Why container runtimes?
Docker is one way to run containers, not the only container runtime: the OCI standards, high-level (containerd, CRI-O) vs low-level (runc) runtimes, Podman, and which runtime Kubernetes uses by default.
Transcript
Docker made containers famous, but Docker is not the only way to run them.
A container is just a Linux process, isolated with namespaces and limited with cgroups. Docker is one tool that sets that up.
The Open Container Initiative sets two standards: the image format and the runtime. Any tool that follows them can run the same image.
A container runtime has two layers. A high-level runtime, like containerd, pulls images and manages containers. A low-level runtime, like runc, creates the process.
Docker Engine is a bundle: the docker command talks to dockerd, which calls containerd, which calls runc.
Other runtimes exist. Podman runs containers without a daemon, CRI-O is built just for Kubernetes, and containerd also works on its own.
Kubernetes builds and runs nothing itself. The kubelet asks a runtime to start containers through one standard API, the Container Runtime Interface.
Since version 1.24, Kubernetes no longer talks to Docker. Today containerd is the default choice: kind, k3s, EKS, GKE and AKS all use it.
OpenShift uses CRI-O. And your Docker images still run everywhere, because they follow the OCI standard.
Standard. Swappable. containerd.

