New to Fly.io, several questions :)

1

Hello there,

Please find questions about Fly services :

  • Most important question : can machines be balanced this way : 1 request = 1 machine, if no machine available, an error is thrown ?
  • Can endpoint be protected ? Our backend, that is not on fly.io, needs to call http endpoint to start machines, but we don’t want that random http request call our endpoint and start machines. I guess I could set up a proxy on fly.io to handle such auth mechanism, but frankly I would prefer avoid it…
  • How many machines/resources can we use ?
  • Is there a NodeJS/Typescript/Javascript client library ?

Thanks everyone !

2

Hi @Biscuit

  • I believe you might be able to to get 1 req = 1 machine if you set soft_limit and hard_limit to 1. I recall a customer who attends this forum does that but I couldn’t find that post.
  • Can you talk more about your use case of endpoints that start machines? I don’t think I fully understand what needs protection
  • We have a soft limit of how many machines a new org can have (set per organization) but if you find that you need to spawn several machines feel free to reach out to us.
  • At the moment we don’t offer official SDKs in any language. In Golang though we ship some needed stuff in our CLI.
3

Thanks for this quick reply. I found this : Concurrent Tasks on Machines · Laravel Bytes which I think explain how to achieve what we need.

Regarding securing endpoint, I would like that our backend calls for example https://myapp.fly.dev . When http request ended, then worker finished its task. But if any random web bot call this URL (as we often see in server logs), it will start a machine, and we will be billed accordingly…

For machines amount, for tests, it’s ok for now, but for production, we will use sometimes dozens of machines (1 proc with 1/2 GB each request).

Regarding SDK, I guess we don’t need it ATM thanks to http endpoint functionnality.

4

I see now!

To prevent random bots from being able to wake up your machines I’d say remove public IP addresses from that machine and work with a Flycast IP (make sure to have a services section!) since it can wake your machine on HTTP.

Its gonna be really easy if it’s a Fly machine calling a Flycast IP because our networking Just Works :tm: but if you need to call that machine from your Not-on-Flyio server you’d probably want to connect it to Fly using Wireguard which is easy and also Just Works :tm:.

Here’s an example of a customer who needed just that: Automatically start and stop internal services

5

Hi, I am building a product that’ll run user code in machines and I can’t predict how many I’d need. So what is this soft limit you talk about? how many?
Do I get email when I’m reaching 80% of this limit?

6

I just checked and the default is 20 running machines at the same time. I don’t think we email but we warn on Flyctl/Machines API. Feel free to reach out to billing[at]fly.io if you need more.

7

Many thanks. And do requests have a max timeout ?

8

Someone can correct me if I’m wrong but I believe our proxy terminates idle connections at 60 seconds. Unfortunately I couldn’t find the post here on our community forum talking about it!

9

This post talks about removing the timeout restrictions and about how connections are dropped if they get backlogged.

10

This seems to answer the question. I’ll do some tests :). Thanks.