The default way of creating an image with Docker is to extend an image by
using the FROM call in the Dockerfile. But this is not the only way to do
this. You can easily create a base image by using only the yum command.
WildFly 8.0.0.Final
was recently released
and is now available in Fedora 20 updates-testing repository. Let’s
create an image that could be used to test this release but make it as minimal
as it could be. The application server does have a lot of dependencies, so do
not expect a 100 MB image. It’ll be closer to 1 GB.
Intro
One difference between virtual machines and Docker images is that a virtual machine need to contain its own kernel to be able to run. This is not the case in Docker since we reuse the host kernel (which is already running).
The following analogy may help:
-
Virtual machines — whole PC with graphic card, CPU, disk, controllers,
-
Docker images — only the disk.
You can move around and launch either, but virtual machines are way heavier.
This means that we can create an image with only the required bits and drop everything else — making it easier to maintain, lighter and of course more secure. An example of such an image is busy box. The current virtual image size is ~2.5 MB. Less than nothing.
To create such a small image we need to use an external tool and not
extend a base image. As I mentioned before in our case we will use the yum
command.
Building the image
I prepared and ran the following script:
#!/bin/bash if [ $UID -ne 0 ]; then echo "You need to run this script as root" exit 1 fi NAME=wildfly-minimal BUILDDIR=`pwd`/build # Removing the earlier build rm -rf $BUILDDIR # Install the required stuff yum -y install wildfly \ --setopt=override_install_langs=en \ --setopt=tsflags=nodocs \ --installroot $BUILDDIR \ --disablerepo=* \ --enablerepo=fedora,updates,updates-testing \ --releasever=20 \ --nogpgcheck # Clean up the cache # and fix the console issue when running the image chroot $BUILDDIR /bin/bash -x <<EOF rm -rf /var/cache/yum/* rm -rf /dev/console ln -s /dev/tty1 /dev/console EOF # Import to Docker tar -C $BUILDDIR -c . | docker import - $NAME
That’s all. Now you can run your image. The image ID was printed after import. In my case it was
05c926b5dc99d5c4edc6eb6bfb7040d6b4bc67e1f00f3f9af4485dffefe21a66. Time to run it:
docker run -i -t 05c926b5 /usr/share/wildfly/bin/standalone.sh -b 0.0.0.0
Now that we have a running WildFly server, let’s point the browser to it. But what’s the address? It’s easy to get this information too!
First execute the docker ps command to get the container ID:
$ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES af59785a5245 wildfly-minimal:latest /usr/share/wildfly/b About a minute ago Up About a minute tender_euclid
Now we can use the docker inspect command to get the IP address:
$ docker inspect -f '{{ .NetworkSettings.IPAddress }}' af59785a5245
172.17.0.2
Great, WildFly is available on http://172.17.0.2:8080.
Update 07.06.2014
I changed a bit the script. Now it installs only the en language files and there
was a mistake in cleaning the YUM cache which is now fixed too. Additionally I
fixed the /dev/console warning messages that were appearing on boot.
Everything above made it possible to decrease the size of the image from 1.6GB to 1GB. Win!