Prometheus ML Metrics

Overview

Instrumenting ML services with Prometheus metrics. This guide covers practical implementation for production ML systems.

Why This Matters in MLOps

Modern ML systems require rigorous operations practices:

Reliability: Models must perform consistently in production

Reproducibility: Experiments must be tracked and reproducible

Scalability: Systems must handle growing data and traffic

Observability: You need visibility into what models are doing

Setup

bash
Install required tools
pip install prometheus mlflow pandas numpy scikit-learn
Or with Docker
docker pull python:3.11-slim

Core Implementation

python
import os
import json
import logging
from datetime import datetime
from pathlib import Path
logger = logging.getLogger(__name__)
class PrometheusMLMetrics:
    """
    Prometheus ML Metrics implementation.
    
    Handles: monitoring
    Tool: prometheus
    """
    
    def __init__(self, config: dict = None):
        self.config = config or self._default_config()
        self._setup()
    
    def _default_config(self) -> dict:
        return {
            "tool": "prometheus",
            "environment": os.getenv("ENVIRONMENT", "development"),
            "log_level": "INFO",
        }
    
    def _setup(self):
        """Initialize prometheus connection and resources."""
        logging.basicConfig(level=self.config.get("log_level", "INFO"))
        logger.info(f"Initialized Prometheus ML Metrics with config: {self.config}")
    
    def run(self, **kwargs) -> dict:
        """Execute monitoring."""
        start = datetime.utcnow()
        
        try:
            result = self._execute(**kwargs)
            
            elapsed = (datetime.utcnow() - start).total_seconds()
            logger.info(f"Prometheus ML Metrics completed in {elapsed:.2f}s")
            
            return {
                "status": "success",
                "result": result,
                "elapsed_seconds": elapsed
            }
        
        except Exception as e:
            logger.error(f"Prometheus ML Metrics failed: {e}")
            return {
                "status": "failed", 
                "error": str(e)
            }
    
    def _execute(self, **kwargs) -> dict:
        """Core monitoring logic. Override to customize."""
        return {"completed": True, "tool": "prometheus"}
Configuration
config = {
    "tool": "prometheus",
    "tracking_uri": os.getenv("MLFLOW_TRACKING_URI", "http://localhost:5000"),
    "artifact_root": "./artifacts",
}
Initialize
processor = PrometheusMLMetrics(config)
result = processor.run()
print(json.dumps(result, indent=2))

PROMETHEUS Integration

python
Specific prometheus integration for monitoring
import subprocess
def setup_prometheus():
    """Configure prometheus for monitoring."""
    
    # Initialize project
    print(f"Setting up prometheus for monitoring...")
    
    # Example configuration
    config = {
        "project": "my-ml-project",
        "tool": "prometheus",
        "specialty": "monitoring",
        "version": "1.0.0"
    }
    
    # Save configuration
    Path(".prometheus").mkdir(exist_ok=True)
    with open(f".prometheus/config.json", "w") as f:
        json.dump(config, f, indent=2)
    
    print(f"prometheus configured for monitoring")
    return config
config = setup_prometheus()

Monitoring and Alerting

python
from dataclasses import dataclass
import time
@dataclass
class MetricSnapshot:
    timestamp: float
    metric_name: str
    value: float
    labels: dict
class MLOpsMonitor:
    """Monitor monitoring metrics."""
    
    def __init__(self):
        self.metrics: list[MetricSnapshot] = []
        self.thresholds = {
            "error_rate": 0.05,
            "latency_p99_ms": 1000,
            "data_drift_score": 0.3
        }
    
    def record(self, metric: str, value: float, labels: dict = None):
        snapshot = MetricSnapshot(
            timestamp=time.time(),
            metric_name=metric,
            value=value,
            labels=labels or {}
        )
        self.metrics.append(snapshot)
        self._check_threshold(metric, value)
    
    def _check_threshold(self, metric: str, value: float):
        threshold = self.thresholds.get(metric)
        if threshold and value > threshold:
            logger.warning(f"ALERT: {metric}={value:.3f} exceeds threshold {threshold}")
monitor = MLOpsMonitor()

CI/CD Integration

yaml
.github/workflows/ml-pipeline.yml
name: ML Pipeline
on:
  push:
    paths: ['src/', 'data/']
jobs:
  train-and-evaluate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      
      - name: Setup Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'
      
      - name: Install dependencies
        run: pip install -r requirements.txt
      
      - name: Run monitoring
        run: python -m src.prometheus_ml_metrics
        env:
          MLFLOW_TRACKING_URI: ${{ secrets.MLFLOW_URI }}
      
      - name: Check model quality
        run: python -m src.validate_model

Best Practices

Version everything — models, data, configs, and code

Automate testing — catch regressions before production

Monitor continuously — don't wait for users to report issues

Document experiments — future you will thank present you

Use feature flags — control rollouts without code changes

Resources

MLflow documentation: https://mlflow.org/docs

DVC documentation: https://dvc.org/doc

Kubeflow documentation: https://www.kubeflow.org/docs

Made With ML MLOps course: https://madewithml.com