Thanks for your comment Josh! These are very good questions, so:

• Because of the business we’re running in my current startup, we have regular sudden spikes of traffic where we go from about 30k rpm to almost 200k in a time of 2–3 minutes. Reaching the AWS support, we did not find a way to be flexible enough on how we could pre-warm the ELB. Also, from another perspective, we wanted to have more control on our LB. As I had already worked with HAProxy before, we went for this solution (this was about 2 months ago). But you’re right, there are other great solutions out there, and I’d be happy to try out Traefik and other options, both for personal knowledge and to see if we might consider another switch at some point.
• I’ll probably do a small update on my post about this point. Although we did not went as far as Yelp on the problem, we did use the “hack” of dropping SYN packets from incoming TCP connections during the time HAProxy reloads (as pointed out in this answer http://serverfault.com/a/580630). Also, we don’t reload at all if the hosts list hasn’t changed.