I deployed a Fullstack application (i.e. Frontend / Backend) on Google Cloud Platform with Kubernetes. And since developers are lazy, I automated everything with Gitlab CI.

It’s a far cry from PHP deployed by hand on an Apache server. If you’re interested in moving up a gear and see how it works, read on.

My application

My project I’m currently working on is divided into three parts: Frontend, Backend, Job Queue.

Each part runs in separate containers Docker. So they all have a `Dockerfile’:

  1. the frontend uses a Node.js image that builds Angular 9 and then a NGINX image that serves the static files
  2. the backend uses a Node.js image that launches a web server on port 3000
  3. the job_queue also uses a Node.js image that launches a Node.js script that communicates with the Postgres database and performs offline actions.

So my project has the following structure:

├── backend
├── docker
├── docker-compose.yml
└── frontend

So I wanted to build these images automatically during a git push and publish it on Google Cloud Platform with Kubernetes. For this, I used Gitlab CI but the logic must be identical with Github’s Circle CI.

The Postgres database is independent of Kubernetes and runs on a [Google Cloud SQL for PostgreSQL] service (https://cloud.google.com/sql/docs/postgres). I won’t talk about it here..

Automation with Gitlab CI

Gitlab allows you to define a workflow of things to do when you push code through a .gitlab-ci.yml file. This is often useful for launching a pipeline (a sequence of actions) that will run unit tests, linter code, etc… ` In my case, I created the following actions :

  1. `test’: launch unit tests
  2. publish: creation of a Docker image and publication in the GCloud’s private image registry
  3. deploy: tells the GCloud to deploy the images previously uploaded.

So the structure of the .gitlab-ci.yml file looks like this:

# .gitlab-ci.yml
# ...
stages: [test, publish, deploy]

job:test:backend:
  stage: test
  # ...

job:publish:backend:
  stage: publish
  # ...

job:deploy-develop:backend:
  stage: deploy
  # ...

I only kept the jobs that involve “backend” to simplify.

I’ll walk you through the steps. Here we go.

test: running unit tests

This is the easiest step. I won’t spend too much time on it because a lot of tutorials exist and that’s not the subject of this article.

So we start two jobs:

  1. job:test:backend will initialize the application, connect a Postgres database and run `yarn test’.
job:test:backend:
  stage: test
  services:
    - postgres:12.3-alpine
  variables:
    POSTGRES_DB: database
    POSTGRES_USER: user
    POSTGRES_PASSWORD: password
    POSTGRES_HOST_AUTH_METHOD: trust
  before_script:
    - cd backend
    - cp .example.env .env
    - yarn install
  script: yarn test
  1. job:test:frontend will initialize the application and run yarn test but from an image containing a Chrome driver to emulate navigation.
job:test:frontend:
  image: weboaks/node-karma-protractor-chrome:alpine
  stage: test
  before_script:
    - cd frontend
    - yarn install
  script: yarn test

There you go.

publish: creating a Docker image and publishing it in the private image registry of the GCloud.

The publish step will build docker images of the different applications and publish them to GCloud - Container Registry. To do this, we will create a job for each image to be published (frontend, backend and worker).

In order to build the images and publish them, we need to use:

It is then possible to use the :

  • gcloud auth to connect to Gcloud in the container.
  • gcloud auth configure-docker to connect Docker to GCloud
  • docker buil/push to create the image and publish it

One last thing to know: to connect to the GCloud in Gitlab CI, you must first log in. To do so, you need to add the GCloud credentials in the Gitlab CI settings in the “Settings > Variables” section. Once this is done, it is possible to log into the job using the following command:

echo $GCLOUD_SERVICE_KEY > ${HOME}/gcloud-service-key.json
gcloud auth activate-service-account --key-file ${HOME}/gcloud-service-key.json

