operator

Operator

This binary is a Kubernetes operator that provides Managed Collection for Google Cloud Prometheus Engine on Kubernetes.

The operator can be run and tested two ways. The first and arguably better way, in lieu of matching actual deployment, is by creating a docker image and pushing it to Google Cloud and telling Kubernetes to fetch and run it. The second is by pushing the configurations to Kubernetes running and instead of Kubernetes running an image, you run the operator locally on your machine.

Running through Google Cloud

As a pre-requisite, ensure that your Kubernetes cluster is setup and you are connected to it via the gcloud CLI. The easiest way is by clicking the Connect button which reveals a command after selecting your cluster in the Google Cloud Console. Ensure that your service account is configured to read images. If you are using the default service account for your Kubernetes node pool, you can find the email via IAM & Admin > Service Accounts and looking for the account that is listed as "Compute Engine default service account".

First, build and push the operator image. In the root directory:

DOCKER_PUSH=1 make operator

Note, that you can configure the Makefile to use certain environment variables, such as PROJECT_ID but these are configured automatically if they are not set. The command will give you the uploaded image URL and update all necessary configurations to use it.

Next, apply the Kubernetes configuration files, starting with the CRDs:

kubectl apply -f cmd/operator/deploy/crds/
kubectl apply -f cmd/operator/deploy/operator/

Finally, wait until the operator starts up. You will see a status of Running for the gmp-operator pod:

kubectl get all -ngmp-system

Run Locally

Deploy all CRDs. In this directory:

kubectl apply -f deploy/crds/

Deploy all of the operator required configurations besides the operator deployment, otherwise you will have an operator deployed in addition to your local one.

kubectl apply -f deploy/operator/00-namespace.yaml
kubectl apply -f deploy/operator/01-priority-class.yaml
kubectl apply -f deploy/operator/02-service-account.yaml
kubectl apply -f deploy/operator/03-role.yaml
kubectl apply -f deploy/operator/04-rolebinding.yaml

Run the operator locally (requires active kubectl context to have all permissions the operator needs):

go run main.go

Finally because the webhooks are configured to the operator apply the rest of the configurations in a separate terminal session:

kubectl apply -f deploy/operator/06-service.yaml
kubectl apply -f deploy/operator/07-operatorconfig.yaml
kubectl apply -f deploy/operator/08-validatingwebhookconfiguration.yaml
kubectl apply -f deploy/operator/09-mutatingwebhookconfiguration.yaml
kubectl apply -f deploy/operator/10-collector.yaml
kubectl apply -f deploy/operator/11-rule-evaluator.yaml

The operator updates the configuration of all collectors after which they start scraping various metric endpoints.

Verify by port-forwarding an arbitrary collector and inspect its UI. You should see various targets being scraped successfully.

kubectl -n gmp-system port-forward --address 0.0.0.0 collector 19090

Go to http://localhost:19090/targets.

Teardown

Simply stop running the operator locally and remove all manifests in the cluster with:

kubectl delete -f deploy/ --recursive