simple-delay-api-k8-auto-scaler

These are two simple APIs which induces some delay for the given number of seconds either by sleep() or by doing some CPU intensive operation. We will deploy these API into Kubernetes and demonstrate how auto scaling works.  
 

Setup

pip install -r requirements.txt

Run locally

python api.py

Docker build & run

docker build -t delay-api:1.0 .
docker run -d -p 8081:8081 delay-api:1.0

OR simply pull from docker hub & run.

docker run -d -p 8081:8081 gireeshp/delay-api:1.0

 

Access the application URL

Hit either of following API with URL parameter 'seconds' - which is the wait you need in number of seconds. If the argument is skipped, default wait will be 10 seconds
http://127.0.0.1:8081/delay?seconds=3
(this induces a CPU intensive wait of 3 seconds)
OR
http://127.0.0.1:8081/sleep?seconds=3
(this induces a simple 3 seconds delay by sleeping)

 
 

Deploying to Kubernetes

This section explains how this application can be deployed to Kubernetes. Using minikube here (see setting up minikube). We are going to create the Deployment, Service, Ingress & a Horizontal Auto Scaler.  

Start minikube

simple-delay-api$ minikube start
simple-delay-api$ minikube addons enable heapster
simple-delay-api$ minikube addons enable metrics-server
simple-delay-api$ kubectl apply -f k8.yaml 

ingress.networking.k8s.io/delay-api created
service/delay-api created
deployment.apps/delay-api created
horizontalpodautoscaler.autoscaling/delay-api created

 

Check the ingress IP.

simple-delay-api$ kubectl get ingress
NAME        HOSTS   ADDRESS         PORTS   AGE
delay-api   *       192.168.64.21   80      2m25s

Now we can access the API over browser using URLs http://192.168.64.21/delay OR http://192.168.64.21/sleep  

Testing the auto-scaling

Use this simulator to simulate parallel hits. Provide inputs needed by the simulator as below.
URL - http://192.168.64.21/delay
Total - 1000
Parallel - 50
 
Check current status before clicking "Start" button

simple-delay-api$ kubectl get hpa
NAME        REFERENCE              TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
delay-api   Deployment/delay-api   0%/25%    1         10        1          17m
simple-delay-api$ kubectl get deploy
NAME        READY   UP-TO-DATE   AVAILABLE   AGE
delay-api   1/1     1            1           17m

Now start simulator & check status after a few seconds

simple-delay-api$ kubectl get hpa
NAME        REFERENCE              TARGETS    MINPODS   MAXPODS   REPLICAS   AGE
delay-api   Deployment/delay-api   201%/25%   1         10        10         20m
simple-delay-api$ kubectl get deploy
NAME        READY   UP-TO-DATE   AVAILABLE   AGE
delay-api   10/10   10           10          20m
simple-delay-api$ kubectl get pods
NAME                        READY   STATUS    RESTARTS   AGE
delay-api-fb78b764b-4xlxl   1/1     Running   0          2m1s
delay-api-fb78b764b-86cf9   1/1     Running   0          2m1s
delay-api-fb78b764b-94s5q   1/1     Running   0          20m
delay-api-fb78b764b-dprgw   1/1     Running   0          2m16s
delay-api-fb78b764b-gqtg5   1/1     Running   0          2m16s
delay-api-fb78b764b-h695t   1/1     Running   0          2m1s
delay-api-fb78b764b-hbf4g   1/1     Running   0          106s
delay-api-fb78b764b-hpmpf   1/1     Running   0          2m1s
delay-api-fb78b764b-hwd7j   1/1     Running   0          106s
delay-api-fb78b764b-kzbs2   1/1     Running   0          2m16s

We can see the deployment is auto scaled as per the load induced by the simulator.