Hacker News new | ask | show | jobs
by rdl 4851 days ago
Even people who have been somewhat screwed by this issue with Heroku tend to love Heroku; Salesforce could make this right with generous payouts to affected people and 2-6mo of engineering to offer a better load balancer as a service (Actually, I think I could build this as a third party provider, even)

EA is basically a lost cause; their business model and culture is the problem. Having better sacks at Dachau wouldn't have helped the Nazi image, either (to auto-Godwin)
2 comments
chubot 4851 days ago
Well that's what they should have done BEFORE the threat of class action lawsuit. It makes them look at a lot worse if it happens after.

I'm pretty surprised they didn't offer refunds already. It was pretty clear they were asked for, and with no update by now, it's safe to say they were refused. That seems unreasonable to me. If you were in the position RapGenius, would you honestly not have expected a refund already?

link
rdl 4851 days ago
I wonder what happened to Heroku. They used to be awesome. They still have elements of being awesome, but seem either resource constrained or self destructive. I find it hard to believe Salesforce wouldn't give them what they need to do things correctly -- ranging from "go multi-AWS Region or move off AWS" to building a decent load balancing system to handling this fiasco.
link
chacham15 4851 days ago
> Actually, I think I could build this as a third party provider, even

I dont use Heroku, but if you read Rap Genius' original paper discussing the queuing problem, they specifically stated that there is nothing that they can do about it.

link
rdl 4850 days ago
RG couldn't do anything while using Heroku's load balancing infrastructure (well, other than rewriting their app to be more constant in performance).

A third-party could do a load balancer entirely outside of Heroku or even AWS, feeding sessions in based on feedback from the application ("intelligent routing"). Heroku even offers a way to bill for that kind of thing. You can open outgoing connections from each dyno and then load-balance over that.

Doing it with ELB would be possible I think, but I don't really like AWS ELB. I think it would make the most sense to do near US-East AWS, but arguably a CDN type service could do it from many locations. Being closer to users (as long as your connectivity to US-East/Heroku is still good) helps performance more.

link