che-incubator/kubernetes-image-puller

To cache images, Kubernetes Image Puller creates a Daemonset on the desired cluster, which in turn creates a pod on each node in the cluster consisting of a list of containers with command sleep 720h. This ensures that all nodes in the cluster have those images cached. The sleep binary being used is golang-based (please see Scratch Images). We also periodically check the health of the daemonset and re-create it if necessary.

The application can be deployed via Helm or by processing and applying OpenShift Templates. Also, there is a community supported operator available on the OperatorHub.

Configuration is done via env vars pulled from ./deploy/helm/templates/configmap.yaml, or ./deploy/openshift/configmap.yaml, depending on the deployment method. The config values to be set are:

The following values can be set:

The following values can be set:

kubectl create namespace k8s-image-puller

helm install kubernetes-image-puller -n k8s-image-puller deploy/helm

To set values, change deploy/helm/values.yaml or use --set property.name=value

OpenShift has a notion of project quotas to limit the aggregate resource consumption per project/namespace. The namespace that the image puller is deployed in must have enough memory and CPU to run each container for each node in the cluster:

(memory/CPU limit) * (number of images) * (number of nodes in cluster)

For example, running the image puller that caches 5 images on 20 nodes, with a container memory limit of 5Mi, your namespace would need a quota of 500Mi.

oc new-project k8s-image-puller

oc process -f deploy/openshift/serviceaccount.yaml | oc apply -f -

oc process -f deploy/openshift/configmap.yaml | oc apply -f -

oc process -f deploy/openshift/app.yaml | oc apply -f -

To change parameters, add -p PARAM=value to the oc process command, before piping to oc apply.

# Build Go binary:
make build
# Make docker image:
make docker
# The above:
make
# Clean:
make clean

The provided Makefile has two parameters:

• DOCKERIMAGE_NAME: name for docker image
• DOCKERIMAGE_TAG: tag for docker image

Build:

CGO_ENABLED=1
BINARY_NAME=kubernetes-image-puller

GOOS=linux go build -v -o ./bin/${BINARY_NAME} ./cmd/main.go
GOOS=linux go build -a -ldflags '-w -s' -a -installsuffix cgo -o ./bin/sleep ./sleep/sleep.go

Make docker image:

DOCKERIMAGE_NAME=kubernetes-image-puller
DOCKERIMAGE_TAG=next

docker build -t ${DOCKERIMAGE_NAME}:${DOCKERIMAGE_TAG} -f ./build/dockerfiles/Dockerfile .

Once built and published to a registry, you can test FIPS compliance using https://github.com/openshift/check-payload#scan-a-container-or-operator-image

To run the unit tests:

make test

End to end tests require kind. Note that kind should not be installed with go get from this repository's directory.

cd $HOME && GO111MODULE="on" go get sigs.k8s.io/[email protected] && cd ~-

./hack/run-e2e.sh

Will start a kind cluster and run the end-to-end tests in ./e2e. To remove the cluster after running the tests, pass the --rm argument to the script, or run kind delete cluster --name k8s-image-puller-e2e.

The image puller also supports pre-pulling the scratch images. Previously the image puller was not able to pull scratch images, as they do not contain a sleep command.

However, the daemonset created by the Kubernetes Image Puller now:

1. creates an initContainer that copies a golang-based sleep binary to a common kip volume.
2. creates containers volumeMounts set to the kip volume, and with command set to /kip/sleep 720h

As a result, every container (including scratch image containers) uses the provided golang-based sleep binary.

"Che" is a trademark of the Eclipse Foundation.