pikkaay / mlops

MLOPS - ML Models in production

Deploying ML models at scale in kubernetes cluster

Includes

1. Simple Flask application which serves http request
   
2. GET method for Health check
   
3. Dockerfile to build docker image
   
4. Configuration yaml files for kubernetes deployments
   

Usage

Working directory ./mlops/src/

docker build ti appname:latest .
docker run -d -p 5000:5000 appname:latest

Test the API using curl

curl localhost:5000/hcheck 

This will output json result with app status

Use the yaml configurations inside to deploy the app to a kubernetes cluster

Apply the deployments with kustomization.yaml using this command

kubectl apply --kustomize=./kubernetes-yaml-configs/ --record=true

Get the status of deployments using this command

kubectl get deployment -n mlops

Get the external ip address using this command

kubectl get service -n mlops

Output will look like this

NAME        TYPE           CLUSTER-IP    EXTERNAL-IP     PORT(S)          AGE
flask-app   LoadBalancer   xx.xx.xx.xx   xx.xx.xx.xx     5000:xxxx/TCP    20s

In the flask app (main.py), add a POST method to get prediction from a real ML model.
Change the kubernetese yaml configuration files to A/B test the model versions and to play with additional nodes and pods as well
Test the latency and load balancer with JMeter