Mlflow

# Deployment Guide

Complete guide to deploying MLflow models to production environments.

## Table of Contents
- Deployment Options
- Local Serving
- REST API Serving
- Docker Deployment
- Cloud Deployment
- Batch Inference
- Production Patterns
- Monitoring

## Deployment Options

MLflow supports multiple deployment targets:

| Target | Use Case | Complexity |
|--------|----------|------------|
| **Local Server** | Development, testing | Low |
| **REST API** | Production serving | Medium |
| **Docker** | Containerized deployment | Medium |
| **AWS SageMaker** | Managed AWS deployment | High |
| **Azure ML** | Managed Azure deployment | High |
| **Kubernetes** | Scalable orchestration | High |
| **Batch** | Offline predictions | Low |

## Local Serving

### Serve Model Locally

```bash
# Serve registered model
mlflow models serve -m "models:/product-classifier/Production" -p 5001

# Serve from run
mlflow models serve -m "runs:/abc123/model" -p 5001

# Serve with custom host
mlflow models serve -m "models:/my-model/Production" -h 0.0.0.0 -p 8080

# Serve with workers (for scalability)
mlflow models serve -m "models:/my-model/Production" -p 5001 --workers 4
```

**Output:**
```
Serving model on http://127.0.0.1:5001
```

### Test Local Server

```bash
# Single prediction
curl http://127.0.0.1:5001/invocations \
  -H 'Content-Type: application/json' \
  -d '{
    "inputs": [[1.0, 2.0, 3.0, 4.0]]
  }'

# Batch predictions
curl http://127.0.0.1:5001/invocations \
  -H 'Content-Type: application/json' \
  -d '{
    "inputs": [
      [1.0, 2.0, 3.0, 4.0],
      [5.0, 6.0, 7.0, 8.0]
    ]
  }'

# CSV input
curl http://127.0.0.1:5001/invocations \
  -H 'Content-Type: text/csv' \
  --data-binary @data.csv
```

### Python Client

```python
import requests
import json

url = "http://127.0.0.1:5001/invocations"

data = {
    "inputs": [[1.0, 2.0, 3.0, 4.0]]
}

headers = {"Content-Type": "application/json"}

response = requests.post(url, json=data, headers=headers)
predictions = response.json()

print(predictions)
```

## REST API Serving

### Build Custom Serving API

```python
from flask import Flask, request, jsonify
import mlflow.pyfunc

app = Flask(__name__)

# Load model on startup
model = mlflow.pyfunc.load_model("models:/product-classifier/Production")

@app.route('/predict', methods=['POST'])
def predict():
    """Prediction endpoint."""
    data = request.get_json()
    inputs = data.get('inputs')

    # Make predictions
    predictions = model.predict(inputs)

    return jsonify({
        'predictions': predictions.tolist()
    })

@app.route('/health', methods=['GET'])
def health():
    """Health check endpoint."""
    return jsonify({'status': 'healthy'})

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5001)
```

### FastAPI Serving

```python
from fastapi import FastAPI
from pydantic import BaseModel
import mlflow.pyfunc
import numpy as np

app = FastAPI()

# Load model
model = mlflow.pyfunc.load_model("models:/product-classifier/Production")

class PredictionRequest(BaseModel):
    inputs: list

class PredictionResponse(BaseModel):
    predictions: list

@app.post("/predict", response_model=PredictionResponse)
async def predict(request: PredictionRequest):
    """Make predictions."""
    inputs = np.array(request.inputs)
    predictions = model.predict(inputs)

    return PredictionResponse(predictions=predictions.tolist())

@app.get("/health")
async def health():
    """Health check."""
    return {"status": "healthy"}

# Run with: uvicorn main:app --host 0.0.0.0 --port 5001
```

## Docker Deployment

### Build Docker Image

```bash
# Build Docker image with MLflow
mlflow models build-docker \
  -m "models:/product-classifier/Production" \
  -n product-classifier:v1

# Build with custom image name
mlflow models build-docker \
  -m "runs:/abc123/model" \
  -n my-registry/my-model:latest

# Build and enable MLServer (for KServe/Seldon)
mlflow models build-docker \
  -m "models:/my-model/Production" \
  -n my-mode