Creating Tests For This Website: Docker Jenkins Nodes

Okay, quick post! Previously, I wrote about how I automated my website tests using Jenkins. When I wrote that post, I had the tests run on any node. I wanted to have the tests run inside a fedora docker container, but ran into issues configuring it. With the problem now long fixed, I decided I would write a quick update post about switching the pipeline to use container nodes.

Why Switch

When I first defined the pipeline, I had it use any node for the agent:

pipeline {
    agent any

    stages {

        // This is where the stages will be defined //

    }
}

This just runs the job on any available node, which for me was just the same VM server I was running Jenkins on. This was fine, but for testing I want to make sure everything in my automation is configured and up to date. The best way to ensure that, is with clean runs.

When using docker for the pipeline agent, a new container is created to run the pipeline in, and then destroyed when completed. This means all packages and dependencies must be defined correctly, or the run will fail. This is what we want.

Using Docker Nodes

First, the obvious: make sure docker is installed on the desired Jenkins nodes. I won’t cover this as it can be different for every user (and I already had docker installed on my Jenkins host).

With docker installed, next make sure the Jenkins server has the Docker Slaves (and possibly Docker Pipeline) plug-in(s) installed.

Lastly, with some docker plug-ins enabled, switch the agent statement to use a container image. I choose to use the fedora:31 image:

pipeline {
    agent {
        docker {
            image 'fedora:31'
        }
    }
    ... // rest of the pipeline
}

Fixing root/sudo error

When I first set the pipeline to use the fedora image, it kept failing. Specifically, the sudo dnf steps would fail because the sudo command didn’t exist in the container. If I removed sudo from the command… it still failed because I didn’t have permissions to run dnf inside the container ಠ_ಠ (yes, the user was root).

After some research, I learned that it wasn’t passing the root permissions to the container, and I could “solve” this issue by providing the -u flag with 0:0 as an arg to the docker agent:

pipeline {
    agent {
        docker {
            image 'fedora:31'
            args '-u 0:0'
        }
    }
    ... // rest of the pipeline
}

I don’t love this solution… but it seems to work.

Conclusion

Like I said, this was just a quick update about switching my test nodes to use docker containers. Honestly, I’d much rather try to use podman containers for my test agents, but I’m sure that would be much more complicated currently. Maybe in the future…