Skip to main content

Developer guide

Operator SDK​

Prerequisites​

Casskop has been validated with :

Install the Operator SDK CLI​

First, checkout and install the operator-sdk CLI:

$ mkdir -p $GOPATH/src/github.com/operator-framework/
$ cd $GOPATH/src/github.com/operator-framework/
git clone -b v0.18.0-forked-pr317 --single-branch git@github.com:erdrix/operator-sdk.git
$ make tidy
$ make install

Note : Use fork from operator-sdk, waiting for PR #317 to be merged.

Initial setup​

Checkout the project.

$ git clone https://github.com/cscetbon/casskop.git
$ cd casskop

Local kubernetes setup​

We use kind in order to run a local kubernetes cluster with the version we chose. We think it deserves some words as it's pretty useful and simple to use to test one version or another

Install​

The following requires kubens to be present. On MacOs it can be installed using brew :

brew install kubectx

The installation of kind is then done with (outside of the cassandra operator folder if you want it to run fast) :

$ GO111MODULE="on" go get sigs.k8s.io/kind@v0.5.1

Setup​

The following actions should be run only to create a new kubernetes cluster.

$ config/samples/kind/create-kind-cluster.sh

or if you want to enable network policies

$ config/samples/kind/create-kind-cluster-network-policies.sh

It creates namespace cassandra-e2e by default. If a different namespace is needed it can be specified on the setup-requirements call

$ config/samples/kind/setup-requirements.sh other-namespace

To interact with the cluster you then need to use the generated kubeconfig file :

$ export KUBECONFIG=$(kind get kubeconfig-path --name=kind)

Before using that newly created cluster, it's better to wait for all pods to be running by continously checking their status :

$ kubectl get pod --all-namespaces -w

Pause/Unpause the cluster​

In order to kinda freeze the cluster because you need to do something else on your laptop, you can use those two aliases. Just put them in your ~/.bashrc or ~/.zshrc :

alias kpause='kind get nodes|xargs docker pause'
alias kunpause='kind get nodes|xargs docker unpause'

Delete cluster​

The simple command kind delete cluster takes care of it.

Build CassKop​

Using your local environment​

If you prefer working directly with your local go environment you can simply uses :

make get-deps
make build

You can check on the Gitlab Pipeline to see how the Project is build and test for each push

Or Using the provided cross platform build environment​

Build the docker image which will be used to build CassKop docker image

make build-ci-image

If you want to change the operator-sdk version change the OPERATOR_SDK_VERSION in the Makefile.

Then build CassKop (code & image)

make docker-get-deps
make docker-build

Run CassKop​

We can quickly run CassKop in development mode (on your local host), then it will use your kubectl configuration file to connect to your kubernetes cluster.

There are several ways to execute your operator :

  • Using your IDE directly
  • Executing directly the Go binary
  • deploying using the Helm charts

If you want to configure your development IDE, you need to give it environment variables so that it will uses to connect to kubernetes.

KUBECONFIG=<path/to/your/kubeconfig>
WATCH_NAMESPACE=<namespace_to_watch>
POD_NAME=<name for operator pod>
LOG_LEVEL=Debug
OPERATOR_NAME=ide

Run the Operator Locally with the Go Binary​

This method can be used to run the operator locally outside of the cluster. This method may be preferred during development as it facilitates faster deployment and testing.

Set the name of the operator in an environment variable

$ export OPERATOR_NAME=casskop

Deploy the CRD

$ kubectl apply -f config/crd/bases/db.orange.com_cassandraclusters.yaml
$ make run

This will run the operator in the default namespace using the default Kubernetes config file at $HOME/.kube/config.

Note: JMX operations cannot be executed on Cassandra nodes when running the operator locally. This is because the operator makes JMX calls over HTTP using jolokia and when running locally the operator is on a different network than the Cassandra cluster.

Deploy using the Helm Charts​

This section provides an instructions for running the operator Helm charts with an image that is built from the local branch.

Build the image from the current branch.

$ export DOCKER_REPO_BASE=<your-docker-repo>
$ make docker-build

Push the image to docker hub (or to whichever repo you want to use)

$ make push

Note: In this example we are pushing to docker hub.

Note: The image tag is a combination of the version as defined in verion/version.go and the branch name.

Install the Helm chart.

$ helm install ./charts/casskop \
    --set-string image.repository=ghcr.io/cscetbon/casskop,image.tag=0.4.0-local-dev-helm \
    --name local-dev-helm

Note: The image.repository and image.tag template variables have to match the names from the image that we pushed in the previous step.

Note: We set the chart name to the branch, but it can be anything.

Lastly, verify that the operator is running.

$ kubectl get pods
NAME                                                READY   STATUS    RESTARTS   AGE
local-dev-helm-casskop-8946b89dc-4cfs9   1/1     Running   0          7m45s

Run unit-tests​

You can run Unit-test for CassKop

make unit-test

Run e2e end to end tests​

CassKop also have several end-to-end tests that can be run using makefile.

You need to create the namespace cassandra-e2e before running the tests.

kubectl create namespace cassandra-e2e

to launch different tests in parallel in different temporary namespaces

make e2e

or sequentially in the namespace cassandra-e2e

make e2e-test-fix

Note: make e2e executes all tests in different, temporary namespaces in parallel. Your k8s cluster will need a lot of resources to handle the many Cassandra nodes that launch in parallel.

Note: make e2e-test-fix executes all tests serially in the cassandra-e2e namespace and as such does not require as many k8s resources as make e2e does, but overall execution will be slower.

