I’ve mentioned earlier the work that our students did on migrating part of the elements of the Database MOOC lab VM to docker.
While docker seems quite cool, let’s face it, participants to the MOOCs aren’t all using Linux where docker can be available directly. Hence the need to use boot2docker, for instance on Windows.
Then we’re back quite close to the architecture of the Vagrang VM, which relies too on a VirtualBox VM to run a Linux machine (boot2docker does exactly that with a minimal Linux which runs docker).
If VirtualBox is to be kept around, then why not stick to Vagrant also, as it offers a docker provider. This docker provider for Vagrant helps configure basic parameters of docker containers in a Vagrantfile, and basically uses the vagrant up command instead of using docker build + docker run. If on Linux, it only triggers docker, and if not, then it’ll start boot2docker (or any other Linux box) in between.
This somehow offers a unified invocation command, which renders a bit more portable the documentation.
Now, there are some tricks when using this docker provider, in particular for debugging what’s happening inside the VM.
One nice feature is that you can debug on Linux what is to be executed on Windows, by explicitely requiring the start of the intermediary boot2docker VM even if it’s not really needed.
By using a custom secondary Vagrantfile for that VM, it is possible to tune some parameters of that VM (like its graphic memory to allow to start it with a GUI allowing to connect — another alternative is to “ssh -p 2222 docker@localhost” once you know that its password is ‘tcuser’).
I’ve committed an example of such a setup in the moocbdvm project’s Git, which duplicates the docker provisioning files that our students had already published in the dedicated GitHub repo.
Here’s an interesting reference post about Vagrant + docker and multiple containers, btw.