Serve cached images upon request timeout

[espadrine]

Suggested here by @seanlinsley.

If a request to a third-party API takes longer than (say) 10 seconds, we might want to serve the last known good image generated.

[seanlinsley]

@espadrine I'm going to take a stab at this.

[espadrine]

@seanlinsley If you want. Look at the cache() function at /server.js:60. I'd like to keep memory bounded, so we should delete cache data when we feel it is unnecessary. We can cache it for a day, for instance (see cacheTimeout, but the check for whether to use the cache needs to change).

[patcon]

Would be great to add a little inconspicious marker, the idea being that it wouldn't register with everyone, but those in the know would understand that it's being cached and not to take it too seriously. I can just imagine people wasting time troubleshooting things otherwise (maybe bugging travis support) or simply swapping back to the official badges if they have no way of knowing when the data isn't fresh.

[espadrine]

I agree that would be preferable, although I am not sure how to indicate that.

[patcon]

Maybe a slight desaturation based on staleness? Project maintainers could easily notice that.

But I think the main issue is that I don't think github gets a cache-control header for a failed image, so it cached the dead img link (suggested in the other thread).

[espadrine]

We now provide the last known good image when the vendor doesn't respond, and we also cache badges locally based on vendor requests.