Minikube Local Deployment

Deploy MLFlow and FastAPI locally using minikube for testing and development. Note that this infrastructure does NOT include all the components of the MLOps pipeline - it is limited to MLFlow and FastAPI and is meant for practice with Kubernetes before moving to the cloud. deployml only generates the manifest (which then may need to be edited) and then creates the deployment in minikube.

Overview

Minikube provides a local Kubernetes environment for testing and development. It's ideal for:

Local testing
No cloud costs
Fast iteration
Learning Kubernetes

Prerequisites

Install minikube following instructures here.

# Verify installation
minikube version

Example

deployml is only used for generating the manifest and then deploying in minikube. In order to deploy MLFlow and FastAPI in minikube, we will need to create the Docker images for both. The corresponding dockerfiles can be found here.

1. Initialize Minikube for MLflow

# Generate MLflow manifests
deployml mlflow-init \
  --output-dir ./manifests/mlflow \
  --image mlflow-demo:latest \
  --backend-store-uri sqlite:///mlflow.db

# Deploy MLflow
deployml mlflow-deploy --manifest-dir ./manifests/mlflow

2. Initialize Minikube for FastAPI

# Generate FastAPI manifests
deployml minikube-init \
  --output-dir ./manifests/fastapi \
  --image fastapi-mlflow-demo:latest \
  --mlflow-uri http://mlflow-service:5000

# Deploy FastAPI
deployml minikube-deploy --manifest-dir ./manifests/fastapi

3. Access Services

# Get service URLs
minikube service mlflow-service --url
minikube service fastapi-service --url

# Or open in browser
minikube service mlflow-service
minikube service fastapi-service

Two-Step Workflow

Step 1: Generate Manifests

# MLflow
deployml mlflow-init \
  --output-dir ./manifests/mlflow \
  --image mlflow-demo:latest \
  --backend-store-uri sqlite:///mlflow.db

# FastAPI
deployml minikube-init \
  --output-dir ./manifests/fastapi \
  --image fastapi-mlflow-demo:latest \
  --mlflow-uri http://mlflow-service:5000

Step 2: Edit Manifests (Optional)

# Edit deployment.yaml to adjust resources
nano ./manifests/mlflow/deployment.yaml

# Common edits:
# - Adjust CPU/memory requests/limits
# - Change replica count
# - Modify environment variables

Step 3: Deploy

# Deploy MLflow
deployml mlflow-deploy --manifest-dir ./manifests/mlflow

# Deploy FastAPI
deployml minikube-deploy --manifest-dir ./manifests/fastapi

Verify Deployment

# Check pods
kubectl get pods

# Check services
kubectl get svc

# View logs
kubectl logs -l app=mlflow --tail=50 -f
kubectl logs -l app=fastapi --tail=50 -f

Testing

# Get MLflow URL
MLFLOW_URL=$(minikube service mlflow-service --url)

# Test health endpoint
curl $MLFLOW_URL/health

# Get FastAPI URL
FASTAPI_URL=$(minikube service fastapi-service --url)

# Test health endpoint
curl $FASTAPI_URL/health

Cleanup

# Delete deployments
kubectl delete deployment mlflow-deployment fastapi-deployment

# Delete services
kubectl delete service mlflow-service fastapi-service

# Stop minikube
minikube stop

# Delete minikube cluster
minikube delete

Troubleshooting

Pod Not Starting

# Check pod status
kubectl get pods
kubectl describe pod POD_NAME

# Check logs
kubectl logs POD_NAME

Image Not Found

# Load image into minikube
minikube image load mlflow-demo:latest
minikube image load fastapi-mlflow-demo:latest

# Or use minikube's Docker daemon
eval $(minikube docker-env)
docker build -t mlflow-demo:latest .

Service Not Accessible

# Check service status
kubectl get svc

# Use minikube service command
minikube service SERVICE_NAME --url