CI/CD Project

Background

In this project you will design and implement CI/CD pipelines for the Object Detection Service, to automate the deployment process for development and production environments.

Infrastructure

Since your service will be deployed in both development and production environments, you have to provision infrastructure for both envs:

• 2 telegram tokens - for dev and prod bots.
• Separate resources in AWS for dev and prod: S3, SQS, DynamoDB, Secret Manager, etc...
• Both dev and prod services would be deployed in the same K8S cluster (in different namespaces).

CI server

Use the Jenkins server we've setup in class, or create a new one if needed (either in k8s, EC2, or any other way). The below guidelines are related to Jenkins, but you can also use any other CI platform (e.g. GitHub Actions, Azure DevOps).

• In your Jenkins server, create dev and prod folders (New Item -> Folder). Production related pipelines and credentials would be created in prod folder, similarly, development related pipelines are in dev folder.
• All pipelines should be running on a containerized agent (the same Docker image for all pipelines). Use the Dockerfile we've presented in class as a baseline.
• No need to run agents on different nodes! All pipelines can be running on the Jenkins server itself.

Build pipelines

Design and implement build pipelines for the polybot and yolo5 services:

• You can simply create 4 Jenkinsfiles as detailed below. Alternatively, if you want to reduce code duplication, use Jenkins shared library.

.
├── yolo5-build-prod.Jenkinsfile
├── yolo5-build-dev.Jenkinsfile
├── polybot-build-prod.Jenkinsfile
├── polybot-build-dev.Jenkinsfile

• Both production and development pipelines should be triggered automatically upon push event.
• Make sure to build only services that their source code have been changed. If you changed the polybot code and pushed it, only the build pipeline of the polybot should be triggered.
• Each new built Docker image should have a different image tag.
• Build pipeline for development environment should be triggered upon changes in branch dev.
• Build pipeline for production environment should be triggered upon changes in branch main.

Deployment to Kubernetes using ArgoCD

ArgoCD is a declarative continuous delivery tool for Kubernetes.

Argo automatically detects changes done in YAML manifests in your GitHub repo, and sync the cluster accordingly. For example, if you change an image tag in a YAML manifest, and commit and push it, Argo can automatically deploy the new version into your cluster (instead of executing the kubectl apply command). This pattern is called GitOps, which means, using Git repositories as the source of truth for defining the desired application state.

• In your K8S cluster, create dev and prod namespaces.

• In your project repo, create YAML manifests for both development and production environments. For example:

  .
  ├── k8s
  │   ├── dev
  │   │   ├── yolo5.yaml
  │   │   └── polybot.yaml
  │   └── prod
  │       ├── yolo5.yaml
  │       └── polybot.yaml
  

  • This is only a suggestion. Feel free to change to any other files layout that works for you.
  • You can use Helm if you want to reduce code duplications.

• Install Argo in your cluster by:

  kubectl create namespace argocd
  kubectl apply -n argocd -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml

  You can access the server by port-forwarding the argocd-server service in argocd namespace.

• Create the below "release" pipelines that are triggered after a successful running of the build pipeline:

  .
  ├── releases-dev.Jenkinsfile
  ├── releases-prod.Jenkinsfile
  

  • The release pipeline receives the new image tag from the build pipeline (similarly as done in class).
  • The pipelines will commit and push the new version of your YAML manifests to the releases branch of your Git repository. Once the new version of the YAML manifests is pushed, ArgoCD automatically detects the changes and initiates the deployment process.

• In your Argo server, create a new application (+ New App) for the polybot and yolo5 for both development and production envs:

  • The apps should use the releases branch as a source for YAML manifests.
  • Deployment to dev namespace should be done automatically.
  • Deployment to prod should be done manually using the Sync button in Argo.

Pull Request testing pipelines (optional)

• Protect branch main so changes can be merged via a PR only.

• Create PR testing pipeline in Jenkins (only for prod, no need to test dev deployments):

  .
  ├── pr-testing.Jenkinsfile
  

• In the pipeline, implement some test, it could be some unittest, a linting check, or any other minimal test.

• Make sure PR merging is blocked when the PR testing pipeline is failed.

Trying it all together

1. From main branch, create a feature branch.
2. Commit some changes.
3. Merge your feature branch into dev, push, wait for Argo to deploy the change to development bot.
4. Create a PR from your feature branch into main, make sure you pass the PR testing pipeline.
5. Merge the PR. After a new version is ready to be released, deploy it manually to production bot.