Cloud Native DevOps on GCP Series Ep2 – Create a CI/CD pipeline with GKE, GCR and Cloud Build

This is the second episode of our Cloud Native DevOps on GCP series. In the previous chapter, we have built a multi-AZ GKE cluster with Terraform. This time, we’ll create a cloud native CI/CD pipeline leveraging our GKE cluster and Google DevOps tools such as Cloud Build and Google Container Registry (GCR). We’ll create a Cloud Build trigger by connecting to GitHub repository to perform automatic build, test and deployment of a sample micro-service app onto the GKE cluster.

For this demo, I have provided a simple NodeJS app which is already containerized and packaged as a Helm Chart for fast K8s deployment. You can find all the artifacts at my GitHub Repo, including the demo app, Helm template/chart, as well as the Cloud Build pipeline code.


  • Access to a GCP testing environment
  • Install Git, Kubectl and Terrafrom on your client
  • Install Docker on your client
  • Install GCloud SDK
  • Check the NTP clock & sync status on your client —> important!
  • Clone or download the demo app repo at here

Step-1: Prepare the GCloud Environment

To begin, configure the GCloud environment variables and authentications.

gcloud init
gcloud config set accessibility/screen_reader true
gcloud auth application-default login

Register GCloud as a Docker credential helper — this is important so our Docker client will have privileged access to interact with GCR. (Later we’ll need to build and push a Helm client image to GCR as required for the pipeline deployment process)

gcloud auth configure-docker

Enable required GCP API services.

gcloud services enable
gcloud services enable
gcloud services enable
gcloud services enable
gcloud services enable

Update Cloud Build service account with an editor role so it will have required permissions to access GKE and GCR within the project.

PROJECT_ID=`gcloud config get-value project`
CLOUDBUILD_SA="$(gcloud projects describe $PROJECT_ID --format 'value(projectNumber)')"
gcloud projects add-iam-policy-binding $PROJECT_ID --member serviceAccount:$CLOUDBUILD_SA --role roles/editor

Step-2: Launch a GKE Cluster using Terraform

If you have been following the series and have already deployed a GKE cluster, you can skip this step and move on to the next. Otherwise you can follow this post to build a GKE cluster with Terraform.

Make sure to deploy an Ingress Controller as there is an Ingress service defined in our Helm Chart!

kubectl apply -f  

Step-3: Initialize Helm for Application Deployment on GKE

As mentioned above, for this demo we have encapsulated our demo app into a Helm Chart. Helm is a package management system designed for simplifying and accelerating application deployment on the Kubernetes platform.

As of version 2, Helm consists of a local client and a Tiller server pod (deployed in K8s cluster) to interact with the Kube-apiserver for app deployment. In our example, we’ll first build a customised Helm client docker image and push it to GCR. This image will then be used by Cloud Build to interact with the Tiller server (deployed on GKE) for deploying the pre-packaged Helm chart — as illustrated in the below diagram.

First let’s configure a service account for Tiller and initialize Helm (server component) on our GKE cluster.

kubectl apply -f ./k8s-helm/tiller.yaml
helm init --history-max 200 --service-account tiller

We’ll then build and push a customised Helm client image to GCR. This might take a few minutes.

cd ./k8s-helm/cloud-builders-community/helm
docker build -t$PROJECT_ID/helm .
docker push$PROJECT_ID/helm

On GCR confirm there is a new Helm (client) image has been pushed through.

Step-4: Review the (Cloud Build) Pipeline Code

Before we move forward, let’s take a moment to review the pipeline code (as defined in the cloudbuild.yaml). There is a total of 4 stages included in our Cloud Build pipeline:

  1. Build a docker image with our demo app
  2. Push the new image to GCR
  3. Deploy Helm chart (for our demo app) to GKE via GCR
  4. Integration Testing

The first two stages are straight forward, we’ll use the Google published Cloud Builder docker image to build the node app image and push it to the GCR repository.

