Operator Sample for CRW

Developer Workspace

Instructions to run the demo

The following paragraphs describe how to setup and run a demo that shows how to develop this operator within Che.

The demo can be preceded by a short Che presentation (slides)

SETUP

Build the operator-sdk image

WARNING: the resulting image is > 4GB!! The build clones the operator source code and does a build of it. During the build all the go dependencies are downloaded. That saves a lot of time during the demo but makes the image uge.

WARNING: this should be done the day before the demo or the same day so that the image has all the operator dependencies already downloaded.

docker build -t mariolet/operator-sdk:0.11.0 -f build/Dockerfile.dev ./build
docker push mariolet/operator-sdk:0.11.0

Once you have build and push it you should replace the image reference in the devfile.yaml

Something that may be useful is to pull the image locally:

SHA256=<the image digest>
eval $(minikube docker-env)
docker pull mariolet/operator-sdk@sha256:${SHA256}

Start the k8s cluster (minikube) and start Che

# Start minikube
mk

# Start che
chectl server:start --platform=minikube --installer=helm

# Change the run as user
kubens che

# Update the default uid: the default is 1724, set it to zero instead
k patch cm/che -p '{"data":{"CHE_INFRA_KUBERNETES_POD_SECURITY__CONTEXT_FS__GROUP":"0", "CHE_INFRA_KUBERNETES_POD_SECURITY__CONTEXT_RUN__AS__USER":"0"}}'

# Every user should have his own namespace
k patch cm/che -p '{"data":{"CHE_INFRA_KUBERNETES_NAMESPACE_DEFAULT":"che-<username>"}}'
k create namespace che-${USERNAME}

# Restart Che
k rollout restart deploy/che

Start a local workspace

# Start the operator-hello-world workspace
DEVFILE=/Users/mloriedo/github/operator-hello-world/devfile.yaml
chectl workspace:start --devfile=${DEVFILE}

Inject the kube config in the workspace

chectl workspace:inject -k -n <worksapce namespace>

Set Che user Preferences (optional)

{
   "workbench.appearance.colorTheme": "light",
   "terminal.integrated.copyOnSelection": true,
   "git.user.name": "l0rd",
   "git.user.email": "mario.loriedo@gmail.com"
}

Locally deploy the podset-operator

# Deploy the service account and setup RBAC
kubectl create -f deploy/service_account.yaml
kubectl create -f deploy/role.yaml
kubectl create -f deploy/role_binding.yaml

# Setup the CRD and deploy the podset-operator
kubectl create -f deploy/crds/app.example.com_podsets_crd.yaml
kubectl create -f deploy/operator.yaml

DEMO

Show the podset operator

# Look for the podset api
kubectl api-resources

# Get the podsets (there should be no podset)
kubectl get podset

# Create a podSet
echo "apiVersion: app.example.com/v1alpha1
kind: PodSet
metadata:
  name: example-podset
spec:
  replicas: 3" | kubectl create -f -

# Get the podset again (there should be one)
kubectl get podset

# Get the pods (there should be a lot)
kubectl get pod

# Delete the podset
kubectl delete podset example-podset

# Scaled down the operator deployment
kubectl scale --replicas=0 deploy podset-operator

Fix the bug using a Che wokrspace

Open the github repo of the operator, show the devfile and start the workspace using the factory

https://github.com/l0rd/operator-hello-world

Switch to the local workspace because che.openshift.io takes time to load and doesn't have enough resources

Show the source code
Build and run the operator using the corresponding task
Change the code
Rebuild and run and verify that the changes to the controller have been taken into account

Old stuff

Build the operator image

# Using the operator SDK
operator-sdk build --image-builder buildah mariolet/podset-operator

# Using buildah
buildah bud --format=docker --isolation=chroot -f build/Dockerfile -t mariolet/podset-operator .

buildah bud --format=docker -f build/Dockerfile -t mariolet/podset-operator .

buildah bud --security-opt label=disable --security-opt seccomp=unconfined --format=docker --isolation=chroot -f build/Dockerfile -t mariolet/podset-operator .

Go build of the operator

go build -o /projects/src/github.com/l0rd/operator-hello-world/build/_output/bin/operator-hello-world \
         -gcflags all=-trimpath=/projects/src/github.com/l0rd \
         -asmflags all=-trimpath=/projects/src/github.com/l0rd \
         github.com/l0rd/operator-hello-world/cmd/manager

Open Issues

Blockers

Investigate why the build of the operator image with buildah doesn't work in a Che container

Nice to have

fix issue described in issue1.md
run on CRW
We should write down instructions to set imagePullStrategy to never for all workspaces components
We should have a script to pre pull all needed images
Setup the powershell PS1
upgrade operator-sdk to latest

ObieBent / operator-hello-world