Django Rest Framework Throttling with API Key | Ddosify: "Canva" of Observability

Introduction

Django Rest Framework Throttling with API Key Preview Image

In this article, we will write an endpoint using the Django REST framework with API Key throttling (rate limiting). We will first use Django default cache configuration Local Memory Caching for throttling and we will see that this method does not work for multiple workers on gunicorn or multiple replicas on Kubernetes. We will change the Django cache configuration to Redis to solve this problem. We will test throttling with the open-source load testing tool Ddosify.

From Django Docs: Each process will have its own private cache instance, which means no cross-process caching is possible. This also means the local memory cache isn’t particularly memory-efficient, so it’s probably not a good choice for production environments. It’s nice for development.

Prerequisites

Python3
Docker for Redis installation
Docker Desktop for single-node Kubernetes
kubectl for accesing the Kubernetes resources
cURL for API testing
Ddosify CLI or Ddosify Docker Extension for load testing

Create a Project

In this section, we will create a Django application with rate limiting based on API key header.

Create Django Environment

Let's create an python environment env and activate it.

Terminal
python3 -m venv env
source env/bin/activate

Create requirements.txt

Django==4.0.7
djangorestframework==3.13.1
gunicorn==20.1.0
redis==4.3.4

Install Django Dependencies

Terminal
pip3 install -r requirements.txt

Create Django Project and Application

Let's create an Django project throttling and application core.

Terminal
django-admin startproject throttling
cd throttling
python3 manage.py startapp core

Update Django Settings

Add rest_framework into INSTALLED_APPS list.
Add REST_FRAMEWORK dictionary configuration.

throttling/settings.py
INSTALLED_APPS = [
    'django.contrib.admin',
    'django.contrib.auth',
    'django.contrib.contenttypes',
    'django.contrib.sessions',
    'django.contrib.messages',
    'django.contrib.staticfiles',

    'rest_framework',
]

REST_FRAMEWORK = {
    'DEFAULT_RENDERER_CLASSES': (
        'rest_framework.renderers.JSONRenderer',
    ),
}

Create a custom throttling class

ConcurrencyThrottleApiKey class uses SimpleRateThrottle from Django REST framework. Last request information for throttling is stored in the default Django cache. For more: Django Docs.

Allow one request per second (1/s).
Use api_key from request url query parameters for the cache key. So requests with the same api_key query parameters will be rate limited.

core/throttling.py
from rest_framework.throttling import SimpleRateThrottle

class ConcurrencyThrottleApiKey(SimpleRateThrottle):
    rate = "1/s"

    def get_cache_key(self, request, view):
        return request.query_params['api_key']

info

You will change this cache implementation later in this article.

Create a basic endpoint

We will create a basic GenericAPIView DjangoThrottlingAPIView endpoint with GET method. This endpoint uses throttling.ConcurrencyThrottleApiKey throttling class for rate limiting.

note

If the request is valid and not throttled, the response will be HTTP 200 OK
If the request is valid and throttled, the response will be HTTP 429 Too Many Requests

core/views.py
from rest_framework import status
from rest_framework.generics import GenericAPIView
from rest_framework.response import Response

from core import throttling


class DjangoThrottlingAPIView(GenericAPIView):
    throttle_classes = [throttling.ConcurrencyThrottleApiKey]

    def get(self, request):
        return Response("ok", status=status.HTTP_200_OK)

Update Endpoint URLs

Update Django application URLs.

core/urls.py
from django.urls import path
import core.views as core_views

urlpatterns = [
    path('', core_views.DjangoThrottlingAPIView.as_view(), name="throtling"),
]

Update Django root URLs.

throttling/urls.py
from django.urls import path

urlpatterns = [
    path('', include('core.urls')),
]

Run gunicorn HTTP Server with 1 worker

Terminal
gunicorn --workers 1 --bind 0.0.0.0:9018 throttling.wsgi

Test Endpoint with cURL

Let's test the endpoint http://localhost:9018/?api_key=1234 with cURL.

tip

We can use random api_key query parameter for the URL. We used 1234 as the api_key in this example. Throttling is determined by the api_key query parameter.

Terminal
curl http://localhost:9018/?api_key=1234

We should see ok in the response of the request.

Load Test Endpoint with Ddosify

Make a load test with Ddosify by sending 100 requests for 10 seconds (10 requests per second).

Terminal
ddosify -t http://localhost:9018/?api_key=1234 -n 100 -d 10

info

Ddosify output shows that 90 requests are throttled, and 10 requests are not throttled, which we expect.

Output
⚙️  Initializing...
🔥 Engine fired.

...

Status Code (Message) :Count
  200 (OK)                   :10
  429 (Too Many Requests)    :90

Run gunicorn HTTP Server with 6 workers

Let's increase the gunicorn worker size from 1 to 6 to be able to serve more users.