  # Build demo app image
  - name:
      - build
      - -t
      - .
  # Push demo app image to GCR
  - name:
      - push

Next we’ll leverage the (previously built) Helm client to interact with our GKE cluster and to deploy the Helm chart (for our node app), with the image repository pointing to the GCR path from the last pipeline stage.

  # Deploy with Helm Chart
  - name:$PROJECT_ID/helm
      - upgrade
      - -i
      - node-app
      - ./k8s-helm/node-app
      - --set
      - -f
      - ./k8s-helm/node-app/values.yaml
      - KUBECONFIG=/workspace/.kube/config
      - TILLERLESS=false
      - TILLER_NAMESPACE=kube-system

Lastly, we’ll run an integration test to verify the demo app status on our GKE cluster. For our node app there is a built-in heath-check URL configured at “/health“, and we’ll be leveraging another Cloud Builder curl image to ping this URL path and expect a return message of <“status”: “ok”> . Note: here we should be polling the internal DNS address for the k8s service (of the demo app) so there is no dependency on IP allocations.

  # Integration Testing
  - name:
    entrypoint: 'bash'
      - '-c'
      - |
        kubectl delete --wait=true pod curl
        kubectl run curl --restart=Never --generator=run-pod/v1 -- http://node-app.default.svc.cluster.local/health
        sleep 15
        kubectl logs curl 
        kubectl logs curl | grep OK
      - KUBECONFIG=/workspace/.kube/config

Step-4: Create a Cloud Build Trigger by Connecting to GitHub Repository

Now that we have our GKE cluster ready and Helm image pushed to GCR, the next step is to connect Cloud Build to the GitHub repository and create a CI trigger. On GCP console, go to Cloud Build —> Triggers, select the GitHub repo as below.

If this is the first time you are connecting to GitHub in Cloud Build, it will redirect you to an authorization page like below, accept it in order to access your repositories.

Select the demo app repository, which also includes the pipeline config (cloudbuild.yaml) file.

Create a push trigger in the next page and you should see a summary like this.

You can manually run the trigger now to kick off the CI build process. However we’ll be running more thorough testing to verify the end-to-end pipeline automation process in the next section.

Step-5: Test the CI/CD Pipeline

It’s time to test our CI/CD pipeline! First we’ll make a “cosmetic” version change (1.0.0 to 1.0.1) to the Helm chart for our demo app.

Commit the change and push to the Git repository.

This (push event) should have triggered our Cloud Build pipeline. You can jump on the GCP console to monitor the fully automated 4-stage process. The pipeline will be completed once the integration test has returned a status of OK.

On the GKE cluster we can see our Helm chart v-1.0.1 has been deployed successfully.

The deployment and node app are running as expected.

Retrieve the Ingress public IP and update the local host file for a quick testing. (Note the Ingress URL is defined as “node-app.local”)

[root@cloud-ops01 nodejs-cloudbuild-demo]# kubectl get ingresses 
NAME       HOSTS            ADDRESS         PORTS   AGE
node-app   node-app.local   80      15m
[root@cloud-ops01 nodejs-cloudbuild-demo]# 
[root@cloud-ops01 nodejs-cloudbuild-demo]# echo "  node-app.local" >> /etc/hosts   

Now point your browser to “node-app.local” and you should see the demo app page like below. Congrats, you have just successfully deployed a cloud native CI/CD pipeline on GCP!

Cloud Native DevOps on GCP Series Ep1 – Build a GKE Cluster with Terraform

This is the first episode of our Cloud Native DevOps on GCP series. Here we’ll be building an Google Kubernetes Engine (GKE) cluster using Terraform. From my personal experience, GKE has been one of the most scalable and reliable managed Kubernetes solution, and it’s also 100% upstream compliant and certified by CNCF.