You can choose to run only 1 test using the args ex:

make e2e-test-fix-arg ClusterScaleDown

Tip: When debugging test failures, you can run kubectl get events --all-namespaces which produce output like:

cassandracluster-group-clusterscaledown-1561640024 0s Warning FailedScheduling Pod 0/4 nodes are available: 1 node(s) had taints that the pod didn't tolerate, 3 Insufficient cpu.

Tip: When tests fail, there may be resources that need to be cleaned up. Run tools/e2e_test_cleanup.sh to delete resources left over from tests.

Run kuttl tests​

This requires kuttl cli to be installed on your machine

You first need to have a Kubernetes cluster set up with kubectl.

Then to run all tests you can simply type :

kubectl kuttl test --config ./test/e2e/kuttl/kuttl-test.yaml ./test/e2e/kuttl/

This will run all testcases in the /test/e2e/kuttl/ directory in parallel on different generated namespaces (with Casskop automatically installed on each).

If you installed only the binary of kuttl, you can omit the kubectl at the beginning

Tip: You can specify a single test case to run by adding --test TestCase where TestCase is the name of one of the directories in /test/e2e/kuttl/*here* (like ScaleUpAndDown for example)

Debug CassKop in remote in a Kubernetes cluster​

CassKop makes some specific calls to the Jolokia API of the CassandraNodes it deploys inside the kubernetes cluster. Because of this, it is not possible to fully debug CassKop when launch outside the kubernetes cluster (in your local IDE).

It is possible to use external solutions such as KubeSquash or Telepresence.

Telepresence launch a bi-directional tunnel between your dev environment and an existing operator pod in the cluster which it will swap

To launch the telepresence utility you can launch

make telepresence

You need to install it before see : https://www.telepresence.io/

If your cluster don't have Internet access, you can change the telepresence image to use to one your cluster have access exemple:

TELEPRESENCE_REGISTRY=you-private-registry/datawire  make debug-telepresence-with-alias

Configure the IDE​

Now we just need to configure the IDE :

and let's the magic happened

Build Multi-CassKop​

Using your docker environment​

cd $(git rev-parse --show-toplevel)/multi-casskop
make docker-build

Run Multi-CassKop​

We can quickly setup a k3d cluster with casskop and multi-casskop to test a PR on multi-casskop.

  • [Build your multi-casskop docker image](#### Using your docker environment) which should print
echo "Generate zzz-deepcopy objects"
Generate zzz-deepcopy objects
...
Successfully built bf57e90615bb
Successfully tagged ghcr.io/cscetbon/multi-casskop:0.5.6-my-pr
  • Create a k3 cluster with 2 namespaces and install casskop
k3d cluster create multi-casskop-qa
cd $(git rev-parse --show-toplevel)
kubectl  apply -f config/crd/bases/
kubectl  create namespace cluster1
kubectl  create namespace cluster2
helm install casskop -n cluster1 oci://ghcr.io/cscetbon/casskop-helm --set debug.enabled=true
helm install casskop -n cluster2 oci://ghcr.io/cscetbon/casskop-helm --set debug.enabled=true
kubemcsa export --context=k3d-multi-casskop-qa casskop --as k8s-cluster2 -n cluster1 | k apply -n cluster1 -f -
  • Update generated secret to use server: https://kubernetes.default.svc/ in its config (We won't need that method anymore and will be able to create 2 different clusters when https://github.com/rancher/k3d/issues/101 is solved)
kubectl  get secrets -n cluster1 k8s-cluster2 -o json|jq -r '.data.config'|base64 -d|pbcopy
# set server to https://kubernetes.default.svc in the output, then copy it in the clipboard and run
pbpaste|base64 -w10000|pbcopy
# You now have to edit the secret and replace config's value by what you have in your clipboard
  • load the docker image you built in the first step into your k3d cluster
k3d image import ghcr.io/cscetbon/multi-casskop:0.5.6-my-pr -c multi-casskop-qa
INFO[0000] Loading images into 'multi-casskop-qa'
INFO[0000] Starting k3d-tools node...
INFO[0000] Saving 1 image(s) from runtime...
INFO[0004] Importing images into nodes...
INFO[0004] Importing images from tarball '/k3d/images/k3d-multi-casskop-qa-images-20200929124019.tar' into node 'k3d-multi-casskop-qa-server-0'...
INFO[0006] Removing the tarball(s) from image volume...
INFO[0007] Removing k3d-tools node...
INFO[0008] Deleted k3d-multi-casskop-qa-tools
INFO[0008] Successfully imported image(s)
INFO[0008] DONE
  • Install multi-casskop using the image you just imported
helm install multi-casskop oci://ghcr.io/cscetbon/multi-casskop-helm --set k8s.local=k3d-multi-casskop-qa \
    --set k8s.remote={k8s-cluster2} --set image.tag=0.5.6-my-pr --set debug.enabled=true -n cluster1 \
    --set image.pullPolicy=IfNotPresent

How this repository was initially build​

Boilerplate CassKop​

We used the SDK to create the repository layout. This command is for memory ;) (or for applying sdk upgrades)

You need to have first install the SDK.

#old version
 operator-sdk new casskop --api-version=db.orange.com/v2 --kind=CassandraCluster
#new version
operator-sdk new casskop --dep-manager=modules --repo=github.com.cscetbon/casskop --type=go

Then you want to add managers:

# Add a new API for the custom resource CassandraCluster
$ operator-sdk add api --api-version=db.orange.com/v2 --kind=CassandraCluster

# Add a new controller that watches for CassandraCluster
$ operator-sdk add controller --api-version=db.orange.com/v2 --kind=CassandraCluster

Useful Infos for developers​

Parsing Yaml from String​

For parsing Yaml from string to Go Object we uses this library : github.com/ghodss/yaml because with the official one not all fields of the yaml where correctly populated. I don't know why..