Terminal
gunicorn --workers 6 --bind 0.0.0.0:9018 throttling.wsgi

Load Test Endpoint with Ddosify

Let's test it again with Ddosify by sending 100 requests for 10 seconds (10 requests per second).

Terminal
ddosify -t http://localhost:9018/?api_key=1234 -n 100 -d 10

danger

Ddosify output shows that 65 requests are throttled, and 35 requests are not throttled, which we do not expect. So why is that? Maybe we can solve this problem by deploying to Kubernetes with 6 replicas and run gunicorn with 1 worker. Let's try.

Output
⚙️  Initializing...
🔥 Engine fired.

...

Status Code (Message) :Count
  200 (OK)                   :35
  429 (Too Many Requests)    :65

Deploy Django Application to Kubernetes

In order to deploy the Django application to Kubernetes, we need to dockerize the application. Create a Dockerfile and entrypoint.sh on the root of the project.

Create Docker Image

entrypoint.sh
#!/bin/bash
cd throttling && gunicorn --workers 1 --bind 0.0.0.0:9018 throttling.wsgi

tip

Make sure to give executable permission to `entrypoint.sh``

Terminal

chmod +x entrypoint.sh

Create Dockerfile with python:3.9.10-slim base.

Dockerfile
FROM python:3.9.10-slim

COPY ./requirements.txt .
RUN pip --no-cache-dir install -r requirements.txt

WORKDIR /workspace
COPY . /workspace

CMD ["/workspace/entrypoint.sh"]

Build the Dockerfile.

Terminal
docker build -t django_throttling .

Run the docker image django_throttling.

Terminal
docker run -it --rm -p 9018:9018 django_throttling

Test the application with cURL.

Terminal
curl http://localhost:9018/?api_key=1234

We should see ok in the response of the request. The docker application is ready. Next step is the deploy the image to Kubernetes.

Create Kubernetes Cluster

info

If you have already installed Kubernetes, you can skip this step.

We will use Docker Desktop to enable single-node Kubernetes cluster.

Docker Desktop Kubernetes

Test Kubernetes with kubectl:

Terminal

kubectl get nodes

We should see the docker-desktop node with Ready status.

Output

NAME             STATUS   ROLES           AGE   VERSION
docker-desktop   Ready    control-plane   99m   v1.24.2

Create Deployment and Service on Kubernetes

We will deploy django application into Kubernetes with 6 replicas.

k8s_django.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: throttling-deployment
  labels:
    app: throttling
spec:
  replicas: 6
  selector:
    matchLabels:
      app: throttling
  template:
    metadata:
      labels:
        app: throttling
        appType: ddosify
    spec:
      containers:
        - name: throttling
          image: django_throttling:latest
          imagePullPolicy: Never
          ports:
            - containerPort: 9018
              name: throttling-port
---
apiVersion: v1
kind: Service
metadata:
  name: throttling-service
spec:
  type: NodePort
  selector:
    app: throttling
  ports:
    - protocol: TCP
      port: 9018
      targetPort: 9018
      nodePort: 31534

Apply the yaml file.

Terminal

kubectl apply -f k8s_django.yaml

Show the pods.

Terminal

kubectl get pods

We should see the pods with Running status. Since we set the replicas to 6 in the yaml file, there should be 6 pods.

Output
NAME                                     READY   STATUS    RESTARTS   AGE
throttling-deployment-6d545fdd88-2c5s2   1/1     Running   0          107s
throttling-deployment-6d545fdd88-5jb9d   1/1     Running   0          107s
throttling-deployment-6d545fdd88-8l48d   1/1     Running   0          107s
throttling-deployment-6d545fdd88-fmpst   1/1     Running   0          107s
throttling-deployment-6d545fdd88-g2d6b   1/1     Running   0          107s
throttling-deployment-6d545fdd88-pskjc   1/1     Running   0          107s

Show the services.

Terminal

kubectl get svc

We should see the service throttling-service with NodePort type.

Output
NAME                 TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)          AGE
kubernetes           ClusterIP   10.96.0.1       <none>        443/TCP          115m
throttling-service   NodePort    10.109.83.155   <none>        9018:31534/TCP   89m

The service is accesible from http://localhost:31534/?api_key=1234.

Let's test the endpoint with cURL.

Terminal
curl http://localhost:31534/?api_key=1234

We should see ok in the response of the request.

Load Test Endpoint with Ddosify

Let's test the endpoint with Ddosify by sending 100 requests for 10 seconds (10 requests per second).

Terminal
ddosify -t http://localhost:31534/?api_key=1234 -n 100 -d 10

Output
⚙️  Initializing...
🔥 Engine fired.

