Data only Docker containers

In this post, we shall take a look at the idea of data only containers - containers whose sole purpose is to exist on the docker host so that other containers can have portable access to a persistent data volume.

Why do we need a persistent data volume?

We will experiment with the jenkins image from the docker hub. Let's run a jenkins container using $ sudo docker run -p 8080:8080 jenkins. Now, if we visit the URL http://docker-host-ip:8080, we will see the familiar Jenkins home page.

By default, a Jenkins installation doesn't come with any authentication configured. Hence, we will first setup a simple authentication setup using Jenkins' own user password database. To do so, we will visit the URL: http://docker-host-ip:8080/configureSecurity/ and check the Enable security checkbox and then select the Jenkins own user database` option, and check the Allow users to sign up check box under Security Realm, select the Logged-in users can do anything option and finally click on Save. This will bring us to the login page from where we can create a new account since we don't have one yet. Now, we will exit out of the container - we can use Ctrl + c combination for that. Now, if we restart the container using the previous command, you will see that none of the configuration changes above has been saved.

The reason for that is because none of the changes we make during a container's lifetime in it's own file system is preserved. So, we need data volumes.

Persistent data with a volume mount

If you look at the Dockerfile you will see the command VOLUME /var/jenkins_home. This essentially means that the mount point /var/jenkins_home points to a location on the docker host. Hence, the changes made in that directory will be available from the host even after you have exited the container. However, the catch here is that every time you run a new container, the host location it mounts to will change and hence as we saw above, the data we wrote (via the configuration changes) were not visible the next time we started a container from the same image. To achieve that, we have to do things slightly differently. We will start the container and give a name to it:

` $ sudo docker run -p 8080:8080 --name jenkins jenkins `

We will perform the same configuration changes above and exit the container using Ctrl + C. Next, we will start the container using sudo docker start jenkins. You will see all your changes have been preserved.

So, now we have a setup of jenkins where our changes are preserved, so long as we make sure we start/stop the containers and not run a new container from the jenkins image. The key point to take away from here is that for a specific container, the host directory the volume maps to is always the same. This leads to the use of what is commonly referred to as data containers for persistent data storage in containers.

Using data containers for persistent storage

The idea here is that you use the same base image from which your actual container will run to only create a container (using docker create), not run it:

$ sudo docker create --name jenkins-data jenkins

We gave the name jenkins-data to this container and it's only purpose is to be there on our filesystem to serve as a source of persistent /var/jenkins_home for other jenkins containers. Let's run a jenkins container now:

$ sudo docker run --volumes-from jenkins-data -p 8080:8080 jenkins

As earlier, we can now go to the Jenkins home page at http://docker-host-ip:8080/configureSecurity/ and make the above configuration changes. You can now exit the container and use the above command to run another jenkins container. The changes will still be visible. We are no more restricted to starting and stopping the same container since our jenkins-data container will have all our changes stored in its /var/jenkins_home. You can have other containers (perhaps a container for backing up your /var/jenkins_home) being able to access the same data by using the same volumes-from option.

Comparison to volume mounting a host directory

The alternative to using data containers is to mount a directory from the host as /var/jenkins_home in the container. This approach will solve our end-goal, but there are two additional steps that one would need to do:

  • Decide which host location to use, perhaps creating it
  • Making sure the container will have appropriate read-write permissions (including SELinux labels).

We don't need to do either of these when using data containers. As long as the image we plan to use has the appropriate VOLUME command in it's Dockerfile, we can adopt the same approach we did here to make sure the data we care about is persisted. For images, which don't, we can easily enough create our own image and add the appropriate VOLUME commands. And hence, this is a portable approach to data persistence - it is not reliant on the host setup.

Conclusion

In conclusion, these are the main reasons why data containers are a good approach to have persistent storage for your containers:

  • No requirement to setup host
  • The permissions are automatically taken care of since we are using the same base image
  • Multiple containers can easily have access to the same data

The following links may be helpful to learn more: