CERN is the home of the Large Hadron Collider (LHC), a 27km circular proton accelerator that generates petabytes of physics data every year. To process all this data, CERN runs an OpenStack Cloud (>300K cores) that helps scientists all around the world to unveil the mysteries of the Universe. The Infrastructure is also used to run all the IT services of the Organization.
Delivering these services, with high performance and reliable service levels has been one of the major challenges for the CERN Cloud engineering team. We have been constantly iterating the architecture and deployment model of the Cloud control plane.
In this presentation we will describe the different control plane architecture models that we relied over the years. Finally, we will describe all the work done to move the OpenStack Cloud control plane from VMs into a kubernetes cluster. We will report about our experience running this architecture at scale, its advantages and challenges.
This presentation will describe how CERN is moving its OpenStack Cloud control plane from VMs to Kubernetes.
We will start with an overview about the CERN OpenStack Cloud. Then we will dive into the infrastructure architecture and describe the different architecture choices for the control plane.
Finally, we will describe all the work done to move the OpenStack Cloud control plane to a kubernetes cluster provisioned by Magnum and deployed in the same infrastructure, report about our experience running this setup at scale, its caveats and how we are mitigating them.