Skip to content

Deploying Django to GCP on k8s using CNC

You've got many options for flavors to deploy with cnc. For many apps on GCP, we recommend Cloud Run due to the low cost of getting started. run-lite flavor will be as cheap as possible, free for many apps. And the run flavor adds a hybrid approach of using some k8s services using GKE Autopilot as well as other enterprise best practices. For bigger teams who have made k8s investments or who want to use k8s, cnc also offers the GKE flavor which uses GKE Autopilot to offer a lot of customization and integration options. This example is illustrative of other languages/frameworks as well, e.g. rails will be similar. - In this example we are also going to customize the deployment.yml and the deploy script to call a custom webhook before deploying.

Before starting, follow the steps at Getting Started. In particular: - cnc installed with pip install cocnc. cnc uses python3 so in some environments you'd run pip3 install cocnc. - terraform installed (see here). - gcloud installed (docs) and have run gcloud auth application-default login (unless you have an alternative auth setup for gcloud). - kubectl installed, possible with gcloud components or using an alternative method. Read more here.

repo setup

Following the instructions at the django tutorial we have set up our app. We're setting up a complex full-stack API that: - uses a Cloud SQL postgres instance - uses redis as a task broker for rq. You mght use celery or another consumer as well in place of this, as well as being able to use a hosted redis like upstash or another broker like RabbitMQ or Kafka as well if those were part of your stack - has a celery worker running as its own k8s deployment

Here's an example repo for this setup: see the repo at github.

As indicated in the tutorial, we update settings with the right configuration to talk to our database. For django with postgres this looks like:

DATABASES = {
    "default": {
        "ENGINE": "django.db.backends.postgresql",
        "NAME": os.environ.get("DB_NAME"),
        "USER": os.environ.get("DB_USER"),
        "PASSWORD": os.environ.get("DB_PASSWORD"),
        "HOST": os.environ.get("DB_HOST"),
        "PORT": "5432",
    }
}

The environment variables refereced will be automatically populated by cnc based on the configuration's resources. In your app, you'd want to handle setting these in development to the right values, having defaults that work locally when these are not set (e.g. in dev), or using an alternate settings.py with different values.

You'll want to add allowed hosts, this can be complex in the cloud due to managed load balancers using their IP as their host header in many cases. There are a variety of solutions to this problem, for more info read this post on SO. For the purpose of this demo, we are going to have you set ALLOWED_HOSTS to * but in production we are sure you will find a better solution.

The repo has a rq worker in tasks.py. You will likely have a few similar processes, maybe using another similar library, and can modify the commands in cnc.yml as needed. Each worker will get its own independent deployment in k8s, with autoscaling and resource allocation able to be overridden as needed.

cnc.yml

You add this to cnc.yml. Read more about the file format here.

services:
  app:
    command: "python manage.py runserver 0.0.0.0:$PORT"
    x-cnc:
      type: backend
      workers:
      - name: default-queue-worker
        # REDIS_URL will be populated by the redis service below by CNC
        command: "rq worker --with-scheduler --url $REDIS_URL"
        system:
          cpus: 1
          memory: 1G
        replicas: 1
    build:
      context: .
  db:
    x-cnc:
      type: database
      version: 15
    image: postgres
  redis:
    x-cnc:
      type: cache
    image: redis

As you can see, for each worker you can set minimum replicas, cpus and memory which is useful if you've got different task resource requirements.

If needed, you can also customize the k8s deployments for the worker and the api server. as much as you'd like, following the AWS example here but using the gke flavor and appropriate filenames instead. You can add more yml objects into those templates as needed, if required.

environments.yml

You add this to environments.yml. Read more about the file format here.

# name this whatever you want
name: django-app
provider: gcp
flavor: gke
version: 1

collections:
# eventually you would likely add a "prod" collection in another AWS account as well by adding another element here
- name: dev
  region: us-east-1
  base_domain: dev.api.mycoolapp.ai
  account_id: "theta-era-421317"
  environments:
  - name: staging
    environment_variables:
    - name: FOO
      value: bar

See more about available options including environment variables in configuration.

add requirements file

You can either write your own Dockerfile and supply it under the build as dockerfile in the cnc.yml or you can use the built in support for nixpacks. As per nixpacks docs we are going to add a requirements.txt to configure as python build. Add the following to requirements.txt:

django
psycopg2
redis
rq
requests

Provision your infra

  • Run cnc provision apply. confirm with yes when requested by terraform, and then wait for the infra to provision.
  • Run cnc info environments and set DNS for the base_domain above.
  • Get the IP by running cnc info environments and copying the Load Balancer IP into a CNAME in your DNS provider for *.basedomain.com where your base_domain in environments.yml is used.
  • Do this once per collection, and then add as many environments without touching DNS record settings again.
  • Your SSL cert will usually work in 20 mins or so, but can take more time depending on DNS record propagaton. Read more here at GCP.

Deploy your app

  • run cnc update perform staging --service-tag app=v1. you can set v1 to whatever you want to use for the release tag, usually the git SHA is a good choice here, can get that with git rev-parse --short HEAD.
  • the first deploy will take the longest, once this runs once locally building the images will be faster due to docker cache in subsequent runs.
    • 15 mins is normal for first time, 3ish mins after that
    • you can also run this step from CI/CD e.g. github actions once it works for you and you want to automate it

visit the URL and test the worker

  • run cnc info environments again, and visit the URL for your staging environment. you will see the index content.
  • you can test the task worker with a command similar to:
curl -X POST http://your_domain/count/ -H "Content-Type: application/json" -d '{"url": "http://example.com"}'

customize your deploy script

See the docs for deploy customization.

Next steps

  • add another environment
  • customize your terraform or k8s templates
  • add additional services
  • explore the toolbox