Vagrant already does this nicely, and works with more than just Docker. For those targeting production outside of Docker my company has made these images available: https://github.com/BashtonLtd/docker-vagrant-images
If you're specifically looking for a convenient way to use Docker for development, don't use Vagrant. It tries too hard to abstract Docker away, so it's not a good choice if you care about compatibility with the rest of the ecosystem, keeping up with future improvements to Docker, etc.
If you don't care about benefitting from Docker specifically, then it's worth evaluating Vagrant for environment devs, it's a good tool.
I quite like Vagrant myself but my primary complaint is that I seem to spend a lot of time waiting for VMs to start up and run a lot of bootstrap code.
Re your complaint: with the Docker provider, Vagrant only spins up _one_ VM (if you're not on Linux). After that, `vagrant up` just executes `docker` directly so there is no waiting around for bootstrapping. It even detects if you have Docker installed locally or not to decide whether it needs to run in the VM or not. And, if you're on Linux, it doesn't use a VM at all, unless you're on a kernel that doesn't support Docker, in which case it automatically/transparently starts the VM.
EDIT: Responding to David's comment below since it would go to deep: Vagrant is basically the same here. It doesn't require Dockerfiles, you can just spin up images.
Awesome :) I think then that the primary difference with Tug is that it can work in a "hybrid" virtualized/local mode if you haven't yet written a Dockerfile/Vagrantfile/other for your app. Tug can start by simply spawning dependencies like Postgres and Redis and let you incrementally move towards a Dockerfile for your app.
Thank you for the kind words and the clarifications Mitchell :)
Tug is a simple tool that I wrote to scratch a personal itch around working with dockerized applications and trying to optimize for startup speed and writing as absolutely little configuration as possible.
I'm hoping to get a few other people to try it out and let me know if it works for them and if it's useful, especially given the existence of so many other tools of this nature.
On Linux, if you're using Docker, then it runs it natively without a VM.
VirtualBox is not "preferred" as much as its the default, because its the easiest to acquire. For example, on Windows 8+, Hyper-V is actually the default for Vagrant because its built in to almost every installation of Windows (Vagrant checks).
Vagrant really just chooses whatever appears to be available on your system.
Vagrant is so slow on my MacBook that I never use it. We always set up Vagrant in our new projects at work but I just run stuff on my host machine because I can't stand waiting for the initial Vagrant setup, processes to start, tests to run, servers to respond, etc.
Hello everyone! We created tug as a tool to help people dockerize applications and run them during development.
Please keep in mind that this is a very alpha project still under heavy development. If you try it I'd love to hear how it goes and whether or not it's useful for you (you can reach me at david@nitrous.io or find me on Freenode as ddollar).
Seems like there's a new wave of projects like these lately.
Fig and Crane (https://github.com/michaelsauter/crane) are the most popular ones but still I see other small things out there.
I'm really excited about the idea of groups that may be coming to Docker. It seems like a great primitive around which to build tools like fig and tug and perhaps other tools to orchestrate production.
One of the features I heavily use from fig is controlling the runtime docker parameters in fig.yml, i.e the VOLUME, ENTRYPOINT, EXPOSE, WORKDIR sort of stuff. I see you guys made the decision to maintain the runtime commands in your Dockerfile, requiring a rebuild to change it. Everything except Environment.
I guess my question is, why did you guys choose to do that?
I wouldn't describe tug as less powerful, it's less complicated from my usage of it. For each service I only need to write one line and tug will figure out (from the Dockerfile) what needs to be forwarded.
The mind blowing moment is when you're using boot2docker (or any case where Docker isn't local) and it still forwards the ports to your local (127.0.0.1) host.
> The mind blowing moment is when you're using boot2docker (or any case where Docker isn't local) and it still forwards the ports to your local (127.0.0.1) host.
That's easily achieved with a Vagrantfile for your Docker VM...
I wouldn't say it's that easy, if you add more containers to your Docker VM and need them forwarded you then need to update your Vagrantfile and reload.
With tug any port you expose on the containers is brought locally to you instantly.
The problem I see with using docker for development is, you still need different rails environments(development, test, etc.), the dockerfiles I've been making are for production, and I'm not sure if they would work in the other environments. Also you have to use tons of commands (rake, rspec, migrate, etc) which at this point requires entering the container, which you really shouldn't have to do unless there is a problem with the container itself.
One thing to note is that you can have multiple fig.yml files. So, you could set slightly different configuration parameters (perhaps use `environment` etc.) for different places you are using fig using the `-f` option to specify a different fig.yml.
I don't think entering a container is bad for development, in fact it's a frequent pattern (to run things like you describe rake etc.). With the release of docker exec this becomes easier too. I think of it kind of like bind-mounting a directory in the container to do code editing without having to rebuild- as long as you are using ADD for every stage afterwards that should be fully automated (QA, test, production) then you are in the clear.
How does it know that web is the one to built with linking from other commands? Also if there is no Dockerfile the web in the README is not running in docker, how is the env being filled by the linking process? Is it hardcoded?