For this demo I have provided a sample Terraform script at here. The target state will look like this:

In specific, we’ll be launching the following GCP/GKE resources:

  • 1x new VPC for hosting the demo GKE cluster
  • 1x /17 CIDR block as the primary address space for the VPC
  • 2x /18 CIDR blocks for the GKE Pod and Service address spaces
  • 1x GKE high availability cluster across 2x Availability Zone (AZ)
  • 2x GKE worker instance groups (2x nodes each)


  • Access to a GCP testing environment
  • Install Git, Kubectl and Terrafrom on your client
  • Install GCloud SDK
  • Check the NTP clock & sync status on your client —> important!
  • Clone the Terraform Repo at here

Step-1: Setup the GCloud Environment and Run the Terrafrom Script

To begin, run below interactive GCloud commands to prepare for the GCP environment

gcloud init  
gcloud config set accessibility/screen_reader true  
gcloud auth application-default login  

Remember to update the terraform.tfvars with your own GCP project_id

project_id = "xxxxxxxx"

Make sure to enable the GKE API if not already

gcloud services enable

Now run the Terraform script:

terraform init
terraform apply

The whole process should be taking about 7~10 mins, and you should get an output like this:

Now register the cluster and update kubeconfig file

[root@cloud-ops01 tf-gcp-gke]# gcloud container clusters get-credentials node-pool-cluster-demo --region australia-southeast1
Fetching cluster endpoint and auth data.
kubeconfig entry generated for node-pool-cluster-demo.

Step-2: Verify the GKE Cluster Status

Check that we can access the GKE cluster and there should be 4x worker nodes provisioned.

[root@cloud-ops01 ~]# kubectl get nodes
NAME                                               STATUS   ROLES    AGE     VERSION
gke-node-pool-cluster-demo-pool-01-03a2c598-34lh   Ready    <none>   8m59s   v1.16.9-gke.2
gke-node-pool-cluster-demo-pool-01-03a2c598-tpwq   Ready    <none>   9m      v1.16.9-gke.2
gke-node-pool-cluster-demo-pool-01-e903c7a8-04cf   Ready    <none>   9m5s    v1.16.9-gke.2
gke-node-pool-cluster-demo-pool-01-e903c7a8-0lt8   Ready    <none>   9m5s    v1.16.9-gke.2

This can also been verified on GKE console

The 4x worker nodes are provisioned over 2x managed instance groups across two different AZs

Run kubectl describe nodes and we can see each node has been tagged with a few customised labels based on its unique properties. These are important metadata which can be used for selective Pod/Node deployment and other use cases like affinity or anti-affinity rules.

Step-3: Deploy GKE Add-on Services

  • Install Metrics-Server to provide cluster-wide resource metrics collection and to support use cases such as Horizontal Pod Autoscaling (HPA)
[root@cloud-ops01 tf-gcp-gke]# kubectl apply -f

Wait for a few seconds and we should have resource stats

[root@cloud-ops01 tf-gcp-gke]# kubectl top nodes
NAME                                               CPU(cores)   CPU%   MEMORY(bytes)   MEMORY%   
gke-node-pool-cluster-demo-pool-01-03a2c598-34lh   85m          4%     798Mi           14%       
gke-node-pool-cluster-demo-pool-01-03a2c598-tpwq   300m         15%    816Mi           14%       
gke-node-pool-cluster-demo-pool-01-e903c7a8-04cf   191m         9%     958Mi           16%       
gke-node-pool-cluster-demo-pool-01-e903c7a8-0lt8   102m         5%     795Mi           14%    
  • Next, deploy a NGINX Ingress Controller so we can use L7 URL load balancing and to save cost by reducing the required numbers of external load balances
[root@cloud-ops01 tf-gcp-gke]# kubectl apply -f  

On GCP console we can see that an external Load Balancer has been provisioned in front of the Ingress Controller. Take a note of the LB address at below — this is the public IP that will be consumed by our ingress services.

In addition, we’ll deploy 2x storage classes to provide dynamic persistent storage support for stateful pods and services. Note the different persistent disk (PD) specs (standard & SSD) for different I/O requirements.

