johnnybeprogramming / deploy-cloud-app

Provisioning and deployment strategies

Exploring different options and strategies for provisioning, publising and deploying applications to cloud provider(s) such as Google, AWS and Azure, using common tooling and application definitions.

A sample application is also provided, deployed with helm to the target Kubernetes cluster, as a bencmark, to compare different environments.

Provisioning a Kubernetes cluster

For each major cloud provider, we can create a functioning Kubernetes cluster (+ backing registry / storage) with these simple commands (they have been tested with Free Tier accounts on each cloud):

# Create, destroy or select a target cloud provider
# It will remember your last choice, default is `local`

./cloud/select < local | aws | azure | google >
./cloud/create [ local | aws | azure | google ]
./cloud/delete [ local | aws | azure | google ]

# Settings defined in `./config/<target>/cloud.env`

# Use docker for desktop for local development...
./cloud/select local 

We also include a special cloud provider called local, that will use your local docker-desktop (if installed) as the target cluster, instead of creating new resources in a cloud.

The CLI tooling used to provision and configure each of these cloud providers differ substantially. Slight variations in deployment settings are also needed per cloud provider. To overcome these differences, we use a combination of shell scripts and configuration files, defined for each provider in ./(cloud|config)/<provider-name>/**.

We focus on provisioning 3 types of cloud resources:

• Storage - Maps a disk volume to a Pod in K8S, and test disk R/W speeds
• Registry - Define a container registry where we can push release artifacts
• Kubernetes - Create and configure fully operational cluster and link to kubectl

Breakdown of resources per cloud provider

We chose to include all the major cloud providers, and used the offerings for each of them:

• aws - EKS cluster, ECR registry and a EFS filesystem volume
• azure - AKS cluster, ACR registry, AzureDisk for storage
• google - GKE cluster, GCR registry, GCE persisted disk

We also include a special provider to test and deploy locally:

• local - docker-desktop, mapped to disk volume $PWD/data

Building, publishing and deploying the sample app

This repository includes a simple React app with a backing python api that we use to test the CI/CD pipeline steps. The app is packaged into a helm chart, container images are pushed to a registry, and then deployed to Kubernetes.

We include following sample pipeline steps to:

• ./build - container images (either locally or remote)
• ./publish - container images to the target registry
• ./deploy - a release to Kubernetes using helm packages
• ./clean - removes installed version(s) of the helm package

Advanced deployment strategies

We include working examples of both canary deployments as well as blue/green deployments.

This is used to illisstrate the main differences in these release strategies:

• blue - Default: Deploys a stable release, exposed to your end users
• green - Deploy a new version, but it's not visible to end users
• canary - Some of your end users will see this version while it's deployed

# Deploy the sample application to the selected cluster
# Usage: 
#   ./deploy [ <target> ] [ <publish> ]
# Where:
# - <target> is [ blue | green | canary ], defaults to 'blue'.
# - <publish> is [ true | false ], defaults to:
#   + blue => (exposed by default)
#   + green => (hidden by default)
#   + canary => (exposed by default)


# Use Case: Deploy blue/green side by side, then promote green
./deploy                # <-- deploy stable (blue) release
./deploy green true     # <-- promote new (green) release
./deploy blue false     # <-- demote current stable release


# Use Case: Create & test canary deployment, then remove it
./deploy                # <-- deploy stable (blue) release
./deploy canary         # <-- expose canary release to users
# ...at this point, canary release is visible...
./clean canary          # <-- Removes canary release