This repo analyzes following problematic configuration, and the needed fixes to make it actually work:
- One frontend gateway app that takes paths under /gateway/my_app and proxies to a rails app. This app is running on port 9292
- The path is changed when proxying, so rails never gets the original prefix.
- Said rails app, running on port 5100, that has one static page showcasing
some urls, and mounts sidekiq under
/sidekiq. - The same rails app, running on port 5555, is started naked, without any proxy configuration.
- Redis should be running (you can use the provided
docker-composefile).
To run this configuration, uncomment the env['PATH_INFO'] = line of the
gateway app (gateway.ru) and remove the block that handles app prefix at the
end of rails' own config.ru.
The app should be accessible at http://localhost:9292/gateway/my-app. One can see that url generation and sidekiq UI (at http://localhost:9292/gateway/my-app/sidekiq) are broken.
On the other side, the naked application at http://localhost:5555/ works.
Sidekiq assets could be fixed by patching the Sidekq::WebHelpers module, see
config/initializers/sidekiq.rb. But still urls in the rails app won't work.
The problem is due to the fact that rails does not really know what was the original path, so it has no way to build the correct links.
The problem is solved by not modifying the path at all in the proxy.
Rails should get the RAILS_RELATIVE_URL_ROOT variable and itself route
starting by that point.
With this configuration, both the application behind the proxy, with env set, and naked, without env set work.
Noticeable files:
- config.ru: Main app rackup file, where rails responds under an optionally configured path
- gateway.ru: Gateway rackup file, proxying to rails without changing the path
- Procfile: Where the app is run with an appropriate environment variable so it responds to namespaced routes