In short
OKD, the community distribution of Kubernetes behind OpenShift, adds what plain Kubernetes leaves out: platform management, an operating system built for containers, and operators that keep the cluster running itself. Ayesha Kaleem, a software engineer on Red Hat's core OpenShift team, explains the two ways to install it, the building blocks of OKD 4, and starts a real installation on VMware vSphere. She ends by inviting viewers to join the OKD and Fedora CoreOS working groups.
Key takeaways
- Distinguish installer-provisioned from user-provisioned infrastructure when planning an OKD cluster.
- Choose CodeReady Containers instead of a full cluster when working on a resource-limited developer workstation.
- Explain how Fedora CoreOS, Ignition, CRI-O and Kubelet fit together in an OKD 4 cluster.
- Prepare an OKD install on VMware vSphere: extract the binaries, build the install config and run the installer.
- Identify the core cluster operators and how OperatorHub and Operator Lifecycle Manager handle optional ones.
What OKD is and how it gets installed
Kaleem starts with the basics. OKD, the OpenShift Kubernetes distribution, is a container platform for developing and running containerized applications. "The foundation of OKD is based on Kubernetes and therefore shares the same technology." (1:17)
A cluster administrator has two install routes. Installer-provisioned infrastructure is largely automated: the installer sets up the infrastructure. User-provisioned infrastructure means configuring the base infrastructure by hand. OKD runs on bare metal, several virtualization platforms and several cloud platforms, so the first decision is where to install it and whether that environment has enough resources.
For developers who only have a laptop, she points to a lighter option: "If you want to install on a typical developer workstation, then code-ready containers may be a better option, as that is a cut-down installation design to run on limited compute and memory resources." (2:17) At the other end of the scale, she says the platform is designed to grow from a few machines to thousands of machines serving millions of clients.
Why OKD 4 goes beyond plain Kubernetes
The core argument of the talk is about the gap between running applications and running a platform. "Although Kubernetes excels at managing your applications, but it doesn't specify or manage platform-level requirements or deployment processes." (3:57) OKD 4 fills that gap with platform management tools and processes. Its major components come from Fedora and related Red Hat technologies, and, as the speaker describes it, it benefits from Red Hat's testing and certification work for enterprise software.
Kaleem then walks through the pieces. Fedora CoreOS is a container-oriented operating system that brings fast installation, operator-based management and simpler upgrades, and it includes Ignition for first-boot machine configuration. CRI-O is the Kubernetes-native container runtime that replaces the Docker engine used in OKD 3. Kubelet is the node agent that launches and monitors containers. One rule matters for planning: "In OKD 4, you must use Fedora CoreOS for all controlled planes machines." (5:02) Worker machines can run Red Hat Enterprise Linux instead, but she warns that this brings more system maintenance.
A live install on VMware vSphere
Kaleem then switches to a virtual machine and begins a real installation, warning that it will still be running when the talk ends. She extracts the openshift-install binary, the oc client and kubectl for the then-current 4.10 release, moves them into the local bin directory, and creates a directory for the cluster.
Next she builds the install config interactively: the vCenter details and credentials, the datastore, the network segment, virtual IP addresses for the API and ingress, the base domain and the cluster name. Then comes a detail that surprises people coming from OpenShift: "Okay, so for the OpenShift OKD installation, because it's a community version, so you don't need a full secret." (10:42) (The transcript hears "full secret" where she means pull secret.) A fake pull secret is enough, as long as it is valid YAML. With the config in place she runs the create cluster command and notes that a full install takes roughly 20 to 30 minutes. If errors come up, her advice is to search for them and fix them.
Operators: putting the cluster on autopilot
While the installer runs, Kaleem returns to the slides for what she calls a really interesting topic. An operator automates what Kubernetes does on a user's behalf and extends Kubernetes to manage the whole lifecycle of an application. "It's a custom Kubernetes controller." (13:17) The Operator SDK lets teams build operators with Go, Helm or Ansible.
In OKD 4, operators manage the cluster itself, not only applications. The Cluster Version Operator makes sure all essential core operators are present and keeps their versions up to date. The Cluster Network Operator installs and upgrades networking components and reconciles the cluster against the desired configuration. The Ingress Operator configures ingress controllers so routes work, a storage operator picks the cloud storage type for the provider, and the monitoring operator manages the Prometheus-based monitoring stack.
Optional operators come from OperatorHub, with Operator Lifecycle Manager acting as a cluster version operator for them: it installs, updates and manages their lifecycle at cluster or namespace scope. Her examples include storage operators based on Rook, and pipeline operators based on Tekton and Argo CD.
An open invitation to the community
Kaleem closes, short on time, by stressing that OKD is built in the open. "The development is open source and the source code is available from public software repository, for example, GitHub. This open collaboration helps rapid innovation and development." (3:57) There are working groups for OKD, CodeReady Containers on OKD, and virtualization, plus a Fedora CoreOS working group. Each has Slack channels and bi-weekly video meetings with a calendar viewers can subscribe to, and she invites everyone to join and collaborate.
Watch the talk
"OKD - Community Version of Openshift" by Ayesha Kaleem, about 19 minutes: https://www.womentech.net/video/okd-community-version-openshift-ayesha-kaleem
About the speaker
Ayesha Kaleem is a Software Engineer at Red Hat, based in Berlin, Germany. She works in Red Hat's core OpenShift team on OKD, the community distribution of Kubernetes, with the aim of building the most efficient product.
Questions this talk answers
- What is OKD and how does it relate to Kubernetes and OpenShift?
- What is the difference between installer-provisioned and user-provisioned infrastructure?
- Which operating system do OKD 4 nodes run?
- Do you need a real pull secret to install OKD?
- What do operators do in OKD 4?
Drafted with AI from the talk transcript. Editor review pending.