offloader

offloader is a lightweight proxy for offloading long-running requests from backends that have limited parallelism. Its behavior is configurable via response headers from the backend. It's designed to be used with Go's httputil.ReverseProxy as a ModifyResponse function.

For example, you may have a Ruby on Rails backend server that can handle a limited number of parallel requests, and the backend server occasionally needs to make long-running requests to another internal service. With offloader sitting in front of your backend server, the backend can defer the long-running work to offloader, freeing up the Rails web workers to handle another client request. In the meantime, offloader handles the long-running request and responds to the original client when the response is ready. Go's excellent support for parallelism means that you can offload a significant number of requests.

  client <--> offloader <--> backend server
                  \
                    <--> offload server 

offloader is inspired by NGINX's X-Accel-Redirect feature and aims to solve a more general version of the same problem.

Getting Started

import "github.com/wistia/offloader"

...

proxy := httputil.NewSingleHostReverseProxy(...)
proxy.ModifyResponse = offloader.Handler

In this example, proxy is an http.Handler and is typically used with an http.Server instance.

Controlling offload behavior

You can control offloader's behavior using response headers from your backend server.

Header	Possible values	Behavior
Offload-Requested	(Any)	If this header is present, then offload is enabled. Otherwise, the backend response is passed directly back to the client.
Offload-Url	(Any valid URL)	The protocol, host, and path for the offload request are determined from this URL.
Offload-Method	GET, HEAD, POST	The request to the offload server uses this HTTP method.
Offload-Forward-Body	(Any)	If this header is present, then the response body from the backend will be sent as the request body to the offload server. This is useful for making POST requests to the offload server.
Offload-X-<your-header-name>	(Any)	Headers in this format will be passed to the offload server as Your-Header-Name.