Connexion Example REST Service with Redis Store

This example application implements a very basic "pet shop" REST service using the Connexion Python library. Compared to hjacobs/connexion-example, this setup focuses more on operational aspects and includes Redis as a persistent storage and Kubernetes deployment manifests.

QUICKSTART: Please follow the steps to deploy to Kubernetes (Minikube).

This example should demonstrate:

• how to map an OpenAPI/Swagger specification to Python code with Connexion
• how to run Connexion with gevent
• how to build a Docker image for Connexion
• how to integrate a simple persistent database (Redis)
• how to set up Kubernetes manifests for deployment and making sure rolling updates do not cause downtime
• how to configure some best practices for production usage

This example is NOT:

• a real world application (unless you need a REST service to store pets)
• a production-ready service
• a best practice setup for a persistent database (Redis has its own issues, but is used here to keep the database part simple)

DISCLAIMER: While trying to show some best practices for production usage, this is still just an example project --- especially the Redis database setup is not highly available (single replica).

Local Development

This requires Pipenv:

pipenv install --dev
pipenv shell
docker run -d --name connexion-example-redis -p 6379:6379 redis:4-alpine
./app.py
xdg-open http://localhost:8080/ui/

Deploying to Kubernetes (Minikube)

First install and then start Minikube:

minikube start

Build the Docker image:

eval $(minikube docker-env)
docker build -t connexion-example:local-minikube .

Deploy to Kubernetes:

kubectl apply -f deploy/

Wait for pods to come up:

kubectl get pod

Create a pet:

url=$(minikube service connexion-example --url)
curl -X PUT $url/pets/susie -d '{"animal_type": "cat", "name": "Susie", "tags": {"color": "black"}}' -H Content-Type:application/json

Get all pets:

curl $url/pets

Simple load test with Vegeta:

echo "GET $url/pets" | vegeta attack -rate 100 -duration 60s | vegeta report

The output should look something like this (depends on your hardware/VM configuration):

Requests      [total, rate]            6000, 100.02
Duration      [total, attack, wait]    1m0.034341468s, 59.98999991s, 44.341558ms
Latencies     [mean, 50, 95, 99, max]  41.344672ms, 46.279545ms, 52.171431ms, 54.320335ms, 72.515124ms
Bytes In      [total, mean]            3156000, 526.00
Bytes Out     [total, mean]            0, 0.00
Success       [ratio]                  100.00%
Status Codes  [code:count]             200:6000
Error Set:

High Availability

While the setup gracefully handles rolling deployments (try it out by changing a pod label in deploy/deployment.yaml and doing kubectl apply -f deploy/), it cannot handle Redis downtimes without causing HTTP errors. Deleting the Redis pod causes a ~10 seconds unavailability for the REST service:

The above latency cap at 5000ms reflects the configured Redis socket timeout of 5 seconds.