Bellow the final result for job:publish:backend :

job:publish:backend:
  stage: publish
  image: google/cloud-sdk:latest
  when: on_success
  services:
    - docker:dind
  tags:
    - dind
  variables:
    DOCKER_HOST: tcp://docker:2375
    DOCKER_TLS_CERTDIR: ""
  before_script:
    - cd backend
    - cp .develop.env .env
    - echo $GCLOUD_SERVICE_KEY > ${HOME}/gcloud-service-key.json
    - gcloud auth activate-service-account --key-file ${HOME}/gcloud-service-key.json
    - gcloud auth configure-docker
  script:
    - docker build --compress -t us.gcr.io/${PROJECT_NAME}/${APP_NAME}-backend:${CI_COMMIT_SHA} .
    - docker push us.gcr.io/${PROJECT_NAME}/${APP_NAME}-backend:${CI_COMMIT_SHA}
    - docker image rm -f us.gcr.io/${PROJECT_NAME}/${APP_NAME}-backend:${CI_COMMIT_SHA}
    - echo "us.gcr.io/${PROJECT_NAME}/${APP_NAME}-backend:${CI_COMMIT_SHA} image build with success and pushed"

I skip over the details of the other two because they’re very similar.

deploy: ask GCloud to deploy the images previously uploaded.

To use kubectl we will also use the google/cloud-sdk image which contains the Kubernetes utility.

But before we start coding the script, we need to create three deployment :

  1. dpl-my-app-frontend on https://my-app/
  2. dpl-my-app-backend at https://api.my-app/.
  3. dpl-my-app-worker that is not exposed

To create them, you can use the GCloud interface or use kubectl directly on your PC (don’t forget to configure cluster access for kubectl).

For the command line version, use kubectl create deployment :

kubectl create deployment dpl-develop-data-frontend --image=us.gcr.io/<your-node>/data-k8s-frontend
kubectl create deployment dpl-develop-data-backend --image=us.gcr.io/<your-node>/data-k8s-backend
kubectl create deployment dpl-develop-data-worker --image=us.gcr.io/<your-node>/data-k8s-worker

We can check that everything went well with kubectl get deployments which should list our three new deployments.

Then, we just have to create the jobs that will update the containers on these deployments and specify the port forwarding.

To do this you also need to use gcloud auth as you saw before. Then, just call the methods :

  • kubectl set image to update the container image.
  • kubectl patch deployment to update container information (e.g. change metadata)

Here is the full version for job:deploy-develop:backend:

job:deploy-develop:backend:
  stage: deploy
  image: google/cloud-sdk:latest
  before_script:
    - echo $GCLOUD_SERVICE_KEY > ${HOME}/gcloud-service-key.json
    - gcloud auth activate-service-account --key-file ${HOME}/gcloud-service-key.json
    - gcloud config set compute/zone us-east1-c
    - gcloud config set project ${GCLOUD_PROJECT_NAME}
    - gcloud container clusters get-credentials ${K8S_CLUSTER_NAME}
    - gcloud auth configure-docker
  script:
    - kubectl set image deployment/${APP_NAME}-backend data-k8s-backend=us.gcr.io/${PROJECT_NAME}/${APP_NAME}-backend:${CI_COMMIT_SHA}
    - kubectl patch deployment ${APP_NAME}-backend -p "{\"spec\":{\"template\":{\"metadata\":{\"annotations\":{\"date\":\"`date +'%s'`\"}}}}}"

There you go.

Then just go to the GCloud interface in the “Kubernetes Engine > Workloads” section, find the corresponding workloads and create the DNS entries on your domain name provider and point them to the corresponding deployments.

Conclusion

For my part I found the Google Cloud Platform approach very confusing but quite well documented. The time invested in learning this technology seems to me well invested because the architecture is really scalable and allows to reduce the infrastructure costs.

The only black point could be that we lock ourselves in a provider but I think that Amazon Web Service or Microsoft Azure share the same terminology because their technology is also based on Kubernetes in my opinion.

See other related posts