Sinatra (and probably other Rack apps) do not properly generate links

[halfbyte]

I'll create a simple test case, but for now, here's the info I gathered:

When nginx proxies the request to the rack app, the sub folder path information gets lost and if you use to/url to generate proper url's you'll always end up with urls that are constructed from the root.

Let's say I created an app "example", and it runs at foo.local/example. Every url constructed with url() or to() in sinatra ends up being wrongly constructed, for example url('/foo') returns /foo instead of /example/foo.

I have not completely understood the underlying problem, as sinatra claims to support reverse proxying and sub folders. One issue seems to be that SCRIPT_NAME is empty, maybe adding that to the proxy headers could help.

As sinatra is a good fit for quick hacks, I consider this a problem :)

I'll update this issue as I gather more information.

[halfbyte]

I have a workaround and a possible fix:

The workaround is to add "rack_injection" to the Gemfile and add this to the config.ru file of your sinatra app (replace $APP with the name of your app, of course):

require 'rack_injection' use RackInjection, 'SCRIPT_NAME' => '$APP'

The fix for experimental platform would be to set SCRIPT_NAME as a header in the nginx config. I hope.

[halfbyte]

Okay, next problem with url(): if running via the protonet proxy (foobar.protonet.info) the X_FORWARDED_PROTO header is not correctly set (I assume it gets lost in the go request router?) which generates http links which makes the protonet proxy redirect to https.

And here I was wondering how somehow magically POST requests were turned into GET requests :)

[tiff]

Thanks @halfbyte we will look into this as soon as possible. The URL issue is something that's very high on our todo list :) The X_FORWARDED_PROTO fix seems like a low hanging fruit that might be fixed even earlier. I'll keep you posted.

[tiff]

@halfbyte https://github.com/experimental-platform/platform-dokku/pull/13 this should fix the X_FORWARDED_PROTO issue. Would this also be the place to put the SCRIPT_NAME or shouldn't this rather be an environment variable?

[tiff]

I would appreciate your feedback @halfbyte :smiley:

[halfbyte]

@tiff I think the nginx config would be the correct place to set the SCRIPT_NAME, yes. Basically, the SCRIPT_NAME must match the path you mount the sinatra app at, so, if the app name is jinglebells, the SCRIPT_NAME should be jinglebells as well, if the app is then to be found at mybox.protonet.info/jinglebells.

I'm not entirely sure it's the correct fix for my problem, but it will be worth a try.

[kdomanski]

@halfbyte can I get some sample code to reproduce the issue?

[tiff]

A lot of popular frameworks only run when they are in the root path (domain.com/).

In the future we will make apps available as multi level subdomains (app-name.box-name.protonet.info). The only challenge we have is that we would need valid dynamically created ssl certificates for that. Let's Encrypt makes it possible. We received our beta access last week and are already digging into it.