If you utilize a GitLab Runner with Docker, the resulting architecture differs from the one in the previous chapter in one way. The Runner binary is executed from inside a Docker container instead of directly on a host system. The following diagram shows this architecture:

The image is the same whether you create a container yourself or you use an existing container from the internet.

If you are going to orchestrate your GitLab Runners using Kubernetes, the architecture is going to look a bit different. You can see that, inside the cluster, a GitLab Runner with the Kubernetes architecture can talk to the cluster's Kubernetes API to scale up the number of Runner instances:

The two architectures both use Docker containers as the core unit of operation. The first needs more management in the field of scaling, upgrading the software, and setting up a network. Much of that has already been arranged with Kubernetes.

Now, we know about the difference between these architectures and that a Runner is deployed without a container, but what does this mean in practice? We will explore this in the following sections.