A Vagrantfile to provisions a boot2docker box, configure shared folders with
translated permissions, install docker-compose, and run docker-compose to
start containers.
The intent is to provide a Vagrantfile which can be dropped into existing projects, minimally edited to configure networking and synced folders.
The Vagrantfile is the only notable object in this project; the other files are just a sort of sample application as a demonstration.
Please note that vagrant 1.7.3 is needed to use VirtualBox 5.0.
Just clone this repository and run:
vagrant up
You can then connect to the containerized webserver at
http://192.168.70.249 You can edit data/index.html and
see the changes reflected after a couple reloads of the page.
The Vagrantfile syncs data/ in the working directory to /data in the
vagrant box, bidirectionally converting the ownership and group permissions of
files in the local directory to 33.33 in the vagrant box.
The working directory with untranslated permissions is also synced to
/vagrant in the box.
Then, docker-compose is run within the Vagrantfile against the sample
docker-compose.yml to spin up two containers: the first a volume container
mapping /data in the vagrant box to /usr/local/apache2/htdocs/ in the
container; the second, a webserver container, grabbing its volumes from the
first.
Since the files ownership is translated in the vagrant box, when the containers
map their volumes to the vagrant box, they see those translated ownerships.
In the webserver container, 33 is the UID and GID of www-data, so the net
effect is that data files can be edited no matter what the platform or
filesystem Vagrant is running on, and the container will still run correctly.
Docker provides excellent encapsulation to unify development and operations; however, especially for web development, it is still sometimes useful to be able edit files "live", and see the changes take effect immediately. This can be done easily with Docker using volumes mapped to local folders.
However, two issues arise: first, docker only runs natively on Linux, and second, the file permissions (ownership and mode) of the local folder may need to be quite different from those mapped in a volume into the container.
To allow diverse development environments, one can use a Linux virtual machine to run docker, and for this there is the boot2docker image running in VirtualBox.
The boot2docker project includes both the ISO for the virtual machine, and a number of installers for different platforms, which hook into VirtualBox to configure the virtual machine. Unfortunately, (at least at the time this Vagrantfile was written), boot2docker does install easily on Windows, is deemed unnecessary on Linux, does not not have a consistent method to fix the permissions problem, and does not have docker-compose installed.
Now, as VirtualBox does have the capability to share folders and translate their permissions, it just becomes a matter of configuring it to to do so, which Vagrant does admirable. Vagrant can also configure the virtual machines it creates, so features that boot2docker lacks may be added.
The result is a solution which will work correctly and consistently on all platforms supported by both VirtualBox and Vagrant, and is also relatively lightweight.
Docker-compose is installed and run on every vagrant reload. If you don't want
to use docker-compose, just change the Vagrantfile so that it doesn't run.
If you have the docker command-line client installed locally, it may be easier to
just connect directly to the docker daemon running in the virtual machine, than
to use vagrant ssh. After the first vagrant up, a some suggested export
lines will be printed which you can just paste into your shell to set up the ENV
variables necessary to connect. Try them until one works.
The docker daemon running in the virtual machine will save images you build or
pull to persistent storage, but if you consistently delete and rebuild the
virtual machine with vagrant destroy and vagrant up, it may speed things to
define DOCKER_IMAGEDIR_PATH in your shell's environment. If defined as a path
to a folder on the box, the Vagrantfile will use that folder as an image cache,
loading tar archives from that folder to the daemon's image store on boot (with
docker load), and saving running images as tar archives to the folder after
docker-compose is run (with docker save). This may speed up initial boot times
by avoiding the need to keep pulling an image from the network.
Windows line endings may be an issue; VirtualBox can synchronize folders and translate permissions, but the data in the files still needs to be correct.
Boot2docker is designed as an essentially read-only image (albeit with some
persistent storage for docker and ssh keys), so to allow Vagrant to add a key to
allow vagrant ssh to work correctly, the Vagrantfile does a number of bind
mounts in on start up.
Python and pip are installed to get docker-compose. Boot2docker is based on Tinycore Linux, so the tce packages are downloaded and installed on the first boot, and reinstalled from persistent storage thereafter. Pip is updated on every boot to get the latest version of docker-compose.