blobcode - on cobalt

So I’ve spent the last little while writing a new project, cobalt

(https://github.com/blobcode/cobalt)

. It’s a minimal reverse proxy that handles http - take a look if it sounds interesting. It’s the descendant of the earlier pine

(https://github.com/blobcode/pine)

, which was a more naive implementation. I figured that talking about the design and implementation would be a good exercise.

the problem space

I wanted a simple reverse proxy that was reasonably performant without bloat. I took a look at

nginx
caddy
haproxy

All of which, with the exception of haproxy, are not first and foremost, proxies. Although the features they offer are great, I prefer having dedicated, configurable solution for everything. Thus, I decided to make cobalt (I was also pretty bored).

design

cobalt is a really simple, based around the tokio async runtime, which runs a tcp server. cobalt expects raw http input, with a host header present. It takes in a single argument, [-c], which is the path to a toml file that defines the config.

cobalt will serialize the hosts segment of the config into a hashmap, which is used by the server.

The server is pretty minimal. It reads a tcpstream, looks for a http host header in the first 1024 bytes, and splits then proxies the stream. All of this is done through the aforementioned tokio, which has been great.

I can get around 100,000 requests a second through it running on an i5-1135, and I plan to test it on a proper server sometime, given it can effectively max out all of my cores and is heavily cpu-bound.

Overall, I’m pretty happy with cobalt, which is a large improvement over pine. If you’d like to see any new features / find any bugs, feel free to submit a pull request or raise an issue.

Thanks!