# Multi-host Docker Networking

Posted on May 30, 2016

A while ago I grew the desire to experiment implementing multi-host docker networking to deploy Capital Match system. This system is made of several interlinked containers and docker-compose does not (did not?) work across several hosts. It seemed to me the official solution based on docker-machine, swarm and service registry was a bit complicated: Our configuration is mostly static, e.g. number, distribution and relationship between containers in known at deploy time. Hence I looked for a simpler solution, something that would be more networky: I am indebted to hashar for suggesting a GRE-based solution and to the following references for actual technical details:

I did some experiment in shell, jotted down a couple of notes in my journal and moved on to other, more urgent duties. I had a couple of hours left on Friday last week and I stumbled on those notes which were sitting there, on my hard disk, and I decided it was a good time to write a blog post about this experiment.

I started writing this post embedding script fragments but I quickly wanted to check what I wrote actually worked, so I began running those scripts fragment. But then it made this experiment non repeatable which is definitely annoying if you make a mistake, want to restart from scratch, change some parameters… So I decided this stuff would warrant a minor project of its own where I could provide all the needed code to configure multi-host networking in docker based on GRE tunnels. I have done quite a share of system configuration and operations management and have been able to use or create some useful tools to streamline ops in Haskell, so it quickly became obvious I would need to write some Haskell code. So what started as a mundane journal cleanup ended up being a full-blown yak-shaving session whose result can be found in this github repository.