Aller au contenu

Container runtimes


OCI Runtime spec

Every OCI-compatible container runtime must implement the runtime-spec

The mandatory components of a OCI container are:

  • the configuration: a config.json file (JSON schema)
  • the filesystem bundle: unix filesystem


The most basic, low-level runtime (donated from Docker to OCI)

run a container with runc

1) runc spec: generates the confg.json file with all the cgroups / namespaces / mounts to create in order to run the container

3) mkdir rootfs; docker export $(docker create busybox) | tar -xf - -C rootfs: creates the filesystem bundle

2) sudo runc run toto: runs a toto container based on the conf + bundle


Built upon runc (communicates with runc via socket)

Adds new features like: pushing and pulling images (registries), volume management, network management (CNI), GRPC API, CLI client (ctr).

ctr oci spec: generates the config.json. Note that it has more features than the basic runc-generated config


containerd and dockerd

dockerd is built upon containerd. dockerd is not a container runtime! It's a developer-facing tool (building images etc)

this is about the same difference between podman and buildah: podman can run containers but not build them directly, buildah can only build.


Built upon runc.

Build upon libpod (golang library + CLI): does not require any daemon (unlike Docker with dockerd). See Docker vs Podman

  • podman CLI is modeled after Docker CLI and can be used as a drop-in replacement for docker
  • podman also exposes a RESTful API when it's running


nvidia-container-runtime is a low-level runtime (modified runc)

allows GPU-enabled containers for CUDA processing. Enjoy melting your bare metal cluster.

Kubernetes CRI

CRI (Container Runtime Interface) is a spec for Kubernetes Kubelet (the daemon running on every K8S node). The goal: making K8S extendable without the dev knowing the kubelet internals && enabling almost any container runtime on kubelet .

To be more specific: CRI is a gRPC/Protobuf API. (Example: a K8S pod makes a gRPC request to the CRI API to start multiple container)


With CRI-O runtime, Kubernetes can run any OCI container.


Historically, K8S supported only Docker as a runtime (not containerd, just docker - when docker itself was a monlithic application)

This is the component which is getting removed in K8S 1.20 release note:


Docker vs Podman

Docker uses a client-server architecture

docker run --rm -it busybox

We see clearly the docker-cli -> dockerd -> containerd architecture (and the socket connecting to runc).

$ ps faux | grep docker
root       18297  0.0  0.0 708436  6284 ?        Sl   23:56   0:00  \_ containerd-shim -namespace moby -workdir /var/lib/containerd/io.containerd.runtime.v1.linux/moby/51cbef396fdf91d549d15011ff8ed12c3fe444a6e7046f1cf702b529edb11f16 -address /run/containerd/containerd.sock -containerd-binary /usr/bin/containerd -runtime-root /var/run/docker/runtime-runc
root         453  0.0  1.5 1452072 118728 ?      Ssl  09:25   0:15 /usr/bin/dockerd -H fd:// --containerd=/run/containerd/containerd.sock
hed        18281  0.3  0.6 1285368 54000 pts/4   Sl+  23:56   0:00          |       \_ docker run --rm -it busybox

Podman manages to run without a central daemon by using a fork() architecture.

podman run --rm -it busybox

We see that there's a parent-child relationship between the podman processes. (we also see that podman containers live in var/lib/containers)

$ ps faux | grep podman
root       17900  0.0  0.0  15560  7160 pts/4    S+   23:55   0:00          |       \_ sudo podman run --rm -it busybox
root       17901  3.0  0.7 1493832 62204 pts/4   Sl+  23:55   0:00          |           \_ podman run --rm -it busybox
hed        14097  0.0  0.2  50428 23172 ?        S    21:46   0:00 podman
root       18026  0.0  0.0  81312  1908 ?        Ssl  23:55   0:00 /usr/bin/conmon --api-version 1 -c f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec -u f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec -r /usr/bin/runc -b /var/lib/containers/storage/overlay-containers/f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec/userdata -p /var/run/containers/storage/overlay-containers/f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec/userdata/pidfile -n priceless_brahmagupta --exit-dir /var/run/libpod/exits --socket-dir-path /var/run/libpod/socket -s -l k8s-file:/var/lib/containers/storage/overlay-containers/f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec/userdata/ctr.log --log-level error --runtime-arg --log-format=json --runtime-arg --log --runtime-arg=/var/run/containers/storage/overlay-containers/f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec/userdata/oci-log -t --conmon-pidfile /var/run/containers/storage/overlay-containers/f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec/userdata/ --exit-command /usr/bin/podman --exit-command-arg --root --exit-command-arg /var/lib/containers/storage --exit-command-arg --runroot --exit-command-arg /var/run/containers/storage --exit-command-arg --log-level --exit-command-arg error --exit-command-arg --cgroup-manager --exit-command-arg systemd --exit-command-arg --tmpdir --exit-command-arg /var/run/libpod --exit-command-arg --runtime --exit-command-arg runc --exit-command-arg --storage-driver --exit-command-arg overlay --exit-command-arg --storage-opt --exit-command-arg overlay.mountopt=nodev --exit-command-arg --storage-opt --exit-command-arg overlay.mountopt=nodev --exit-command-arg --events-backend --exit-command-arg journald --exit-command-arg container --exit-command-arg cleanup --exit-command-arg --rm --exit-command-arg f2a418363107e7a7e6bbaeb2a499d11d0939e20dae8350627b43babb887c07ec