Docker is a platform designed to help developers build, share, and run container applications.
In gVisor, all basic docker commands should function as expected. The host network driver and the bridge network driver are supported.
| Docker Version | GKE standard cluster | GKE autopilot cluster |
|---|---|---|
| Docker v27 | 1.29.0 and later | 1.33.2 and later |
| Docker v28 | 1.35.3 and later | 1.35.3 and later |
| Docker v29 | not supported | not supported |
Install a GKE standard cluster and deploy a node pool with gVisor enabled. You can view the full documentation here.
Alternatively to GKE standard cluster, you could run docker in gVisor at GKE autopilot cluster. GKE autopilot is a fully managed mode of operation in GKE that automates cluster infrastructure management, including node provisioning, scaling, and security.
When creating the autopilot cluster, please add the option
--workload-policies=allow-net-admin to allow NET_ADMIN capability that will be
granted by the gVisor sandbox which is needed by docker daemon.
An example command to start an GKE autopilot cluster will be:
gcloud container clusters create-auto [CLUTER_NAME] --workload-policies=allow-net-admin --location=[LOCATION]
Prepare a container image with pre-installed Docker:
$ docker build -t docker-in-gvisor images/basic/docker/
$ docker push {registry_url}/docker-in-gvisor:latest
Create a Kubernetes pod YAML file (docker.yaml) with the following content:
apiVersion: v1
kind: Pod
metadata:
name: docker-in-gvisor
spec:
runtimeClassName: gvisor
containers:
- name: docker-in-gvisor
image: {registry_url}/docker-in-gvisor:latest
securityContext:
capabilities:
# NOTE: This setting does **not** grant any capabilities on the host.
# gVisor *always* runs with zero privileges and zero capabilities,
# regardless of this setting.
add: [
# Required for Docker daemon to work in the sandbox.
# This does *NOT* grant any host capabilities. See below.
NET_ADMIN,SYS_ADMIN,
# Default set of capabilities granted to any container:
AUDIT_WRITE,CHOWN,DAC_OVERRIDE,FOWNER,FSETID,KILL,MKNOD,NET_BIND_SERVICE,NET_RAW,SETFCAP,SETGID,SETPCAP,SETUID,SYS_CHROOT,SYS_PTRACE,
]
volumeMounts:
- name: docker
# NOTE: This does *not* expose any host directory;
# this just mounts a tmpfs here.
mountPath: /var/lib/docker
volumes:
- name: docker
emptyDir: {} # tmpfs
NOTE: gVisor never runs with capabilities on the host Linux kernel, even when the above
securityContext.capabilitiesfields are specified. These fields only control the capabilities perceived by the in-sandbox application (in this case, the in-sandbox Docker daemon). This does not provide the sandboxed application, nor the gVisor sandbox itself, any host privileges.
This YAML file defines a Kubernetes Pod named docker-in-gvisor that will run a single container from the {registry_url}/docker-in-gvisor:latest image.
Apply the pod YAML to your GKE cluster using the kubectl apply command:
$ kubectl apply -f docker.yaml
Verify that the docker-in-gvisor pid is running successfully: shell $ kubectl
get pods | grep docker-in-gvisor
You can access the container by executing a shell inside it. Use the following command:
kubectl exec -it docker-in-gvisor -- bash
Now, we can build and run Docker containers.
$ mkdir whalesay && cd whalesay
$ cat > Dockerfile <<EOF
FROM ubuntu
RUN apt-get update && apt-get install -y cowsay curl
RUN mkdir -p /usr/share/cowsay/cows/
RUN curl -o /usr/share/cowsay/cows/docker.cow https://raw.githubusercontent.com/docker/whalesay/master/docker.cow
ENTRYPOINT ["/usr/games/cowsay", "-f", "docker.cow"]
EOF
$ docker build -t whalesay .
....
Successfully tagged whalesay:latest
$ docker run -it --rm whalesay "Containers do not contain, but gVisor-s do!"
_________________________________________
/ Containers do not contain, but gVisor-s \
\ do! /
-----------------------------------------
\ ## .
\ ## ## ## ==
## ## ## ## ===
/""""""""""""""""\___/ ===
~~~ {~~ ~~~~ ~~~ ~~~~ ~~ ~ / ===- ~~~
\______ o __/
\ \ __/
\____\______/