"At the edge" is going to mean different things on different clouds. Cloudflare has hundreds of points of presence. What's more edgy about fly.io that you don't get from Cloudflare, App Engine, or AWS Lambda?

Oh hello! (disclosure, I work on Fly.io).

We're more like Fargate than Lambda, but Fly apps scale and balance across regions. I wouldn't really call it "more edgy", it's just built specifically for running app servers close to users (vs emulating a traditional single region datacenter).

We actually started with a JS runtime like CloudFlare has, but it was wrong for our customers. They're better off just running Docker images.

There's a lot more in our HN launch thread: https://news.ycombinator.com/item?id=22616857

Sorry if this is somewhat unrelated to the main thread, but considering that someone working on Fly.io is here, I wanted to ask some questions that I couldn't find the answers to in your documentation.

1. What's the average cold start time when spinning up a new instance? For comparison, AWS Fargate often has cold starts of around 1 minute.

2. Is there a limit on the container size created by the Dockerfile?

3. Is there a limit on the number of concurrent instances a single user can run?

1. Cold start times are wildly variable. They can be <1s for optimized, fast booting docker images. We actually mask cold start time as much as possible, though, so you'll almost never see one.

2. There's no hard limit. We've had 8GB images running (that I know of). Cold starts can be pretty bad on those just because pulling down an 8gb filesystem is slow.

3. Also no hard limit. I'm curious what you're doing though! Feel free to email me: kurt at fly.io

I'm using Go, so the default no-config container is about 140MB. But it's possible to use a multi step build to bring that down to about 10MB (build it in a full container and copy into a very lightweight scratch container). That's in the range of Lambda or CloudFlare worker sizes, so startup speeds shouldn't be an issue.

Yeah, Fly isn't "edge" as defined in the US, but sort of is as defined in India.

CloudFlare, CloudFront etc tend to have one or two edge PoPs in India, and one in Singapore, and Fly has one (upcoming) in India, which is coincidentally in the city where I live. So for me, Fly and CloudFlare and CloudFront have servers a few blocks away.

Same case for Singapore — everyone has an edge there.

But of course for the rest of the world it depends. The full blown CDNs will try to put edges in every city, but Fly and AWS Outposts will likely be few per continent / one per timezone / country.

Yeah, I actually moved a project from Workers to Fly. It's more expensive to run but it's still very affordable and you solve all the limitations workers have.