Why would these instances not be behind ELB?

There are service architectures where putting them behind an ELB is not an ideal scenario (an example would be RabbitMQ; works great with haproxy in front of it, not so much with ELBs). ELBs aren't always a perfect fit, and depending on what sort of connection lifecycle or scaling velocity you require, direct addressing of service instances might be required.

WebSocket. ELB adds latency and is harder to debug.

Have you considered using NLBs instead?

That was before NLBs. But yeah, NLBs might be a good choice now.

Reply on news.ycombinator.com