Check out the new Buildah project logo. Isn’t it sweet?
I’ve been working with Linux containers at Red Hat since OpenShift used cartridges. With the help of some colleagues in Redwood City, I built a Red Hat MRG Messaging (Qpid) broker cartridge. When OpenShift moved to Docker, in 2013, I contributed all the original Docker man pages. I also got to contribute some of the original content to the Project Atomic web site (this site).
I’m very excited to see the Project Atomic efforts in the area of the Open Container Initiative (OCI) and specifically Buildah. After trying out the great work that the team have contributed, I discovered I had the start of a pretty useful introductory tutorial, and decided to see if I could contribute that content upstream. That pull request was merged recently. So let’s get started.
The purpose of this tutorial is to demonstrate how Buildah can be used to build container images compliant with the Open Container Initiative (OCI) image specification. Images can be built from existing images, from scratch, and using Dockerfiles. OCI images built using the Buildah command line tool (CLI) and the underlying OCI based technologies (e.g. containers/image and containers/storage) are portable, and can therefore run in a Docker environment.
In brief: the
containers/image project provides mechanisms to copy, push, pull, inspect and sign container images. The
containers/storage project provides mechanisms for storing filesystem layers, container images, and containers. Buildah is a CLI that takes advantage of these underlying projects and therefore allows you to build, move, and manage container images and containers.
The first step is to install Buildah:
dnf -y install buildah
After installing Buildah we can see there are no images installed. The
buildah images command will list all the images:
We can also see that there are also no containers by running:
When you build a working container from an existing image, Buildah defaults to appending ‘-working-container’ to the image’s name to construct a name for the container. The Buildah CLI conveniently returns the name of the new container. You can take advantage of this by assigning the returned value to a shell variable using standard shell assignment :
container=$(buildah from fedora)
It is not required to assign a shell variable. Running
buildah from fedora is sufficient. It just helps simplify commands later. To see the name of the container that we stored in the shell variable:
What can we do with this new container? Let’s try running bash:
buildah run $container bash
Notice we get a new shell prompt because we are running a bash shell inside of the container. It should be noted that
buildah run is primarily intended for helping debug during the build process. A runtime like runc or a container interface like CRI-O is more suited for starting containers in production.
Be sure to
exit out of the container and let’s try running something else:
buildah run $container java
Oops. Java is not installed. A message containing something like the following was returned.
container_linux.go:274: starting container process caused "exec: \"java\": executable file not found in $PATH"
Lets try installing it using:
buildah run $container -- dnf -y install java
-- syntax basically tells Buildah: there are no more
buildah run command options after this point. The options after this point are for inside the containers shell. This syntax is required if the command we specify includes command line options which are not meant for Buildah.
buildah run $container java will show that Java has been installed. It will return the standard Java
One of the advantages of using
buildah to build OCI compliant container images is that you can easily build a container image from scratch, and therefore exclude unnecessary packages from your image. For example, most final container images for production probably don’t need a package manager like
Let’s build a container from scratch. The special image name
scratch tells Buildah to create an empty container. The container has a small amount of metadata about the container but no real Linux content.
newcontainer=$(buildah from scratch)
You can see this new empty container by running:
You should see output similar to the following:
CONTAINER ID BUILDER IMAGE ID IMAGE NAME CONTAINER NAME 82af3b9a9488 * 3d85fcda5754 docker.io/library/fedora:latest fedora-working-container ac8fa6be0f0a * scratch working-container
Its container name is working-container by default and it’s stored in the
$newcontainer variable. Notice the image name (IMAGE NAME) is “scratch”. This just indicates that there is no real image yet: it is containers/storage but there is no representation in containers/image. So when we run:
We don’t see the image listed. There is no corresponding scratch image. It is an empty container.
So does this container actually do anything? Let’s see.
buildah run $newcontainer bash
Nope. This really is empty. The package installer
dnf is not even inside this container. It’s essentially an empty layer on top of the kernel. So what can be done with that? Thankfully there is a
buildah mount command.
scratchmnt=$(buildah mount $newcontainer)
$scratchmnt we can see the path for the overlay image, which gives you a link directly to the root file system of the container.
# echo $scratchmnt /var/lib/containers/storage/overlay/b78d0e11957d15b5d1fe776293bd40a36c28825fb6cf76f407b4d0a95b2a200d/diff
Notice that the overlay image is under
/var/lib/containers/storage as one would expect. (See above on
containers/storage or for more information see containers/storage.)