...

Status Code (Message) :Count
  200 (OK)                   :39
  429 (Too Many Requests)    :61

danger

Ddosify output shows that 61 requests are throttled, and 39 requests are not throttled, which we do not expect. Since all the pods run a separate Django application, the default Django caches are not global. You need a global cache for throttling.

Change the Throttling Cache

We need a global cache for consistent throttling. Deploy a redis to Kubernetes with 1 replica.

k8s_redis.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: redis-deployment
  labels:
    app: redis
spec:
  replicas: 1
  selector:
    matchLabels:
      app: redis
  template:
    metadata:
      labels:
        app: redis
        appType: ddosify
    spec:
      containers:
        - name: redis
          image: redis:6.2.7
          imagePullPolicy: Always
          ports:
            - containerPort: 6379
              name: redis-port
---
apiVersion: v1
kind: Service
metadata:
  name: redis-service
spec:
  type: NodePort
  selector:
    app: redis
  ports:
    - protocol: TCP
      port: 6379
      targetPort: 6379
      nodePort: 31514

Apply the yaml file.

Terminal

kubectl apply -f k8s_redis.yaml

Update Django Settings

throttling/settings.py
CACHES = {
    'alternate': {
        'BACKEND': 'django.core.cache.backends.redis.RedisCache',
        'LOCATION': f'redis://redis-service:6379',
    },
}

Update the custom throttling class

Change the default Django cache to Redis cache.

core/throttling.py
from rest_framework.throttling import SimpleRateThrottle
from django.core.cache import caches

class ConcurrencyThrottleApiKey(SimpleRateThrottle):
    cache = caches['alternate']
    rate = "1/s"

    def get_cache_key(self, request, view):
        return request.query_params['api_key']

Let's deploy to Kubernetes with 6 gunicorn workers. Update workers to 6 in entrypoint.sh.

entrypoint.sh
#!/bin/bash
cd throttling && gunicorn --workers 6 --bind 0.0.0.0:9018 throttling.wsgi

Build the Dockerfile.

Terminal
docker build -t django_throttling .

Delete and apply the yaml file.

Terminal
kubectl delete -f k8s_django.yaml && kubectl apply -f k8s_django.yaml

Load Test Endpoint with Ddosify

Let's test the endpoint with Ddosify by sending 100 requests for 10 seconds (10 requests per second).

Terminal
ddosify -t http://localhost:31534/?api_key=1234 -n 100 -d 10

Output
⚙️  Initializing...
🔥 Engine fired.

...

Status Code (Message) :Count
  200 (OK)                   :10
  429 (Too Many Requests)    :90

info

Ddosify output shows that 90 requests are throttled, and 10 requests are not throttled, which we expect. The results show that the global Redis cache solved the problem.

Conclusion

In this article, we built a simple Django application with throttling using API key. First, the default Django cache is used for throttling. If we use more than 1 worker on gunicorn or more than 1 replica on Kubernetes deployment, the throttling results were unexpected. Then a global cache with Redis is used and tested again with multiple workers on gunicorn and multiple replicas on Kubernetes deployment. The global Redis cache solved the problem.

Share on social media:

Introduction​

Prerequisites​

Create a Project​

Create Django Environment​

Create requirements.txt​

Install Django Dependencies​

Create Django Project and Application​

Update Django Settings​

Create a custom throttling class​

Create a basic endpoint​

Update Endpoint URLs​

Run gunicorn HTTP Server with 1 worker​

Test Endpoint with cURL​

Load Test Endpoint with Ddosify​

Run gunicorn HTTP Server with 6 workers​

Load Test Endpoint with Ddosify​

Deploy Django Application to Kubernetes​

Create Docker Image​

Create Kubernetes Cluster​

Create Deployment and Service on Kubernetes​

Load Test Endpoint with Ddosify​

Change the Throttling Cache​

Update Django Settings​

Update the custom throttling class​

Load Test Endpoint with Ddosify​

Conclusion​

Introduction

Prerequisites

Create a Project

Create Django Environment

Create requirements.txt

Install Django Dependencies

Create Django Project and Application

Update Django Settings

Create a custom throttling class

Create a basic endpoint

Update Endpoint URLs

Run gunicorn HTTP Server with 1 worker

Test Endpoint with cURL

Load Test Endpoint with Ddosify

Run gunicorn HTTP Server with 6 workers

Load Test Endpoint with Ddosify

Deploy Django Application to Kubernetes

Create Docker Image

Create Kubernetes Cluster

Create Deployment and Service on Kubernetes

Load Test Endpoint with Ddosify

Change the Throttling Cache

Update Django Settings

Update the custom throttling class

Load Test Endpoint with Ddosify

Conclusion