It's about total cost of ownership

A: Imgix = cost of imgix subscription + integration time * hourly rate (it's dead simple)

B: Rolling your own = dev time * hourly rate + maintenance/ops time * hourly rate

For most orgs B > A. For most individual programmers since hourly rate is not a factor A < B.

That's certainly true. I'm not saying that this service doesn't have any value. Just wanted to share my experience with writing similar solution.

It's good to be dev and have spare time for stuff like that. You learn stuff while making such projects, you save money and it just works.

About money, judging by the pricing on imgix page (and if I understand pricing correctly, I'm saving around $300 per month ($50 for cheapest plan + $250 for traffic).

If you have to pay for hardware to run it on, that would be part of B.

What was the max throughput your server could handle? We ended up using Go because we serve 400 - 600 reqs/sec and Python/Ruby solutions didn't have high throughput.

I've never tested its throughput - there was no reason to do that because all of the processed images are saved and stored as long as those files are accessed at least once a week.

So, the answer is pretty much this: it'll handle as many rps as nginx serving static files can handle.

If we are talking about unique requests each requesting to process unique image - I'm pretty sure linode server CPU it's running on will choke.