Now that we have a new empty container we can install or remove software packages or simply copy content into that container. So let’s install
coreutils so that we can run bash scripts. This could easily be
nginx or other packages needed for your container.
dnf install --installroot $scratchmnt --release 26 bash coreutils --setopt install_weak_deps=false -y
Let’s try it out (showing the prompt in this example to demonstrate the difference):
# buildah run $newcontainer bash bash-4.4# cd /usr/bin bash-4.4# ls bash-4.4# exit
Notice we have a
/usr/bin directory in the newcontainer’s image layer. Let’s first copy a simple file from our host into the container. Create a file called runecho.sh which contains the following:
#!/bin/bash for i in `seq 0 9`; do echo "This is a new container from ipbabble [" $i "]" done
Change the permissions on the file so that it can be run:
chmod +x runecho.sh
buildah files can be copied into the new image and we can also configure the image to run commands. Let’s copy this new command into the container’s
/usr/bin directory and configure the container to run the command when the container is run:
buildah copy $newcontainer ./runecho.sh /usr/bin buildah config --cmd /usr/bin/runecho.sh $newcontainer
Now run the container:
# buildah run $newcontainer This is a new container from ipbabble [ 0 ] This is a new container from ipbabble [ 1 ] This is a new container from ipbabble [ 2 ] This is a new container from ipbabble [ 3 ] This is a new container from ipbabble [ 4 ] This is a new container from ipbabble [ 5 ] This is a new container from ipbabble [ 6 ] This is a new container from ipbabble [ 7 ] This is a new container from ipbabble [ 8 ] This is a new container from ipbabble [ 9 ]
It works! Congratulations, you have built a new OCI container from scratch that uses bash scripting. Let’s add some more configuration information.
buildah config --created-by "ipbabble" $newcontainer buildah config --author "wgh at redhat.com @ipbabble" --label name=fedora26-bashecho $newcontainer
We can inspect the container’s metadata using the
buildah inspect $newcontainer
We should probably unmount and commit the image:
buildah unmount $newcontainer buildah commit $newcontainer fedora-bashecho buildah images
And you can see there is a new image called
fedora-bashecho:latest. You can inspect the new image using:
buildah inspect --type=image fedora-bashecho
Later when you want to create a new container or containers from this image, you simply need need to do
buildah from fedora-bashecho. This will create a new container based on this image for you.
Now that you have the new image you can remove the scratch container called working-container:
buildah rm $newcontainer
buildah rm working-container
Let’s test if this new OCI image is really portable to another OCI technology like Docker. First you should install Docker and start it. Notice that Docker requires a daemon process (that’s quite big) in order to run any client commands. Buildah has no daemon requirement.
dnf -y install docker systemctl start docker
Let’s copy that image from where containers/storage stores it to where the Docker daemon stores its images, so that we can run it using Docker. We can achieve this using
buildah push. This copies the image to Docker’s repository area which is located under
/var/lib/docker. Docker’s repository is managed by the Docker daemon. This needs to be explicitly stated by telling Buildah to push to the Docker repository protocol using
buildah push fedora-bashecho docker-daemon:fedora-bashecho:latest
Under the covers, the containers/image library calls into the containers/storage library to read the image’s contents, and sends them to the local Docker daemon. This can take a little while. And usually you won’t need to do this. If you’re using
buildah you are probably not using Docker. This is just for demo purposes. Let’s try it:
docker run fedora-bashecho This is a new container from ipbabble [ 0 ] This is a new container from ipbabble [ 1 ] This is a new container from ipbabble [ 2 ] This is a new container from ipbabble [ 3 ] This is a new container from ipbabble [ 4 ] This is a new container from ipbabble [ 5 ] This is a new container from ipbabble [ 6 ] This is a new container from ipbabble [ 7 ] This is a new container from ipbabble [ 8 ] This is a new container from ipbabble [ 9 ]
OCI container images built with
buildah are completely standard as expected. So now it might be time to run:
dnf -y remove docker
What if you have been using Docker for a while and have some existing Dockerfiles. Not a problem. Buildah can build images using a Dockerfile. The
bud for short, takes a Dockerfile as input and produces an OCI image.
Find one of your Dockerfiles or create a file called Dockerfile. Use the following example or some variation if you’d like:
# Base on the Fedora FROM fedora:latest MAINTAINER ipbabble email buildahboy @ notreal.oi # Update image and install httpd RUN echo "Updating all fedora packages"; dnf -y update; dnf -y clean all RUN echo "Installing httpd"; dnf -y install httpd # Expose the default httpd port 80 EXPOSE 80 # Run the httpd CMD ["/usr/sbin/httpd", "-DFOREGROUND"]
buildah bud with the name of the Dockerfile and the name to be given to the image (e.g. fedora-httpd):
buildah bud -f Dockerfile -t fedora-httpd
buildah bud defaults to Dockerfile (note the period at the end of the example):
buildah bud -t fedora-httpd .
You will see all the steps of the Dockerfile executing. Afterwards
buildah images will show you the new image. Now we need to create the container using
buildah from and test it with
httpcontainer=$(buildah from fedora-httpd) buildah run $httpcontainer
While that container is running, in another shell run:
You will see the standard Apache webpage.
Why not try and modify the Dockerfile? Such as: instead of installing httpd, ADD the runecho.sh file and have it run as the CMD.
Well done! You have learned a lot about Buildah using this short tutorial. Hopefully you followed along with the examples and found them to be sufficient. Make sure to look through the Buildah man pages to see what other commands you can use.
If you have any suggestions or issues please post them at the ProjectAtomic Buildah Issues page.
For more information on Buildah and how you might contribute please visit the Buildah home page on Github.