We put it behind Fastly. We could have used Cloudflare too, but we were racing against time (and angry customers!). We set the cache timeout to 1 hour, and that's pretty much it.
Nice! Why did you choose to implement it as a proxy instead of using it as a primary source and just fetch from S3 on-the-fly while caching the generated image? If capacity is the problem then making it HA should solve that issue?
That's actually what we did. From pilbox's docs:
> It is not intended to be the primary source of images, but instead acts as a proxy which requests images and resizes them as desired.
We fetched from S3 on-the-fly while caching the generated image using a CDN we already had at hand. This is a band-aid for the night (we're in India, it's 2 AM here) and not for handling daytime traffic. We're obviously hoping imgix will fix their issues in the next several hours. Even if not, at least our customers get a degraded experience than a broken one.