 [root@cloud-ops01 tf-gcp-gke]# kubectl create -f ./storage/storageclass/  

Step-4: Deploy Sample Apps onto the GKE Cluster for Testing

  • We’ll first deploy the famous Hipster Shop app, which is a cloud-native microservice application developed by Google.
kubectl apply -f  

wait for all the Pods up and running

[root@cloud-ops01 tf-gcp-gke]# kubectl get pods 
NAME                                     READY   STATUS    RESTARTS   AGE
adservice-687b58699c-fq9x4               1/1     Running   0          2m16s
cartservice-778cffc8f6-dnxmr             1/1     Running   0          2m20s
checkoutservice-98cf4f4c-69fqg           1/1     Running   0          2m26s
currencyservice-c69c86b7c-mz5zv          1/1     Running   0          2m19s
emailservice-5db6c8b59f-jftv7            1/1     Running   0          2m27s
frontend-8d8958c77-s9665                 1/1     Running   0          2m24s
loadgenerator-6bf9fd5bc9-5lsrn           1/1     Running   3          2m19s
paymentservice-698f684cf9-7xbjc          1/1     Running   0          2m22s
productcatalogservice-789c77b8dc-4tk4w   1/1     Running   0          2m21s
recommendationservice-75d7cd8d5c-4x9kl   1/1     Running   0          2m25s
redis-cart-5f59546cdd-8tj8f              1/1     Running   0          2m17s
shippingservice-7d87945947-nhb5x         1/1     Running   0          2m18s

check the external frontend service, you should see a LB has been deployed by GKE with a public IP assigned

[root@cloud-ops01 ~]# kubectl get svc frontend-external 
NAME                TYPE           CLUSTER-IP      EXTERNAL-IP     PORT(S)        AGE
frontend-external   LoadBalancer   80:32408/TCP   5m32s

You should be able to access the app via the LB public IP.

  • Next, we’ll deploy the sample Guestbook app to verify the persistent storage setup.
[root@cloud-ops01 tf-gcp-gke]# kubectl create ns guestbook-app  
[root@cloud-ops01 tf-gcp-gke]# kubectl apply -f ./demo-apps/guestbook/  

The application requests 2x persistent volumes (PV) for the redis-master and redis-slave pods. Both PVs should be automatically provisioned by the persistent volume claims (PVC) with the 2x different storage classes as we deployed earlier. You should see the STATUS reported as “Bound” between each PV and PVC mapping.

Retrieve the external IP/DNS for the frontend service of the Guestbook app.

[root@cloud-ops01 tf-gcp-gke]# kubectl get svc frontend -n guestbook-app 
NAME       TYPE           CLUSTER-IP        EXTERNAL-IP    PORT(S)        AGE
frontend   LoadBalancer   80:31006/TCP   23m

You should be able to access the Guesbook app now. Enter and submit some messages, and try to destroy and redeploy the app, your data will be kept by the redis PVs.

  • Lastly, we’ll deploy a modified version of the yelb app to test the NGINX ingress controller
[root@cloud-ops01 tf-gcp-gke]# kubectl create ns yelb  
[root@cloud-ops01 tf-gcp-gke]# kubectl apply -f ./demo-apps/yelb/

You should see an ingress service deployed as per below.

Retrieve the external IP for the ingress service within the yelb namespace. As mentioned before, this should be the same address of the external LB deployed for the ingress controller.

[root@cloud-ops01 tf-gcp-gke]# kubectl get ingresses -n yelb 
NAME           HOSTS        ADDRESS       PORTS   AGE
yelb-ingress   yelb.local   80      6m47s

Also, notice the ingress URL path is defined as “yelb.local”. This is the DNS entry that will be redirected by the http ingress service. So we’ll update the local host file (with the ingress public IP) for a quick testing.

[root@cloud-ops01 tf-aws-eks]# echo "  yelb.local" >> /etc/hosts  

and that’s it, the incoming requests to “yelb.local” are now routed via the ingress service to the yelb frontend pod running on our GKE cluster.