MLOps on Kubernetes.md

Deploy a Simple MLOps Application to Kubernetes

Requirements

System Requirements

1. macOS, Linux, or Windows (with WSL2 recommended for Windows)

Software Requirements

1. Python 3.8+. For model training and serving (FastAPI)
2. pip. For installing Python packages
3. Docker. For building and pushing container images
4. Minikube

Python Package Requirements

fastapi
uvicorn
scikit-learn
joblib
numpy
pydantic

(These are installed in the Dockerfile.)

Cloud/Registry Requirements

1. Docker Hub username and password/token for image push/pull

Kubernetes Manifests

deployment.yaml
service.yaml

Optional:

1. curl or Postman for testing the API endpoint

Implementation

Prepare Your App

1. Create a directory and files

mkdir ml-demo
cd ml-demo
touch train.py app.py Dockerfile

2. Edit Model Training Script (train.py)

# train.py
from sklearn.datasets import load_iris
from sklearn.ensemble import RandomForestClassifier
import joblib

iris = load_iris()
X, y = iris.data, iris.target
clf = RandomForestClassifier()
clf.fit(X, y)
joblib.dump(clf, "model.joblib")

3. Edit Model Serving App (app.py)

# app.py
from fastapi import FastAPI
from pydantic import BaseModel
import joblib
import numpy as np

app = FastAPI()
model = joblib.load("model.joblib")

class IrisInput(BaseModel):
    data: list

@app.post("/predict")
def predict(input: IrisInput):
    arr = np.array(input.data).reshape(1, -1)
    pred = model.predict(arr)
    return {"prediction": int(pred[0])}

3. Edit Dockefile

FROM python:3.10-slim

WORKDIR /app
COPY app.py train.py ./
RUN pip install fastapi uvicorn scikit-learn joblib

# Train the model at build time (for demo)
RUN python train.py

EXPOSE 8000
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"]

Build and Push Docker Image

1. Build the image locally

docker build -t <your-dockerhub-username>/mlops-demo:latest .

2. Push to Docker Hub

docker login
docker push <your-dockerhub-username>/mlops-demo:latest

Install Minikube and kubectl

1. Install Minikube. Please refer to here.

2. Install kubectl. Please refer to here.

3. Start Minikube

minikube start

4. 5. Create Kubernetes Manifests (deployment.yaml)

apiVersion: apps/v1
kind: Deployment
metadata:
  name: mlops-demo
spec:
  replicas: 1
  selector:
    matchLabels:
      app: mlops-demo
  template:
    metadata:
      labels:
        app: mlops-demo
    spec:
      containers:
      - name: mlops-demo
        image: <your-dockerhub-username>/mlops-demo:latest
        ports:
        - containerPort: 8000

2. Edit services.yaml

apiVersion: v1
kind: Service
metadata:
  name: mlops-demo
spec:
  type: NodePort
  selector:
    app: mlops-demo
  ports:
    - protocol: TCP
      port: 80
      targetPort: 8000

6. Deploy to Minikube

kubectl apply -f deployment.yaml
kubectl apply -f service.yaml

7. Access the App

minikube service mlops-demo --url

8. Test the Endpoint. You’ll get a URL like http://127.0.0.1:XXXXX.

curl -X POST -H "Content-Type: application/json" \
  -d '{"data": [5.1, 3.5, 1.4, 0.2]}' \
  http://127.0.0.1:XXXXX/predict