Live data from Hacker News

Show HN: An API to provision custom hostnames with SSL

news.ycombinator.com

1–10 of 38 posts

Show HN: An API to provision custom hostnames with SSL

#1
We launched an API for custom hostnames today. A simple HTTP post with a hostname gets a temp hostname with CNAME/a record instructions, as soon as DNS is setup we issue SSL and start accepting traffic for that hostname. It's the quickest way we know of to provision custom hostnames: https://fly.io/mix/custom-hostnames/

When we started building Fly, we figured it was too late to "make SSL easy and the customers will come". Lets Encrypt changed the world, and we knew many companies rolling their own SSL support with nginx, Traefik, etc, etc. It seemed like a solved problem.

It turns out that SSL is still a pain, particularly companies with lots of hostnames pointed at them ... anyone hosting apps/content on behalf of their own customers. Distributing certificates, keeping them renewed, and — most of all‚ making https fast is still hard. Devs can solve the basic problem in a week or so with a proxy, but then it sits, and no one feels very comfy with untouched infrastructure. And it's usually something distracting devs from more important work that's core to their own customers.

So this is our way of solving that problem. We can handle any number of hostnames for applications, devs can spend more timeon what makes their apps special. It's a relatively "simple" use of what we've built, but solves a really fundamental problem for a many companies. It's the most fun thing we've discovered this year.

So, if you're building an app, and serve stuff on behalf of your users, we can make your life easier:

https://fly.io/mix/custom-hostnames/

Re: Show HN: An API to provision custom hostnames with SSL

#3
post #2

I'm not sure how this is easier than a wildcard certificate and a wildcard CNAME...

It's not, it's meant for cases where people need entirely different custom hostnames. Subdomains are quick and easy, but people have become quite a bit more demanding about how their stuff is branded, and tend to want it all on mydomain.com (instead of me.service.com).

Actually I lied, it's probably a little easier than a wildcard cert. :D But not enough to matter. People who just need wildcard subdomains might want to use us for other reasons. We handle all load balancing, global SSL termination, etc. These things are stuff you'd use even if you were just using a wildcard cert.

Re: Show HN: An API to provision custom hostnames with SSL

#4
How do you work around the rate limits imposed by Let's Encrypt?

https://letsencrypt.org/docs/rate-limits/

If you're getting certificates issued for subdomains of your domain, you're limited to 20 certificates per week of up to 100 SANs per certificate.

Are you just not hitting this limit yet, or do you load balance certificate requests across a pool of domains?

Re: Show HN: An API to provision custom hostnames with SSL

#5
post #4

How do you work around the rate limits imposed by Let's Encrypt? https://letsencrypt.org/docs/rate-limits/ If you're getting certificates issued for subdomains of your domain, you're limited to 20 certificates per week of up to 100 SANs per certificate. Are you just not hitting this limit yet, or do you load balance certificate requests across a pool of domains?

As I understand it, this is for domains provided by their users. The rate limits would not be problematic unless the domain uses Let's Encrypt for a lot of other subdomains.

I'm guessing they're using a wildcard certificate for the temporary hostname.

Re: Show HN: An API to provision custom hostnames with SSL

#6
Timely. I launched a SaaS a few months ago and have found early customers have expected custom domain names with SSL. It seems that not offering this is a deal breaker these days.

I didn't even consider that third parties such as yourself might offer a solution.

Edit: I tried to sign up. I have to already choose lots of detailed config options? No option for AWS elastic beanstalk or EC? Asking for my AWS secret key without explaining why?

Re: Show HN: An API to provision custom hostnames with SSL

#7
post #4

How do you work around the rate limits imposed by Let's Encrypt? https://letsencrypt.org/docs/rate-limits/ If you're getting certificates issued for subdomains of your domain, you're limited to 20 certificates per week of up to 100 SANs per certificate. Are you just not hitting this limit yet, or do you load balance certificate requests across a pool of domains?

pfg is correct, the rate limits don't really come into play when you're issuing certificates with a lot of different root domains. We'll get the rate limits raised if we end up issuing a lot of subdomains, but for now we recommend people bring their own wildcard certs, and as soon as LE starts issuing wildcards we'll just use those.

Re: Show HN: An API to provision custom hostnames with SSL

#8
I am trying to understand what you offer, compared to AWS. Does it kind of substitute the combination of API Gateway + CloudFront + Certificate Manager ?

I have some Lambda funcs running and just today starting looking into this AWS combination to access Lambda from the outside world. So I wonder if Fly would do the same.

If its the same then it certainly is simpler to set up. And might be cheaper as well, but the pricing of AWS is unclear as always (meaning always having some unexpected aspects).

Re: Show HN: An API to provision custom hostnames with SSL

#10
post #8

I am trying to understand what you offer, compared to AWS. Does it kind of substitute the combination of API Gateway + CloudFront + Certificate Manager ? I have some Lambda funcs running and just today starting looking into this AWS combination to access Lambda from the outside world. So I wonder if Fly would do the same. If its the same then it certainly is simpler to set up. And might be cheaper as well, but the pr…

Fly is a good alternative to API Gateway + Cloudfront + Certificate Manager. We can invoke lambda functions directly.

And yes, AWS pricing is a monstrous headache. I'm really not sure if we're cheaper at scale, but we're easy to predict at the very least (and the first 2 million requests are free).

Post reply on HN