Seems very inspired by Molly White's page [0]
Serverless Horrors
21–30 of 174 posts
Re: Serverless Horrors
#22A kill switch seems a practical and trivial solution. E.g. something toggled when a monthly bill exceeds a configurable amount. I’m guessing many hobby users would set that to $100, end of story. I don’t understand why it’s so unpopular to offer that. I’d imagine the providers would benefit too in the long run.
It would be practical for the user, but not for the company.
Why does this not exist after years (more than a decade since EC2) of cloud computing?
Because it is not good for the VC investors.
Re: Serverless Horrors
#23A kill switch seems a practical and trivial solution. E.g. something toggled when a monthly bill exceeds a configurable amount. I’m guessing many hobby users would set that to $100, end of story. I don’t understand why it’s so unpopular to offer that. I’d imagine the providers would benefit too in the long run.
You can set a hard budget on most cloud providers, including Azure etc. I actually did this thinking I was just setting a budget alert and that was a new and interesting way to interrupt production.
Re: Serverless Horrors
#24Earlier quoted context omitted.
It used to be we will take your server down. Now its we'll DDOS your serverless website and leave you a 100k bill. I'm not sure how sustainable such business model is. When you owned the server, you could unplug it. Now you have no way of knowing if somebody is going to hit your /api a million times per minute
true. At what point could you say they are accomplices to the attackers
Ah the Yelp biz model:
"would be a shame if somebody read a bad review about your biz, pay us to remove it"
to Netlify
"would be a shame if somebody DDOS'd your serverless website, subscribe to our DDOS protection plan"
Re: Serverless Horrors
#25I don't think the Netlify case has happend due to serverless architecture. Yes, there are many technical issues with serverless architectures, but getting a huge bill for incoming traffic is independent from a serverless architecture. You could also put your own hardware in a colo and pay for traffic. If your data center provider does not protect you from DDoS and simply bills by the TB, then the same situation would…
If your stuff scales to infinity your money burn rate also scales to infinity. A single server in a colo can only do so much damage to your wallet before it’s saturated.
Sure, but according to the article, he received the 100k bill because of peak traffic of 60TB of bandwidth per day over several days. If you have a dedicated server in a colo, then a 10GbE connection is very common these days. With a 10GbE connection, you can serve 100TB of traffic in a single day. Hence, if the colo had the same prices as Netlify for traffic, then a single server with a 10GbE connection would be sufficient for such a high bill, no serverless and infinity scaling required, just absurdly high traffic costs.
Re: Serverless Horrors
#26Everyone in this thread is talking about how this is not a Serverless problem, but a cloud problem. That's correct, but the point stands. Bandwidth is pure margin and to me, it's a bit scummy to a) charge this much for bandwidth and b) offer no tools to combat attacks that aren't in the customer's control. According to this thread[1], Netlify doesn't even have an option to take a site down for a while if it's under DDoS attack.
Netlify has multiple options to solve the problem: billing controls, request throttling, setting bandwidth limit, and waving of overages if it was a genuine DDoS, but it doesn't seem they are interested because that'd be like killing the golden goose. Just look at all the things bunny.net CDN offers [2].
It's disappointing how we acquiesce to explanations of cloud providers that are motivated by nothing other than greed.
[1]: https://answers.netlify.com/t/limiting-bandwidth-traffic-to-...
[2]: https://support.bunny.net/hc/en-us/articles/360014190440-Und...
Re: Serverless Horrors
#27How is no spending limit a serverless problem? It's a cloud problem. Edit: Well, if you have small cloud servers they might go down under an attack. And if you pay only outgoing like in AWS that's good for you.
It’s an architecture failing. Setting maximums is critical. If you choose a platform that doesn’t support that more fool you. It’s implicitly easier to have a ‘circuit breaker’ in a traditional environment, but still warrants consideration. Load test for success and failure.
Re: Serverless Horrors
#28A kill switch seems a practical and trivial solution. E.g. something toggled when a monthly bill exceeds a configurable amount. I’m guessing many hobby users would set that to $100, end of story. I don’t understand why it’s so unpopular to offer that. I’d imagine the providers would benefit too in the long run.
You can set a hard budget on most cloud providers, including Azure etc. I actually did this thinking I was just setting a budget alert and that was a new and interesting way to interrupt production.
This is a highly welcomed feature however, I don’t think AWS has it.
Re: Serverless Horrors
#29Bandwidth charging models from hosts are crazy. They don’t model their upstream costs in any way. Transit providers charge based on 95th% and ports scale based on highest burst not volume passed. I’ve always hated it.
Re: Serverless Horrors
#30Earlier quoted context omitted.
If your stuff scales to infinity your money burn rate also scales to infinity. A single server in a colo can only do so much damage to your wallet before it’s saturated.
Sure, but according to the article, he received the 100k bill because of peak traffic of 60TB of bandwidth per day over several days. If you have a dedicated server in a colo, then a 10GbE connection is very common these days. With a 10GbE connection, you can serve 100TB of traffic in a single day. Hence, if the colo had the same prices as Netlify for traffic, then a single server with a 10GbE connection would be suf…
No Colo charges the same insane prices for Traffic as Cloud Providers.
Hetzner charges 1€/TB over 20TB for their 10GbE Servers