Terminology: Container Engine vs container runtime

CarlosValencia

Hello,

want to get my terminology right. Could someone clarify the difference between container engine and container runtime? And examples of each?

From the official documentation I read:
"All nodes run the kubelet and kube-proxy, as well as the container engine, such as Docker or cri-o, among several options. Other management daemons are deployed to watch these agents or provide services not yet included with Kubernetes."

And from RedHat page I read:
A container engine is a piece of software that processes user requests, including command line options and image pulls. The container engine uses a container runtime, also called a lower-level container runtime, to run and manage the components required to deploy and operate containers. You likely will not need to interact with the container engine or container runtime.

So ok....cri-o is a container engine, which seems to take care of requests and leverage container runtimes like containerd to do the actual container instantiation? When I list nodes I can see what my kubernetes nodes use containerd as the container runtime. How can I know what container engine is being used?

Now, what is crictl ? Just a tool to interact with the container runtime?

ubuntu@master:~$ kubectl get nodes -o wide
NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME
master Ready control-plane 17h v1.27.1 192.168.122.186 Ubuntu 20.04.6 LTS 5.4.0-165-generic containerd://1.6.24
worker1 Ready 16h v1.27.1 192.168.122.131 Ubuntu 20.04.6 LTS 5.4.0-165-generic containerd://1.6.24
worker2 Ready 16h v1.27.1 192.168.122.99 Ubuntu 20.04.6 LTS 5.4.0-165-generic containerd://1.6.24

The reason for my question is that when i asked chat gpt it says otherwise:

I apologize for any confusion. CRI-O is indeed a container runtime, not a container engine. It is designed specifically to serve as a container runtime for Kubernetes. CRI-O is an implementation of the Kubernetes Container Runtime Interface (CRI), which allows Kubernetes to interact with container runtimes to create and manage containers on nodes within a cluster. While CRI-O performs the low-level tasks of running containers, it is not a high-level service like Docker, which can be considered more of a container engine.

To clarify:

A container engine (like Docker) is a higher-level tool that often includes features for managing container images, building containers, and more.
A container runtime (like CRI-O) is a lower-level component responsible for the core tasks of creating and running containers.

In Kubernetes, CRI-O is used as the container runtime to handle the execution of containers, and it interacts with Kubernetes through the CRI to ensure container orchestration within the cluster.

dcruzd

Take my comment with a grain of salt as I am still a student. If I am wrong anywhere, please do tell me.

I think the line between container engine and container runtime is very blurry right now. It depends on how you define each and so it depends on who you ask. I think it is more important to understand the stack, what each component does, what you can plug-in/out and some key definitions like CRI, OCI-compliant runtime, etc.

From top-down, in Kubernetes you need to plug-in anything that implements/satisfies CRI (Container Runtime Interface). This interface was created by Kubernetes to standardize and off-load container creation/management. Kubernetes specifically defines these components as container runtimes and lists some examples such as containerd, cri-o.

These "CRI container runtimes"/container engines, such as containerd and cri-o, use, in turn, OCI (Open Container Initiative) compliant runtimes to run the containers. OCI was created to define standards/specs for containers. OCI-compliant runtimes are low-level runtimes that effectively run the containers. Containerd, by default, uses runc (an OCI-compliant runtime) and spawns instances of runc for each container. On cri-o you can plug in any OCI runtime.

crictl, initially developed for debugging, is a tool that interacts with anything that implements the CRI (containerd, cri-o, etc). Given that Docker itself uses containerd, theoretically you could use crictl to interact with/manage containers created with Docker, as long as you provide the correct runtime endpoint (e.g. unix:///var/run/docker/containerd/containerd.sock).

On the labs we have:
Kubernetes -----CRI----> containerd -----> runc (for each container)

On the internet and some documentation I've seen containerd defined as a container engine (example Podman docs). RedHat, as per the definition you showed, defines containerd as a container engine and runc a container runtime as "you likely will not need to interact" directly with both (you can interact with containerd with Docker, nerdctl, cri-o). Some (or most, don't know which) also define anything to the left of containerd as a container engine (Kubernetes, Docker). Basically container runtime=OCI-compliant runtime only.

On the other hand Kubernetes, on its docs, defines anything that implements CRI, a container runtime. To add more to the confusion, Kubernetes lists Docker Engine as a CRI compatible container runtime.

So, given all of this, the exact answer to your question might be "complicated". Personally, when specifying container runtime I would always prefix OCI/CRI to avoid any confusion.

CarlosValencia

but runc is a container runtime as well isnt it? I thought out it would do the same function as containerd. Head spinning.

On the labs we have:
Kubernetes -----CRI----> containerd -----> runc (for each container)

CarlosValencia

@CarlosValencia said:
but runc is a container runtime as well isnt it? I thought out it would do the same function as containerd. Head spinning.

On the labs we have:
Kubernetes -----CRI----> containerd -----> runc (for each container)

Oh now that I read more closely on containerd and runc:

containerd supports the OCI container image specification and the OCI runtime specification by utilizing runc as its low level OCI runtime with the possibility to extend it with plugins to support the Kubernetes’ Container Runtime Interface (CRI) as well. containerd adds implementation for some missing, yet desired, capabilities of runc, such as support for container image pull and push operations, and network interfaces and network namespaces management operations.